| We propose and a stereo matching method based on OpenCL.We adopt the optimization strategy of data parallelism.We design different kernel functions for hardware conditions and various models that need to be realized,which greatly reduces the time of running the program,and on different platforms can be run smoothly.In this paper,we use the method of SLIC super-pixel segmentation,divide the image into different super-pixel plane blocks,obtain the plane label information of each pixel by using the characteristics of the super pixel plane,and use RANSAC sampling to calculate each superpixel plane equation,using the plane equation to fit the parallax graph,improve the accuracy of the parallax value,determine the super-pixel plane boundary type,define the relationship between two adjacent super pixel planes,and smooth the fitting plane.This paper uses OpenCL parallel programming language,for semi-global stereo matching and ultra-pixel segmentation algorithm has greatly improved the speed of operation,the host side has also use the OpenMP optimization,saving the running time,the use of ultra-pixel segmentation to fit parallax Figure as a means of optimization,but also improve the accuracy.In contrast to some popular matching algorithms on Kitti,the algorithm adopted in this paper has been balanced in speed and accuracy,and has achieved very good results. |
PDF Full Text Request