Can we recognize multiple human group activities using ultrasonic sensors?

Human activity recognition is a challenging research area having possible implementation in various aspects of our life. It can be beneficial in automation, surveillance, vigilance, assisted living, etc. Most of the recent works done in this area have focused on identifying single-person activities. But in real scenario, many of these activities are performed in groups. The kind of technologies used are combination of computer vision and wearable devices. The use of such techniques may interfere with one's privacy. In this context, our work intends to identify group activities using a non-invasive, non-intrusive, non-wearable sensor based system. The proposed system identifies activities such as walking, standing, sitting, gathering etc. comprising a group of people. We have used customized multiple ultrasonic sensor grids for data capturing. For classifying primary activities, decision tree technique has been used. Other concurrent group activities were identified using Hidden Markov Model (HMM). The proposed system achieved an accuracy up to 90% in most of the cases.