Drone-Based Human Surveillance using YOLOv5 and Multi-Features

Drone use has expanded recently to include airborne photography, human action recognition (HAR), search and rescue (SAR), and surveillance. Drones have a distinct edge in human action recognition (HAR) due to their bird's-eye vision, but this also brings challenges such as shifting backgrounds and obscured perspectives, which call for creative solutions. To address these issues, we propose a method that processes video frames using preprocessing techniques, detects human presence using YOLO, and human silhouettes are extracted using morphological erosion. We focus on key skeletal landmarks such as the head, wrists, ankles, and neck to analyze human movement patterns. We enhance feature extraction with methods like MSER and HOG. Classification is performed using Naïve Bayes algorithm, with feature optimization achieved through the Grey Wolf optimizer. Our method is tested on the Drone-Action dataset and shows itself to be rather effective, detecting human activity with an astounding 84% accuracy.