Privacy-Preserving Human Activity Recognition System for Assisted Living Environments
In today's ultra-competitive professional environment, people seldom have time to look after their ageing parents. Shortage of cheap labor further exacerbates the issue of caring for the aged. In such circumstances, use of technology for effective monitoring of assisted living environments becomes imperative. Sensors have generously been used for monitoring such spaces but they have several limitations in terms of accuracy and discomfort. Vision sensors overcome these limitations but they cannot be used owing to privacy concerns. Recent endeavours harness depth cameras for capturing distorted views of such spaces but the practical efficacy of such solutions is limited. In this article, we propose novel approaches to process depth camera images for superior monitoring accuracy. The processing leads to the creation of specific descriptors for articulating bone angles and joint positions of the monitored individual more effectively. In addition to this, the monitoring accuracy is further improved by using a newly proposed hybrid deep learning architecture.