Animal Behavior Monitoring Based on Sequence to Sequence Model

Object tracking is a core task in the field of computer vision, with wide-ranging applications in scenarios such as video surveillance and autonomous driving. This paper focuses on the application of the SeqTrack method, a sequence-to-sequence (Seq2Seq) model-based approach for object tracking, to animal behavior monitoring video datasets. The SeqTrack method leverages an encoder-decoder architecture to capture the temporal dependencies of objects, enabling accurate object localization and bounding box prediction. To validate the performance of SeqTrack in practical scenarios, we conducted comparative experiments on a self-made Adelaide Zoo dataset, comparing SeqTrack with three other mainstream object tracking algorithms: ToMP, OSTrack, and MixFormer. The experimental results show that SeqTrack outper-forms in terms of both precision and success rate, demonstrating its superior capability in handling challenges such as fast object movements, complex backgrounds, and partial occlusions. These findings indicate that SeqTrack is well-suited for high-precision tracking scenarios like animal behavior monitoring.