A WiFi-based System for Recognizing Fine-grained Multiple-Subject Human Activities

Device-free human activity recognition has become a topic of much interest in recent years. While there is much existing work on course-grained human activity recognition, the recognition of fine-grained human activities is still a research challenge. In this paper, we propose a new approach using CSI and RSSI WiFi data to recognize fine-grained human activities. We selected 4 different fine-grained human activities from a human-to-human interaction dataset and defined some frequency features over CSI and RSSI data to use as input to our classification model. Using some classification methods and the K Nearest Neighbors (KNN) classifier, we achieved 97.5% of accuracy in fine-grained human activity recognition.

I. Introduction

Human Activity Recognition (HAR) can be defined as the ability to recognize human activities based on the received data from different sensors. HAR has its great potential for various applications [1]. For example in healthcare [2], where detecting patients’ activities helps caregivers to easier monitor them [3]– [6]. Detecting anomalies in daily routines for security [7]–[9] is another example use case of HAR. Also, within the context of human-computer interaction, HAR has been used by researchers to capture users’ movements, gestures, or actions, like detecting teacher’s activities in a class [10] or kids’ activities in different environments [11].