Matching Algorithm And Its Application Based On Binocular Stereo Vision

Binocular vision as an important branch of computer vision is primarily responsible for the restoration of two-dimensional projection images into three-dimensional real scene, and it has important applications in many areas. This dissertation focuses on the key of binocular vision----stereo matching, and we proposed the improved local and global stereo matching algorithms by analyzing various existing stereo matching algorithms, and through the specific application of binocular vision----head detection method to illustrate the usability of the research content of this dissertation, as follows:1) proposed an improved local stereo matching algorithm based on adaptive weighting, this algorithm is an optimized method based on colour/distance weighting information, to adjust the weighting by using its colour/distance information as the basis of image segmentation, and the obtained initial disparity map can be further improved by using a new combinatorial optimization strategy.2) proposed an improved global stereo matching algorithm based on graph cuts, using the result of census transform as the data item of the energy function, the smooth item is calculated by the colour gradient method, and uses a simplified figure network model to structure the corresponding energy function, then using the maximum flow/minimum cut method to obtain the minimum energy function, which is the best disparity distribution.3) proposed an improved head detection method based on binocular vision under complex scene, as a practical application of stereo matching algorithm, this method uses a rapid SAD stereo matching algorithm based on disparity piecewise continuous to get the whole scene, and then identifies the heads by contour detection algorithm after extracting the depth layer contains the information of heads. The method can effectively detect the heads in a complex scene, the recognition rate reached98%.