Deep Learning-Based Human Activity Recognition via Power Lines

This paper presents a novel method for recognizing human activities using ubiquitous power lines. With the assistance of a Universal Software Radio Peripheral (USRP), high-frequency electromagnetic signals are injected into power lines. These signals are then radiated from and received by a power line while individuals move nearby. Doppler spectrograms of the received signals are processed and used for activity recognition. A DenseNet-201 network is trained to classify the types of human activity using the Doppler spectrograms, achieving accurate predictions of activities such as walking, running, and jumping.