A NEW STRATEGY FOR DSM GENERATION FROM HIGH RESOLUTION STEREO SATELLITE IMAGES BASED ON CONTROL NETWORK INTEREST POINT MATCHING

Image matching is the key for automatic DSM generation. Currently two kinds of image matching method represent two different trends: area-based method and feature-based method. However, both of them seriously rely on the gray distribution of the image. Therefore, they share a common drawback: ambiguity in homogeneous areas, such as grass land, heavily forested area, highway, and building roofs etc. To solve this problem, a popular way is to apply different conditions to image matching to limit the search window, so that reduce the ambiguity in the homogeneous area. In the photogrammetric community, eppipolar condition is widely used in image matching to reduce the search window from two dimensions to one dimension. Unfortunately, in many cases, the eppipolar condition is not applicable. In order to solve this problem, the control network interest point matching algorithm was recently developed. This method constructs a control network based on the prominent feature points and the spatial information is provided to the image matching to limit the search window. In this paper, we proposed a new strategy based on the control network interest point matching to generate digital surface model from high resolution satellite stereo images. We commence our paper with a brief review of current research on image matching. We then introduce the proposed algorithm in detail and describe experiments with high resolution satellite images. Through experiment, the digital surface model was successfully created from the stereo images. The experiment results show that the proposed algorithm can successfully process local distortion in high resolution satellite images and can avoid ambiguity in the homogeneous areas.