Fishery monitoring system with AUV based on YOLO and SGBM

AUV has gained more and more attention with the growing importance of the ocean. The visual intelligence extends the sensing ability of the AUV. Prior work on visual intelligence of AUV mainly focuses on the object classification and serving areas underwater. Instead, we focus on a better understanding of images and get more useful information from binocular cameras. This paper introduces a new fisheries application of the visual intelligence of AUV and proposes a new approach to monitor the growth of fish with binocular cameras. We construct a deep network with YOLO architecture to detect the bounding boxes of fishes and extract the edges of fishes from the bounding boxes. Then we use the SGBM method to predict the depth map, which is used to estimate the length and width of the fishes. The YOLO object detection system and SGBM algorithm are both low-cost with accurate real-time results in experiments. This fishery monitoring system is efficient, real-time and accurate compared with the manual method or dual-frequency sonar.