Improved fall detection algorithm based on yolov8s

Abstract

For the problem of physical injuries caused by untimely fall relief, this paper proposes an improved fall detection algorithm based on YOLOv8s.First, through many experiments, in the YOLOv8 algorithm model, select the YOLOv8s framework. Second, in the backbone part insert SE (Squeeze-and-Excitation) attention mechanism, strengthen the network feature extraction ability and adaptive ability, makes the network in the process of feature extraction network can pay more attention to the target, improve the network in the fall scene, improve the model accuracy. Lastly, Spatial and channel reconstruction convolution (SCConv) and C2f module are introduced into the network, reconstructed into C2f_SCConv module to integrate multi-scale features and reduce spatial and channel redundancy in the convolutional neural network, so as to improve the efficiency and accuracy and improve the representation ability of the model. After the improved YOLOv8s algorithm, the mAP increased from the initial 57.6% to 65.1%, which provided a good reference value for the subsequent study of the fall detection algorithm.