Research On Space Target Manipulate System Relative Navigation Algorithm Based On Monocular Vision

Due to the increasing frequency of human space activities,a series of space debris are brought along,which has a great threat to astronauts and space,seriously affecting the development of human space.The development of space debris removal technology is of great significance to the operation environment,communication services and national defense security of China's satellites.Laser debris removal technology based on monocular vision is a clean,safe and effective active debris removal technology,especially for small target debris removal,which is often a large number of debris,and the orbit is random.At present,this technology mainly uses a monocular camera to capture the image sequence of the target,and the navigation computer to calculate the position and attitude of the target,track the target and clear the target.In this design,the visual system of the space target control system adopts FPGA + DSP architecture,uses FPGA to realize the parallel processing of the image,extracts the image features,and DSP carries out the direct output results of pose solution,so as to realize the parallel working mode of FPGA and DSP,and improves the performance of the monocular visual system.This paper mainly studies the vision system in the space target control system,including the design of the hardware circuit of the vision system,the FPGA image preprocessing and algorithm design to improve the visual data update rate,the monocular vision image processing technology and the space target relative navigation algorithm.The specific research work is as follows:The hardware structure of vision system is designed.On this basis,the method to improve the data update rate is obtained by comparing the working time of traditional serial working mode and parallel pipelining working mechanism.At present,most of the visual processing systems are serial mode,that is,image exposure,image storage,image processing,feature extraction and pose calculation,which obviously can not meet the data update rate.Compared with the traditional serial mode,the parallel pipeline can synchronize the image exposure,image storage and image processing,so as to design a three-level parallel processing mode and improve the real-time performance of the vision system.According to the pinhole imaging principle,the projection model from three-dimensional space to two-dimensional plane is established.Considering the distortion of the optical lens in the process of processing,the internal parameters of the vision system are obtained by Zhang's calibration method.Through the design of the target surface pattern and feature points,the stability and accuracy of the positioning of the spatial target feature corner points and convex polygon feature points are compared.The design algorithm processes the random irregular feature points,and uses the PNP algorithm based on the coplanar feature points of a single image to solve the pose.Through the research on the solution of the classical PNP problem of vision,the direct linear transformation method and EPNP method in the closed solution of the PNP problem,the post algorithm and L-M algorithm in the iterative solution are studied.Finally,L-M method is used for iterative optimization,and the pose results are output through data communication.In order to further verify the credibility and reliability of the design method in this paper,the hardware structure of the visual system based on FPGA + DSP is designed,and the image exposure,image storage and image processing parallel computing are run in FPGA.Because the L-M method of PNP problem involves floating-point operation,the L-M method is run in DSP.Finally,the visual system based on FPGA + DSP is adopted ground system The system is verified.