Design Of Binocular Stereo Vision System Based On ZYNQ

In recent years,computer vision technology has been developed rapidly and its theory and applications have been greatly enriched.Binocular vision,as a major branch of computer vision,has important research significance.Binocular vision is based on the principle of parallax,using imaging devices to capture images of the same scene from different locations,and using the position deviation of the same point in two images to obtain the depth information of the current scene.Compared with active depth measurement such as structured light,binocular vision has a simpler structure and a wider range of application scenarios.The challenges facing binocular vision: 1.As a real-time system,binocular stereo vision is computationally intensive,in which the binocular matching algorithm has a high time complexity and requires more hardware resources to implement the system.2.Binocular vision systems are mostly used for depth measurement on mobile platforms,which have extremely demanding requirements for the power consumption of their mounted systems.This paper addresses these challenges.This paper addresses these challenges by designing a real-time,low-power binocular vision system based on ZYNQ,placing a large number of parallelizable logic and basic operations in PL,while placing complex floating-point operations in PS,giving full play to the logic resources of all parts of ZYNQ,improving the operational efficiency of the whole system,and finally realizing image acquisition,binocular correction,binocular stereo matching and parallax map output.The system implements binocular polar correction and aberration correction at the algorithm level,uses Census transform and Hamming distance to obtain the matching cost,optimizes the aggregation path of the semi-global matching algorithm for the characteristics of ZYNQ,aggregates the matching cost,eliminates the false matching points by left-right consistency check,and achieves a resolution of 640×480 and a frame rate of 60 fps with a power consumption of 2.176 W after median filtering.480 resolution,60 fps frame rate and 64 parallax map output.After comparison,the difference between the parallax map calculated by this system and the PC with the same parameters is only 0.5 pixel points,and the deviation of the optimized aggregated path from the standard path is less than 30%.