Digitization of Free-Swimming Fish Based on Binocular Stereo Vision

With the rapid development of computer vision, the digitization of free-swimming fish has drawn increasing attention. Research on the digitization of fish has many important applications in aquaculture, such as length measurement and species recognition. The aim of digitization of free-swimming fish is acquiring three dimensional point cloud of free-swimming fish and carrying out three-dimension stereo-reconstruction. Calibration and stereo rectification are fulfilled before stereo matching to get intrinsic and extrinsic parameters of cameras and row-aligned image pairs. In order to reduce the effects of non-uniform illumination in the underwater environment, an improved non-local cost aggregation algorithm based on gradient magnitude and phase information is proposed to generate the disparity map in stereo matching step. Then the 3D point cloud of free-swimming fish can be obtained based on triangulation and parallax principle. At last a new way is presented to measure the length of fish by the 3D point cloud of the fish and the error is 5.1% compared with the real length.