Research On Vision-based Pose Estimation Method For UAV Landing And Cooperative Target Optimization

The technology of vision-based UAV landing is in focus because of its practical value and development prospects. One of the research hotspots is the pose estimation method based on cooperative target. This paper mainly researched on the vision-based pose estimation method for UAV landing and the optimization method of cooperative target shape.Firstly, addressing the lack of contrast research on common pose estimation algorithms in unified experiment conditions, this paper researched on principles of common pose estimation algorithms,including P3 P algorithm, P4 P algorithm, RPnP algorithm, Tsai algorithm and orthogonal algorithm,and designed simulation experiments for comparison research on common algorithms' stability,precision and real-time. Then concluded that RPnP algorithm is suitable for vision-based UAV landing,and verified this conclusion with physical experiment.Then, addressing the problem that the stability curve of P3 P algorithm showed bimodal, this paper discussed the causes of the problem. Addressing the multiple solving methods of the initial solution of orthogonal iteration(OI) algorithm, by contrasting the OI algorithms based on different initial solutions, this paper concluded that the OI algorithm based on the solution of RPnP algorithm had better performance. Addressing the Tsai algorithm's problem that the accuracy of solving rotation matrix R was high but the accuracy of solving translation vector T was low, this paper put forward the improved Tsai algorithm. According to the proportion of pinhole imaging model, this paper improved the method of solving translation vector T with the redundant information about the distances between feature points and the focal length of the camera. The accuracy of the improved Tsai algorithm increased by 10% and the real-time increased by 50%.Finally, addressing the problem that the accuracy of pose estimation was affected by the shape of cooperative target, this paper researched on the shape optimization method of cooperative target. In one aspect, this paper researched on the shape optimization method of cooperative target based on genetic algorithm, and verified this optimization method with physical experiment. Thereinto,addressing the restrictions on the shapes in present researches, this paper put forward a new method of coding and decoding. Addressing the problem that the present evaluation index of pose estimation can't unify the dimensions and magnitudes of 6 pose parameters, and according to the evaluation method of regression equation significance in the regression analysis, this paper put forward a newevaluation index based on the formula of regression equation significance test, and constructed the fitness function of genetic algorithm with this new evaluation index. In another aspect, addressing the obvious compression along the vertical direction of the cooperative target' projection when the airborne camera descended along the small angle path, this paper researched the slope cooperative target which make a certain angle with the ground, and researched on the pose estimation accuracy affected by different angles, in order to providing the theoretical basis when using the slope cooperative target in engineering application.