Intelligent Elderly People Fall Detection Based on Modified Deep Learning Deep Transfer Learning and IoT Using Thermal Imaging-Assisted Pervasive Surveillance

Abstract

Early detection of knee osteoarthritis and poor balance can decrease falls in the elderly. Thus, automatic fall detection is an essential system for assuring the safety and health of the elderly. However, the use of the Visible Imaging System (VIS) installed in homes can affect people’s privacy. Compared to visible imaging, thermal imaging involves people’s privacy less and allows various incidents to be identified based on machine vision. A novel two-step framework through thermal imaging videos is introduced in this paper, including tracking humans and deep learning-based for recognizing the fall incidents. In the first step, the Kalman filter is employed to distinguish people’s positions. Then, the novel modified ShuffleNet is utilized to refine the obtained bounding boxes of people at risk of falling. The proposed approach is implemented using the Internet of Things (IoT) deployment. The publicly thermal fall dataset analyses reveal the superior outcomes achieved with an average of less than 7% error compared to the conventional fall detection models. Besides, the IoT platform helps to process the incidents data and more efficiently, real-time monitoring, manage energy usage, and healthcare management.