PowerPoint: Tracking and Measuring of Catch Events in Stereo Video for Longline Fisheries
Automated video analyses in monitoring fishery activities have drawn increasing attention due to its scalability and capability. Stereo videos, compared to monocular videos, can capture the depth of information in addition to color and texture; thus, it can be more robust in monitoring and capable of measuring the length of fish. In this work, we present a reliable tracking and measurement approach to stereo videos for catch events in longline fisheries. First, we combine background subtraction method with image object detector to detect fish in a stereo frame. Using the location and disparity information in the stereo frame, we can thus back-project the fish and track it in the 3D space. With the inferred 3D information, we separate the environmental noises from the fish being tracked. Based on the tracking results, we can know whether the fish catch is retained or discarded. We can also measure the length of the fish based on the depth map calculated from stereo matching. Our experiment results demonstrate that our method can work reliably under wild sea environment.
