microscopic behavioural analysis of cyclists and …...draft alsaleh, hussein, sayed 4 1 the freedom...

Draft

Microscopic Behavioural Analysis of Cyclists and Pedestrians Interactions in Shared Space

Journal: Canadian Journal of Civil Engineering

Manuscript ID cjce-2018-0777.R2

Manuscript Type: Article

Date Submitted by the Author: 03-May-2019

Complete List of Authors: Alsaleh, Rushdi; University of British Columbia, Civil Engineering Hussein, Mohamed; Mcmaster University Department of Civil EngineeringSayed, Tarek; University of British Columbia, Civil Engineering

Keyword: pedestrian-cyclist interactions, shared space, shared space modeling, microscopic behavioural analysis, cyclist behaviour

Is the invited manuscript for consideration in a Special

Issue? :Not applicable (regular submission)

https://mc06.manuscriptcentral.com/cjce-pubs

Canadian Journal of Civil Engineering

Draft

1 Microscopic Behavioural Analysis of Cyclists and Pedestrians 2 Interactions in Shared Space345 Rushdi Alsaleh, M.Sc. (Corresponding Author)6 Ph.D. Student and Research Assistant7 Department of Civil Engineering8 University of British Columbia9 6250 Applied Science Lane,

10 Vancouver BC, Canada V6T 1Z411 Email: [email protected] Mohamed Hussein, Ph.D.,15 Assistant Professor,16 Department of Civil Engineering17 McMaster University18 1280 Main Street West19 Hamilton, ON, Canada, L8S 4L720 Email: [email protected] Tarek Sayed, Ph.D., P.Eng.24 Professor25 Department of Civil Engineering26 University of British Columbia27 6250 Applied Science Lane,28 Vancouver BC, Canada V6T 1Z429 Email: [email protected] Submission Date: May. 8th, 2019

Page 1 of 45



mailto:%[email protected]

mailto:[email protected]

mailto:%[email protected]

Draft

Alsaleh, Hussein, Sayed 2

1 ABSTRACT

2 This study investigates the microscopic interaction behaviour between cyclists and pedestrians in

3 shared space environments. Video data was collected at the Robson Square shared space in

4 downtown Vancouver, British Columbia. Trajectories of cyclists and pedestrians involved in 208

5 interactions (416 Trajectories) were extracted using computer vision algorithms. The extracted

6 trajectories were used to define different indicators for the analysis. The indicators included the

7 speed and acceleration profiles and the longitudinal and lateral distances between road users during

8 different phases of the interactions. The study also investigated the collision avoidance

9 mechanisms employed by road users to avoid collisions with other shared space users. The

10 collision avoidance mechanisms included changing the walking/cycling speed and changing the

11 movement direction. The results showed that the collision avoidance mechanisms depend on the

12 shared space density and the space available for road users. The study identified a set of parameters

13 that can be used to calibrate a microscopic cyclist-pedestrian modeling platform to represent the

14 behaviour of pedestrians and cyclists in shared space environments.

15

16

17 Keywords: shared space, pedestrian-cyclist interactions, shared space modeling, microscopic

18 behavioural analysis, cyclist behaviour.

1920

Page 2 of 45



Draft


1 INTRODUCTION

2 Many cities and road authorities are adopting policies that aim at promoting active modes

3 of transportation such as walking and cycling. Encouraging active transportation supports the cities’

4 sustainability goals and reduces traffic-induced air pollution and congestion. As well, active modes

5 of transportation help road users to adopt a healthy lifestyle and increase their level of physical

6 activity which have numerous health benefits. The "Open Streets" program (Ciclovia) represents

7 one of the policies that are gaining popularity in many cities in North America to encourage active

8 road users of all ages and abilities to share the road. The program involves the temporary closure

9 of selected city streets to motorized traffic and creating non-motorized shared space areas for social

10 and recreational activities. By 2016, around 122 cities in the United States have hosted an “Open

11 Street” program, including New York City and Los Angeles that have more than 100 thousands

12 participants per event (Hipp et al. 2017). Similar “Open Street” programs were hosted in other

13 cities such as Vancouver, Calgary, and Vienna.

14 The “Open Streets” program was inspired by the emerging concept of “shared space”. The

15 shared space concept is established based on the recent findings in behavioral and environmental

16 psychology and the evolution of the risk compensation theory (Adams 1995; Hamilton-Baillie

17 2008). Shared spaces are areas where there is no obvious segregation between road users, so that

18 the right of way is shared by all road users. The access of motorized vehicles to the shared space

19 area can be prohibited to create non-motorized shared spaces, which provides a safe and

20 comfortable environment for non-motorized road users. Despite the emerging popularity of the

21 shared space concept, limited studies have investigated the interaction behavior of active road

22 users in such facilities. In addition to the difficulty of analyzing the behaviour of active road users

23 in general, studying the behaviour of non-motorized road users in shared space environments is

24 more challenging. Unlike conventional roads, the shared space concept provides road users with

Page 3 of 45



Draft


1 the freedom to move in the whole area of the facility without being restricted to use predefined

2 paths (e.g. sidewalk or bike lanes). Thus, the frequency of interactions between road users and the

3 behavior during interactions in shared spaces can differ significantly than the corresponding

4 behaviour in conventional streets.

5 Microscopic simulation modeling of road user dynamics has been advocated as a promising

6 tool to analyze road user behavior in different facilities. Adopting a valid simulation model to

7 analyze road user interactions in shared space areas can provide a powerful tool that helps planners

8 and engineers to assess the level of service and safety of road users of shared space facilities.

9 However, existing simulation packages are inadequate to model the microscopic behaviour of road

10 users in shared space facilities (Gibb 2015). Recently, several agent-based simulation models that

11 better reflect road user behaviour have been developed (Hussein and Sayed 2015; Hussein and

12 Sayed 2017; Liu et al. 2014). For example, a pedestrian simulation model was developed at the

13 University of British Columbia to study the behavior of pedestrians during interactions with other

14 pedestrians and objects in the walking environment (Hussein and Sayed 2015; Hussein and Sayed

15 2017). The model relies on the identification of the rules of interactions between pedestrians,

16 which are usually extracted from comprehensive analysis of pedestrian behaviour. The results

17 reported in a recent study that applied the model to analyse pedestrian interactions on the Brooklyn

18 Bridge pedestrian/bike path (Hussein and Sayed 2018) suggests that the model can be expanded

19 to address pedestrians-bikes interactions in non-motorized shared spaces. However, such a task

20 should be preceded by a comprehensive study to analyze road user behaviour in shared spaces.

21 Therefore, this study investigates the microscopic interaction behaviour between cyclists

22 and pedestrians in a non-motorized shared space. Video data were collected at the shared space of

23 Robson Square in downtown Vancouver. Computer vision algorithms were applied to detect and

24 track pedestrians and cyclists from video footage. The extracted trajectories were used to define

Page 4 of 45



Draft


1 different indicators that are used to analyze interaction behaviour of road users, including speed

2 and acceleration profiles, longitudinal and lateral distances between road users during different

3 phases of the interaction. Furthermore, the study identifies several parameters that can be used to

4 model such interaction in the microsimulation model.

5 PREVIOUS WORK

6 Shared Space Concept

7 Shared space is an emerging concept of urban design that supports pedestrian and cyclist

8 movements with slower vehicles (typically, 30 km/hr) and was shown to have safety, economic,

9 and quality of life benefits (Swinburne 2005; Akkar 2005). These schemes of street design

10 encourage the integration of road users (pedestrian-friendly environment) by reducing the

11 segregation between road users and decreasing the dominance of motorized vehicles (Kaparias et

12 al. 2012). Several studies showed documented benefits of shared spaces. For example, converting

13 Kensington High Street, a busy route in London, to a shared space area was shown to have a

14 significant improvement in pedestrian safety (pedestrian collisions were reduced by about 64%)

15 and increase in pedestrian activity levels (Swinburne 2005). As well, the conversion of a five-leg

16 intersection in Oosterwolde, Netherlands to a paved open area for all users resulted in a reduction

17 in severe collisions despite the increase in traffic volume at the intersection (Hamilton-Baillie

18 2008).

19 The benefits of shared space schemes are also extended to improve the attractiveness of

20 the urban realm (Hamilton-Baillie 2008), and reduce the traffic-related air pollution (Shu et al.

21 2016). As well, economic and urban revitalization and investments growth were clearly observed

22 after a remodeling of the Grey’s Monument area, Newcastle into a public space area (Akkar 2005).

23 Despite the benefits of shared space areas, concerns were raised about the severity of interactions

Page 5 of 45



Draft


1 between non-motorized road users. Collisions between the vulnerable road users (cyclist-

2 pedestrian collision) can lead to severe consequences, especially for pedestrians, who are more

3 likely to suffer from severe injuries that can lead to death if involved in collisions with cyclists

4 (Graw and Konig 2002).

5 Road User Behaviour in Shared Spaces

6 Only few studies have investigated the behaviour of road users in shared spaces. For

7 example, Hussein et al. (2016) studied the safety of pedestrians and cyclists in the shared space of

8 Park Avenue during the summer street program in New York City. The study utilized computer

9 vision techniques in extracting the trajectory of road users and investigating their speeds and gait

10 parameters during the summer street program. Schonauer et al. (2012) studied the behaviour of

11 road users before and after converting a complex roundabout in Austria into a shared space area.

12 The study concluded that the speed distributions of all modes became narrower in the shared space

13 area, which can be explained by the smoother movements and less stop and go conditions in shared

14 spaces compared to complex roundabouts. Kiyota et al. (2000) studied the behaviour and the safety

15 of pedestrians and cyclists in a non-motorized shared space in Japan and found that the perceived

16 risk by pedestrians becomes higher as the passing distance between the cyclists and pedestrians

17 decreases. Cyclist speed was shown to be affected by pedestrian density, and lower cyclist speeds

18 were observed at higher pedestrian densities (Essa et al. 2018). Beitel et al. (2018) studied the

19 behaviour and safety of pedestrians and cyclist on a non-motorized shared space area in McGill

20 University, Montreal. The study classified the conflicts between the cyclists and pedestrians based

21 on two criteria: the angle between their intersecting trajectories and who reaches first to the conflict

22 point. The results showed that speed of cyclists decreases as pedestrian density increases, and the

23 conflict rate between cyclists and pedestrians increases with pedestrian density. Furthermore,

24 previous studies used several parameters to describe the behaviour of pedestrians or cyclists in

Page 6 of 45



Draft


1 shared spaces or segregated roads/paths. These parameters include speed and acceleration profiles

2 (Twaddle and Grigoropoulos 2016; Ma and Luo 2016; Afghari et al. 2014), lateral distance and/or

3 headway (Luo et al. 2015; Hussein and Sayed 2017; Hoogendoorn and Daamen, 2016), pedestrian

4 gait parameters (Alsaleh et al. 2018; Hussein and Sayed 2015; Guo et al. 2016; Hediyeh et al.

5 2014).

6 The literature review shows that most studies did not investigate the microscopic

7 interaction behaviour between cyclists and pedestrians in non-motorized shared spaces.

8 Investigating such interaction behaviour is important for modeling these interactions in

9 microsimulation models. Therefore, this study aims to 1) investigate the microscopic interaction

10 behaviour between cyclists and pedestrians in shared space environments; 2) investigate the

11 collision avoidance mechanisms and the factors affecting the selection of the collision avoidance

12 mechanisms employed by road users to avoid collisions with other shared space users; and 3)

13 identify several parameters that can be used to model cyclists and pedestrians interactions in

14 microsimulation models.

15 METHODOLOGY

16 Video data was collected at the busy shared space of Robson Square in downtown

17 Vancouver. Robson square is a non-motorized shared space area that is characterized by high

18 pedestrian and cyclist activities. Pedestrians and cyclists were automatically tracked and their

19 trajectories were extracted from the video footage using Computer Vision (CV) algorithms. The

20 interacting road users were classified based on the interacting angle into three categories: [i]

21 interactions with shared space users moving in the same direction, [ii] interactions with opposing

22 shared space users, and [iii] interactions with crossing shared space users. The speed and

23 acceleration profiles and relative spatial profiles were used to analyze the interaction behaviour of

24 pedestrians and cyclists in different cases. The following sub-sections summarize the details of

Page 7 of 45



Draft


1 each of the following tasks: data collection, road user tracking, extraction of spatial, speed and

2 acceleration profiles, and interaction identification and analysis.

3 Data Collection

4 Video data were collected at a busy non-motorized shared space, located in Robson

5 Square in downtown Vancouver, as shown in Figure 1. The city of Vancouver considered the

6 permanent closure of Robson Street, between Hornby Street in the west and Howe Street in the

7 east, for motorized traffic in order to provide a comfortable and safe plaza for vulnerable road

8 users in the heart of downtown. The area is a place for many recreational and business facilities,

9 including the Vancouver Art Gallery, the Robson Square ice rink, the Vancouver Supreme Court,

10 among others. The area is an active environment for walking and cycling especially in the summer

11 season. Pedestrian-cyclist interactions are frequently observed in the shared space area. Eighteen

12 hours of video data were collected in daylight over several days in August and September 2016.

13 Video data were collected using a video camera installed near the western entrance of the shared

14 space, as shown in Figure 1.

15 Road user Tracking

16 The automated extraction of road user trajectories from video footage was conducted using

17 a video analysis system that has been developed at the University of British Columbia (Saunier

18 and Sayed 2006). The system employs computer-vision algorithms to detect, track, and classify

19 road users in traffic scenes. The detection of the road users is carried out using an implementation

20 of the Kanade–Lucas–Tomasi (KLT) feature-tracking algorithm (Lucas and Kanade 1981)

21 (Tomasi and Kanade 1991). The algorithm detects distinct points (features) on moving objects in

22 the video scene. The algorithm is capable of differentiating between features that belong to road

23 users (e.g. pedestrians, cyclists) and that are part of the background.

Page 8 of 45



Draft


1 Typically, multiple features are tracked on each road user each video frame. Clustering of

2 features is important to determine which group of features belongs to the same road user. The

3 clustering algorithm uses different cues to cluster the tracked features, including the spatial

4 proximity of features and their dynamical similarities. The subsequent step in the analysis is the

5 classification of road users. The classification process is then used to discriminate between

6 trajectories that belong to different categories of road users (e.g. pedestrians and cyclists). The

7 main cues in the classification of road users’ trajectories are the frequency harmonics extracted

8 from their speed profiles (Zaki and Sayed 2013). Several previous studies validated the accuracy

9 of the system and found to be satisfactory (Ismail et al. 2010; Zaki and Sayed 2013; Ismail et al.

10 2013).

11 However, an essential task prior to the aforementioned procedures is to create a mapping

12 between video image plane coordinates and the real-world coordinate system, so that the spatial

13 and temporal information of the tracked trajectories are known in the actual coordinate system of

14 the location being analyzed. This task is known as the camera calibration, and is described in

15 details in Ismail et al. (Ismail et al. 2013). The overall analysis procedures are illustrated in Figure

16 2.

17 One of the limitation of the using KLT algorithm for tracking cyclists and pedestrians is

18 the over-grouping problem which occurs when road users (e.g. pedestrians) move closely together

19 or when one feature is detected moving consistently with two other distinct groups (Saunier and

20 Sayed 2006). Moreover, the KLT tracking algorithm may not handle occlusion very well as the

21 trajectories may be disrupted by occlusions (Saunier and Sayed 2006). In this study, the tracking

22 parameters were carefully selected and tuned to minimize tracking errors. As well, the tracking

23 was checked manually for each road user involved in the interactions to ensure the accuracy of the

24 tracking of each interaction. Other tracking techniques and a benchmark for the evaluation of

Page 9 of 45



Draft


1 different multi-object tracking methods can be found in (Milan et al. 2016).

2 Extraction of Spatial, Speed and Acceleration Profiles

3 The road user trajectories record the spatial coordinates and the instantaneous speed at each video

4 frame (1/30 second). The trajectory is defined along the trajectory lifetime (n video frame) as a finite set of

5 tuples (Ismail et al. 2013), as shown in Equation 1:

67 (1)𝑇 = {(𝑋1,𝑌1,𝑉𝑋1,𝑉𝑌1, … ,𝑋𝑖,𝑌𝑖,𝑉𝑋𝑖,𝑉𝑌𝑖, … ,𝑋𝑛,𝑌𝑛,𝑉𝑋𝑛,𝑉𝑌𝑛)}89 Where is a discrete temporal index, are the spatial coordinates of the road 𝑖 = {1,…,𝑛} 𝑋𝑖 and 𝑌𝑖

10 user at the frame ( ), and are the corresponding velocities. A speed profile (S) for each road 𝑖 𝑉𝑋𝑖,𝑉𝑌𝑖

11 user is defined along the trajectory lifetime as , with and are the velocity 𝑆 = 𝑛𝑜𝑟𝑚(𝑉𝑥, 𝑉𝑦) 𝑉𝑥 𝑉𝑦

12 vectors of length n, for the and coordinates, respectively. 𝑋 𝑌

13 Estimation of pedestrian and cyclist acceleration profiles from the noisy walking speed

14 signal obtained from the video data required first smoothing of the speed time series. The road

15 user acceleration profiles can be derived from their smoothed speed profile as follows:

1617 (2)𝑎 =

∆(𝑆)∆𝑡 =

𝑆𝑖 ― 𝑆𝑖 ― 𝛿

𝛿1819 Where is the acceleration profile and it is defined by a time series of length ( ), 𝑎 𝑛 ― 𝛿 𝑖 = {𝛿 + 1,…

20 is a discrete temporal index, and is the smoothen speed at the video frame. The analysis ,𝑛} 𝑆𝑖 𝑖𝑡ℎ

21 procedures of road user trajectory are illustrated in Figure 2.

22 A validation of the accuracy of the road user trajectory extraction process was carried out

23 by comparing a sample of 100 trajectories with the manually measured speeds of the road users

24 based on the time required to traverse a known distance in the video scene. The mean percentage

25 absolute error was 6.6 % (absolute error of 0.11 m/s for pedestrians and 0.18 m/s for cyclists),

26 which is considered low and to not affect the findings. In this study, similar to many previous

Page 10 of 45



Draft


1 studies of shared spaces as (Beitel et al. 2018), road user classification was carried out manually

2 to avoid any misclassification of road users in the shared space area.

3 Interaction Identification and Analysis

4 Video data were manually reviewed in order to define different types of interactions between

5 pedestrians and cyclists in the shared space. Observed interactions were classified into three main

6 categories based on the interacting angle (Beitel et al. 2018):

7 Interactions with slower shared-space users moving in the same direction: These

8 interactions involve cyclists and pedestrians moving in the same directions (difference in

9 movement direction = 0° ± 30°).

10 Interactions with opposing shared-space users: These interactions involve cyclists and

11 pedestrians moving in the opposite directions (difference in movement direction = 180° ±

12 30°).

13 Interactions with crossing shared-space users: These interactions include all other cyclist-

14 pedestrian interactions.

15 A total number of 208 interactions were observed during the 18 hours of the analyzed video

16 data. The trajectories, speed profiles, acceleration profiles of all pedestrians and cyclists involved

17 in at least one of the previous interactions were considered for the analysis. In the analyzed

18 interactions, an evasive action was taken by at least one of the road users involved in an interaction

19 to resolve the conflict. Generally, two types of evasive actions were observed: 1) changing

20 cycling/walking speed, and 2) swerving maneuvers.

21 For each shared-space user involved in the interaction, the start and end of each interaction

22 were identified manually form both the speed profile and the trajectory of the road user. The road

23 user trajectories were used to compute the relative longitudinal and lateral distance between the

Page 11 of 45



Draft


1 interacting road users at different phases of the interaction. Test of significance (two-sample t-test)

2 is used to investigate the significance of change in road user behaviour during the interaction.

3 DISCUSSION OF RESULTS

4 Interactions with Shared-Space Users Moving in the Same Direction

5 This type of interactions occurs when a faster shared-space user is hindered by another

6 slower user, moving in the same direction (difference in movement direction = 0° ± 30°). It is

7 frequently observed in shared-space areas due to the significant difference in the operational speed

8 between different categories of shared-space users (Department of Transport and Main Roads of

9 Queensland State 2014). Faster shared-space users (typically, cyclists) tend to apply at least one

10 of the following two collision avoidance mechanisms to avoid collision with the slower shared-

11 space user (pedestrians, in most cases): [i] reduce their speeds to keep adequate space with the

12 slower shared-space users (following -maneuver), or [ii] apply swerving maneuvers to overtake

13 the slower shared-space users (overtaking maneuvers). In most cases, the overtaking maneuvers

14 are associated with a change in speed of the faster shared-space user. The factors affecting the

15 selection of the collision avoidance mechanism and a detailed description of the two strategies are

16 discussed in more details in the following subsections.

17 I) Selection of the Collision Avoidance Strategy

18 As discussed earlier, road users involved in interactions with other road users moving in

19 the same direction may choose to follow slower road users or overtake them. Reviewing the

20 interactions suggested that the selection of the collision avoidance mechanism taken by the faster

21 road users mainly depends on two factors; [i] the available lateral space for the faster shared-space

22 user, and [ii] the shared space density during the interaction. When the density of the shared space

23 increases, it becomes more difficult for faster road users to swerve and overtake slower pedestrians.

Page 12 of 45



Draft


1 Similarly, when the available lateral space for the faster road users is limited, due to barriers or

2 other street furniture, faster road users may be forced to reduce their speed and follow slower road

3 users.

4 A binary logit model was developed to assess the probability of selecting each of the two

5 collision avoidance strategies. One hundred and five cyclist-pedestrian same-direction interactions

6 were reviewed manually, and the collision avoidance strategy adopted by the faster road users

7 were recorded. The model is expressed as follow:

8 (3)𝑃𝑜𝑣𝑒𝑟𝑡𝑎𝑘𝑖𝑛𝑔 =1

1 + 𝑒―(𝑎0 + 𝑎1.𝐷𝑖𝑠𝑡 + 𝑎2.𝐷𝑒𝑛)

9 Where is the probability of performing an overtaking maneuver, is the maximum 𝑃𝑜𝑣𝑒𝑟𝑡𝑎𝑘𝑖𝑛𝑔 𝐷𝑖𝑠𝑡

10 available lateral distance for the faster road during the interaction (in meters), is the average 𝐷𝑒𝑛

11 shared space density (in road user / m2), and a0, a1, and a2 are model parameters. The density is

12 calculated over the area of 12.5 m2 that covers the interacting road users. This area is selected to

13 reflect the modelled area and the presence of sitting tables for pedestrians and is consistent with

14 other studies (Hussein and Sayed 2018; Liu et al. 2014).

15 The estimated model parameters are presented in Table 1. The coefficients of the two

16 explanatory variables were significant at 95%. The estimated parameters confirm that probability

17 of the faster cyclist to execute overtaking maneuvers increases as the density of the shared space

18 decreases and the available space increases as shown in Figure 3. The average value of the shared

19 space density and available space for the two collision-avoidance strategies are presented in Table

20 2.

21 a) Description of the Following Strategy

22 Sixty-five pedestrian-cyclist interactions that involved following maneuvers were analyzed

23 in this study. The speeds of all shared-space users involved in the maneuvers were extracted, as

Page 13 of 45



Draft


1 well as the longitudinal distances between each pair of shared-space users in conflict. Figure 4

2 shows an example of a following maneuver that involves a faster cyclist and slower pedestrian.

3 The cyclist’s and pedestrian’s speed profiles, acceleration profiles, and the longitudinal distance

4 profile between the cyclists and the pedestrian are shown in Figure 5. As shown in the figure, the

5 interaction can be classified into three phases:

6 1. Phase 1: the cyclist is travelling at a normal cycling speed (4.33 m/s in this example) as

7 he/she is approaching a slower walking pedestrian.

8 2. Phase 2: at a specific distance from the slower pedestrian (4.68 meters in the current

9 example), the cyclist started to reduce the speed to avoid collision with the slower

10 pedestrian. The average longitudinal distance at which cyclists started to reduce speed for

11 the 65 interactions analyzed was found to be 4.46 m, with a standard deviation of 1.29 m.

12 The average deceleration value for the 65 interactions was found to be 0.81 m/s2, with a

13 standard deviation of 0.40 m/s2. This phase can be considered as an evasive action taken

14 by the cyclist to avoid collision with the slower pedestrians during the high-density periods

15 in the shared space.

16 3. Phase 3: the cyclist approaches a safe following distance (2.02 meters in the current

17 example) so that he/she keeps travelling at a reduced speed to maintain this distance. The

18 average following distance was found to be 2.36 m, with a standard deviation of 0.50 m

19 for the 65 interactions analyzed.

2021 Table 3 shows the mean and the standard deviation of the cycling speed, and the walking

22 speed during the following maneuver, compared to corresponding values during free moving

23 periods. As shown in the table, the cycling speed was significantly reduced by about 39% when

24 following slower pedestrians in the shared space. The pedestrian speed was almost constant during

25 the maneuver, which was expected since pedestrians did not apply any evasive actions and, in

Page 14 of 45



Draft


1 some cases, they were not even aware of the following cyclist.

2 Based on the previous analysis, it is suggested to define three parameters to model this type

3 of interaction in a micro-simulation model:

4 1. The longitudinal distance between the pedestrian and the cyclist at which the cyclists start

5 to reduce their speeds.

6 2. The desired distance, which the cyclists prefer to keep from slower leading pedestrians

7 (following distance).

8 3. The cyclists’ deceleration needed to reduce speed when approaching the slow pedestrian.

9 b) Description of the Overtaking Strategy

10 Fourty interactions that involve an overtaking maneuver were observed in the analyzed

11 data set. The maneuver typically includes cyclists changing their movement direction at a specific

12 distance from the slower pedestrians to avoid collision. As well, an increase in cycling speed was

13 observed during the overtaking phase. Figure 6 shows an example of a cyclist’s overtaking

14 maneuver. The cyclists and pedestrians speed and acceleration profiles, and the longitudinal

15 distance profile between the cyclist and the pedestrian are shown in Figure 7. As shown in the

16 figure, this type of interaction can be described by the following three phases:

17 1. Phase 1: the cyclist is moving at the normal cycling speed (or at a reduced speed if the

18 cycling is following a slower pedestrian as was noticed in many cases).

19 2. Phase 2: the cyclist decides to execute a swerving maneuver to overtake the slower

20 pedestrian. This usually occurs when an adequate space becomes available for the cyclists

21 to perform the maneuver. The cyclist accelerates and swerves in order to overtake the

22 slower pedestrian. The average acceleration observed in the 40 interactions was found to

23 be 0.87 m/s2, with a standard deviation of 0.35 m/s2. According to the data analyzed in this

Page 15 of 45



Draft


1 study, the swerving maneuver is usually initiated simultaneously with the cyclist’s

2 acceleration. The average longitudinal distance between the overtaking cyclists and the

3 overtaken pedestrians at the start of the overtaking maneuver was found to be 3.00 m, with

4 a standard deviation of 1.65 m.

5 3. Phase 3: the cyclist reaches the desired speed and the desired lateral distance that he/she

6 prefers to keep from the slower pedestrian. Consequently, the cyclist maintained the

7 desired speed and continues cycling in the original movement direction to overtake the

8 slower pedestrian. The average minimum lateral distance between the overtaking cyclists

9 and the overtaken pedestrians during the analyzed maneuvers was found to be 1.23 m, with

10 a standard deviation of 0.25 m.

11 Table 3 shows the mean and the standard deviation of the cycling speed, and the walking

12 speed during the overtaking maneuver, compared to corresponding values before the interaction.

13 As shown in the table, the cycling speed increased significantly during the overtaking process. On

14 average, cyclist speeds were increased by about 41% during the overtaking of the slower

15 pedestrians in the shared space. Similar to the previous interaction, it was found that the pedestrian

16 speed was almost constant during the maneuver, which was expected since pedestrians did not

17 apply any evasive action and, in some cases, they were not even aware of the following cyclist.

1819 Based on the previous analysis, it is suggested to define three parameters to model this type

20 of interaction in a micro-simulation model:

21 1. The longitudinal distance between the cyclist and the pedestrian at which at the swerving

22 maneuver starts.

23 2. The lateral distance between the overtaking cyclists and the overtaken pedestrians at the

24 passing point.

25 3. The cyclists’ acceleration needed to execute the overtaking maneuver.

Page 16 of 45



Draft


1 Interactions with Shared-Space Users Moving in Opposing Direction (Head-on Interactions)

2 This type of interaction occurs when a pedestrian and a cyclist approach each other, with

3 an angle of 180° ± 30°, in time and space so that a collision would occur if none of them takes an

4 evasive action. In the analyzed video data, sixty-two head-on interactions were observed. This type

5 of interaction involves at least one of the shared-space users in conflict changing direction to avoid

6 the collision. It was observed that cyclists tend to change direction when the shared space density

7 is relatively low, where a sufficient lateral distance is available for bikes to perform the swerving

8 maneuver. While at higher densities, pedestrians are more likely to take the evasive action. The

9 average value of the shared space density and the available lateral space for road users during the

10 swerving maneuver are presented in Table 4.

11 It was observed that most of the maneuvers are associated with a reduction in the cyclist’s

12 speed. Table 5 shows the mean and the standard deviation of the cycling speed, and the walking

13 speed in both free flow conditions and during the interaction. As shown in the table, cycling speed

14 is reduced significantly during the interaction, regardless whether the cyclist executes a swerving

15 maneuver or not. The average reduction in speed during the interaction was found to be 20%. On

16 the other hand, the speed of pedestrians involved in this type of interaction did not change during

17 the interaction regardless of whether they change the movement direction. This agrees with the

18 results reported in (Hussein and Sayed 2015), which indicated that pedestrians do not change speed

19 when changing direction to avoid an opposing pedestrian. However, other studies found that

20 pedestrians may change their speed while changing the direction (Daamen et al. 2014). Figure 8

21 and Figure 9 show a typical example of a head-on interaction along with the corresponding speed

22 and acceleration profiles of the two shared space users involved in the interaction. As shown in

23 Figure 9, this type of interaction can be described by the following three phases:

Page 17 of 45



Draft


1 1. Phase 1: cyclist is moving at the preferred speed, depending on the current density of the

2 shared space environment.

3 2. Phase 2: the cyclist approaches the opposing pedestrian and decides to swerve and reduce

4 speed to avoid collision. The average deceleration value for the cyclists was found to be

5 1.14 m/s2, with a standard deviation of 0.75 m/s2. The average longitudinal distance at

6 which the cyclist or pedestrian starts to execute the evasive action was found to be about

7 6.39 m, with a standard deviation of 2.31 m.

8 3. Phase 3: once the cyclist ensures that he/she has an adequate lateral distance with the

9 opposing pedestrian, the cyclist starts to accelerate to the normal cycling speed at the

10 current density level of the shared space. The average acceleration value for the cyclists

11 was found to be 1.11 m/s2, with a standard deviation of 0.56 m/s2. The average lateral

12 distance between the cyclists and pedestrians at the crossing point was found to be 1.24 m,

13 with a standard deviation of 0.32 m. This lateral distance is significantly higher than the

14 corresponding distance in pedestrian head-on interactions. Hussein and Sayed reported that

15 the average lateral distance between two opposing pedestrians on a typical crosswalk in

16 Vancouver was 0.75 m (Hussein and Sayed 2017). The difference can be explained by the

17 larger size and the higher speed of bicycles, compared to pedestrians.

1819 Based on the previous analysis, it is suggested to define three parameters to model the

20 head-on interaction between pedestrians and cyclists in a micro-simulation model:

21 1. The longitudinal distance between the cyclist and the pedestrian at which one of them

22 starts to swerve to avoid the collision.

23 2. The lateral distance between the cyclists and the pedestrians at the crossing point.

24 3. The cyclists’ acceleration and deceleration needed to execute the overtaking maneuver.

Page 18 of 45



Draft


1 Bicycles Interaction with Crossing Pedestrians

2 This type of interaction occurs when a pedestrian and a cyclist approach each other from

3 the side in time and space so that a collision can occur if no action is taken by either of the shared-

4 space users. The shared space does not provide a right of way for one user over the other. Rather,

5 all users share the right-of-way so that it is not always clear which shared-space user should yield

6 during these interactions. Fourty-one cyclist-pedestrian crossing interactions were observed in the

7 data set. At least one shared space user involved in this type of interaction changed the speed

8 and/or the direction to avoid a collision. The 41 interactions were classified into two categories; 1)

9 interactions where the pedestrian crosses first (cyclist yields to the pedestrian), and 2) interactions

10 where the cyclist crosses first (pedestrian yields for the cyclist). If the cyclist yields to pedestrians,

11 pedestrians tend to increase their walking speed to cross and clear the conflict area. As shown in

12 Table 6, the pedestrian speed increased significantly during the interaction if the cyclist yields to

13 the pedestrian. On average, the speed of pedestrians increased by 25.7% during the interaction.

14 The cycling speed was reduced significantly by about 23.4% in order to provide right of way for

15 the pedestrians. If the pedestrian yields to the cyclist, pedestrians reduce their walking speed. On

16 average, pedestrians significantly reduced their speed by 63.5% in order to provide the right of

17 way to the crossing bicycles. As well, the cyclist speed was reduced significantly during the

18 interaction. The average reduction in cycling speed during the interaction was found to be 18.8%.

19 Values reported in Table 6 suggest that cyclists tend to reduce speed when approaching a crossing

20 pedestrian. However, the decision is made by the pedestrian whether to speed up and cross first or

21 slow down and allow the cyclist to cross first. The decision of the pedestrian depends on the

22 approaching bike speed as shown in Table 7, where the approaching speeds of cyclists are

23 significantly higher for the cases where the pedestrian yields (bike cross first) compared to the

24 cases where pedestrian crosses first. If the approaching cyclist speed is high, pedestrians prefer to

Page 19 of 45



Draft


1 yield for the cyclist. Otherwise, the pedestrian accelerates to cross before the cyclists.

2 Figure 10 and Figure 11 show a typical example of a crossing interaction that involves a

3 cyclist yields to the crossing pedestrian along with the corresponding speed and acceleration

4 profiles of the two shared space users involved in the interaction. As shown in the figure, the

5 interaction can be described by the following three phases:

6 1. Phase 1: the cyclist and pedestrian are moving at their preferred cycling and walking

7 speeds, depending on the current density of the shared space environment.

8 2. Phase 2: the pedestrian decides to cross first. The pedestrian increases his/her walking

9 speed to safely clear the conflict course. The cyclist decelerates in order to allow the

10 pedestrian to cross the conflict course safely. The average deceleration of cyclists yielded

11 to pedestrian was found to be 0.92 m/s2, with a standard deviation of 0.42 m/s2. While the

12 average deceleration of the cyclists when the pedestrians yield to cyclists was found to be

13 1.21 m/s2, with a standard deviation of 0.68 m/s2. The average longitudinal distance

14 between the cyclists and pedestrians along the crossing point at which the cyclist or

15 pedestrian behavior starts to be affected by this interaction is about 5.95 m with a standard

16 deviation of 2.73 m. At the end of this phase, the pedestrian has safely crossed the conflict

17 course.

18 3. Phase 3: as the pedestrian clears the conflict course, the cyclist starts to bike at a cycling

19 speed that is suitable to the current density level of the shared space.

20 Based on the previous analysis, it is suggested to define three parameters to model the

21 crossing interaction between pedestrians and cyclists in a micro-simulation model. However, a

22 more detailed analysis for a larger data set is required to determine the factors that affect

23 pedestrian’s decision to yield or to accelerate and cross first, and the factors that affect the change

24 of road user directions. The three recommended parameters are summarized as follows:

Page 20 of 45



Draft


1 1. The percentage of the change in pedestrian’s speed for the two possible evasive actions

2 (pedestrian yields to bike or crosses first).

3 2. The longitudinal distance between the pedestrian and the cyclist along the crossing point

4 at which they start to modify speed to resolve the conflict.

5 3. The cyclists’ deceleration as they approach the crossing pedestrian, for the two possible

6 evasive actions.

7 CONCLUSION AND FUTURE RESEARCH

8 This study investigates the microscopic interaction behaviour of cyclists and pedestrians in

9 a non-motorized shared space in Vancouver, British Columbia. Video data were collected at the

10 Robson Square shared space in downtown Vancouver. Trajectories of cyclists and pedestrians

11 involved in 208 interactions (416 Trajectories) in the shared space area were extracted using

12 computer vision algorithms. The extracted trajectories were used to define the speed and

13 acceleration profiles of road users, longitudinal and lateral distances between road users during the

14 different phases of the interactions. The study demonstrates the use of speed and acceleration

15 profiles, and relative spatial profiles in describing the microscopic interaction behaviour between

16 cyclists and pedestrians. The interactions between cyclists and pedestrians were categorized into

17 three categories; [i] interactions with shared-space users moving in the same direction, [ii]

18 interactions with shared-space users moving in opposing direction (head-on interactions), and [iii]

19 bicycles interaction with crossing pedestrians. Different collision avoidance mechanisms were

20 applied by the road users to avoid collision with other road users during the interactions. The

21 collision avoidance mechanisms involved either a change in the movement direction or

22 walking/cycling speed or both of them to avoid collision with other shared-space users. The results

23 showed that the collision avoidance mechanisms adopted by road users depend on several factors

24 including the shared space density and the space available for road users.

Page 21 of 45



Draft


1 The results obtained in this study represent the basis for developing the behaviour rules

2 required to model cyclist and pedestrian interactions in shared space areas. The study identifies

3 several parameters that can be used to model the microscopic behaviour of cyclists and pedestrians

4 in the microsimulation models. Table 8 summarizes the parameters that describe cyclist-pedestrian

5 interactions in shared space areas and the values reported in the current study. Modeling road user

6 interactions can be beneficial in evaluating the safety and the facility performance under various

7 designs of shared space areas.

8 Limitations of the current work are mainly related to the sample size of the interactions

9 analyzed. Thus, future research should consider investigating shared-space users’ interactions in

10 different environments and layouts of shared space areas using larger datasets. Investigating the

11 changes in pedestrian speed while changing direction in head-on interactions and the factors

12 effecting the change of road user directions in crossing interactions are also important. As well, it

13 is recommended to develop statistical models to describe some of the interaction behaviour

14 discussed in this study. This includes developing a statistical model to predict which shared space

15 user yield in crossing interactions. Also, a model that addresses which shared space user changes

16 direction in head-on conflicts is required.

17 REFERENCES

18 Adams J. 1995. Risk. Routledge, London.

19 Afghari, A. P., Ismail, K., Saunier, N., Sharma, A., and Miranda-Moreno, L. 2014. Pedestrian-cyclist

20 interactions at bus stops along segregated bike paths: a case study of Montreal. In Proceedings of the

21 93rd Annual Meeting of the Transportation Research Board, Washington, D.C..

22 Akkar, M. 2005. The changing ‘publicness’ of contemporary public spaces: a case study of the Grey's

23 Monument Area, Newcastle upon Tyne. Urban Design International. 10(2): 95-113.

24 Alsaleh, R., Sayed, T., and Zaki, M. H. 2018. Assessing the Effect of Pedestrians’ Use of Cell Phones

25 on Their Walking Behavior. A Study Based on Automated Video Analysis. Transportation Research

Page 22 of 45



Draft


1 Record: Journal of the Transportation Research Board, 2672(35): 46-57.

2 Beitel, D., Stipancic, J., Manaugh, K., and Miranda-Moreno, L. 2018. Assessing safety of shared space

3 using cyclist-pedestrian interactions and automated video conflict analysis. Transportation research part

4 D: transport and environment, 65: 710-724.

5 Daamen, W., Hoogendoorn, S., Campanella, M., and Versluis, D. 2014. Interaction behavior between

6 individual pedestrians. In Pedestrian and Evacuation Dynamics 2012. Edited by Weidmann U., Kirsch

7 U., Schreckenberg M. Springer, Cham. pp. 1305-1313.

8 Department of Transport and Main Roads. 2014. Speed management on shared paths. Technical Note

9 TN-130. State of Queensland, Australia. https://www.tmr.qld.gov.au/-

10 /media/busind/techstdpubs/Technical-notes/Traffic-engineering/TN130.pdf?la=en

11 Essa, M., Hussein, M., and Sayed, T. 2018. Road Users’ Behavior and Safety Analysis of Pedestrian-

12 Bike Shared Space: Case Study of Robson Street in Vancouver. Canadian Journal of Civil Engineering,

13 45(12): 1053-1064.

14 Gibb, S. 2015. Simulating the Streets of Tomorrow: An Innovative Approach to Modeling Shared

15 Space. CH2M Hill Inc.

16 Graw, M., and Konig, H. 2002. Fatal pedestrian–bicycle collisions. Forensic Science International,

17 126(3): 241-247.

18 Guo, Y., Sayed, T., and Zaki, M. H. 2016. Automated analysis of pedestrian walking behaviour at a

19 signalised intersection in China. IET Intelligent Transport Systems, 11(1): 28-36.

20 Hamilton-Baillie, B. 2008. Shared space: Reconciling people, places and traffic. Built environment,

21 34(2): 161-181.

22 Hediyeh, H., Sayed, T., Zaki, M. H., and Mori, G. 2014. Pedestrian gait analysis using automated

23 computer vision techniques. Transportmetrica A: Transport Science, 10(3): 214-232.

24 Hipp, J. A., Bird, A., van Bakergem, M., and Yarnall, E. 2017. Moving targets: Promoting physical

25 activity in public spaces via open streets in the US. Preventive medicine, 103: S15-S20.

26 Hoogendoorn, S., and Daamen, W. 2016. Bicycle headway modeling and its applications.

27 Transportation Research Record: Journal of the Transportation Research Board, 2587: 34-40.

Page 23 of 45



Draft


1 Hussein, M., and Sayed, T. 2017. A bi-directonal agent based pedestrian microscopic model.

2 Transportmetrica A: Transport Science, 13(4): 326-355.

3 Hussein, M., and T. Sayed. 2015. A Uni-directional agent based pedestrian microscopic model.

4 Canadian journal of civil engineering, 42(12): 1114-1124.

5 Hussein, M., and Sayed, T. 2015. Microscopic Pedestrian Interaction Behavior Analysis Using Gait

6 Parameters. Transportation Research Record: Journal of the Transportation Research Board, 2519: 28-

7 38.

8 Hussein, M., and Sayed, T. 2018. Validation of an agent-based microscopic pedestrian simulation

9 model at the pedestrian walkway of Brooklyn Bridge. Transportation Research Record: Journal of the

10 Transportation Research Board, 2672(35): 33-45.

11 Hussein, M., Popescu, B., Sayed, T., and Kim, L. 2016. Analysis of Road User Behavior and Safety

12 During New York City’s Summer Streets Program. Transportation Research Record: Journal of the

13 Transportation Research Board, 2586: 120-130.

14 Ismail, K., Sayed, T., and Saunier, N. 2013. A methodology for precise camera calibration for data

15 collection applications in urban traffic scenes. Canadian Journal of Civil Engineering, 40(1): 57-67.

16 Ismail, K., Sayed, T., and Saunier, N. 2010. Automated analysis of pedestrian-vehicle: Conflicts context

17 for before-and-after studies. Transportation Research Record: Journal of the Transportation Research

18 Board, 2198: 52-64.

19 Tomasi, C. and Kanade, T. 1991. Detection and Tracking of Point Features. Carnegie Mellon University.

20 Technical Report CMU-CS-91-132.

21 Kaparias, I., Bell, M. G., Miri, A., Chan, C., and Mount, B. 2012. Analysing the perceptions of

22 pedestrians and drivers to shared space. Transportation research part F: traffic psychology and

23 behaviour, 15(3): 297-310.

24 Kiyota, M., Vandebona, U., Katafuchi, N., and Inoue, S. 2000. Bicycle and pedestrian traffic conflicts

25 on shared pavements. In Proceedings of the 14th Velo-city International Conference, Munich, Germany.

26 Liu, S., Lo, S., Ma, J., and Wang, W. 2014. An agent-based microscopic pedestrian flow simulation

27 model for pedestrian traffic problems. IEEE Transactions on Intelligent Transportation Systems, 15(3):

28 992-1001.

Page 24 of 45



Draft


1 Lucas, B. D. and Kanade, T. 1981. An Iterative Image Registration Technique with an Application to

2 Stereo Vision. In Proceedings of the International Joint Conference on Artificial Intelligence. pp. 674-

3 679.

4 Luo, Y., Jia, B., Liu, J., Lam, W. H., Li, X., and Gao, Z. 2015. Modeling the interactions between car

5 and bicycle in heterogeneous traffic. Journal of advanced transportation, 49(1): 29-47.

6 Ma, X., and Luo, D. 2016. Modeling cyclist acceleration process for bicycle traffic simulation using

7 naturalistic data. Transportation research part F: traffic psychology and behaviour, 40: 130-144.

8 Milan, A., Leal-Taixe, L., Reid, I., Roth, S., and Schindler, K. 2016. MOT16: A benchmark for multi-

9 object tracking. arXiv preprint arXiv:1603.00831v2.

10 Saunier, N., and Sayed, T. 2006. A feature-based tracking algorithm for vehicles in intersections. In

11 Proceedings of the 3rd Canadian Conference of Computer and Robot Vision, Quebec, Canada.

12 Schonauer, R. S. M., Schrom-Feiertag, H., and Mensik, K. 2012. Social and Spatial Behavior in Shared

13 Spaces. In Proceedings of the 17th International Conference on Urban Planning and Regional

14 Development in the Information Society, Schwechat, Austria, 2012. pp. 759-767.

15 Shu, S., Batteate, C., Cole, B., Froines, J. and Zhu, Y. 2016. Air quality impacts of a CicLAvia event

16 in Downtown Los Angeles, CA. Environmental Pollution, 208: 170-176.

17 Swinburne, G. 2015. Report on Road Safety in Kensington High Street. Royal Borough of Kensington

18 and Chelsea, London.

19 Twaddle, H., and Grigoropoulos, G. 2016. Modeling the Speed, Acceleration, and Deceleration of

20 Bicyclists for Microscopic Traffic Simulation. Transportation Research Record: Journal of the

21 Transportation Research Board, 2587: 8-16.

22 Zaki, M. H., and Sayed, T. 2013. A framework for automated road-users classification using movement

23 trajectories. Transportation research part C: emerging technologies, 33: 50-73.

242526

Page 25 of 45



Draft


1 List of Figures: 23 1. FIGURE 1 The data collection site (a) world image, (b) camera Image.

4 2. FIGURE 2 Trajectory extraction process, and speed, acceleration and relative distance profiles extraction.

5 3. FIGURE 3 Logit Model for the Selection of the Collision Avoidance Mechanism in the Same-Direction

6 Interaction

7 4. FIGURE 4 Example of a typical following strategy by cyclists hindered by a slower pedestrian moving in

8 the same direction.

9 5. FIGURE 5 Example of a typical following strategy by cyclists hindered by a slower pedestrian moving in

10 the same direction expressed by the speed, acceleration and longitudinal distance profiles.

11 6. FIGURE 6 Example of a typical overtaking strategy by cyclists hindered by slower pedestrians moving in

12 the same direction.

13 7. FIGURE 7 Example of a typical overtaking strategy by cyclists hindered by a slower pedestrian moving in

14 the same direction expressed by the speed, acceleration, and longitudinal distance profiles.

15 8. FIGURE 8 Example of a typical head-on interaction between cyclists and pedestrians (opposing direction

16 interaction).

17 9. FIGURE 9 Example of a typical head-on interaction between pedestrians and cyclists expressed by the

18 speed and acceleration profiles.

19 10. FIGURE 10 Example of a typical crossing interaction between cyclists and pedestrians (case of the

20 pedestrian crosses first).

21 11. FIGURE 11 Example of a typical crossing interaction between pedestrians and cyclists expressed by the

22 speed and acceleration profiles (case of the pedestrian crosses first).

23

Page 26 of 45



Draft


1 TABLE 1 Logit Model for the Selection of the Collision Avoidance Mechanism (Same-Direction Interaction)2

Model Model Parameter value

Intercept 5.045b

(0.09)

Dist (m) 2.504a

(<0.01)

Den (road user/m2) -29.888a

(<0.01)

Goodness of fitDF 102

R2 0.801

3 Note: Values in parentheses represent the p-value. 1 m = 3.28 ft. 4 aStatistically significant (at 95% confidence level).5 bStatistically significant (at 95% confidence level).678

Page 27 of 45



Draft


1 Table 2 Features Significance across different Collision Avoidance Strategies2

Mean [SD], Collision Avoidance MechanismParameters Following Strategy Overtaking Strategy

1.28[0.65]

3.53a

[1.58](<0.01)

0.52[0.16]

0.25a

[0.10]

Max. Available Distance (m)

Shared space Density (object/m2)

(<0.01)

3 Note: Values in parentheses represent the p-value of the t-test. SD = standard deviation. 1 m = 3.28 ft. 4 aStatistically significant difference (at 95% confidence level) compared with the cell directly to the left.56

Page 28 of 45



Draft


1 TABLE 3 Features Significance for Cyclist and Pedestrians by Interaction2

Mean [SD], Bikes change speed

Bicycle-Ped Interaction Type

Road Users involved in interaction Parameter

Normal/Before interaction

During Interaction

Bikes Speed (m/s) 3.12 1.85a

[0.74] [0.45](<0.01)

Ped

Following Strategy(n=65)

Walking Speed (m/s)

1.67[0.36]

1.68[0.39]

Speed (m/s) 2.30 3.14a

[1.10] [0.63]Bikes

(<0.01)

Overtaking Strategy(n=40)

Ped Walking Speed (m/s)

1.24[0.69]

1.18[0.67]


Page 29 of 45



Draft


1 Table 4 Features Significance across Head-on Interaction Collision Avoidance Strategies2

Mean [SD], Bikes Mean [SD], PedestriansParameters Swerve Not Swerve Swerve Not Swerve

0.26[0.10]

0.40a

[0.12]0.35

[0.12]0.23a

[0.1](<0.01) (<0.01)

Shared-Space Density (object/m2)

Max. Available lateral distance for Bikes (m) 2.17[0.97]

1.07a

[0.39]1.52

[0.76]2.30a

[1.13](<0.01) (<0.01)

Max. Available lateral distance for Pedestrians (m) 2.08[1.02]

2.71a

[1.45]2.50

[1.26]1.93a

[1.04](0.01) (0.01)

3 Note: Values in parentheses represent the p-value of the t-test. SD = standard deviation. 1 m = 3.28 ft. 4 aStatistically significant difference (at 95% confidence level) compared with the cell directly to the left.

Page 30 of 45



Draft


1 TABLE 5 Features Significance across Cyclists and Pedestrians in Head-on Interaction2

Mean [SD], Bikes Swerve Mean [SD], Ped Swerve Mean [SD], Both Swerve Mean [SD], Total

Bicycle-Ped Interaction Type

Road Users involved in interaction Parameter

Normal/Before

interaction

During Interaction

Normal/Before

interaction

During Interaction

Normal/Before

interaction

During Interaction

Normal/Before

interaction

During Interaction

Bikes 3.70 2.98a 3.00 2.33a 2.86 2.26a 3.45 2.74aSpeed (m/s) [1.40] [1.07] [0.96] [0.68] [0.62] [0.71] [1.29] [0.99]

(0.01) (0.01) (0.04) (<0.01)

Ped 1.67[0.43]

1.70[0.48]

1.57[0.41]

1.57[0.43]

1.74[0.31]

1.78[0.32]

1.66[0.40]

1.68[0.44]

Interaction with Opposing flow (Head-on) (n=62)

Walking Speed (m/s)


Page 31 of 45



Draft


1 TABLE 6 Features Significance across Cyclist and Pedestrians in Crossing Interaction 2

Mean [SD], Pedestrian Cross First

Mean [SD], Bike Cross First Mean [SD], Bike Swerve Mean [SD], All

Bicycle-Ped Interaction

Type

Road Users involved in interaction

Parameter

Normal/Before Interaction

During Interaction


During Interaction


During Interaction


During Interaction

3.36 2.57b 4.61 3.65a 3.52 2.80a 4.02 3.14aBikes Speed (m/s) [1.44] [1.12] [1.60] [1.00] [1.10] [0.89] [1.63] [1.18]

(0.08) (0.03) (0.05) (0.01)

1.32[0.28]

1.67a

[0.44]1.40

[0.41]0.51a

[0.44]1.24

[0.25]0.98

[0.74]1.37

[0.35]1.07a

[0.73]

Interaction with crossing flow (n=41)

Ped Walking Speed (m/s) (0.01) (<0.01) (0.02)

3 Note: Values in parentheses represent the p-value of the t-test. SD = standard deviation. 1 m = 3.28 ft.4 aStatistically significant difference (at 95% confidence level) compared with the cell directly to the left.5 bStatistically significant difference (at 95% confidence level) compared with the cell directly to the left.

Page 32 of 45



Draft


1 Table 7 Features Significance across Crossing Interaction2

Mean [SD], Bikes Mean [SD], Pedestrians

Parameters Pedestrian Cross First Bike Cross First Pedestrian Cross First

Bike Cross First

3.36[1.44]

4.61a

[1.60]1.32

[0.28]1.41[0.4]

(0.01)

Approaching speed (m/s2)


Page 33 of 45



Draft


1 TABLE 8 Parameters to Model Cyclist-Pedestrian Microscopic Interaction Behaviour2

Interaction Evasive action

Factors affecting choice of the evasive action

Parameters needed to model

Values obtained from the current study Mean [SD]

(1) Same direction movement interaction (following strategy)

Changing Speed (cyclist)

Available lateral distance for bike

Shared space density

1. Longitudinal distance at start of reducing speed

2. Following distance3. Cyclist deceleration

profile

4.46 [1.29] m

2.36 [0.5] m0.81 [0.40] m/s2 (average value)

(2) Same direction movement interaction (overtaking strategy)

Changing movement direction (cyclist)

Changing speed (cyclist)



1. Longitudinal distance at start of overtaking maneuver

2. Lateral distance3. Cyclist acceleration

profile

3.00 [1.65] m

1.23 [0.25] m0.87 [0.35] m/s2 (average value)

(3) Opposing direction movement interaction (head-on interaction)

Changing movement direction (cyclist or pedestrian or both)

Changing speed (cyclist)



Available lateral distance for pedestrians

1. Longitudinal distance at start of interaction

2. Lateral distance3. Cyclist deceleration

profile4. Cyclist acceleration

profile

6.39 [2.31] m

1.24 [0.32] m 1.14 [0.75] m/s2 (average value)1.11 [0.56] m/s2 (average value)

5.95 [2.73] m

Cyclist yield to Ped25.7%

Ped yield to cyclist-63.5%

(4) Interaction with Crossing road users

Changing movement direction (cyclist or pedestrian or both)

Changing in cyclist speed

Changing in pedestrian walking speed

Speed of the approaching bike

1. Longitudinal distance at start of interaction

2. Percentage change in pedestrian’s speed

3. Cyclist deceleration profile

0.92 [0.42] m/s2 (average value)

1.21 [0.68] m/s2 (average value)

3

Page 34 of 45



Draft

FIGURE 1 The data collection site (a) world image, (b) camera Image.

118x58mm (300 x 300 DPI)

Page 35 of 45



Draft

FIGURE 2 Trajectory extraction process, and speed, acceleration and relative distance profiles extraction.

120x84mm (300 x 300 DPI)

Page 36 of 45



Draft

FIGURE 3 Logit Model for the Selection of the Collision Avoidance Mechanism in the Same-Direction Interaction

112x84mm (300 x 300 DPI)

Page 37 of 45



Draft

FIGURE 4 Example of a typical following strategy by cyclists hindered by a slower pedestrian moving in the same direction

331x157mm (300 x 300 DPI)

Page 38 of 45



Draft

FIGURE 5 Example of a typical following strategy by cyclists hindered by a slower pedestrian moving in the same direction expressed by the speed, acceleration and longitudinal distance profiles.

329x436mm (300 x 300 DPI)

Page 39 of 45



Draft

FIGURE 6 Example of a typical overtaking strategy by cyclists hindered by slower pedestrians moving in the same direction.

354x157mm (300 x 300 DPI)

Page 40 of 45



Draft

FIGURE 7 Example of a typical overtaking strategy by cyclists hindered by a slower pedestrian moving in the same direction expressed by the speed, acceleration, and longitudinal distance profiles.

329x437mm (300 x 300 DPI)

Page 41 of 45



Draft

FIGURE 8 Example of a typical head-on interaction between cyclists and pedestrians (opposing direction interaction).

344x154mm (300 x 300 DPI)

Page 42 of 45



Draft

FIGURE 9 Example of a typical head-on interaction between pedestrians and cyclists expressed by the speed and acceleration profiles.

329x258mm (300 x 300 DPI)

Page 43 of 45



Draft

FIGURE 10 Example of a typical crossing interaction between cyclists and pedestrians (case of the pedestrian crosses first).

397x155mm (300 x 300 DPI)

Page 44 of 45



Draft

FIGURE 11 Example of a typical crossing interaction between pedestrians and cyclists expressed by the speed and acceleration profiles (case of the pedestrian crosses first).

329x319mm (300 x 300 DPI)

Page 45 of 45



microscopic behavioural analysis of cyclists and …...draft alsaleh, hussein, sayed 4 1 the freedom...

Documents