DenseDuckMOT: A Real-Time Detection-Tracking Coupled Counting Framework for Complex Avicultural Environments.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Animal Science

Background:

• High-density farming of protected breeds like the Liancheng White Duck presents challenges for automated monitoring.
• Occlusion, motion blur, and flock aggregation hinder accurate target detection and behavior recognition in poultry farms.

Purpose of the Study:

• To develop an integrated detection-tracking framework, DenseDuckMOT, for practical, real-time monitoring of ducks in farm environments.
• To improve the accuracy and efficiency of flock monitoring and counting using existing surveillance infrastructure.

Main Methods:

• Proposed DenseDuckMOT framework combining an improved DuckNet detector (based on YOLOv11 with BiFPN, GLSA, ESDH) and AKFTrack tracker.
• DuckNet achieved high precision (98.19%) and recall (97.72%) with a lightweight design.
• AKFTrack incorporated adaptive Kalman prediction and a two-stage association scheme for robust tracking.

Main Results:

• DuckNet demonstrated superior performance metrics including precision, mAP@0.75, F1-score, and recall.
• AKFTrack outperformed or matched state-of-the-art trackers (DeepSORT, StrongSORT, ByteTrack) in MOTA, IDF1, and recall, especially in crowded and occluded scenarios.
• Experimental results and visualizations confirmed the effectiveness of the integrated framework in handling occlusion and rapid motion.

Conclusions:

• DenseDuckMOT offers an accurate, efficient, and stable solution for real-time monitoring in dynamic poultry farms.
• The framework provides a scalable approach for intelligent farming, addressing limitations of manual monitoring.
• The study highlights the complementary benefits of specific architectural components in DuckNet and the robustness of AKFTrack.