Algorithm Of Relative Pose Estimation For Space Target Based On Monocular-vision

The Space rendezvous and docking technology is a key technology for theimplementation of the manned space mission, vision-based measurement system has theadvantages of simple structure, stable performance, good anti-interference, and thusclose target visual measurement techniques have been a hot research which the mainobjective is to study visual measurement algorithm whose performance and stability,measurement accuracy and speed meet the actual needs. The methods of measurementthrough image collecting, image preprocessing, feature detection and extraction andpose solver.The research on visual measurement algorithm of this thesis is based on thebackground of the close rendezvous and docking which implements on semi-physicalflotation simulation platform. This paper summarizes the development status of thetarget measurement technology and research progress on the basis of design of themonocular visual pose measurement algorithm based on cooperation and non-cooper--ative targets.In the target feature detection stage, in order to improve the speed of featureextraction, we use the feature detection algorithm based on the comparison, particularlyfeature detection operator based on a random tree which can transfer feature extractiontime to the training phase and effectively improve the speed of its online run. Thisarticle also made the corresponding improvement of the algorithm to run moreefficiently;In Target position and attitude measurement phase, we design vision measurementalgorithm based on the PnP algorithm. For the cooperative target, a position andorientation measurement algorithm based on the P3P algorithm is put forward.P3P problem has multiple solutions, often using the fourth feature in practicalapplications to verify the obtained solution, so for practical application we discuss theunique solution of P3P problem. By the geometric definition of the problem andcontrolling the layout of the control point, geometric conditions for a unique solution isput forward, compared to the complex and not intuitive algebraic conditions. Throughsimulation analysis and experimental verification, we prove the validity of the algorithmwhich meets system design requirements. For the non-cooperative target, assuming thatthe target model is known, improved PnP algorithm based on an non-iterative method isput forward.With the virtual control point, PnP algorithm can be calculated in O(n) complexity.Compared to iteration-PnP algorithm, it has the advantage of rapid calculation speedand precision to meet the requirements of practical applications, meanwhile theimproved non-iterative PnP algorithm make it a better noise robustness. Using the targetdetection and extraction, simulation analysis results show the algorithm speed andaccuracy to meet the basic requirements and provide a valid reference for the next experiments.