Human Recognition from High-altitude UAV Camera Images by AI based Body Region Detection

The usage of unmanned aerial vehicles (UAVs) has rapidly increased during the recent past. The application areas for UAVs include search and rescue operations, tracking humans for surveillance, disaster management etc. In most such situations, the UAV's ability to automatically recognize humans, body regions, and other objects is crucial. Yet visibility issues may exist in UAV images resulting in challenges for object detection. In the case of recognizing humans from the high-altitude UAV camera images, conventional methods mostly recognize humans, based on the features of the entire body. Such methods are less effective when the parts of the body appear in the images, especially during high-altitude flights. In this paper, we develop a framework for human recognition by detecting the human body regions with YOLOv5 and Haar cascade classifier. The body part detection is done using YOLOv5, and then the output is forwarded to Haar cascade classifiers to classify body regions such as head, upper body, lower body. Human availability is confirmed, if the classified body part belongs to one of above body regions. We conducted experiments using the VisDrone data set. Preliminary results show promise on the effectiveness of the method.