Elderly People Activity Recognition Based on Object Detection Technique Using Jetson Nano

Regular activity monitoring and recognition are closely associated with healthcare benefits. This paper aims to detect the activities of elderly people to support them live independently. This work focuses on developing an elderly people activity recognition system with Jetson Nano. Two datasets have been created. The setup of hardware includes a Jetson Nano 2GB, monitor/display, and web camera. The YOLO network is trained with the proposed dataset using Google Colab. The trained weights are dumped in the Jetson Nano 2GB and the testing have been carried out in a real-time environment. The accuracy achieved is very satisfactory with more than 80% in both simulation and hardware implementation. The real challenge is creating a hardware-based activity recognition system since most of the researchers have worked only on software-based activity recognition systems and are also less focused on elderly people’s activity recognition.