A Robust Method for Semi-Automatic Extraction of Road Centerlines Using a Piecewise Parabolic Model and Least Square Template Matching

In this paper, we present a semi-automatic road extraction method based on a piecewise parabola model with 0-order continuity. The piecewise parabola model is constructed by seed points coarsely placed by a human operator. In this case, road extraction actually becomes a physical problem of solving of each piece of parabola with only two or three unknown parameters by using image constraints. We have used a least square template matching to solve the parabola parameters. The template is deformable developed based on the automatic detection of dual road edges. In addition, a method of flexible observation weight evaluation has also been developed in this matching method. Extensive testing experiments on various image sets demonstrate that the method is able to extract road centerlines reliably. It offers much higher efficiency in contrast to manual digitizing process. We also discuss some issues about semiautomatic road extraction and future work for improving the reliability and extending the availability of our method. Introduction In the past decades, numerous methods in road extraction from spatial imagery (aerial and satellite) have been presented. There are many significant advances in the development of methods and algorithms for road extraction. Due to the complexity of the automation process, the human operator still plays a principle role in extracting road information from imagery. The key issue in automatic object extraction is reliable object identification. In resent years the human machine cooperation strategy for object extraction has been an active research area. In this strategy, the object of interest is first identified by a human operator and some seed points are often provided, the object is then delineated automatically and accurately by the computer algorithm. This is also called semiautomatic extraction. The bottle-neck problem in identifying the object is thus eliminated, and the given approximations serve as strong constraints to control the extraction process, offering more reliable and accurate extraction result. For the semi-automatic road extraction, there are basically two ways to provide the approximations. One is to give an initial point and an approximate direction for the subsequent automatic extraction. In Nevatia and Babu (1980), edge analysis A Robust Method for Semi-Automatic Extraction of Road Centerlines Using a Piecewise Parabolic Model and Least Square Template Matching Xiangyun Hu, Zuxun Zhang, and C. Vincent Tao is a key clue for road finding from some given seeds. The profile matching combined with the Kalman filtering algorithm is presented in Vosselman and de Knecht (1995). A road tracking system developed by Mckeown and Denlinger (1988) uses both edge clues and the profile matching techniques as a cooperative strategy for road extraction from aerial imagery. Tao (2000) used a multiple-image matching strategy for object measurement from mobile mapping image sequences of road corridors. In his method, an object point in one image is measured manually as the initial information. Couloigner and Ranchin (2000) presented a semi-automatic method of street extraction from the urban area. It makes use of a waveletbased, multi-scale representation of the image. When a set of seed points are input as the approximation, the least square template matching method (Agouris, et al., 2000) and the differential snakes (Agouris, et al., 2001) are used for road extraction. From a viewpoint of practical operations, sequentially inputting the seed points for road extraction can be integrated with the manual digitizing process naturally. Gruen and Li (1995) presented a model-driven method which uses dynamic programming for road extraction. They also developed a method that is based on the snakes and the least square template matching (LSB-SNAKES). This method can be used to perform the high precision linear feature (for example, the road centerline) tracking in two-dimensional image space or threedimensional object space (Gruen and Li, 1997; Li, 1997). In this paper, we present a concise geometric representation of a road centerline, piecewise parabola with 0-order continuity, for semi-automatic road extraction. The objective of the method is to develop a robust and user friendly tool that can be integrated into a digitizing production with high reliability, efficiency, and accuracy. Overall Strategy of Semi-Automatic Road Centerlines Extraction The semi-automatic method attempts to integrate the intelligence of our human visual system with an ability to recognize the object robustly and the computer system with an ability to perform fast feature extraction and accurate shape representation. To ensure that the semi-automatic road extraction can be applied in a real operational environment, the method needs to guarantee better performance in terms of: • Reliability: Extraction is not sensitive to the noise (shadows and occlusions) and the slight variation of position of the input approximations, i.e., re-doing or correcting (editing) of the extracted result should be largely minimized. P H OTO G R A M M E T R I C E N G I N E E R I N G & R E M OT E S E N S I N G December 2004 1 3 9 3 Xiangyun Hu and C. Vincent Tao are with the Department of Earth and Atmospheric Science, York University, 4700 Keele Street, Toronto, ON, Canada M3J 1P3 ([email protected] and [email protected]). Zuxun Zhang is with the School of Remote Sensing and Information Engineering, Wuhan University, Luoyu Road 129#, Wuhan, China, 430079 ([email protected]). Photogrammetric Engineering & Remote Sensing Vol. 70, No. 12, December 2004, pp. 1393–1398. 0099-1112/04/7012–1393/$3.00/0 © 2004 American Society for Photogrammetry and Remote Sensing LFX-531.qxd 11/9/04 16:11 Page 1393