Research on Human Fall Detection Based on Tiny-yolov3 Algorithm

Most families today have experienced a fall problem with an elderly person or child. The number one cause of accidental injury to an elderly person or child is a fall. When there is a danger, it is difficult for the elderly to save themselves in time. Children, on the other hand, have the characteristic of being active, which makes them fall frequently during running resulting in different degrees of injuries to children. In recent years, deep learning algorithms have become a research hotspot in the field of graphic image processing, and the YOLO series of algorithms [1-3], as a classical algorithm in the single-stage target detection algorithm, has the accuracy comparable to the two-stage algorithm and the speed beyond the two-stage algorithm, this paper is using the Tiny-yolov3 algorithm model to detect human falls, which has the advantages of high detection accuracy and low computational complexity. By monitoring the trajectory of the human body in real time through the camera loaded on the trolley, the abnormal posture of the human body is detected and stays in certain positions for a long time to serve as a criterion for whether the human body has fallen or not. A better result is achieved, which can provide some reference for fall monitoring related research.