Deep Learning of CSI for Efficient Device-free Human Activity Recognition

Over the years, wireless sensing is gaining popularity in the applications of indoor localization and human activity recognition (HAR). As wireless signals are sensitive to human motion, they reflect and scatter in different directions depending on the activities performed by people. The channel state information (CSI) stores the combined effect of changes in the environment, and such stored pattern is utilized to recognize different human activities such as walking, standing, and sitting. Prior studies on activity recognition mostly differentiate human activities by classifying one complete series into an activity. However, these approaches require massive datasets to give accurate results in real-time scenarios, and the classification is in fact based on short-term activity samples instead of the complete activity series. In this paper, highly accurate sample-level activity recognition is achieved by exploiting a special type of convolutional neural network (CNN), U-Net. The data collection setup does not require manual feature extraction and can efficiently classify short-term activity samples. Our experimental results indicate that the proposed architecture can classify different levels of human activities with an accuracy of 98.57%, which outperforms conventional Deep Neural Network by 14.67% for the same dataset.