KS-FALL: Indoor Human Fall Detection Method Under 5GHz Wireless Signals

Abstract

Abstract In modern society, it has become the main threat to the elderly fall’s health or even death in the elderly. The real-time and reliable fall detection system can save the fall and hurt the elderly in time. In this paper, a human fall detection method KS-FALL based on Channel State Information and in 5G environment is proposed. KS-FALL uses Atheros commercial NIC equipment to map the amplitude information in the wireless signal to the human body’s fall action, and does not require the user to wear any equipment. Compared with the traditional 2.4 GHz signal, the 5 GHz signal provides richer sub-carrier frequency domain information, which better reflects the relationship between human motion and wireless signals, thereby more effectively distinguishing and recognizing walking, squatting, falling, etc. action, filter the environmental interference through powerful denoising method, use K-means to cluster different action data, combine SVM classifier to construct fine-grained offline fingerprint database, and use SoftMax regression model to correct SVM classification in real-time detection stage. And real-time test in two different scenarios, and the detection accuracy of the fall reached 92.3%, realizing the device-free, non-invasive, high-precision human fall detection.