Wifienabled Nonintrusive Realtime Fall Detection System


Aishwarya Laturkar

Oral Defence Date: 

Thursday, July 11, 2019 - 10:00


TH 434


Assistant Prof. Hao Yue and Prof. Bill Hsu


Human Falls is the leading cause of injury-related visits to medical emergency departments and accidental deaths among people over the age of 65. The health outcome of a fall highly depends on rapid response and rescue of the immobile person. Therefore, an alarm system that can provide accurate and real-time fall detection will dramatically improve the health outcomes of the elders and reduce the associated healthcare cost after a fall. Recent technologies in wireless sensing have shown that Wi-Fi signals contain rich information that characterizes the surrounding environment they propagate through, which can be used to achieve human localization and activity recognition. In this project, we develop a new system that can provide accurate and real-time human fall detection using commercial Wi-Fi devices. We extract Channel State Information (CSI) from the Wi-Fi signals collected in the environments with and without human falls. We then construct and train a machine learning classifier consisting of an Autoencoder and SoftMax regression with the CSI data to recognize falls. We conduct numerous experiments in different environment settings and the best accuracy of the system in detecting human falls is 92%. The results show great promise that wireless sensing technologies can be used for detecting human falls and alerting the emergency medical services to provide the required emergency help.

Aishwarya Laturkar

Computer Networks, Channel State Information, Machine Learning, Autoencoder, Softmax regression