Research On Stereo Matching Algorithm Based On Multi-measure Fusion

Binocular stereo vision is an important part of computer vision,which realizes the acquisition of three-dimensional information from two-dimensional images by simulating the principle of human vision.Stereo matching refers to the process of finding the corresponding points of the left image in the right image in the left and right images taken by the binocular stereo vision system.It is the core link of binocular stereo vision.The accuracy and speed of stereo matching algorithm directly affect the effect of binocular stereo vision,so the research on stereo matching algorithm has become a hot topic in the field of binocular stereo vision.The research on stereo matching algorithm can be traced back to the 1980s.The visual theory framework proposed by Marr plays a key role in the study of binocular vision.The stereo matching algorithm can be divided into global stereo matching and local stereo matching according to different constraint ranges.The stereo matching algorithm mainly includes four relatively independent parts of original matching cost calculation,cost aggregation,initial disparity calculation and disparity optimization,and then on the stereo matching algorithm.Most of the research is conducted on one or several parts of the above framework.In recent years,the stereo matching algorithm for multi-measure fusion has achieved rich results,but in the choice of multiple similarity measures,relying on subjective experience to choose,lacking a specific selection method,based on this problem,a basis is proposed.A method of selecting the similarity measure of the measure complementarity coefficient.The measure complementarity coefficient is defined according to the correct matching point set of the measure.The degree of complementarity of different measures is described.Firstly,the correct rate of each measure is used alone,and the measure with the highest correct rate is selected as the initial measure,and the measure of the measure is calculated.The complementarity coefficient is selected by the measure with the largest complementarity coefficient of the measure,and then the measure weight is determined by the maximum match rate of the average correct rate on the data set.The fusion measure is used as the initial measure,and the complementary coefficient with the rest of the measure is calculated again.The maximum measure is fused,and the above steps are repeated until the matching effect is not significantly improved after the fusion measurement.After experimental verification,the measure complementarity coefficient can effectively describe the degree of complementarity between different measures.Based on this,the choice of fusion measure can efficiently select the complementary measure.The semi-global stereo matching algorithm uses the randomly initialized disparity map as the initial disparity to iteratively calculate on multiple scales until the original scale is restored,and the final disparity map is obtained,resulting in low stability and efficiency of the algorithm.Aiming at this problem,the disparity map based on SURF feature is proposed as the initial disparity map and the iterative calculation on multi-scale is cancelled to improve the performance of the algorithm.The experimental results show that the improved algorithm has better matching effect and efficiency.The similarity measure fusion based on the measure complementarity coefficient selection is used as the matching cost,and the disparity map based on the SURF feature is used as the initial disparity map for cost aggregation.After the parallax calculation and the parallax optimization step such as the left and right consistency test and the reliability test,the final disparity map is obtained.Experiment shows that the proposed algorithm can achieve better stereo matching.