Enhancing Safety in Smart Homecare System Through Deep Learning Based Fall Detection

Falls are a major global health concern that can affect people physically and psychologically in the short- and long-term, especially the elderly. This initiative aims to monitor human activities in an indoor environment and identify falling incidents without requiring individuals to wear a device or sensor on their body. If a fall is not prevented in time, it can significantly reduce an elderly person's mobility, independence, and quality of life. When a fall is detected, the system immediately responds by raising an alert and sending messages to the agencies in charge of caring for the elderly. To achieve accurate fall detection in an indoor environment, this work incorporates the deep learning Tiny YOLO algorithm. Tiny YOLO is a real-time object detection algorithm that is a smaller and faster version of the original YOLO algorithm. By leveraging the power of (CNNs), in Tiny YOLO it consists of three layers such as fully linked, max-pooling, and convolutional layers. To be able to detect objects, these layers work together to recognize patterns and features in the incoming image. CNNs are well-suited for image-related tasks, as they can capture hierarchical features in images, making them effective for object detection, classification, and localization. In the context of fall detection, the Tiny YOLO algorithm analyses video streams captured by cameras placed strategically in the indoor environment. Tiny YOLO can effectively identify human poses, movements, and actions with high precision. Extract the key points from the human pose estimation and classify them using Tiny YOLO and detect the fall or not fall. If fall is happened then alert the family members using twilio platform. Otherwise, detects as fall warning or no fall. Tiny YOLO is known for its speed and efficiency while maintaining reasonable accuracy in object detection tasks.