The elderly population in the world is continuing to grow at an unparalleled rate and it is important to keep in mind about the safety of the aging population. Falls have become one of the major reasons leading to injury to them and if timely help is not provided, it could lead to serious complications. There are various traditional methods that have been used to detect human falls and wearable devices are one of these methods that contain sensors like accelerometer, gyroscope, barometers, etc. These devices are highly instrumental in computing various parameters based on which the fall will be detected. However, these devices have certain limitations, as they are complex for the elderly people to understand and use. The accuracy of such devices is very low and there are high chances for false alarms as well. Hence, applying a vision-based object detection algorithm for detecting these human falls is highly significant in order to overcome such challenges faced by wearable devices. In this research work, an Object detection based Automated Fall Detection System has been proposed wherein, the YOLOv7 (You Only Look Once) pose model is used to discriminate the fall and non-fall activity. Given a video, the YOLOv7 model will first distinctly separate all the video frames and then pre-process these frames. This pre-processed data is sent for estimating the pose of the person. This resulting output is further subjected to classification as fall or non-fall activity. The dataset used in our approach are self-generated videos that cover a set of daily human activities. The strength of the proposed method has been proved through various performance measures like precision and classification accuracy.
