Chinese Journal of Computers   Full Text
Title	Road Extraction Algorithm for Aerial Images Based on Total Variations and Mathematical Morphology
Authors	LI Shu-Xiao CHANG Hong-Xing
Address	(Integrated Information System Research Center, Institute of Automation, Chinese Academy of Sciences, Beijing 100080)
Year	2007
Issue	No.12(2173—2180)
Abstract &   Background	Abstract Roads in high resolution aerial images appear to be narrow areas and this creates an opportunity for classification based methods. A new approach based on classification to road extraction for aerial images is proposed in this paper. The method is based on total variation and mathematical morphology analysis. This approach firstly classifies the image into road and non-road pixels by appropriate thresholds based on neighbor total variations and histogram analysis, and then uses a criterion based on connected area total variations, geometric attributes and its pattern spectrum to find an appropriate threshold for morphological trivial opening. Finally, morphological trivial opening is adopted to avoid noises including objects that have similar spectral characteristics to road surfaces. Strict experiments show that this algorithm is robust and is capable of coping with partial occlusion and extracting roads with different spectral characteristics in the same image. keywords road detection； total variations； morphological trivial opening； variation geometry criterion； pattern spectrum background The problem in object detection with moving camera has been far beyond resolved because of the complexity of itself. There has been no object detection system which has the high adaptability that could fit for the general situation in this case till now. We usually have to solve these problems according to different specific problems. Pilotless aircrafts that have their own vision can have extensive potential applications such as traffic surveillance from aircraft videos, frontier patrolling and emergency managing and so on. And this stimulates the work in this paper. The main technologies on computer vision involved in this project are road detection and tracking, vehicle detection and tracking, man-made object recognition, singularity detection and vision matching etc. The authors and other members in this project have made a great deal of work on object tracking, man-made object recognition and vision matching. This paper mainly focuses on the research of the road detection techniques for high resolution aerial images.