A Stereo Correspondence Algorithm Based On Image Segmentation And Multiple Windows

Stereo vision is one of the most active subjects in computer science recent years. It iswidely and commonly used in city mapping, vehicle navigation, industrial test, objectrecognition, virtual reality and so on. Stereo match is the key stept of stereo vision, which isalso a restrict for the development and application of stereo vision technology. How to getmore accurate disparity maps with less time cost is a pursuit for researchers all of the world.The paper is aim to improve the stereo correspondence algorithms, using gray stereomaps to generate dense disparity map. The algorithm in this paper takes the basic principles ofstereo correspondence, digital image processing and computer programs into account, so italso be used in color maps if make some modification. And the algorithms are available forvariety of platforms and application situations. The improvement can be found in theoreticalcalculation and the experimental results.This paper analyzes the theories in stereo vision and stereo correspondence first, andthen summarizes the basic principles in matching, the causes may lead to errors, and thecommon methods that can improve the speed and the accuracy in matching process. Afteranalyzing local stereo matching algorithms’ performance in grey images, this paper putsforward a fast multiple windows matching algorithm and the method how to calculate a betterbase-height ratio aspect ratio in the applications.To solve the textureless area problem, we introduce a method based on imagesegmentation and multi-scale theory. This method takes the results of image segmentation asaggressive windows, and puts the whole process into image multi-scale space. Comparing tothe traditional algorithms, our method can get better results.At last, this paper also analyzes subpixel theory, lists the subpixel precision functionusing interpolation and fitting methods. After introducing a improved linear interpolationalgorithm, this paper proves effect is obvious through the theoretical calculation andexperimental. According to the image registration and digital signal processing knowledge,this paper puts forward a simpler method to generate stereo maps test data with a turedisparity map.