Fish tracking based on YOLO and ByteTrack

Computer vision has been increasingly applied in the study of fish behavior. However, accurate and robust detection and tracking of fish schools remains a challenging problem due to appearance changes and occlusion during fish locomotion. To address these issues, we first propose a fish detection algorithm based on Yolov5, which utilizes automatic annotation to construct a detection model for fish head regions. Then, inspired by the ByteTrack tracking algorithm, we associate high and low-scoring bounding boxes separately to resolve tracking errors caused by partial occlusion. Finally, we use the position and direction to build a cost function to tackle trajectory fragmentation caused by long-term occlusion. Experimental results on zebrafish videos demonstrate that our proposed method achieves better performance in terms of detection and tracking compared to existing methods.