Indoor Group Identification and Localization Using Privacy-Preserving Edge Computing Distributed Camera Network

Abstract

Social interaction behaviors change as a re-1 sult of both physical and psychiatric problems, and it is 2 important to identify subtle changes in group activity en-3 gagements for monitoring the mental health of patients in 4 clinics. This work proposes a system to identify when and 5 where group formations occur in an approximately 1,700m 2 6 therapeutic built environment using a distributed edge-7 computing camera network. The proposed method can lo-8 calize group formations when provided with noisy positions 9 and orientations of individuals, estimated from sparsely 10 distributed multi-view cameras, which run a lightweight 11 multi-person 2D pose detection model. Our group identifi-12 cation method demonstrated an F1 score of up to 90% with 13 a mean absolute error of 1.25m for group localization on 14 our benchmark dataset. The dataset consisted of 7 subjects 15 walking, sitting, and conversing for 35 minutes in groups of 16 various sizes ranging from 2 to 7 subjects. The proposed 17 system is low-cost and scalable to any ordinary building to 18 transform the indoor space into a smart environment using 19 edge computing systems. We expect the proposed system 20 to enhance existing therapeutic units for passively moni-21 toring the social behaviors of patients when implementing 22 real-time interventions.