real-timechestcompressionqualitymeasurementsby...

Research ArticleReal-Time Chest Compression Quality Measurements bySmartphone Camera

Oslashyvind Meinich-Bache 1 Kjersti Engan 1 Tonje Soslashraas Birkenes2

and Helge Myklebust2

1University of Stavanger Kjell Arholmsgate 41 4036 Stavanger Norway2Laerdal Medical Tanke Svilands Gate 30 4002 Stavanger Norway

Correspondence should be addressed to Oslashyvind Meinich-Bache oyvindmeinich-bacheuisno

Received 31 January 2018 Accepted 18 July 2018 Published 28 October 2018

Academic Editor Emiliano Schena

Copyright copy 2018OslashyvindMeinich-Bache et al(is is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

Out-of-hospital cardiac arrest (OHCA) is recognized as a global mortality challenge and digital strategies could contribute toincrease the chance of survival In this paper we investigate if cardiopulmonary resuscitation (CPR) quality measurement usingsmartphone video analysis in real-time is feasible for a range of conditions With the use of a web-connected smartphoneapplication which utilizes the smartphone camera we detect inactivity and chest compressions and measure chest compressionrate with real-time feedback to both the caller who performs chest compressions and over the web to the dispatcher who coachesthe caller on chest compressions (e application estimates compression rate with 05 s update interval time to first stablecompression rate (TFSCR) active compression time (TC) hands-off time (TWC) average compression rate (ACR) and totalnumber of compressions (NC) Four experiments were performed to test the accuracy of the calculated chest compression rateunder different conditions and a fifth experiment was done to test the accuracy of the CPR summary parameters TFSCR TCTWC ACR and NC Average compression rate detection error was 27 compressions per minute (plusmn50 cpm) the calculated chestcompression rate was within plusmn10 cpm in 98 (plusmn55) of the time and the average error of the summary CPR parameters was 45(plusmn36) (e results show that real-time chest compression quality measurement by smartphone camera in simulated cardiac arrestis feasible under the conditions tested

1 Introduction

With a yearly number of out-of-hospital cardiac arrest(OHCA) incidents around 370000-740000 in Europe aloneand a low average survival rate of 76 [1] OHCA isrecognized as a major mortality challenge [2] (e timefrom collapse to care is crucial and there is a high focuson low response times of emergency medical services (EMS)[3] A majority of EMS treated OHCAs are witnessed [4]and quality cardiopulmonary resuscitation (CPR) untilEMS arrives can have positive effects on survival [5ndash7] (ewitness is often in close relation with the patient and couldexperience the situation as extremely stressful [8] Studieshave shown that telephone-assisted CPR (T-CPR) hasa positive effect by getting more callers to start CPR andcoaching callers to provide quality CPR [9ndash11]

Furthermore CPR feedback has been shown to improveCPR quality [12ndash15] Combining T-CPR with CPR feedbackmay improve CPR quality and survival from OHCA

In the recent statement from the America Heart Associ-ation (AHA) the use of digital strategies to improve healthcarein general and to document its effect is encouraged [16 17]Devices providing the bystander with CPR quality measure-ment by utilizing an accelerometer to measure CPR metricsare available [18ndash20] A challenge with these devices is to getthe users to carry it with them at all times Smartwatches hasa built-in accelerometer and has been suggested as a tool formeasuring CPR metric [21ndash23] However a very small per-centage of the population wears a smartwatch at all times (esmartphone on the contrary is a digital device most peoplecarry with them In recent years smartphone applications havebeen developed for CPR quality measurement and to support

HindawiJournal of Healthcare EngineeringVolume 2018 Article ID 6241856 12 pageshttpsdoiorg10115520186241856

learning [24 25] and to help communicate the location of anemergency [26] In addition there are publications describingthe use of the accelerometer in smartphones to measure CPRmetrics [25 27ndash30] Smartphone solutions utilizing the ac-celerometer require the smartphone to be held on the patientrsquoschest or strapped to the bystanderrsquos arm while performingCPR(ese solutions may be more suited for training than foractual emergencies since buttons causing phone connectioninterruptions with the emergency unit can accidentally bepressed when performing the compressions

Our research group has earlier presented an applicationQCPR cam-app 10 utilizing the smartphone camera toestimate the chest compression rate and provide feedback toboth the bystander and the dispatcher while the phone isplaced flat on the ground [31] Besides from a small offlinestudy by Frisch et al [32] we have found no other publishedwork or products that utilize the smartphone camera whenmeasuring compression rate QCPR cam-app 10 demon-strated accuracy issues when challenged with bystandershaving long loose hair and in cases of people moving aroundthe emergency scene In this paper we present test results ofQCPR cam-app 20 improved to handle this but also toprovide more information by calculating a CPR summaryreport after CPR has ended (ese parameters can be used toevaluate each session and to generate data that can be usedfor dispatcher-caller quality improvement and research

2 Materials and Methods

(e application QCPR cam-app 20 captures CPR move-ments utilizing the smartphone camera while the smart-phone is placed flat on the ground next to the patient Fromthe detected motions the algorithm estimates the chestcompression rate and hands-off time and provides (1) real-time objective feedback to the bystander (2) real-time ob-jective feedback to the dispatcher during the emergency calland (3) a CPR summary report

21 Illustration of Bystander and Dispatcher Use An illus-tration of the application in use can be seen in Figure 1(a) withscreenshots in Figure 1(b) By clicking the emergency buttonthe application activates speaker mode establishes telephoneconnection with the dispatcher and sends GPS location andreal-time compression data to a web server available for thedispatcher (e bystander then places the smartphone at theopposite side of the patient see Figure 1(a)(e preview framesfrom the front camera are shown to the bystander allowinghim to position himself and to keep track of the ongoingactivity in the field of view of the camera (Figure 1(b)) Aspeedometer is displayed next to the preview frame allowingthe bystander to keep track of the applied compression rate

A live sequence example of the proposed web serversolution monitored by the dispatcher is shown in Figure 1(c)A 20 seconds sliding window providing the developmentand history of the compression rate in real-time is shownwhere different colors are used to make the interpretationeasier Green dots correspond to compression rates in thedesired range of 100ndash120 cpm and yellow outside Above the

graph a circular color indicator provides information aboutthe certainty of the reported compression rates If the de-tections are carried out in low noise the indicator is greenbut if high noise conditions are present ie some cases oflong loose hair and from large disturbances the indicatorshifts to yellow (e bystanderrsquos GPS location is provided tothe dispatcher as seen in Figure 1(c)

22 Technical Description QCPR cam-app 20 was designedto handle the disturbance issues observed in QCPR cam-app10 [31] and the technical description of the improvements arepresented in more detail in the appendix In short All theestimations are performed on the smartphone and the mainsteps in detection of compression rate are illustrated in Fig-ure 2 In step 1 difference frames g(i j) are generated bythresholding the differences between subsequent input framesf(i j) from the camera A dynamic region of interest (ROI) isestablished from the largest connected moving object and isupdated each half second by checking the activity in the blocksaround the ROI boundary By using a dynamic ROI we allowothers to move around in the emergency scene without dis-turbing the detections In step 2 we generate a signal d(l)from the activity in the ROI and for each half second timestepn a short time Fourier transform (STFT) is performed on thethree last seconds of d(l) A sliding Hanning window is ap-plied to d(l) prior to the STFT In step 3 the power spectrumdensity Dn(ω) found from the STFT is studied and a decisionthree is used to separate compression rates from noise (edecision three recognizes a system in the Dn(ω) for cases ofbystanders with long loose hair thus solve the detection issuesobserved inQCPR cam-app 10 [31] for these cases If a CR(n)

is detected it further undergoes some postprocessing stepsindicated in step 4 Figure 2(ese steps filter out and suppressnoise by performing smoothing and removing short detectionpauses caused by compression stops or disturbances In step 5the detected and filtered compression rate CRf(n) (cpm) isdisplayed on the smartphone and sent to the web server anddisplayed to the dispatcher providing the real-time feedback toboth bystander and dispatcher

23 CPR Summary Report After completion of a callersession a set of CPR summary parameters are calculated byQCPR cam-app 20 (e parameters which are both shownon the smartphone screen for the bystander and saved on theweb server for the dispatcher are as follows

(i) TFSCR (s) time from start of phone call to start offirst stable compression rate A compression rate isdefined as stable if CRf(n)gt 40 and |CRf(n)minusCRf(nminus 1)|lt 20 is true for at least 6 seconds

(ii) TC (s) total active compression time (e timewhere CRf(n)gt 0 for t(n)gtTFSCR and contin-uously for more than 2 s

(iii) TWC (s) time without compressions TDPC-TCwhere TDPC (s) is the duration of the phone call

(iv) ACR (minminus1) average compression rate An averageof all CRf(n)gt 0 for t(n)gtTFSCR and continu-ously for more than 2 s

2 Journal of Healthcare Engineering

(v) NC total number of compressions Estimated byACR lowast (TC60)

24 Data Material and Evaluation Measures All experi-ments were performed on a Resusci Anne QCPR manikine QCPR cam-app 20 algorithm was implemented inAndroid Studio and the experiments were performed witha Sony Xperia Z5 Compact (Sony Japan) A reference signal

for the compression rate was provided by an optical encoderembedded in the Resusci Anne QCPR A three-second longsliding window frequency analysis was performed on thesignal each half second providing the reference dataCRtrue(n) with the same sample rate as the compression ratedetection CRf(n) from the app

To evaluate the results dierent measurements wereused Average error (E) Performance (P) Relative errorparameter (REpar) and Bland Altman plots used to visualize

(a) (b)

(c)

Live session - CPR technology from Laerdal and University of Stavanger

Figure 1 (a) Illustration photo of the smartphone application in use in a simulated emergency situation (b) Screenshots of the smartphoneapplication Front page to the left and bystander feedback example to the right (c) Screenshot of the web server available for the dispatcher

DynamicROI

finder

Slidinghanningwindow

STFT

PSD model(a) Noise(b) Hair(c) OK

Afterprocessing

Dn(ω) CR(n)d(l)g(ij)f(ij) CRf(n)

Video frames Difference frames Difference signal Power spectral density

40ndash160 cpm

Output comp rate

Figure 2 Simplied block scheme of the proposed system for chest compression rate measurement Image frames from the smartphonefront camera is used as input and output is the detected compression rate CRf(n)

Journal of Healthcare Engineering 3

the agreement between data provided by QCPR cam-app 20and the reference data provided by Resusci Anne QCPRmanikin E is given in compressions per minute (cpm) and isthe average error of the sequence defined as

E[cpm] 1N

1113944

N

n0CRf(n)minusCRtrue(n)

11138681113868111386811138681113868

11138681113868111386811138681113868 (1)

where N is the number of samples of the sequence For se-quences containing discontinuity in the reference data ie 30 2 session we allowed errors in a plusmn1 s interval around theautomatically detected discontinuities (is reduced the in-fluence of insignificant delays on the errormeasureP is definedas the percentage of time where |CRf(n)minusCRtrue(n)|ltΔAccording to guidelines [33ndash35] the acceptable compressionrate is between 100 and 120 cpm thusΔ 10 (cpm) was chosenas an acceptance criterion REpar measures the performance ofthe CPR summary parameters listed in Section 23 REpar isgiven in percentage and defined as

REpar[] ParD minusParR

11138681113868111386811138681113868111386811138681113868

ParR

100 (2)

where ParD is a CPR summary parameter estimated by theapp and ParR the corresponding CPR parameter found fromthe reference signal If the test contained more than onesequence the results are presented with mean and standarddeviations ie μE(σE) μP(σP) and μREpar(σREpar) foundover the result values of the sequences

Desired detection results provide a low Average error E

a low Relative error parameter REpar and a high Perfor-mance P

25 Experiments (e performance of the QCPR cam-app20 was tested in various conditions that could occur in realemergencies (e experiments were divided into five dif-ferent testsmdashSmartphone position test Outdoor test Dis-turbance test Random movement test and CPR summaryreport test Altogether this sums up to approximately 162minutes of CPR Specifications for the subtests included ineach test are listed in Table 1

(e Smartphone position test included seven test per-sonsmdashtwo with short hair (SH) two with medium lengthloose hair (MLLH) ie chinshoulder length and three withlong loose hair (LLH) ie chest length Each of the testpersons performed 8 subtests carried out indoor

(e result for subtest RateP1 Table 1 was presented inMeinich-Bache et al [36] to verify that QCPR cam-app 20 isable to estimate correct compression rate for test objectswith various hair lengths and for different compressionrates which were an issue in QCPR cam-app 10 [31]

(e subtest D1R110P1 included a person that walksaround and behind the bystander during CPR leaning overthe patient waving his arms and thus causing disturbances(ese results were also presented in Meinich-Bache et al[36] to verify improvements of QCPR cam-app 20 overQCPR cam-app 10 where sometimes disturbances couldtake over the dynamic ROI [31] (e results of subtestsRateP1 and D1R110P1 are repeated here for the reader to

experience all the various tests that QCPR cam-app 20 hasbeen exposed to

Various other conditions were also tested in theSmartphone position test (ree camera positions were in-cluded next to shoulder (Pos1) 20 cm away from shoulder(Pos2) and next to head (Pos3) (e camera positions areshown in Figure 3 30 2 sessions were carried out for camerapositions Pos 1 subtest 30 2P1 and Pos 3 subtest 302P3Pos3 was included to see if the algorithm provides falsedetection when the bystander is still visible in the imageframe when performing rescue breaths Since the bystanderis not visible in the image frame while performing rescuebreaths when the camera is positioned in Pos2 this positionis not relevant for the 302 sessions and therefore not in-cluded Pos2 is used to measure the algorithmrsquos ability todetect when only a small part of the bystander is visible in theimage frame and used in subtests R100P2 and R150P2 (ealgorithm was also tested in low lighting conditions 7 lux insubtest LightP1

(e Outdoor test included three test persons one witheach hair length SH MLLH and LLH (e detections werecarried out in cloudy (C) and sunny (S) weather both withand without noisy background (B) ie trees

(e purpose of the Disturbance test was two-fold (1) tomeasure the algorithmrsquos ability to detect compression ratewhen there is a large disturbance present ie anothermoving person and to (2) quantify the disturbance sizerelative to the bystander performing the compressions whenthe algorithm fails to detect due to too much noise A secondSony Xperia Z5 Compact (Sony Japan) phone was used tocapture video recordings of the test and the video is studiedoffline to perform the quantification (e bystander carriedout continuous compressions during the sequence (edisturbing personmoved around the patient waving arms indifferent frequencies standing behind and over the by-stander while waving arms stepping over patient etc

In the Random movement test no CPR was performedon the manikin and the purpose of the test was to measurethe algorithmrsquos resilience to false detections (e randommovement included checking breathing and pulse of patientturning patient unzipping jacket walking around wavingfor help etc (ree test persons were included

(e CPR summary report test is an evaluation of thesession summary parameters (e test included five differenttest persons with different hair lengths and the following testprotocols

(i) (e bystander sits next to patient with the smart-phone in his hands Heshe presses the emergencycall button and places the smartphone flat on theground For approximately 20 seconds the by-stander checks for patientrsquos pulse and respirationbefore starting performing chest compressions

(ii) Next four intervals of 120-second continuous com-pressions and 20-second pauses while checking forrespiration are followed

(iii) (e total sequence time is approximately 580 swhich is a typical response time for medical assis-tance [37ndash40]


e CPR summary parameters evaluated are the pa-rameters explained in Section 23 TFSCR TC TWC ACRand NC

3 Results

e error measurement results of all ve tests are summarizedin Table 2 e average compression rate detection error Ewas 27 compressions per minute (plusmn50 cpm) the perfor-mance P accepted detections in 98 (plusmn55) of the time andthe relative error of the CPR summary parameters REpar were45 (plusmn36) In subtest R150P2 from the Smartphone position

test the results reveal some weaknesses when only a small partof the bystander is visible to the camera the compression rateis as high as 150 cpm and the person performing compressionhasMLLHor LLH In the two sequences with poor resultsP of562 and 802 the bystander is only present in 46 and69 of the image frame and an example from the largest oneis shown in Figure 4(a) Figures 4(b)ndash4(d) also show examplesfrom the subtests (B) low lighting conditions LightP1 (C)LLH in noisy outdoor conditions OSBR110P1 and (D) thesmallest disturbance occupying 34 times the size of the areaoccupied by the bystander that cause the algorithm to fail todetect for a short period of time in D2R110P1

Table 1 Detailed description of the subtests included in the 5 tests performed to bothmeasure the accuracy of QCPR cam-app 20rsquos ability todetect the compression rate under various conditions and to evaluate the CPR summary parameters calculated after an ended session

Subtest name Compressionrate (cpm)

Duration(s)

Cameraposition Lighting Measures

Smartphone position test (n 7)

RateP126 Normal 60 100120 150 60 x 4 Pos1 480 lux μE μP

D1R110P126 Disturb person 110 120 Pos1 480 lux μE μP302P1 302 110 90 Pos1 480 lux μE μPLightP1 Dimmed light 110 60 Pos1 7 lux μE μPR100P2 Small part of image frame (position Change) 100 60 Pos2 480 lux μE μPR150P2 Small part of image frame (position Change) 150 60 Pos2 480 lux μE μP302P3 302 (position Change) 110 90 Pos3 480 lux μE μPR100P3 Normal (position Change) 100 60 Pos3 480 lux μE μPOutdoor test (n 3)

OCBR110P1 Cloudy with noisy (threes) background 110 60 Pos1 Cloudyweather μE μP

OCR110P1 Cloudy with no background 110 60 Pos1 Cloudyweather μE μP

OSBR110P1 Sunny with noisy (threes) background 110 60 Pos1 Sunnyweather μE μP

OSR110P1 Sunny with no background 110 60 Pos1 Sunnyweather μE μP

Disturbance test (n 1)

D2R110P1 Disturbing person 110 180 Pos1 Normalindoor μE μP

Random movement test (n 3)

RanMovP1 Random movements mdash 150 Pos1 Normalindoor μP

CPR summary report test (n 5)

CPRsrR110P1 Compressions with pauses 110 580 Pos1 Normalindoor μREpar

R rate P position D disturbance O outdoor Bnoisy background C cloudy S sunny CPRsrCPR summary report

Pos 1 Pos 2 Pos 3

Figure 3 Dierent camera positions used in smartphone position test


(e Bland Altman plot in Figure 5 shows the agree-ment between reference data and detection data forSmartphone position test Outdoor test and the Distur-bance test Each analysis in all the test sequences are here

included (e subtests with poorer results 302P1R150P2 and 302P3 is marked with the colors red yellowand purple respectively (e total number of samples inthe plot is 11718 and the number of samples with larger

Table 2 Detection results for all of the 5 tests included in the experiments

μE(σE) (cpm) (0-gt) μP(σP) () (0ndash100) μREpar(σREpar) ()Smartphone position test (n 7)RateP126 13 (03) 997 (03) mdashD1R110P126 18 (13) 995 (12) mdash302P1 45 (38) 959 (37) mdashLightP1 11 (03) 100 (0) mdashR100P2 30 (34) 981 (37) mdashR150P2 114 (149) 898 (164) mdash302P3 33 (14) 960 (21) mdashR100P3 11 (02) 999 (04) mdashOutdoor test (n 3)OCBR110P1 17 (03) 100 (0) mdashOCR110P1 15 (03) 100 (0) mdashOSBR110P1 14 (04) 997 (05) mdashOSR110P1 11 (04) 100 (0) mdashDisturbance test (n 1)D2R110P1 58 960 mdashRandom movement test (n 3)RanMovP1 mdash 896 (25) mdashCPR summary report test (n 5)

CPRsrR110P1

TFSCR mdash 61 (33)TC mdash 28 (26)TWC mdash 100 (91)ACR mdash 18 (12)NC mdash 16 (10)

Total (all tests) 27 (50) 980 (55) 45 (36)(e results are given in mean Average error μE mean Performance μP and mean Relative error parameter μREpar Standard deviations are shown inparenthesis R rate P position D disturbance O outdoor B noisy background C cloudy S sunny CPRsrCPR summary report

(a) (b)

(c) (d)

Figure 4 (a) Screenshot of a MLLH bystanderrsquos position in image frame when algorithm provided poor detection results for compressionrate of 150 cpm R150P2 (b) Screenshots of low lighting conditions LightP1 (c) Screenshot of LLH and noisy outdoor backgroundOSBR110P1 (d) Screenshot of the disturbance size when the algorithm failed to detect the compression rate in D2R110P1


deviation than plusmn10 cpm compared to reference data is 180(153)

In Figure 6 the Bland Altman plots show the agreementbetween the summary parameters calculated from the de-tection data and the summary parameters calculated fromthe reference data in the CPR summary report test

4 Discussion

e results presented in this paper show that the camera ina smartphone can be used to measure chest compressionrates and hands-o times under various conditions withgood accuracy Our proposed method allows for real-timefeedback to both the bystander and to a dispatcher in realemergencies which could improve CPR quality

41 Challenges Although the algorithm works well with onlya small part of the bystander being visible under low noisesituations we discovered reduced accuracy in two of thesequences where the bystander had long loose hair com-pressed with a very high rate and were visible only in a smallpart of the image frame In these sequences the loose hair issometimes almost the only thing visible in the image frameand QCPR cam-app 2 interprets this as compression in therate the visible hair is bouncing inese two cases explain theyellow samples in R150P2 that caused disagreement in theBland Altman plot Figure 5 To avoid these false detectionsthe bystander should position the smartphone such that mostof the head and shoulders are captured in the image frame

We also experienced that repetitive random movementsduring compression pauses could cause the algorithm todetect a false stable-low compression rate causing QCPRcam-app 2 to calculate a longer TC and a shorter TWC Itcould be observed that during compression pauses peopleoften bend towards and away from the patient in a sometimevery repetitive movement and on a few occasion when thebystander had long loose hair these movements caused thealgorithm to interpret the movements as a stable but verylow compression rate lasting a minimum of 5 secondsesefalse stable-low compression rate detections did not occur inthe Random Movement Test when the test persons wereasked to perform all kinds of dierent tasks that could becarried out before compression starts Deactivating thedynamic rate range could solve this problem but a conse-quence of this would be that compressions rates below70 cpm would not be detected

e samples that show disagreement between the de-tections data and the reference data in Figure 5 for subtest 30 2P1 (red) and 30 2P3 (purple) occurs in the transitionsbetween compression and compression pauses when per-forming 302 and do not signicantly aect the visual pre-sentation of the detected signal that is shown to the dispatcher

42 Further Work e proposed system allows the by-stander to have both hands free with compression feedbackon the smartphone screen visible next to the patient which isdierent from accelerometer-based smartphone solutionsthat require the smartphone to be held on the patientrsquos chestor strapped to the bystanderrsquos arm [24 25 27ndash30] isadvantage could make the proposed solution suited for realemergencies where the phone is also used as a life line to theemergency unit Studies comparing the proposed solutionwith the accelerometer-based solutions in simulated emer-gencies should be considered

Testing of QCPR cam-app 20 in simulated real emer-gencies must be carried out in order to conclude if thismethod could be suited for real emergencies In additionstudies with the aim of documenting the usability of theapplication safety of the method and eectiveness on theCPR quality also need to be carried out as suggested byRumsfeld et al [16] If QCPR cam-app 20 shows a well-documented positive eect on the CPR quality it may besubject to appropriate medical device regulations and madeavailable for clinical use [41 42]

e detected and stored compression rate signal and theCPR summary report provide further opportunity forevaluation debrieng and quality improvement of thedispatcher-caller interaction e stored data and the visualdispatcher feedback system can be used to provide con-tinuing education in T-CPR for dispatchers as AHA rec-ommends in T-CPR guidelines [43] In addition thesemeasurements can provide the EMS arriving at the scenewith detailed information about the treatment the patienthas received A feature which records audio and video will beconsidered integrated in QCPR cam-app 2 A possible so-lution could be to let the recordings be automaticallyuploaded to a cloud storage when available bandwidth

40 60 80 100 120 140 160Average of reference and detection data

ndash40

ndash20

0

20

40

60

80

100

120

Diff

eren

ce b

etw

een

refe

renc

e and

det

ectio

n da

ta

+196 SDndash196 SDMean diffZero

302P1R150P2302P3Others

Figure 5 Bland Altman plot to compare reference data fromResusci Anne manikin with detection data fromQCPR cam-app 20for the tests Smartphone position testOutdoor test andDisturbancetest All together 11718 compared samples Dierent colors are usedto dierentiate the subtests 302P1 R150P2 and 302P3 from therest


would allow it Still images video and audio could be madeavailable for the dispatcher and allow for a better un-derstanding of the emergency situations Audio recordingsmay also be analyzed with respect to chest compression rateand inactivity to further improve measurement accuracysince most dispatcher protocols include prompting andcounting loud while compressing on the chest

e collected data could also be utilized in a machinelearning framework providing potential decision support infuture systems

We are currently investigating camera-based methods formeasurement of compression depths [44] In future work wewill try to develop a robust depth algorithm that could beimplemented together with the proposed method Animplementation of depth measurement would make thissolution a complete CPR quality measurement and feedbackdevice Although the proposed solutionrsquos main idea is to assistlaypersons in real emergencies we have also developeda training version of the solution called TCPR Link available

onApp Store andGoogle Play [45 46] in selected countries AsAHA has announced CPR feedback devices will also be re-quired to use in all AHA CPR courses by February 2019 [47]

Studies have also shown that both laypersons andprofessionals could benet from objective feedback duringCPR In a study presented by Abella et al [48] the CPR-certied rescuers performed chest compression rateslt80 cpm in 369 of the CPR segments included in the studyand rates of 100plusmn 10 cpm in only 314 of the segmentsclearly suggesting that CPR-certied rescuers could alsobenet from the proposed solution

43 Study Limitations

(i) e validity testing of the QCPR cam-app 20 wasassessed with a manikin in a simulated cardiac arrest

(ii) e QCPR cam-app 20 does not measure chestcompression depth

25 26 27 28 29

ndash3

ndash2

ndash1

0

1Time to first stable compression rate

440 445 450 455 460 465 470ndash40

ndash20

0

20Total active compression time

110 115 120 125 130 135

0

20

40

Time without compressions

110 112 114 116 118

0

2

4

Average compression rate

830 840 850 860 870 880ndash40

ndash20

0

20

Total number of compressions

Test persons+196 SDndash196 SD

Mean diffZero

Figure 6 Bland Altman plots of the agreement between the summary parameters calculated from theQCPR cam-app 20 detection data andthe summary parameters calculated from the Resusci Anne manikin reference data in the CPR summary report test


(iii) (e bystanders used in the validity testing wereaware of CPR and the QCPR cam-app 2

5 Conclusion

Real-time chest compression quality measurement bysmartphone camera is feasible for a range of bystanderscompression rates camera positions and noise conditions(is technology may be used to measure and improve thequality of telephone CPR and minimize hands-off times

Appendix

Method

(is appendix provides a pseudocode description of methodfor measurement of chest compression rate More details canbe found in [31 36] (e application is called TCPR link andis available on App Store [43] and Google Play [44]

Let the input fl(i j) represent video frames l where(i j) corresponds to row index i and column index j Outputis the filtered compression rate measurement CRf(n) foreach 05 sec analysis interval n

Input fl(i j) Output CRf(n)

while receiving image frames do

Activity measurement

(1) Generating difference framefor All pixels in frame doif |fl(i j)minusflminus1(i j)|lt ε then

gl(i j)⟵0else

gl(i j)⟵fl(i j)minusflminus1(i j)

end ifend for

(2) Dividing gl(i j) into non-overlapping blocks andfinding the sum of change in region block Rk overthe received frames L in the last half secondSL

Rk(n)⟵1113936

LLminus 1SRk

(m)⟵1113936 LLminus 11113936(ij)isinRk

|gm(i j)|

(3) Marks the blocks with SLRk

(n)gt than the averageblock activity S

L

R(n) with an indicator functionIRk

(n)if SL

Rk(n)gt S

L

R(n) thenIRk

(n)⟵1else

IRk(n)⟵0

end ifif ROIestablished FALSE then

Establishing ROI

(4) Establishes a temporary ROIRk isin TminusROIn1113864 1113865 if1113936

nmnminus 3IRk

(m)ge 3(5) Fills block-gaps in the temporary ROI

Rk isin TFminusROIn1113864 1113865 if Rk is a gap in a connectedobject in T-ROI

(6) Choses the largest connected object LCO in theTF-ROI to be the established ROIRk isin ROIn1113864 1113865 if Rk isin TFminusROILCOn1113966 1113967

ROIestablished TRUE

end ifwhile ROIestablished TRUE do for each halfsecond

Activity signal from ROI

(7) Generate difference signal at time point ld(l) 1113936RkisinROIn

1113936(ij)isinRkg(i j)

Frequency analysis

(8) STFTis performed on overlapping blocks of d(l) withblocklength Lf corresponding to 3 sec updatedevery 05 sec A sliding Hanning window is usedprior to the STFTCe PSD Dn(w) is estimated bythe periodogram calculated from the STFTDn(w) 1Lf|FM dhf(l)1113966 1113967|2 l (nminus 1)Lf nLf

where FM denotes M point FFT and dhf(l)

denotes the Hanning filtered difference signal

PSD modelling

(9) Decision tree Recognizes and handle cases of longloose hair and separate compressions from noiseRelevant frequency range is 40-160 (cpm)

Attributes found from Dn(w)(1) Amplitude of the first significant peak ap1(n)(2) Amplitude of the second significant peak

ap2(n)(3) Frequency of the first significant peak fp1(n)(4) Frequency of the second significant peak

fp2(n) and(5) Mean amplitude hight of PSD aPSD(n)

CR(n)⟵decisionTree(ap1(n) ap2(n) fp1(n)

fp2(n) aPSD(n)

Post processing

CRf(n) CR(n)

(10) Short spikedrop removalif |CRf(nminus 1)minusCRf(nminus 1minus k)|ltTsd1 forallkle 2thenif |CR(n)minusCRf(nminus 1)|gtTsd2 then

CRf(n) CR(nminus 1)

i i + 1if i 4 then

CRf(nminus 3 n) CR(nminus 3 n)

i 0end if

elsei 0

end ifend if

(11) Smoothing mean filterforj 1 3 do

K argmaxJ |CRf(n)minusCRf(nminus j)|ltTmf

foralljle J

end forCRf(n) 1113936

Kk0ak CRf(nminus k)

where ak is the filter coefficients 1113936Kk0ak 1

and aj ai forall i j


(12) Dynamic rate rangeCRdrr(n) CRf(n)if CRdrr(n)lt 70 then

for j 1 10 doK argmaxJ |CRdrr(n)minusCRdrr(nminus j)|ltTdrr foralljle J

end forif K10 then

CRf(nminus 10 n) CRdrr(nminus 10 n)

elseCRf(n) 0

end ifend if

ROI update

(13) Add and remove blocks in ROIif SL

boi(n)gt 05 middot S

L

R(n) thenRboi isin ROIn1113864 1113865

end ifif SL

bii(n)lt 05 middot S

L

R(n) thenRbii notin ROIn1113864 1113865

end ifwhere Rboi denote block i on the outside of theROIn boundary and Rbii denote block i inside theROIn

(14) Freq analysis if ROI is divided into multiple areasif of connected areas AROI isin ROIn1113864 1113865gt 1 then

for i 1 of AROI doPerform step 7 8 and 9 andif CRAROIi(n) is in range of 40ndash160 cpm then

AROIi isin ROIn1113864 1113865

elseAROIi notin ROIn1113864 1113865

end ifend for

end ifif of Rbii isin ROIn1113864 1113865lt 2 then

ROIestablished FALSE

end ifend while

end while

Data Availability

(e data used in the evaluation of the study are included assupplementary materials in this published article Our systemdoes not capture and store the videos but only detects thusthe videos cannot be made available

Disclosure

(e results and the application were presented at EmergencyCardiovascular Care Conference (ECCU) December 2017New Orleans USA [49]

Conflicts of Interest

Tonje Soslashraas Birkenes and Helge Myklebust are employeesof Laerdal Medical

Acknowledgments

We want to thank Solveig Haukas Haaland (LaerdalMedical) for help with planning and execution of experi-ments and (omas Hinna (BI Builders) and Daniel Vartdal(Webstep Stavanger) for help with real-time implementationof the algorithm in an android app (e study and appli-cation development was funded by University of Stavangerand Laerdal Medical

Supplementary Materials

(e supplementary material consists data files for each of thefive presented experiments For the first four experimentsthe data are presented for each test person as Matlab mat-files containing both the compared reference data and thedetection data For the fifth test CPR summary report testthe data is presented as ssx-files (reference data) and xls-files(detection data) A readme-file explaining the preprocessingperformed prior to the comparison between reference dataand detection data is here also included (SupplementaryMaterials)

References

[1] L L Bossaert G D Perkins H Askitopoulou et al ldquoEu-ropean resuscitation council guidelines for resuscitation2015rdquo Resuscitation vol 95 pp 302ndash311 2015

[2] O F Norheim P Jha K Admasu et al ldquoAvoiding 40 of thepremature deaths in each country 2010ndash30 review of nationalmortality trends to help quantify the un sustainable devel-opment goal for healthrdquo Ce Lancet vol 385 no 9964pp 239ndash252 2015

[3] J P Pell J M Sirel A K Marsden I Ford and S M CobbeldquoEffect of reducing ambulance response times on deaths fromout of hospital cardiac arrest cohort studyrdquo BMJ vol 322no 7299 pp 1385ndash1388 2001

[4] M Wissenberg F K Lippert F Folke et al ldquoAssociation ofnational initiatives to improve cardiac arrest managementwith rates of bystander intervention and patient survival afterout-of-hospital cardiac arrestrdquo JAMA vol 310 no 13pp 1377ndash1384 2013

[5] K B Kern R W Hilwig R A Berg A B Sanders andG A Ewy ldquoImportance of continuous chest compressionsduring cardiopulmonary resuscitationrdquo Circulation vol 105no 5 pp 645ndash649 2002

[6] S Steen Q Liao L Pierre A Paskevicius and T Sjobergldquo(e critical importance of minimal delay between chestcompressions and subsequent defibrillation a haemody-namic explanationrdquo Resuscitation vol 58 no 3 pp 249ndash258 2003

[7] P A Meaney B J Bobrow M E Mancini et al ldquoCardio-pulmonary resuscitation quality Improving cardiac re-suscitation outcomes both inside and outside the hospitalrdquoCirculation vol 128 no 4 pp 417ndash435 2013

[8] R Swor I Khan R Domeier L Honeycutt K Chu andS Compton ldquoCPR training and CPR performance do CPR-trained bystanders perform CPRrdquo Academic EmergencyMedicine vol 13 no 6 pp 596ndash601 2006

[9] T S Birkenes H Myklebust A Neset and J Kramer-Johansen ldquoHigh quality cpr with optimized rescuer-


dispatcher teamworkrdquo Circulation vol 128 no 22 articleA15393 2016

[10] T S Birkenes H Myklebust A Neset T M Olasveengenand J Kramer-Johansen ldquoVideo analysis of dispatcherndashrescuer teamworkndasheffects on cpr technique and perfor-mancerdquo Resuscitation vol 83 no 4 pp 494ndash499 2012

[11] M Akahane T Ogawa S Tanabe et al ldquoImpact of telephonedispatcher assistance on the outcomes of pediatric out-of-hospital cardiac arrestlowastrdquo Critical CareMedicine vol 40 no 5pp 1410ndash1416 2012

[12] J Kramer-Johansen H Myklebust L Wik et al ldquoQuality ofout-of-hospital cardiopulmonary resuscitation with real timeautomated feedback a prospective interventional studyrdquoResuscitation vol 71 no 3 pp 283ndash292 2006

[13] A J Handley and S A J Handley ldquoImproving CPR per-formance using an audible feedback system suitable for in-corporation into an automated external defibrillatorrdquoResuscitation vol 57 no 1 pp 57ndash62 2003

[14] C W Yang H C Wang W C Chiang et al ldquoInteractivevideo instruction improves the quality of dispatcher-assistedchest compression-only cardiopulmonary resuscitation insimulated cardiac arrestslowastrdquo Critical Care Medicine vol 37no 2 pp 490ndash495 2009

[15] A Neset T S Birkenes H Myklebust R J MykletunS Odegaard and J Kramer-Johansen ldquoA randomized trial ofthe capability of elderly lay persons to perform chest com-pression only CPR versus standard 302 CPRrdquo Resuscitationvol 81 no 7 pp 887ndash892 2010

[16] J S Rumsfeld S C Brooks T P Aufderheide et al ldquoUse ofmobile devices social media and crowdsourcing as digitalstrategies to improve emergency cardiovascular carerdquo Cir-culation vol 134 no 8 pp e87ndashe108 2016

[17] M E Kleinman E E Brennan Z D Goldberger et al ldquoPart5 adult basic life support and cardiopulmonary resusci-tation qualityrdquo Circulation vol 132 no 18 pp S414ndashS4352015

[18] C Buleon J J Parienti L Halbout et al ldquoImprovement inchest compression quality using a feedback device(cprmeter) a simulation randomized crossover studyrdquoAmerican Journal of Emergency Medicine vol 31 no 10pp 1457ndash1461 2013

[19] CPR measurement device CPR meter 2 httpwwwlaerdalcomusproductsmedical-devicescprmeter-2

[20] A Cheng L L Brown J P Duff et al ldquoImproving cardio-pulmonary resuscitation with a CPR feedback device andrefresher simulations (CPR cares study) a randomizedclinical trialrdquo JAMA Pediatrics vol 169 no 2 pp 137ndash1442015

[21] Smartwatch-app W-CPR httpsplaygooglecomstoreappsdetailsidcommelabw_cpr

[22] Smartwatch-app CPR wear httpsplaygooglecomstoreappsdetailsidcomsackcpwear

[23] Y Song Y Chee J Oh C Ahn and T H Lim ldquoSmartwatchesas chest compression feedback devices a feasibility studyrdquoResuscitation vol 103 pp 20ndash23 2016

[24] Smartphone-app ICPR httpicprit[25] F Semeraro F Taggi G Tammaro G Imbriaco L Marchetti

and E L Cerchiari ldquoICPR a new application of high-qualitycardiopulmonary resuscitation trainingrdquo Resuscitationvol 82 no 4 pp 436ndash441 2011

[26] Norsk luftambulanse-app Hjelp 113-GPS httpsnorskluftambulansenoapper

[27] T Amemiya and T Maeda ldquoPoster depth and rate estimationfor chest compression CPR with smartphonerdquo in Proceedings

of IEEE Symposium on 3DUser Interfaces (3DUI) pp 125-126Piscataway NJ USA March 2013

[28] N K Gupta V Dantu and R Dantu ldquoEffective CPR pro-cedure with real time evaluation and feedback using smart-phonesrdquo IEEE Journal of Translational Engineering in Healthand Medicine vol 2 pp 1ndash11 2014

[29] M Kalz N Lenssen M Felzen et al ldquoSmartphone apps forcardiopulmonary resuscitation training and real incidentsupport a mixed-methods evaluation studyrdquo Journal ofMedical Internet Research vol 16 no 3 p e89 2014

[30] Y Song J Oh and Y Chee ldquoA new chest compression depthfeedback algorithm for high-quality CPR based on smart-phonerdquo Telemedicine and e-Health vol 21 no 1 pp 36ndash412015

[31] K Engan T Hinna T Ryen T S Birkenes andH Myklebust ldquoChest compression rate measurement fromsmartphone videordquo BioMedical Engineering OnLine vol 15no 1 2016

[32] A Frisch S Das J C Reynolds F d L Torre J K Hodginsand J N Carlson ldquoAnalysis of smartphone video footageclassifies chest compression rate during simulated CPRrdquoAmerican Journal of Emergency Medicine vol 32 no 9pp 1136ndash1138 2014

[33] I Fernandez Lozano C Urkıa J B Lopez Mesa et alldquoEuropean resuscitation council guidelines for resuscitation2015 key pointsrdquo Revista Espantildeola de Cardiologıa (EnglishEdition) vol 69 no 6 pp 588ndash594 2016

[34] R W Neumar M Shuster C W Callaway et al ldquoPart 1executive summaryrdquo Circulation vol 132 no 18 pp S315ndashS367 2015

[35] S P Chung T Sakamoto S H Lim et al ldquo(e 2015 re-suscitation council of asia (RCA) guidelines on adult basic lifesupport for lay rescuersrdquo Resuscitation vol 105 pp 145ndash1482016

[36] Oslash Meinich-Bache K Engan T S Birkenes andH Myklebust ldquoRobust real-time chest compression ratedetection from smartphone videordquo in Proceedings of 10thInternational Symposium on Image and Signal Processing andAnalysis Ljubljana Slovenia September 2017

[37] G Nichol E (omas C W Callaway et al ldquoRegional var-iation in out-of-hospital cardiac arrest incidence and out-comerdquo JAMA vol 300 no 12 pp 1423ndash1431 2008

[38] S D Shin M E Hock Ong H Tanaka et al ldquoComparison ofemergency medical services systems across pan-asian coun-tries a web-based surveyrdquo Prehospital Emergency Carevol 16 no 4 pp 477ndash496 2012

[39] H Yasunaga H Miyata H Horiguchi et al ldquoPopulationdensity call-response interval and survival of out-of-hospitalcardiac arrestrdquo International Journal of Health Geographicsvol 10 no 1 p 26 2011

[40] I Hasselqvist-Ax G Riva J Herlitz et al ldquoEarly cardio-pulmonary resuscitation in out-of-hospital cardiac arrestrdquoNew England Journal of Medicine vol 372 no 24pp 2307ndash2315 2015

[41] Mobile medical applications-FDA httpswwwfdagovmedicaldevicesdigitalhealthmobilemedicalapplicationsdefaulthtm

[42] Medical devices - the council of the european communitieshttpseur-lexeuropaeuLexUriServLexUriServdouriCELEX31993L0042ENHTML

[43] Telephone CPR (T-CPR) program recommenda-tions and performance measures httpcprheartorgAHAECCCPRAndECCResuscitationScienceTelephoneCPRRecommendationsPerformanceMeasuresUCM_477526_


Telephone-CPR-T-CPR-Program-Recommendations-and-Performance-Measuresjsp

[44] Oslash Meinich-Bache K Engan T Eftestoslashl and I AustvollldquoDetecting chest compression depth using a smartphonecamera and motion segmentationrdquo in Proceedings of Scan-dinavian Conference on Image Analysis pp 53ndash64 TromsoslashNorway May 2017

[45] TCPR link on app store httpsitunesapplecomusapptcpr-linkid1314904593mt8

[46] TCPR link on google play httpsplaygooglecomstoreappsdetailsidnolaerdalglobalhealthtcprlinkamphlen

[47] Use of feedback devices in adult CPR training httpminnstateedusystemasaworkforcemrtcupdatesdocumentsuse_of_feedback_devices_in_aha_adult_cpr_training_coursespdf

[48] B S Abella N Sandbo P Vassilatos et al ldquoChest com-pression rates during cardiopulmonary resuscitation aresuboptimalrdquo Circulation vol 111 no 4 pp 428ndash434 2005

[49] Emergency cardiovascular care update (ECCU) 2017presentation httpscitizencprorgwp-contentuploads201801CPR-Measurement-and-Feedback-by-Smartphone-Camerapdf


International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018


Active and Passive Electronic Components

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of



Journal ofEngineeringVolume 2018

SensorsJournal of



RotatingMachinery


Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation


Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom

learning [24 25] and to help communicate the location of anemergency [26] In addition there are publications describingthe use of the accelerometer in smartphones to measure CPRmetrics [25 27ndash30] Smartphone solutions utilizing the ac-celerometer require the smartphone to be held on the patientrsquoschest or strapped to the bystanderrsquos arm while performingCPR(ese solutions may be more suited for training than foractual emergencies since buttons causing phone connectioninterruptions with the emergency unit can accidentally bepressed when performing the compressions

Our research group has earlier presented an applicationQCPR cam-app 10 utilizing the smartphone camera toestimate the chest compression rate and provide feedback toboth the bystander and the dispatcher while the phone isplaced flat on the ground [31] Besides from a small offlinestudy by Frisch et al [32] we have found no other publishedwork or products that utilize the smartphone camera whenmeasuring compression rate QCPR cam-app 10 demon-strated accuracy issues when challenged with bystandershaving long loose hair and in cases of people moving aroundthe emergency scene In this paper we present test results ofQCPR cam-app 20 improved to handle this but also toprovide more information by calculating a CPR summaryreport after CPR has ended (ese parameters can be used toevaluate each session and to generate data that can be usedfor dispatcher-caller quality improvement and research

2 Materials and Methods

(e application QCPR cam-app 20 captures CPR move-ments utilizing the smartphone camera while the smart-phone is placed flat on the ground next to the patient Fromthe detected motions the algorithm estimates the chestcompression rate and hands-off time and provides (1) real-time objective feedback to the bystander (2) real-time ob-jective feedback to the dispatcher during the emergency calland (3) a CPR summary report

21 Illustration of Bystander and Dispatcher Use An illus-tration of the application in use can be seen in Figure 1(a) withscreenshots in Figure 1(b) By clicking the emergency buttonthe application activates speaker mode establishes telephoneconnection with the dispatcher and sends GPS location andreal-time compression data to a web server available for thedispatcher (e bystander then places the smartphone at theopposite side of the patient see Figure 1(a)(e preview framesfrom the front camera are shown to the bystander allowinghim to position himself and to keep track of the ongoingactivity in the field of view of the camera (Figure 1(b)) Aspeedometer is displayed next to the preview frame allowingthe bystander to keep track of the applied compression rate

A live sequence example of the proposed web serversolution monitored by the dispatcher is shown in Figure 1(c)A 20 seconds sliding window providing the developmentand history of the compression rate in real-time is shownwhere different colors are used to make the interpretationeasier Green dots correspond to compression rates in thedesired range of 100ndash120 cpm and yellow outside Above the

graph a circular color indicator provides information aboutthe certainty of the reported compression rates If the de-tections are carried out in low noise the indicator is greenbut if high noise conditions are present ie some cases oflong loose hair and from large disturbances the indicatorshifts to yellow (e bystanderrsquos GPS location is provided tothe dispatcher as seen in Figure 1(c)

22 Technical Description QCPR cam-app 20 was designedto handle the disturbance issues observed in QCPR cam-app10 [31] and the technical description of the improvements arepresented in more detail in the appendix In short All theestimations are performed on the smartphone and the mainsteps in detection of compression rate are illustrated in Fig-ure 2 In step 1 difference frames g(i j) are generated bythresholding the differences between subsequent input framesf(i j) from the camera A dynamic region of interest (ROI) isestablished from the largest connected moving object and isupdated each half second by checking the activity in the blocksaround the ROI boundary By using a dynamic ROI we allowothers to move around in the emergency scene without dis-turbing the detections In step 2 we generate a signal d(l)from the activity in the ROI and for each half second timestepn a short time Fourier transform (STFT) is performed on thethree last seconds of d(l) A sliding Hanning window is ap-plied to d(l) prior to the STFT In step 3 the power spectrumdensity Dn(ω) found from the STFT is studied and a decisionthree is used to separate compression rates from noise (edecision three recognizes a system in the Dn(ω) for cases ofbystanders with long loose hair thus solve the detection issuesobserved inQCPR cam-app 10 [31] for these cases If a CR(n)

is detected it further undergoes some postprocessing stepsindicated in step 4 Figure 2(ese steps filter out and suppressnoise by performing smoothing and removing short detectionpauses caused by compression stops or disturbances In step 5the detected and filtered compression rate CRf(n) (cpm) isdisplayed on the smartphone and sent to the web server anddisplayed to the dispatcher providing the real-time feedback toboth bystander and dispatcher

23 CPR Summary Report After completion of a callersession a set of CPR summary parameters are calculated byQCPR cam-app 20 (e parameters which are both shownon the smartphone screen for the bystander and saved on theweb server for the dispatcher are as follows

(i) TFSCR (s) time from start of phone call to start offirst stable compression rate A compression rate isdefined as stable if CRf(n)gt 40 and |CRf(n)minusCRf(nminus 1)|lt 20 is true for at least 6 seconds

(ii) TC (s) total active compression time (e timewhere CRf(n)gt 0 for t(n)gtTFSCR and contin-uously for more than 2 s

(iii) TWC (s) time without compressions TDPC-TCwhere TDPC (s) is the duration of the phone call

(iv) ACR (minminus1) average compression rate An averageof all CRf(n)gt 0 for t(n)gtTFSCR and continu-ously for more than 2 s






(a) (b)

(c)



DynamicROI

finder


STFT


Afterprocessing



40ndash160 cpm

Output comp rate




E[cpm] 1N

1113944

N


11138681113868111386811138681113868

11138681113868111386811138681113868 (1)



11138681113868111386811138681113868111386811138681113868

ParR

100 (2)



















3 Results





Duration(s)
















Pos 1 Pos 2 Pos 3







CPRsrR110P1



(a) (b)

(c) (d)





4 Discussion









ndash40

ndash20

0

20

40

60

80

100

120

Diff

eren

ce b

etw

een

refe

renc

e and

det

ectio

n da

ta













25 26 27 28 29

ndash3

ndash2

ndash1

0


440 445 450 455 460 465 470ndash40

ndash20

0


110 115 120 125 130 135

0

20

40


110 112 114 116 118

0

2

4


830 840 850 860 870 880ndash40

ndash20

0

20



Mean diffZero




5 Conclusion


Appendix

Method







gl(i j)⟵0else


end ifend for


Rk(n)⟵1113936

LLminus 1SRk


|gm(i j)|



L


(n)if SL

Rk(n)gt S

L

R(n) thenIRk

(n)⟵1else

IRk(n)⟵0


Establishing ROI


nmnminus 3IRk




ROIestablished TRUE





Frequency analysis




PSD modelling






fp2(n) aPSD(n)

Post processing

CRf(n) CR(n)


CRf(n) CR(nminus 1)

i i + 1if i 4 then


i 0end if

elsei 0

end ifend if



foralljle J


Kk0ak CRf(nminus k)






end forif K10 then


elseCRf(n) 0

end ifend if

ROI update



L


end ifif SL


L







end ifend for



end ifend while

end while

Data Availability


Disclosure




Acknowledgments




References


























































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia






(a) (b)

(c)



DynamicROI

finder


STFT


Afterprocessing



40ndash160 cpm

Output comp rate




E[cpm] 1N

1113944

N


11138681113868111386811138681113868

11138681113868111386811138681113868 (1)



11138681113868111386811138681113868111386811138681113868

ParR

100 (2)



















3 Results





Duration(s)
















Pos 1 Pos 2 Pos 3







CPRsrR110P1



(a) (b)

(c) (d)





4 Discussion









ndash40

ndash20

0

20

40

60

80

100

120

Diff

eren

ce b

etw

een

refe

renc

e and

det

ectio

n da

ta













25 26 27 28 29

ndash3

ndash2

ndash1

0


440 445 450 455 460 465 470ndash40

ndash20

0


110 115 120 125 130 135

0

20

40


110 112 114 116 118

0

2

4


830 840 850 860 870 880ndash40

ndash20

0

20



Mean diffZero




5 Conclusion


Appendix

Method







gl(i j)⟵0else


end ifend for


Rk(n)⟵1113936

LLminus 1SRk


|gm(i j)|



L


(n)if SL

Rk(n)gt S

L

R(n) thenIRk

(n)⟵1else

IRk(n)⟵0


Establishing ROI


nmnminus 3IRk




ROIestablished TRUE





Frequency analysis




PSD modelling






fp2(n) aPSD(n)

Post processing

CRf(n) CR(n)


CRf(n) CR(nminus 1)

i i + 1if i 4 then


i 0end if

elsei 0

end ifend if



foralljle J


Kk0ak CRf(nminus k)






end forif K10 then


elseCRf(n) 0

end ifend if

ROI update



L


end ifif SL


L







end ifend for



end ifend while

end while

Data Availability


Disclosure




Acknowledgments




References


























































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



E[cpm] 1N

1113944

N


11138681113868111386811138681113868

11138681113868111386811138681113868 (1)



11138681113868111386811138681113868111386811138681113868

ParR

100 (2)



















3 Results





Duration(s)
















Pos 1 Pos 2 Pos 3







CPRsrR110P1



(a) (b)

(c) (d)





4 Discussion









ndash40

ndash20

0

20

40

60

80

100

120

Diff

eren

ce b

etw

een

refe

renc

e and

det

ectio

n da

ta













25 26 27 28 29

ndash3

ndash2

ndash1

0


440 445 450 455 460 465 470ndash40

ndash20

0


110 115 120 125 130 135

0

20

40


110 112 114 116 118

0

2

4


830 840 850 860 870 880ndash40

ndash20

0

20



Mean diffZero




5 Conclusion


Appendix

Method







gl(i j)⟵0else


end ifend for


Rk(n)⟵1113936

LLminus 1SRk


|gm(i j)|



L


(n)if SL

Rk(n)gt S

L

R(n) thenIRk

(n)⟵1else

IRk(n)⟵0


Establishing ROI


nmnminus 3IRk




ROIestablished TRUE





Frequency analysis




PSD modelling






fp2(n) aPSD(n)

Post processing

CRf(n) CR(n)


CRf(n) CR(nminus 1)

i i + 1if i 4 then


i 0end if

elsei 0

end ifend if



foralljle J


Kk0ak CRf(nminus k)






end forif K10 then


elseCRf(n) 0

end ifend if

ROI update



L


end ifif SL


L







end ifend for



end ifend while

end while

Data Availability


Disclosure




Acknowledgments




References


























































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



3 Results





Duration(s)
















Pos 1 Pos 2 Pos 3







CPRsrR110P1



(a) (b)

(c) (d)





4 Discussion









ndash40

ndash20

0

20

40

60

80

100

120

Diff

eren

ce b

etw

een

refe

renc

e and

det

ectio

n da

ta













25 26 27 28 29

ndash3

ndash2

ndash1

0


440 445 450 455 460 465 470ndash40

ndash20

0


110 115 120 125 130 135

0

20

40


110 112 114 116 118

0

2

4


830 840 850 860 870 880ndash40

ndash20

0

20



Mean diffZero




5 Conclusion


Appendix

Method







gl(i j)⟵0else


end ifend for


Rk(n)⟵1113936

LLminus 1SRk


|gm(i j)|



L


(n)if SL

Rk(n)gt S

L

R(n) thenIRk

(n)⟵1else

IRk(n)⟵0


Establishing ROI


nmnminus 3IRk




ROIestablished TRUE





Frequency analysis




PSD modelling






fp2(n) aPSD(n)

Post processing

CRf(n) CR(n)


CRf(n) CR(nminus 1)

i i + 1if i 4 then


i 0end if

elsei 0

end ifend if



foralljle J


Kk0ak CRf(nminus k)






end forif K10 then


elseCRf(n) 0

end ifend if

ROI update



L


end ifif SL


L







end ifend for



end ifend while

end while

Data Availability


Disclosure




Acknowledgments




References


























































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia






CPRsrR110P1



(a) (b)

(c) (d)





4 Discussion









ndash40

ndash20

0

20

40

60

80

100

120

Diff

eren

ce b

etw

een

refe

renc

e and

det

ectio

n da

ta













25 26 27 28 29

ndash3

ndash2

ndash1

0


440 445 450 455 460 465 470ndash40

ndash20

0


110 115 120 125 130 135

0

20

40


110 112 114 116 118

0

2

4


830 840 850 860 870 880ndash40

ndash20

0

20



Mean diffZero




5 Conclusion


Appendix

Method







gl(i j)⟵0else


end ifend for


Rk(n)⟵1113936

LLminus 1SRk


|gm(i j)|



L


(n)if SL

Rk(n)gt S

L

R(n) thenIRk

(n)⟵1else

IRk(n)⟵0


Establishing ROI


nmnminus 3IRk




ROIestablished TRUE





Frequency analysis




PSD modelling






fp2(n) aPSD(n)

Post processing

CRf(n) CR(n)


CRf(n) CR(nminus 1)

i i + 1if i 4 then


i 0end if

elsei 0

end ifend if



foralljle J


Kk0ak CRf(nminus k)






end forif K10 then


elseCRf(n) 0

end ifend if

ROI update



L


end ifif SL


L







end ifend for



end ifend while

end while

Data Availability


Disclosure




Acknowledgments




References


























































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia




4 Discussion









ndash40

ndash20

0

20

40

60

80

100

120

Diff

eren

ce b

etw

een

refe

renc

e and

det

ectio

n da

ta













25 26 27 28 29

ndash3

ndash2

ndash1

0


440 445 450 455 460 465 470ndash40

ndash20

0


110 115 120 125 130 135

0

20

40


110 112 114 116 118

0

2

4


830 840 850 860 870 880ndash40

ndash20

0

20



Mean diffZero




5 Conclusion


Appendix

Method







gl(i j)⟵0else


end ifend for


Rk(n)⟵1113936

LLminus 1SRk


|gm(i j)|



L


(n)if SL

Rk(n)gt S

L

R(n) thenIRk

(n)⟵1else

IRk(n)⟵0


Establishing ROI


nmnminus 3IRk




ROIestablished TRUE





Frequency analysis




PSD modelling






fp2(n) aPSD(n)

Post processing

CRf(n) CR(n)


CRf(n) CR(nminus 1)

i i + 1if i 4 then


i 0end if

elsei 0

end ifend if



foralljle J


Kk0ak CRf(nminus k)






end forif K10 then


elseCRf(n) 0

end ifend if

ROI update



L


end ifif SL


L







end ifend for



end ifend while

end while

Data Availability


Disclosure




Acknowledgments




References


























































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia










25 26 27 28 29

ndash3

ndash2

ndash1

0


440 445 450 455 460 465 470ndash40

ndash20

0


110 115 120 125 130 135

0

20

40


110 112 114 116 118

0

2

4


830 840 850 860 870 880ndash40

ndash20

0

20



Mean diffZero




5 Conclusion


Appendix

Method







gl(i j)⟵0else


end ifend for


Rk(n)⟵1113936

LLminus 1SRk


|gm(i j)|



L


(n)if SL

Rk(n)gt S

L

R(n) thenIRk

(n)⟵1else

IRk(n)⟵0


Establishing ROI


nmnminus 3IRk




ROIestablished TRUE





Frequency analysis




PSD modelling






fp2(n) aPSD(n)

Post processing

CRf(n) CR(n)


CRf(n) CR(nminus 1)

i i + 1if i 4 then


i 0end if

elsei 0

end ifend if



foralljle J


Kk0ak CRf(nminus k)






end forif K10 then


elseCRf(n) 0

end ifend if

ROI update



L


end ifif SL


L







end ifend for



end ifend while

end while

Data Availability


Disclosure




Acknowledgments




References


























































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



5 Conclusion


Appendix

Method







gl(i j)⟵0else


end ifend for


Rk(n)⟵1113936

LLminus 1SRk


|gm(i j)|



L


(n)if SL

Rk(n)gt S

L

R(n) thenIRk

(n)⟵1else

IRk(n)⟵0


Establishing ROI


nmnminus 3IRk




ROIestablished TRUE





Frequency analysis




PSD modelling






fp2(n) aPSD(n)

Post processing

CRf(n) CR(n)


CRf(n) CR(nminus 1)

i i + 1if i 4 then


i 0end if

elsei 0

end ifend if



foralljle J


Kk0ak CRf(nminus k)






end forif K10 then


elseCRf(n) 0

end ifend if

ROI update



L


end ifif SL


L







end ifend for



end ifend while

end while

Data Availability


Disclosure




Acknowledgments




References


























































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia




end forif K10 then


elseCRf(n) 0

end ifend if

ROI update



L


end ifif SL


L







end ifend for



end ifend while

end while

Data Availability


Disclosure




Acknowledgments




References


























































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia

















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia


real-timechestcompressionqualitymeasurementsby...

Documents