True Detect: Deep Learning-Based Device-Free Activity Recognition Using WiFi

摘要

Activity recognition has taken an important role in smart devices and has a wide range of applications. Traditional approaches to activity recognition either require the target to carry electronic sensors or require dedicated devices such as software-defined radios, which may not be feasible in many situations. This paper provides an overview of device-free activity recognition techniques and proposes a convolutional neural network-based approach for classifying different human activities by leveraging the channel state information (CSI) obtained using off-the-shelf hardware. The design and implementation of the deep learning framework, to classify activities in a live scenario, are also discussed. The results show that, in controlled environments, a high activity recognition accuracy can be achieved using CSI and that the framework can also be implemented in live scenarios.