Computer Vision for Wildlife Monitoring: Detecting Brown Howler Monkeys using YOLO

Urban expansion and habitat fragmentation pose severe threats to arboreal species, such as the brown howler monkey (Alouatta guariba). Canopy bridges are a critical mitigation strategy, but their effectiveness must be monitored using camera traps. This monitoring generates massive volumes of video data, much of which contains false positives (e.g., moving vegetation), creating a significant bottleneck for conservationists who must manually review footage. This study explores the use of computer vision to automate the detection of howler monkeys in these videos, specifically addressing the challenge of limited manually annotated training data.

The researchers utilized the YOLOv10 object detection framework. To overcome the scarcity of annotated real-world images, they developed a hybrid training strategy. They created a synthetic dataset using the Unity game engine, incorporating 3D models of howler monkeys and various environmental conditions (lighting, weather, and camera angles). They then fine-tuned YOLOv10 using different proportions of real-world images and synthetic data. Finally, they implemented an automated video triage tool that classifies 30-second clips as containing or not containing howler monkeys based on a frame-count threshold.

The study found that incorporating synthetic data significantly enhances the model's performance compared to training on limited real-world data alone. The best-performing model, trained on a combination of 10,000 real images and 10,000 synthetic images, achieved superior F1-score and mAP@0.5 metrics compared to models trained solely on real-world data. The video triage tool successfully identified howler monkey presence with an F1-score of 0.762, demonstrating that automated systems can effectively reduce the manual workload for conservationists. The authors noted that while the model struggles with very brief or partial animal appearances, it remains a viable tool for large-scale monitoring.