FishMOT: A Simple and Effective Method for Fish Tracking Based on IoU Matching
TL
TL;DR
Experimental results demonstrate FishMOT outperforms state-of-the-art multi-object trackers and specialized fish tracking tools in terms of MOTA (Multiple Object Tracking Accuracy), accuracy, computation time, memory consumption, etc and exhibits excellent robustness and generalizability for varying environments and fish numbers.
Abstract
HighlightsExisting tracking methods face challenges in accuracy and robustness.FishMOT (Multiple Object Tracking for Fish) is a novel fish-tracking approach combining object detection and IoU matching.FishMOT outperforms state-of-the-art multi-target trackers and specialized fish-tracking tools on several metrics.The simplified workflow and strong performance make FishMOT a highly effective fish-tracking approach.Abstract. Fish tracking plays a vital role in understanding fish behavior and ecology. However, existing tracking methods face challenges in accuracy and robustness due to morphological changes in fish, occlusion, and complex environments. This article proposes FishMOT (Multiple Object Tracking for Fish), a novel fish tracking approach combining object detection and IoU (Intersection over Union) matching, including a basic module, an interaction module, and a refind module. Wherein, a basic module performs target association based on the IoU of detection boxes between successive frames to deal with the morphological change of fish; an interaction module combines the IoU of detection boxes and the IoU of fish entities to handle occlusions; and a refind module uses spatio-temporal information to overcome the tracking failure resulting from the missed detection by the detector in a complex environment. FishMOT reduces the computational complexity and memory consumption since it does not require complex feature extraction or identity assignment per fish and does not need a Kalman filter to predict the detection boxes of successive frames. Experimental results demonstrate FishMOT outperforms state-of-the-art multi-object trackers and specialized fish tracking tools in terms of MOTA (Multiple Object Tracking Accuracy), accuracy, computation time, memory consumption, etc. Furthermore, the method exhibits excellent robustness and generalizability for varying environments and fish numbers. The simplified workflow and strong performance make FishMOT a highly effective fish-tracking approach. The source codes and pre-trained models are available at: https://github.com/gakkistar/FishMOT. Keywords: Behavior analysis, Fish behavior, Fish tracking, Multiple object tracking, Object detection.