Fall detection algorithm based on improved Yolov7

Aiming at the problems of slow detection speed and low accuracy in current human fall detection tasks, a fall detection algorithm based on Yolov7 was proposed. ODConv-ELAN module was constructed in Yolov7 backbone network to replace the original ELAN structure and enhance the ability of extracting target features. Secondly, the more advanced EIoU function is used as the new boundary frame loss function, which improves the convergence speed and efficiency of the prediction frame in the process of model training. Finally, CA attention mechanism is introduced into the output terminal of the network to improve the detection performance of human fall behavior. In addition, a fall detection data set in the campus environment was created. The accuracy P of the improved algorithm in this data set reached 94.34%, the recall rate R reached 92.34%, and the average accuracy mAP reached 94.65%, which realized the demand for more accurate human fall detection.