Study Of The Area_based Stereo Matching Algorithm Based On The Census Transform

Stereo vision has always been a hot research topic in the field of computer vision.Stereo vision using two or more cameras shooting the same scene,it restors theoriginal scene of the three-dimensional information from the two-dimensional imageinformation according to the geometric principle. A stereo vision system consists offive parts, image acquisition, camera calibration, stereo matching,3D reconstructionand post-processing. Stereo matching is important and has far-reaching significancein stereo vision. Binocular stereo matching is based on the principle of human vision,it is observed from two different perspectives of the same visual target. Binocularstereo matching matches the perceived image.Binocular stereo matching is mainlyused in the field of robot navigation, parameter detection of micro operating system,three-dimensional measurement and virtual reality.By comparing the accuracy and the processing speed,the article chooses thealgorithm based on Census Transform as the stereo matching algorithm.Thisalgorithm has good robustenss,low computational complexity and it is easy toimplement in the hardware.The implementation platform in this article is FPGA.In order to improve the accuracy of the algorithm,this paper improved theCensus Transform.The neighbor pixels has to compare with both of the center pixeland the average of the neighbor pixels in the improved Census Transform.The resultsof the comparison has two binary. In costs aggregation,the method of adaptiveweights is used to further improve the matching accuracy.In cost calculation,mutualinformation with image gradient information is used as a new similaritymeasure.Compare with the original algorithm of the Hamming distance,the new onecan improve the accuracy of stereo matching,especially in the part of the objects’edges.In the part of the hardware design,this system uses the shift register of Quartus IIas a window buffer of the image.And the system achieved the modules,like thesub-pixel module,the LRC check module,the mutual information calculation module,the gradient information calculation module.In order to ensure the real-timestereo matching,this paper uses the parallel computing structure of the FPGA.Toimprove the computational efficiency of the part of the Census Transform and the partof stereo matching the pipeline technology is used.At the last,this artical test the proposed algorithm by using the stereo matchingimages of the Middlebury website and analysis the experimental results in matchingaccuracy and the processing speed.