Multi-Person Fall Detection in Complex IoT-Assisted Living Environments
Amit Kumar Singh, Prescilla Koshy, B. S. B. S. Manoj · 2022 IEEE 19th India Council International Conference (INDICON) · 2022
Fall is one of the major concerns of the elderly population. Falls can cause fractures that lead to immobility in the elderly. Internet of Things (IoT) based solutions can help detect falls, thereby, ensuring health care services are delivered in a timely manner. Automating fall detection is a challenging process in retirement homes, nursing homes, and other senior care centers. Although, there exist several studies on the detection of single-person falls, only limited research has focused on detecting multi-person falls. Single-person fall detection uses ambient as well as body-based sensors along with a computer vision approach for capturing and detecting falls. This paper proposes a new approach for the detection of multi-person falls in a complex and challenging environment. The proposed approach involves four steps: (i) Object Detection and Image Segmentation using Region-Based Convolutional Neural Networks (RCNN) and Human Class Bounding Box Extraction, (ii) Pose estimation of individual segments and extraction of coordinates of important key points, (iii) Features extraction, and (iv) Classification to determine the occurrence of fall. We created a new data set as part of this work, including the frames created from videos of complex environments and collecting various YouTube video clips where fall incidents take place while several persons are present in a frame. Our approach for detecting falls in a complex environment and with multiple people in a frame, is much more robust and can detect and classify fall events accurately. Our method demonstrates the high accuracy of multi-person fall detection.