Abstract
Abstract
This paper focuses on human activity recognition using LSTM (Long Short-Term Memory) for handling time series data issues. Because LSTM can better deal with the correlation and long-term correlation of time-series data, the model is utilized to build a human-computer interaction action recognition system. The experimental steps include the following: first, the bracelet sensor obtains the human body acceleration signal to obtain a data set containing a variety of motion postures and preprocesses the data. Then, LSTM is used to extract the features of the dataset, and the feature vector is utilized as the input to train the model iteratively. Finally, to improve the accuracy of recognition, a variety of machine learning models will be applied to compare the performance with the LSTM model. The experimental results indicate that the proposed model can better capture the acceleration changes of human motion in various life scenes, and can accurately distinguish various sudden action patterns of the human body in daily life. This also makes the human movement behavior recognition system proposed in this study more comprehensive.