Tracking and positioning of industrial objects using CAD models

A 3D target tracking and localization algorithm based on monocular vision configuration is studied to provide high precision positional information for the construction of virtualreal fusion scenes for the needs of complex aerospace products such as auxiliary assembly and quality inspection. The 3D model is rendered in real time using OpenGL, and the corresponding projection view is generated using virtual space imaging technology to extract the edge features of the virtual image; meanwhile, the video of the actual assembly scene is captured in key frames to extract the point, line and contour features of the key frame image; the best viewpoint evaluation function is constructed to recommend the best camera pose for the initial alignment of the model; the contour feature matching method is used to A non-linear optimization algorithm is used to improve the stability of camera pose estimation. The experimental results show that the algorithm can accurately track and locate the 3D target with the accuracy within 0.5mm in the plane position, 10mm in the depth direction, and within 0.5° in the pose accuracy control.