Stack-CNN: An End-to-end Target Relative Pose Recognition Method

With the rapid development of the aerospace industry in the world in recent years, the precise measurement of the relative position and attitude of spacecraft has become one of the hot spots in the development of space technology. Among them, the observation technology of computer vision based on optical imaging is the basis for realizing the automation of on-orbit spacecraft position and attitude measurement. It is required that the vision system can measure the relative position and attitude between the observer and the target online. In recent years, a large number of target pose measurement methods based on deep learning have emerged, which are also partially applied in the aerospace field. However, these methods generally have the problems of only recognizing a single image, large amount of calculation, and ignoring the information between images. In order to solve the problems above, we propose an end-to-end target relative pose recognition method based on CNN. This method uses the information between multiple consecutive images to realize the end-to-end recognition of relative pose and absolute pose, and does not need subsequent pose calculation. The advantages of this method, such as low computation, high precision and fast speed, ensure that this method has good application value in the intelligent application of mobile terminals.