Stereo Vision System Based On Zynq-7000 SoC

With the rapid development of computer science and technology and maturing of the computer vision theory,computer vision has been rapid development in the past three decades,the stereo vision as an important branch of computer vision,also be more and more applied in various fields.Stereo vision processes the images that captured by a number of different perspectives of camera in the same scene to obtain the depth information of the scene,and then reconstruct the three-dimensional structure information of the scene.In aspects of planet cars,autonomous vehicles,unmanned aerial vehicles,robots,positioning and navigation,distance measurement,industrial applications and so on,stereo vision have widely application.Because the stereo vision algorithms are particularly complex,in order to obtain faster processing speed,most stereo vision systems are implemented on FPGA or the architecture of FPGA + DSP.In this paper, using Zedboard board that contains Xilinx’s the new generation of 28 nm Zynq-7000 SoC(System on Chip) chip designed a set of stereo vision system. This chip adopts the architecture of FPGA+ARM, and integrates two Cortex A9 processor and a lot of Artix-7 FPGA resources, this architecture has higher performance, lower power consumption and greater design flexibility.In this paper,image correction and the average image reduced algorithms are implemented in the FPGA part of the Zynq-7000 SoC,and the stereo matching is implemented in the ARM part of the Zynq-7000 Soc by using the OpenCV video library.In the course of image correction,for traditional correction methods,first,needs to off-line calculate the correction parameters and the index coordinate and then store them in an external Flash memory;Secondly,in order to solve the problem of slow read speed,the system needs to load correction parameters and index coordinate to the SDRAM during the period of system startup;Finally,the correction module reads the correction parameters and index coordinate in the SDRAM to complete the process of image correction.In this paper,an improved image correction was adopted,this algorithm does not need to calculate the correction parameters and index coordinate beforehand,but calculate the correction parameters and index coordinate in the image correction module in real time,improve the efficiency of image correction algorithm.At the time of image reduction,using the average reduced algorithm reduce the image in real time from an resolution of 1080?1920 to an resolution of 270?960,the purpose is to reduce the amount of computation in the stereo matching algorithm.In the course of stereo matching,this paper uses the watershed algorithm to extract image feature points, then using the Block Matching(regional matching)stereo matching algorithm to obtain the disparity map and calculate the correction parameter and feedback to the image correction module based on the match results.The experimental results showed that this implementation has achieved satisfactory results.