Self-Supervised WiFi-Based Activity Recognition

Traditional approaches to activity recognition involve the use of wearable sensors or cameras in order to recognise human activities. In this work, we extract fine-grained physical layer information from WiFi devices for the purpose of passive activity recognition in indoor environments. While such data is ubiquitous, few approaches are designed to utilise large amounts of unlabelled WiFi data. We propose the use of selfsupervised contrastive learning to improve activity recognition performance when using multiple views of the transmitted WiFi signal captured by synchronised receivers deployed in different positions. We conduct extensive Human Activity Recognition (HAR) experiments in two furnished office rooms, whereby six participants of different age groups performed six day-to-day activities. We compare the proposed contrastive learning system with conventional non-contrastive systems and observe significant improvement on the task of WiFi based activity recognition under few-shot learning scenarios. Namely, contrastively pretraining with an AlexNet-based backbone encoder led to a 22% increase in macro F
1
score when only 1.29% of labelled training samples are considered in the fine-tuning stage.