Research on Fall Detection Algorithm Based on CNN and LSTM
Gang Yang, Jie Zhao, Jian Wen Guo · 2021
At present, the use of sensors for fall detection mainly uses manual methods to extract characteristic signals, which limits the accuracy of the algorithm. In order to improve the accuracy of fall recognition, a fall detection algorithm based on fusion convolutional neural network (CNN) and long short-term memory network (LSTM) is designed. The algorithm uses the efficient feature extraction capabilities of CNN to automatically extract the feature values of the data, and uses them as the input information of the LSTM network, and then uses the ability of LSTM to describe the long-term dependence to obtain the overall time series, and finally uses the Soft-max classifier to perform the fall and Identification of daily activities. In order to verify the feasibility of the algorithm, designed a comparative experiment. By comparing with the latest fall detection algorithm, the fusion algorithm has a fall detection accuracy rate of 98.8%, and the accuracy of human activity recognition and loss accuracy are 98.7% and 0.18%, respectively. The experimental results show that the algorithm proposed in this paper has higher recognition accuracy and lower loss accuracy, can effectively identify daily activities and fall behavior, and provide timely assistance for the elderly.