A Fall Detection Algorithm Based on Improved MobileNetV2
Mengqi Gao, Jiangjiao Li, Bin Li · 2022
With the aggravation of the current social aging problem, the health problems of the elderly living alone have become the focus of attention in the medical and health field. Due to a large number of network parameters, the current deep learning-based fall detection algorithm takes a long time to train. To make the algorithm meet the demands of both lightweight and high accuracy, a new method for fall detection is proposed in this article. We improve the block of MobileNetV2 by means of adding channel attention and spatial attention mechanisms before channel convolution and spatial convolution in the block, respectively. The above operation can improve the performance of capturing essential information without additional network computation, which enhances the effectiveness of network feature extraction and detection performance. The computer simulation shows that the detection accuracy of the proposed algorithm on the public datasets Le2i and UR is 98.8% and 99.7%, respectively, which can accurately detect the falling behavior in the image frame column.