Research And Implementation Of Real-time Stereo Matching Algorithm Based On FPGA

A very important research direction in the field of computer vision is binocular stereo matching. The 21st century is an intelligent era.Binocular stereo vision system can give the robot similar to the human eye’s ability to make it easy to quickly access their own various types of information around. Environmental information obtained through computer vision has been applied to real-world scenarios such as unmanned aerial vehicles, vehicle-assisted driving, robot positioning and navigation, map reconstruction, smart home and smart city. In the practical application process, the real-time performance of this more stringent requirements. Therefore, this paper studies the real-time performance of binocular stereo matching algorithm in computer vision,and studies the existing algorithm to accelerate the program. The goal is to reduce the time consumption of stereo matching algorithm to a reasonable level for the subsequent development of SLAM Study lay the foundation.First of all, we read a large number of authoritative scholars on the binocular stereo matching algorithm research papers, according to their own understanding of the binocular stereo matching algorithm to solve the problem, and established the research objectives of this paper. And then according to the advantages and disadvantages of binocular stereo matching algorithm for a selective study, especially on the performance of real-time indicators. At the same time, the method of hardware acceleration is researched, and the advantages and disadvantages of each acceleration method are compared. Finally, FPGA is selected as the carrier of acceleration.Secondly, according to the characteristics of FPGA parallelism, the semi-global binocular stereo matching algorithm is tested on the software level, and the final binocular vision algorithm is selected. Then the algorithm is applied to the algorithm complexity calculation, function division, modular design and algorithm optimization. Finally, the binocular matching system is developed.Finally, the final output of the system image resolution is 752 X 480,transmission frame rate of 60fps, the transmission of images around three deep, the test scene: indoor, outdoor, woods, architecture, sunny, cloudy and so on. Later, the collected data sets are compared and analyzed, and the problems of the algorithm are summarized,and the algorithm is iterated. For resource use, power consumption, timing and other issues were optimized. At the same time to carry out the late SLAM research to reserve a certain amount of resources and data interface.