Researching and Implementing the Posture Recognition Algorithm of the Elderly on Jetson Nano

Abstract

Falls are a common phenomenon among the elderly. Falling not only causes serious physiological injuries such as fractures, head injuries, etc., but also causes psychological damage to the elderly. In addition to prevention, detecting falls in a timely manner can help limit the consequences of falls. In this paper, we present a fall detection method for the elderly using a neural network on Jetson Nano. The fall recognition model is built based on the Convolutional Neural Network (CNN) deep learning model. The model has functions like object body shape recognition, body recognition, and integrates a trained OpenPose algorithm model that allows receiving human body parts from which allows to predict object behavior through a Feed-Forward Networks (FFN). The experimental results on the real data set collected by us show that the proposed model is suitable for detecting falls in the elderly with an accuracy of 89.07% and the frame per second (FPS) on the Jetson Nano is 2.49.