Automated Detection and Threshold-Based Alerting of Stray Dogs

The free-roaming dog population of India poses a serious public safety, health, and animal welfare challenge. Approximately 20 million dog-bite cases and 20,000 rabies-related deaths occur each year in the country. Robust and widespread solutions are required for this problem, and monitoring stray animals using traditional methods could be challenging and inefficient. This study explores an AI-driven solution using advanced deep learning techniques for this problem. We developed a custom-trained YOLOv11s model using transfer learning based on real-world dataset obtained from various locations of Srinagar and Ganderbal district of Jammu and Kashmir, India. This model detects and counts both dogs and humans in images taken from urban streets and campuses, particularly focusing on stray dog situations in Kashmir. Our system achieved precision rates of 0.995 for dogs and 0.991 for humans, with recall rates of 0.966 for dogs and 0.915 for humans. The mean average precision (mAP@0.5) was 0.973. This model was integrated into a user-friendly web application, enabling detection and the setting of configurable threshold-based alerts. This application characterizes the deployable use of YOLOv11s for urban stray dog surveillance in India. AI-based digital solutions provide scalable and minimally intrusive solutions for public safety and One Health.