Posture Detection Based on Kinect and YOLO Technologies

The increasing aging of society has led to a growing focus on the health care of the elderly. Fall among the elderly pose a significant threat, emphasizing the urgent need for timely detection and assistance. With the continuous evolution of intelligent monitoring technology, the importance of fall detection systems in the field of human safety monitoring is increasingly evident. Our research, based on the capabilities of Kinect and the You Only Look Once Version5 (YOLOv5) object detection model, which help in improving the precision and real-time functionalities of fall detection in video stream. Initially, we delve into the depth perception capabilities of Kinect and the lightweight efficiency of the YOLOv5 object detection model. Subsequently, through system integration and data preparation, we establish a comprehensive video stream fall detection system. In experiments, we utilize a diverse fall datasets and conduct evaluations of the system performance in different scenarios. Our results demonstrate a significant performance advantage in the combination of Kinect and YOLOv5 for video stream fall detection tasks. Across various lighting conditions, background scenarios, and diverse fall behaviors, the system exhibits remarkable robustness and generalization capability.