ISFDNet: Using Low-Resolution Infrared Array Sensors to Detect Elderly Fall in Home Elevator Scenarios

Falls are a significant cause of injury in older people, and in the home elevator environment, the consequences of a fall can be more severe due to the confined and narrow space. This article focuses on the home elevator scenario. First, we construct a fall detection dataset based on infrared array sensors, considering nonhuman factors such as pets in a home environment. Second, we propose a background filtering method based on Otsu’s method to eliminate the interference of nonhuman factors in the background. In addition, we introduce a new and more generalized single-frame fall detection model, ISFDNet, which shows higher fall detection precision than traditional models and achieves a precision rate of 96.64%. Finally, we adopt a single-frame detection approach combined with a sliding window-based fall detection event triggering mechanism, which ensures fall detection accuracy, reduces memory requirements, and improves inference speed; the method obtained a 100% precision rate of fall event detection.