On automatizing recognition of multiple human activities using ultrasonic sensor grid

Human activity recognition is an important problem in health care, ambient-assisted living, surveillance-based security, etc. and has crucial applications in smart environment. A non-invasive, automated system for monitoring human activity using array of heterogeneous ultrasonic sensors has been proposed in this work. Ultrasonic sensors are widely used for distance measurement in many applications. In the proposed system experiments have been conducted using ten volunteers in a controlled laboratory environment. The data collection unit has two kinds of setups of ultrasonic sensors: the former with five HC-SR04 sensors, and the latter with four HC-SR04 ultrasonic sensors and an LV-MaxSonar-EZ0 sensor. The proposed method is found capable of detecting standing, sitting and falling of a person, and also the movements in different directions. Based on the collected data, we have performed classification analysis using multiple machine learning algorithms. The experimental results show 81% to 90% correct detection of different activities of the volunteers.