Mobile robot visual image hierarchical matching algorithm based on deep reinforcement learning and orthogonal matching pursuit

Traditional mobile robot visual image hierarchical matching algorithms can only complete rough matching, which leads to low matching accuracy and long matching time. Therefore, this paper proposes a mobile robot visual image hierarchical matching algorithm based on deep reinforcement learning and orthogonal matching pursuit. Firstly, the strategy network and value network in the deep reinforcement learning network structure are used to guide the floating image to move to the reference image in the right direction. Secondly, the color feature rough matching is realized by designing the reward function in the process of rough matching. Finally, on the basis of rough matching, the orthogonal matching pursuit is used to extract local features of the images to be matched, and hierarchical matching of the mobile robot visual images is carried out according to the similarity. The experimental results show that the proposed algorithm can effectively realize the rough matching and fine matching of images, and has high stability in feature detection under different angles and scales, high matching accuracy and short time, and better image quality after matching, which improves the practical application effect of mobile robots.