Abstract: The existing vision-based dynamic parameter measurement methods of satellite flexible solar array often rely on the deployment of cooperative markers on the surface of solar array. Such methods affect the dynamic parameter characteristics of solar array and are difficult to operate on orbit. In order to realize the non-contact measurement of solar array motion information without identification points, a dual-branch exposure recovery and denoising network based on U-net structure is used to improve the image quality of spaceborne camera and optimize the image exposure problem. Then, a high-precision measurement method based on visual model is proposed. Based on the CoTracker algorithm, position coding and Gaussian downsampling with Laplace residual compensation are introduced to improve the tracking and measurement accuracy of the visual model in the case of extremely dark light or occlusion interference. Finally, based on the experimental and rendering image data sets, it is verified that the visual model can achieve sub-millimeter-level target feature point position measurement accuracy.

Key words: flexible solar array, binocular vision, point tracking, image processing, non-contact measurement