Wi-Fi Imaging Based Segmentation and Recognition of Continuous Activity

Automatic segmentation and action recognition have been a long-standing problem in sensorless sensing. In this paper, we propose CHAR, a continuous human activity recognition system to solve these problems in a different way. We’ve noticed that these challenges have been solved in image processing field, so CHAR could effectively perform action segmentation and recognition after WiFi imaging. The key idea behind Wi-Fi imaging is that different body part reflects transmitted signal, the receiver receives the combination of them, and then we separate the received signals from different directions and get the signal intensity in each direction to get the heat map showing the shape of the object. The imaging sequence contains multiple pictures records a continuous action at different time, and we can easily separate and recognize the action based on (image classification), a classification framework we proposed. We implement CHAR using commodity WiFi devices to evaluate its performance under different environment. The results show that the imaging result is better than prior works, facilitating CHAR to achieving an average recognition accuracy, i.e., >95%.