1 registration of road network to imagery jose l. flores and jie shan geomatics engineering, school...

1

Registration of Registration of Road Network to Road Network to

ImageryImageryJose L. Flores and Jie ShanJose L. Flores and Jie Shan

Geomatics Engineering, School of Civil Geomatics Engineering, School of Civil Engineering Engineering

Purdue UniversityPurdue University

2

OutlineOutline

Problem descriptionProblem description ss

3

Problem StatementProblem Statement Accurate & current data essential to GIS applications Accurate & current data essential to GIS applications Growing availability of HR satellite imagery & aerial photos Growing availability of HR satellite imagery & aerial photos

shortens the update cycle. shortens the update cycle. This has been labor intensive w/ great resource. This has been labor intensive w/ great resource. Roads are one of the most often changing urban features in a Roads are one of the most often changing urban features in a

sustainable development society.sustainable development society. More automated process is needed. More automated process is needed.

Up to this point, must of the effort in the scientific Up to this point, must of the effort in the scientific community have concentrated in locating roads with community have concentrated in locating roads with automated approaches automated approaches

There are other authors that concentrate their efforts in the There are other authors that concentrate their efforts in the location of road intersections. (Haverkamp 2002) follows a location of road intersections. (Haverkamp 2002) follows a low-level pixel-based techniques with higher-level reasoning low-level pixel-based techniques with higher-level reasoning to extract the intersections.to extract the intersections.

One slide with figures to identify the problem. One slide with figures to identify the problem.

4

Our ApproachOur Approach Our approach uses the existing road networks, Our approach uses the existing road networks,

their geometry, and relative location/topology.their geometry, and relative location/topology. However, it is not assumed that this However, it is not assumed that this

information is complete, accurate; rather it is information is complete, accurate; rather it is accurate enough to concentrate the extraction accurate enough to concentrate the extraction process in confined segments of the image process in confined segments of the image located around intersection points in the road located around intersection points in the road networks. networks.

The located intersection points in the image The located intersection points in the image will guide the registration or re-registration will guide the registration or re-registration process of the road network with reference to process of the road network with reference to the image. the image.

5

Methodology Methodology Locate intersections on the vector road layersLocate intersections on the vector road layers Preprocess image with common filters, like image smooth with Preprocess image with common filters, like image smooth with

edges enhance.edges enhance. Use canny edge detector to find the road edges.Use canny edge detector to find the road edges. Use the Hough Transform (HT) to parameterize the edges into Use the Hough Transform (HT) to parameterize the edges into

lines.lines. Use only the lines that fit a two lane road pattern, and calculate Use only the lines that fit a two lane road pattern, and calculate

the center line of the road.the center line of the road. Create a point representing an intersection where two center lines Create a point representing an intersection where two center lines

meet.meet. Compare these intersection points with the vector layer Compare these intersection points with the vector layer

intersection points to find the appropriate transformation intersection points to find the appropriate transformation parameters (4) that give the highest yield of matched points. parameters (4) that give the highest yield of matched points.

Using the matched points on both points sets, find using least Using the matched points on both points sets, find using least squares, higher order transformation parameters (6 – 8).squares, higher order transformation parameters (6 – 8).

6

Locate Intersections from Locate Intersections from Road NetworkRoad Network

Find all vertices on a line (there is duplicity in the points generated)

L1

L2

L3 L4L5

L6

1 2,3,11,12 4

5

6 7,8,10 9

13

Find all nodes in the intersections using all vertices within 1m of each other as criteria.

1

2

L714

15

Find the intersecting points between specified lines that are at a distance less than 1m of point.

1

2

Select only the points that are within extent of the intersected line and 1m of the intersecting point.

3

2

7

Image PreprocessingImage Preprocessing In order to minimize false road edges in a In order to minimize false road edges in a

image caused by shadows, vegetation, texture image caused by shadows, vegetation, texture within an area the image should be within an area the image should be preprocessed prior to using the Canny edge preprocessed prior to using the Canny edge detector.detector.

Some recommended processing may include:Some recommended processing may include: Eliminating preselected areas that may represent Eliminating preselected areas that may represent

vegetation. This is possible using multispectral vegetation. This is possible using multispectral images.images.

Use of common image processing filters, like and Use of common image processing filters, like and even changing image contrast.even changing image contrast.

What you did ???What you did ??? image smoothing, edge enhancersimage smoothing, edge enhancers Select the pixels whose pixels beyond certain rangeSelect the pixels whose pixels beyond certain range

8

Effect of PreprocessingEffect of Preprocessing

Canny edge without image preprocessingCanny edge with image preprocessing

9

Edge detection from Edge detection from imageimage

Edge detection by Edge detection by Canny edgeCanny edge

Canny filter was Canny filter was preferred because it preferred because it is robust to noise, is robust to noise, and more likely to and more likely to detect true weak detect true weak edges.edges.

This step does not This step does not provide parameters provide parameters to define lines, so to define lines, so another step will another step will needed.needed.

10

The Moving Box for HTThe Moving Box for HT

Then move to next row still with 50% overlap

First move box along row with 50% overlap

Blue – from imageGreen – from vector

11

Facts of the Moving Box Facts of the Moving Box for HTfor HT

Using the moving box methods guaranties Using the moving box methods guaranties searching all possible areas where there could an searching all possible areas where there could an intersection.intersection.

The size of the box is 100x100m. (1 m per pixel)The size of the box is 100x100m. (1 m per pixel) Moving the box with a 50% overlap, allows for the Moving the box with a 50% overlap, allows for the

box to position itself to accommodate over an box to position itself to accommodate over an intersection without cropping an incoming road intersection without cropping an incoming road from either side of the mention intersection. – not from either side of the mention intersection. – not missing possible road across at box edgemissing possible road across at box edge

However, it may yield duplicates for one However, it may yield duplicates for one intersection – increase TP; in the meantime, FPintersection – increase TP; in the meantime, FP

This has little consequence given that one of the This has little consequence given that one of the overlapping points will be ignored when the overlapping points will be ignored when the relative positions with other found intersections is relative positions with other found intersections is compared with the vector intersection points.compared with the vector intersection points.

12

Hough Transform (HT)Hough Transform (HT) Using the Hough transform Using the Hough transform

to search for edges that to search for edges that represent the border of a represent the border of a road, is more effective if road, is more effective if made locally.made locally.

HT prioritizes the edges HT prioritizes the edges from the longest set of from the longest set of pixels that may represent a pixels that may represent a line to the shortest.line to the shortest.

So it will provide with most So it will provide with most likely set of lines that may likely set of lines that may represent an road end.represent an road end.

However it will not However it will not discriminate between discriminate between different cultural object like different cultural object like road, parking lots or road, parking lots or buildings or even natural buildings or even natural objects like tree lines. objects like tree lines. Within One box 100x100

13

Finding Intersection Finding Intersection from HT linesfrom HT lines

Using the HT lines to Using the HT lines to find pairs that best find pairs that best represent a two lane represent a two lane road (~7meters).road (~7meters).

Then calculate the Then calculate the center lines for each pair center lines for each pair (dashed lines). (dashed lines).

Find each intersection Find each intersection between two center lines between two center lines within the moving box.within the moving box.

On the image on the On the image on the right, the edges of a right, the edges of a structure had the same structure had the same dimension of a road and dimension of a road and this provided a false this provided a false intersection point.intersection point.

14

Pair Up SetsPair Up Sets The procedure to pair up the The procedure to pair up the

vector point with the raster vector point with the raster points is simple.points is simple.

It will pick the first point It will pick the first point select the second and select the second and compared distances from compared distances from sets. If distance match, it sets. If distance match, it determine a set of four determine a set of four parameters for parameters for transformation.transformation.

Then it will check how many Then it will check how many points match, then repeat the points match, then repeat the process, and keep the set of process, and keep the set of parameters that provide the parameters that provide the most match ups. most match ups.

For instance, it will take 1-2 For instance, it will take 1-2 red, compare to 1-5 blue, no red, compare to 1-5 blue, no match. The go with 1-2 blue, match. The go with 1-2 blue, which will match, and get the which will match, and get the parameters. parameters.

1

3

2

4

3

1

5

2 4

No

YesNo

15

Pairing Raster & Vector Pairing Raster & Vector PointsPoints

Green points represent the vector points after matching up with the raster points (blue points) using a four parameter transformation.

16

Registering the Road Registering the Road LayerLayer

Once the higher order parameters Once the higher order parameters are obtained these parameters are are obtained these parameters are used to transform the nodes and used to transform the nodes and each vertices of the lines in the road each vertices of the lines in the road layer.layer.

This will finish the process of This will finish the process of registering or re-registering the registering or re-registering the road layer to the georeferenced road layer to the georeferenced image.image.

17

Roads OverlaidRoads Overlaid

Original road Original road layer.layer.

Roads after Roads after matching with matching with raster points, raster points, using a 4 using a 4 parameter parameter transformation.transformation.

Roads after Roads after applying a applying a higher higher transformation transformation using the using the matched points matched points of the previous of the previous points.points.

18

Closer look at Closer look at IntersectionsIntersections

Original road layer.Original road layer. Roads after matching with raster points, using a Roads after matching with raster points, using a

4 parameter transformation.4 parameter transformation. Roads after applying a higher transformation Roads after applying a higher transformation

using the matched points of the previous points.using the matched points of the previous points.

19





No raster points in this area (SE)

20





21

Parameters UsedParameters Used Image is a 1 meter resolution panchromatic.Image is a 1 meter resolution panchromatic. The HT algorithm connected pixels into lines with The HT algorithm connected pixels into lines with

a minimum of 7 pixels, and allowed gaps of 2 a minimum of 7 pixels, and allowed gaps of 2 pixels.pixels.

The maximum number of lines that the HT The maximum number of lines that the HT algorithm detected was 20.algorithm detected was 20.

Lines to be considered parallel have to be within Lines to be considered parallel have to be within 5 degrees from each other.5 degrees from each other.

For a pair of parallel lines be considered as road For a pair of parallel lines be considered as road edges they have to be 8(±3) meters apart. This is edges they have to be 8(±3) meters apart. This is for a 2 lane rural road.for a 2 lane rural road.

To consider to points a match they should be lest To consider to points a match they should be lest that 20m apart.that 20m apart.

22

ReferencesReferences Canny, J. (1986). "A Computational Approach To Edge Detection." Canny, J. (1986). "A Computational Approach To Edge Detection." IEEE IEEE

Transaction Pattern Analysis and Machine IntelligenceTransaction Pattern Analysis and Machine Intelligence 88: 679-714.: 679-714. Fischler, M. A. and R. C. Bolles (1981). "Random sample consensus: a Fischler, M. A. and R. C. Bolles (1981). "Random sample consensus: a

paradigm for model fitting with applications to image analysis and paradigm for model fitting with applications to image analysis and automated cartography." automated cartography." Communications of the ACMCommunications of the ACM 2424(6): 381-395.(6): 381-395.

Gautama, S. and W. Goeman (2004). Gautama, S. and W. Goeman (2004). Robust Detection of Road Junctions Robust Detection of Road Junctions in VHR Images Using an Improved Ridge Detectorin VHR Images Using an Improved Ridge Detector. Proceedings ISPRS . Proceedings ISPRS XXth Congress, Istanbul.XXth Congress, Istanbul.

Habib, A. F., R. I. Al-Ruzouq, et al. (2004). Habib, A. F., R. I. Al-Ruzouq, et al. (2004). SEMI-AUTOMATIC SEMI-AUTOMATIC REGISTRATION AND CHANGE DETECTION USING MULTISOURCE REGISTRATION AND CHANGE DETECTION USING MULTISOURCE IMAGERY WITH VARYING GEOMETRIC AND RADIOMETRIC IMAGERY WITH VARYING GEOMETRIC AND RADIOMETRIC PROPERTIESPROPERTIES. XXth ISPRS Congress, Istanbul, International Archives of . XXth ISPRS Congress, Istanbul, International Archives of Photogrammetry and Remote Sensing.Photogrammetry and Remote Sensing.

Haverkamp, D. (2002). "Extracting Straight Road Structure in Urban Haverkamp, D. (2002). "Extracting Straight Road Structure in Urban Environments using IKONOS Satellite Imagery." Environments using IKONOS Satellite Imagery." Societyof Photo-Optical Societyof Photo-Optical Instrumentation EngineersInstrumentation Engineers 4141(9): 2107-2110.(9): 2107-2110.

Mena, J. B. (2003). "State of the art on automatic road extraction for GIS Mena, J. B. (2003). "State of the art on automatic road extraction for GIS update: a novel classification." update: a novel classification." Pattern Recognition LettersPattern Recognition Letters 2424(16): 3037-(16): 3037-3058.3058.

Tóth, Z. and A. Barsi (2005). Tóth, Z. and A. Barsi (2005). Analyzing Road Junctions by Geometric Analyzing Road Junctions by Geometric TransformationsTransformations. High-Resolution Earth Imaging for Geospatial . High-Resolution Earth Imaging for Geospatial Information, Hannover, ISPRS.Information, Hannover, ISPRS.

1 registration of road network to imagery jose l. flores and jie shan geomatics engineering, school...

Documents