INTELLIGENT FALL DETECTION SYSTEM FOR DRONES IN AGRICULTURAL FIELD SCENARIOS

Abstract

Aiming at the high fall risk of elderly farmers during field operations and the lack of suitable intelligent monitoring technology for complex agricultural scenarios, this paper proposes an intelligent UAV-based fall detection and alarm system dedicated to farmland environments. We constructed a dedicated fall posture dataset covering diverse field vegetation, lighting conditions, and clothing styles, and expanded the sample size to 3,360 images through rich data augmentation. A fall detection model based on the YOLOv8 algorithm is designed to realize real-time and accurate positioning of fall events. Once a fall is detected, the system automatically triggers an on-board buzzer alarm and sends a timely notification email to emergency contacts via the SMTP protocol. Experimental results demonstrate that the proposed model achieves 92.6% mAP@50 and 92.3% precision, with an average system response time of 16 seconds and email notification completed within 40 seconds. The system maintains favorable robustness and generalization ability under complex conditions such as low illumination, and can effectively adapt to fall detection tasks in actual farmland scenes. It reduces the delay of manual monitoring and expands the coverage of safety protection, providing a practical and reliable technical approach for safety monitoring and emergency assistance in smart agriculture.