FPGA Based Binocular Image Acquisition System And Stereo Matching

Vision recognition is always a very important sensing method for human beings to know the world.Although researchers have done a lot of research from different perspective,but nowadays it is still a state of art research topic.Vision system can acquire the 2 dimension information of environment.It can get 3 dimension information if use two different vision sensors.After getting these information,it can be used on measuring the position and size of the objects,also can be used on recognizing and recording the surrounding senses.Combining current research trend and industrial application trend,this thesis chooses to research the binocular vision acquisition system and stereo matching algorithm on FPGA platform.The image sensor capturing method with ZYNQ7020 as the core is studied.The software program for image sensor configuration and video data transfer is developed.The vision system is calibrated by chessboard and MATLAB toolbox.The distorted image is rectified by bi-linear interpolation method.Then the video data is forwarded to next step matching algorithm.The framework of stereo matching algorithm is investigated.One sum of absolute difference cost calculation method and one census transform cost calculation method are studied.Comparing our method with other researcher's method,we use one linear weight fusion method to get a better result.For cost aggregation,SGM method is chosen as a base path.Based on the predefined cost calculation,one parameter selection strategy is proposed for cost aggregation penalty.We use guided filter to fine tune the cost after aggregation.Comparing the disparity maps generated from different algorithm steps,the output results gradually rise up.The average matching accuracy is higher than 80% with our algorithm framework.In this thesis,we also research the algorithm implemented on FPGA hardware platform.After the optimization,the design running time on FPGA is 60 times less than non-optimized design.The results from personal computer and FPGA platform are also compared.The consistency is close to 90%.