Advancing Zebrafish Behavioral Analysis: A Synergistic YOLOv8-DeepSORT Framework for Precision Tracking

Multiple Object Tracking (MOT) faces significant challenges in accurately tracking small, fast-moving, and visually similar objects like zebrafish (Danio rerio) i n complex environments marked by occlusions, lighting variations, and unpredictable movements. An integrated MOT system proposed in this research uses K-means clustering for unsupervised trajectory classification, DeepSORT with a refined ResNet50 for dependable multi-object tracking, and YOLOv8 for robust object recognition to improve zebrafish behavioral a nalysis. With a mean Average Precision (mAP@0.5) of 96.5% in detection, an ID Switch Rate (IDSR) of 2.4% in tracking, and a silhouette score of 0.68 for trajectory clustering, this framework demonstrates strong performance. Heatmap analysis highlights important meeting places and clarifies spatial mobility patterns. The method improves real-time tracking and behavioral analysis by reducing occlusion and ID switching.