An RGB camera-based fall detection algorithm in complex home environments
Zhiyu Tian, Liang Zhang, Guoan Wang, Xuefeng Wang · Interdisciplinary Nursing Research · 2022
Abstract Objectives: Accidental falls are a threat to the well-being of older people. This study aimed to develop a real-time human fall detection system to detect fall behaviors and provide timely medical treatment for older adults. Methods: An RGB camera-based fall detection system is designed and it can send alarm messages when a fall occurs. This fall detection system consists of two design aspects: a hardware and a software algorithm. The fall detection algorithm includes (1) algorithm initialization phase to obtain environmental parameters; (2) 2-dimensional pose detection to identify human targets and human joint locations; and (3) limb-length and multiframe fall judgment to confirm the occurrence of falls based on its practical features. Results: By combining fall detection algorithms with a hardware system, the test results in complex home environments showed that the system sensitivity was 94.2%, the specificity was 96%, and the accuracy was 94.5%. Conclusion: The proposed method is more robust compared with the algorithm based exclusively on action recognition. Using only a monocular camera is cost-friendly and can realize real-time fall detections, and help older people to get timely and effective care after a fall.