Development of an intelligent interactive detection system for fishery industry based on improved stereo matching technology

Against the global fishery industry's intelligent transformation, efficient underwater target detection and 3D information acquisition are vital for resource monitoring and aquaculture. Traditional stereo matching faces challenges like optical distortion and low contrast, causing poor edge/weak-texture accuracy and low real-time efficiency. This paper proposes a multi-feature optimized binocular stereo vision method suitable for underwater scenarios. Contour maps are generated through probabilistic Hough transform and edge detection, and regions with small areas are filtered out to generate regions of interest, thus reducing the computational load. The edge disparity and map quality are enhanced through a three-stage sub-pixel refinement process. The system integrates point cloud generation and interactive measurement functions for real-time 3D coordinate analysis. Experiments show that in on-land / shallow-water environments, the matching accuracy is significantly improved, providing a reliable solution for underwater navigation and detection. It has strong adaptability to optical distortion and low-contrast conditions, offering practical value for intelligent fishery detection.