LAUSR

Imaging sonar, as an efficient sensor for underwater observation and measurement, plays a very important role in the positioning and mapping of autonomous underwater vehicles (AUVs). Compared to visual simultaneous localization and mapping (SLAM) susceptible to the water quality and distance, sonar SLAM based on acoustic images has a better perception range and immunity to the surrounding fluid characteristics. However, due to the complexity of underwater environments, imaging sonar is often severely affected by noise, and sonar images usually exhibit low resolution, insufficient contrast, and blurred target edges, resulting in severe errors in positioning and mapping. To solve these problems, a sonar image preprocessing method is proposed to effectively eliminate noise, while feature edges preserved. First, the proposed image preprocessing method employs a threshold filter to eliminate potential noises; then, the contrast-limited adaptive histogram equalization algorithm is used to enhance the image features, and the guided filter (GF) is used to eliminate the residual noise. In addition, we introduce a factor graph-based sonar SLAM method, and the preprocessed sonar images are integrated into the SLAM process. Through numerical evaluation, the effectiveness of our proposed preprocessing method is validated, and the results show that it improves the processing speed, detection precision, and matching precision by about 73.9%, 30.8%, and 20.0%, respectively. Furthermore, simulation experiments and real experiment demonstrate that our proposed method improves both positioning accuracy and stability of the AUV.