DM6446 -based non- cooperative target Pose Estimation Algorithm

With the development of information science and technology, computer vision in more and more industries are widely used. Computer vision can be divided according to the target species and non-cooperative targets vision measurement cooperative target vision measurement, measurement of non-cooperative target visual field in recent years become a research hotspot. This work focuses on two main research areas: one is given two non-cooperative target pose more practical solver algorithm, these two algorithms through analysis and simulation experiments, validation is feasible and efficient than existing There are way high. The other is in the design of experimental platform to achieve the above pose a solver algorithms. For non-cooperative target monocular measurement gives two algorithms. One is to improve monocular visual similarity iterative algorithm, improvement in the final pose calculation, the predecessors through the singular value decomposition method for solving, this article is by building a new coordinate system to solve, this approach is relatively simple , the programming will be a lot easier. The idea of ??this algorithm is the relationship between the 2D-3D conversion of the relationship between the 3D-3D, and through solving the depth of field and the position and orientation solution absolute azimuth. Another is based on the linear characteristic monocular vision pose solver, the algorithm also in previous studies done on the basis of the corresponding improvement to take the initial value is given by solving analytical solution, the final pose is calculated Newton iterative method is used, the simulation experiments verify this algorithm is correct, and accuracy of the results in the range of allowable error. Experimental design of a hardware platform, the algorithm used to verify the validity and rationality. For the actual application requirements, the first choice of the hardware to do a feasibility analysis, taking into account future software upgrades, hardware choice can also stay in a certain amount of leeway. Entire hardware system, including the target body, CCD camera, DM6446 processor, LCD liquid crystal display, PC, is easy to detect, but also has representative body designed as a rectangular target; taking into account the large amount of image data processing, you need high performance processor, so I chose TI's DM6446 processor, this chip performance is very high, is designed to do image processing. Implemented in the experimental platform in front of the algorithm. Respectively, in image filtering, edge detection, Hough transform, pose calculation to do the experiment, from repeated experiments, summed up the image pre-processing is very important, this part of the direct impact on the accuracy of calculation results pose behind; there is to write the program to consider the running time, try to optimize the process, make the time as short as possible. Experimental results show that: In the DM6446 experimental platform to achieve non-cooperative target position and attitude measurement, to achieve the desired results, to meet actual needs.