An Improved Image Enhancement Method for Underwater Robot SLAM

With the development of technology, SLAM technology has been widely used in AUV trajectory tracking, where precise trajectory is the basis for AUV to perform underwater tasks such as 3D underwater environment reconstruction and navigation. Visual SLAM extracts image feature points to achieve robot trajectory tracking. However, due to the poor quality of underwater images, the effect of visual SLAM in underwater environments is not ideal. In this context, an improved image enhancement method for underwater robot SLAM is proposed. The method introduces a grayscale image enhancement algorithm to improve the quality of underwater grayscale images. The high-quality images enable the robot to extract more effective feature points for pose estimation. Considering the difference in object features between underwater and ground environments, an ORB vocabulary suitable for underwater environments is trained for ORB-SLAM3 loop closure, which associates the current data with all underwater data to perform relocalization and provides more effective data for backend optimization to eliminate cumulative errors. The effectiveness of the image enhancement is verified by comparing the number and matching of ORB feature points before and after enhancement. AQYACOL dataset is used to run ORB-SLAM3 in both structured and unstructured underwater environments, and the results show that the proposed method improves the robustness of ORB-SLAM3 in underwater operation.