Towards smart conservation: AI-based bird detection and tracking

Abstract

Wildlife monitoring is essential for conservation and bird conservation, yet traditional methods relying on manual observation remain labour intensive and inefficient. Despite the increasing adoption of computer vision for animal monitoring, a standardised approach for integrating detection and tracking into an automated pipeline remains underdeveloped. By processing camera trap footage and structuring detection results into actionable insights, automated systems can significantly enhance and speed up data-driven conservation strategies. This report presents an end-to-end automated pipeline that integrates state-of-the-art object detection (YOLOv11) andmultiple tracking algorithms (DeepSORT, ByteTrack, BoT-SORT) to identify, track, and count animals of interest. The object detection model achieved an mAP50 of 0.9513 and an mAP50-95 of 0.768, indicating high detection accuracy across varying IoU thresholds. The generated output is processed into log files and visualised through an interactive dashboard website, providing an intuitive platform for stakeholders to analyse trends and behaviours. Comparative analysis results of the tracking algorithms indicate that BoT-SORT achieved the highest Multiple Object Tracking Accuracy (MOTA) score of 0.716, potentially reaching higher values if provided videos had less erratic movements and occlusions. Overall, this research contributes to the development of scalable, autonomous wildlife monitoring methods.