Deep Learning Based Computer Vision Bound Fall Detection for Elderly People

Due to the devastating effects, such as increased mortality and morbidity, falls among the elderly are a significant health issue globally. Supporting the elderly in their daily activities is a significant problem for society because they are the age group with the largest rate of population growth. An automated fall detection system has gained popularity in the healthcare sector due to the social and financial benefits it offers. Vision-based fall detection systems have a great promise now that Smart Homes are a growing trend and there are more cameras in our everyday environments. An autonomous real-time fall detection system based on vision is given in this paper. In this research, we suggested a 3D RGB camera-based method for fall detection. Machine learning algorithms, which were trained on characteristics retrieved by a deep learning-based computer vision algorithm, are used to distinguish between occlusion, falling, and routine everyday activities. Deep learning algorithms are used to identify people in videos. Two datasets—one developed locally and the other made public—were used for the experimental validation of the suggested methodology. Several assessment metrics are computed for evaluation. This analysis demonstrated the efficacy of the suggested solution. Temporal characteristics in a video series are captured using motion history images, and spatial features are then effectively retrieved for classification using a depth-wise convolutional neural network.