Research On Active Binocular Matching Algorithm Based On FPGA

With the continuous development of computer vision and artificial intelligence technology,3D vision technology has gradually become a hotspot in the field of computer vision.How to provide accurate and reliable depth information is a core issue of 3D vision technology.3D vision technology is wildly used in mobile robot navigation,object recognition,pedestrian tracking and other research field.In addition to high precision requirements for depth information,these application field also need real-time algorithm.This paper takes the active binocular vision as the research topic,aim to improve the robustness and accuracy of depth information output by depth camera,and focuses on analysis and resear-ch of clustering algorithm,active binocular matching algorithm and FPGA(Field Programmable Gate Array)design of the algorithm.The main work of this paper is as follows:1.Research on clustering algorithm.The clustering algorithm in the machine learning is used to classify each pixel in binocular images,extract features and map them into binarization feature vectors.According of the specific application of binocular matching,suitable framework is selected,clustering algorithm flow is designed,and algorithm parameters are optimized,correspondence clustering effect evaluation method is designed.2.Research on active binocular matching algorithm.According to the characteristics of the active binocular matching algorithm,select local matching algorithm as the foundation,using the disparity propagation algorithm improve the effect of algorithm in the less structured light texture(image edge,the matching of distant obj ects),improve traditional subpixel interpolation using bilinear interpolation method,reduce the uneven distribution of three-dimensional point cloud.3.FPGA design and implementation of matching algorithm.Active binocular matching algorithm has a large amount of computation,and the general computing platform cannot meet the requirement of real-time output depth information.FPGA has unique parallel computing ability,which can accelerate the algorithm.According to the characteristics of FPGA,the matching algorithm is redesigned to realize real-time output on embedded platform.4.Design and experimental study of matching algorithm accuracy evaluation scheme.Design and build an active binocular camera image acquisition device to collect data sets.An effective evaluation method for the accuracy of active binocular matching algorithm is designed,and the effects and precision of several matching algorithms are evaluated and compared.