Human Postures Classification and Fall Detection Using Convolutional Neural Networks

Abstract

Accidental fall is a major problem among the elderly people aged 65 and above. The weakness of a person is related to aging. As the age of the person increases the ratio of falling increases day by day. Accidental falls affecting a large number of people, which causes some serious injuries and even death in some cases, it is also affecting the health care of countries badly. The timely alert could be helpful in this case as it enables instant delivery of medical service to the injured and minimizes the negative consequences of falls. To address this problem, we designed a Convolutional Neural Networks (CNN) model which detects the fallen body by reading the frames. A Convolutional Neural Network also called ConvNets is a Deep Learning algorithm that is used to recognize and classify image features and be able to differentiate one from the other, it is widely used for analyzing visual images. In this project, we present the novel approach for the fallen detection completely based on visual data. We introduce the mechanism which can learn how to detect the fallen body from the input frames. The CNN-based fall detection model is trained and validated on a large dataset that categorized into "falling" and "not falling" categories to enhance its performance and later tested on different data to check its accuracy. In testing a new dataset is provided which is the set of labeled images, we did not use this data in the training process. This data run through the trained model and compares the output to the actual labels. This process is needed to evaluate the neural network.