Urban Road Network Extraction from Spaceborne SAR Image

Transcription

1 Progress In Electromagnetics Research Symposium 005, Hangzhou, hina, ugust Urban Road Network Extraction from Spaceborne SR Image Guangzhen ao and Ya-Qiu Jin Fudan University, hina bstract two-step method is developed for the extraction of road network from spaceborne SR image: road candidates detection and connection. In the road candidates detection, classification of fused infrared and microwave SR images effectively reduces the noise in the edge detection, and also well removes possible confusion of no-road objects, i.e. linearly-featured rivers in the edge image. Possible road candidates are further processed using the morphological thinning algorithm. Road candidates connection is carried out hierarchically according to road models we established. Finally, the main road network is established from the SR image successfully. s an example, using the ERS- SR image data, automatic detection of main road network in Shanghai Pudong area is presented. Introduction Urban road network extraction from spaceborne SR image has been one of most important applications in remote sensing technology. For example, it is greatly helpful of urban transportation mapping, planning and management and city GIS database etc. uring recent two decades, some approaches for automatic or semiautomatic detection of road from the optic or radar images have been developed [-]. However, due to some difficulties such as roads irregularity, multiplicative speckles, and complicated distribution of various objects in the urban area, these approaches do not seem to be well tractable to process the radar image, especially for distinguishing linearly featured objects, e.g. water body and roads. In this paper, a constant false alarm rate (FR) edge detector is first applied to extracting the potential roads as candidates. Then, classification of fused infrared and microwave SR images is used to reduce the noise of the edge detection with logic N fusion operation. Meanwhile, confusion of no-road objects such as linearly-featured rivers is removed. Further, the morphological thinning algorithm is employed to make the width of road candidates to be one pixel. To reduce complexity and time consuming, all roads candidates are classified into groups based on their orientations. To avoid possible loss of some useful road segments, the road candidates are linked and extended based on the thinning results and reference of original SR images. s an example, using the ERS- SR image data, automatic detection of main road network in Shanghai Pudong area is presented. Extraction of the Road andidates from a SR Image Three steps are applied to extraction of road candidates from a SR image: edge detection, speckle reduction and edge thinning. The coefficient of variation detector based on the speckle model and statistics with FR has been employed to radar image edge detection []. Its threshold evaluating the homogeneity of the image is chosen as []. To reduce the confusion of no-road objects, such as grass, flat field and water body in the edge image, the logic N fusion operation is carried out between it and the classification result of fused infrared and SR images []. If one pixel as the road candidate in the edge image is classified as grass, or flat field, or water in the classification image, it should be re-assigned as the background instead of road. nd the final road candidates are obtained by thinning the edges with morphological thinning algorithm. Linking Road andidates fter road candidates clustering based on their orientations, road linking and extension carried out hierarchically according to the characteristics of the main roads as well as their connections. To realize road candidates linking of the similar orientation, a model based on the characteristics of the single road in dense urban area and the connections between different roads is established as follows:

2 60 Progress In Electromagnetics Research Symposium 005, Hangzhou, hina, ugust -6 () The length and curvature of road candidates are within the given thresholds; () onnection takes place between two different road candidates; () The distance between road candidates satisfying () should be smaller than the given threshold; () The slope of the road formed with the endpoints satisfying () has little difference from the slopes of road candidates, which the two endpoints are belong to. haracteristic () indicates the local smoothness of the road, and () () () show the road continuity. s shown in Fig., only Point is found to satisfy all demands, and the road candidate is added and the road candidate is reserved. Endpoints of road candidates Road candidates and labels Road candidates refused Road candidates added Figure : Linking those road candidates with same orientation In order to fill the gap between different road candidates, a further linking is made based on all possible road candidates without classes considered. The linking should follow the rules as follows: (a) (b) Road candidate detected Possible road candidates Road candidate added Figure : Linking the road candidates Figure : n ERS- SR image Figure : Edge image using the coefficient of the variation edge detector

3 Progress In Electromagnetics Research Symposium 005, Hangzhou, hina, ugust -6 6 () Linking should be carried out between one detected road candidate and one possible road candidate or between two different detected road candidates; () The distance between the neighbor endpoints of the two candidates satisfying () should be limited by the given threshold; () The slope of the road added should either be close to both of the two candidates satisfying () or be close to one of them and be about 90 difference from the other, as explained in Fig.. The line, and in Figures (a, b) satisfy the above demands, respectively. Therefore, they are linked together. However, the line, and do not meet the demands and no connection is made between them. It is similar to the line, and. () The slope of the road formed with the endpoints satisfying () has little difference from the slopes of road candidates, which the two endpoints are belong to. Figure 5: The fusion result of the edge detection Figure 6: The classification of road candidates according to their orientation Figure 7: ombination of all linked road candidates after linking in different directions Figure 8: Final result of the road detection overlapped on the ERS- image fter the above works, the key clues about the location and orientation of main road network in SR image have been retrieved. Now, we superpose the result over the original SR image to make further road extension based on the amplitude of SR image. Extension is realized by linking the endpoint of current road candidate with the one in SR image satisfying the following demands: () It is located in a small range restricted by current road orientation; () Its distance from current endpoint is restricted by the threshold; () Its amplitude is close to that of the road in SR image. Then, the pixel acts as a new extension. This process is iteratively operated until no such pixel can be found.

4 6 Progress In Electromagnetics Research Symposium 005, Hangzhou, hina, ugust -6 Example n ERS- SR data (5.GHz, vv polarization, spatial resolution.5m) over the entral Park of Shanghai Pudong istrict, hina on pril 9, 00 is taken as an example. fter speckle filtering, the image of pixels is shown in Fig.. Fig. -Fig. 7 are some medium results in the road candidates detection and linking. The final result of the main road network is overlapped over the ERS- SR image, presenting a good matching between the detected roads and real roads, as shown in Fig. 8. onclusion This paper focuses on the urban road network extraction from spaceborne SR image. Hierarchical connection and extension of road candidates are developed. () The coefficient of variation detector, as a constant false alarm rate edge detector, is a good tool to detect possible road candidates in dense urban area. () lassification of fused infrared and microwave SR images effectively reduce the noise in the edge detection and removes some confusion of no-road objects such as linearly featured water body. () Hierarchical linking and extension of road candidates take account of characteristics and connections of main road network without complicated model or construction of time consuming cost function. cknowledgement This work was supported by hina State Major asic Research Project (00090) and the Optical Science Project of Shanghai, hina (06050). REFERENES. Shackelford,. K. and. H. avis, Urban Road Network Extraction from High-resolution Multispectral ata, nd GRSS/ISPRS Joint Workshop on Remote Sensing and ata Fusion over Urban reas, -6, - May 00.. Yagoub, M. M., Urban Road Network etection from Satellite Imagery, nd GRSS/ISPRS Joint Workshop on Remote Sensing and ata Fusion over Urban reas, 88-9, - May 00.. ell,. F. and P. Gamba, etection of Urban Structures in SR Images by Robust Fuzzy lustering lgorithms: the Example of Street Tracking, IEEE Transactions on Geoscience and Remote Sensing, Vol. 9, No. 0, 87-97, Oct Touzi, R.,. Lopes and P. ousquet, Statistical and Geometrical Edge etector for SR Images, IEEE Transactions on Geoscience and Remote Sensing, Vol. 6, 76-77, ovik,.., On etecting Edges in Speckle Imagery, IEEE Transactions on Signal Processing, Vol. 6, No. 0, 68-67, Tupin, F., H. Maitre, J. F. Mangin, J. M. Nicolas and E. Pechersky, etection of Linear Features in SR Images: pplication to Road Network Extraction, IEEE Transactions on. Geoscience and Remote Sensing, Vol. 6, -5, Mar Rianto, Y., Road Network etection from SPOT Satellite Image Using Hough Transform and Optimal Search, PS sia-pacific onference on ircuits and Systems, Vol., 77-80, 8- Oct urns, J.,. Hanson and E. Riseman, Extracting Straight Lines, IEEE Transactions on Pattern nalysis and Machine Intelligent, Vol. 8, No., 5-55, Wang, Ying and Qinfen Zheng, Recognition of Roads and ridges in SR Images, Record of the IEEE 995 International Radar onference, 99-0, 8- May aumgartner,., et al., utomatic Road Extraction ased on Multi Scale, Grouping, and ontext, Photogrammetric Engineering & Remote Sensing, Vol. 65, No. 7, , Jeon,. K., J. H. Jang and K. S. Hong, Road etection in Spaceborne SR Images Using a Genetic lgorithm, IEEE Transactions on Geoscience and Remote Sensing, Vol. 0, No., -9, Jan. 00.

5 Progress In Electromagnetics Research Symposium 005, Hangzhou, hina, ugust Gruen,. and H. Li, Linear Feature Extraction with LS-snakes from Multiple Images, Int. rch. Photo. Remote Sensing, Vol., 66-7, Park, J. M., W. J. Song and W.. Pearlman, Speckle Filtering of SR Images ased on daptive Windowing, IEE Proceedings Vision, Image and Signal Processing, Vol. 6, No., 9-97, ao, G. and Y. Q. Jin, Hybrid P-NN/G lgorithm for lassification of Urban Terrain Surfaces with Fused ata of Landsat ETM+ and ERS- SR, International Journal of Remote Sensing, accepted.