The Research On Visual Positioning Technology Based On Machine Vision

As an important research direction of machine vision,vision positioning technology can play an important role in unmanned driving,Robot Perception and behavior control,indoor positioning and other fields by collecting visual information to track and predict dynamic targets,reconstruct environmental models and plan motion trajectories.The development of visual positioning has limited its use in embedded systems due to the complexity of its operation.This paper proposes a specific scheme of hardware acceleration of visual positioning algorithm using FPGA chip.Tests show that this scheme can greatly improve the efficiency of operation and meet the requirements of real-time operation in embedded systems.Firstly,various kinds of visual localization algorithms are studied and compared,and the frame and flow of the related algorithms are determined.The flow of the algorithm is to first distort the input image,then extract the feature points of the image,then track the feature points by using optical flow algorithm,calculate the rotation and translation matrix of the camera according to the change of the position of the feature points,and finally update the moving track of the camera.Trace.Secondly,the paper introduces the operation framework of the whole visual positioning system,which uses the ZCU102 development board integrated with ARM and FPGA as the hardware platform and the SDSOC system compiler as the development tool.ARM processor is responsible for the whole program flow,including the input and output of video stream,information exchange between the system and the FPGA,and the FPGA is responsible for image processing and feedback calculation results.The whole system runs based on the Linux operating system.Thirdly,the paper introduces the whole operation framework of the visual localization algorithm accelerated by the FPGA,and gives the operation logic of some algorithms,including the establishment of image pyramid,feature point detection and optical flow algorithm.At the same time,it also gives the optimization method to improve the computing speed of the underlying hardware resources.At last,the KITTI data set is used to test the algorithm.The results show that the acceleration method based on FPGA can achieve the desired acceleration effect on the premise of guaranteeing the accuracy.It provides a train of thought for the realization of real-time visual positioning in embedded system.