A Privacy Protected Fall Detection IoT System for Elderly Persons Using Depth Camera
Xiangbo Kong, Zelin Meng, Lin Meng, Hiroyuki Tomiyama · 2018
The proportion of the elderly persons in the world is constantly on the rise, and fall accidents have become a serious problem, especially for those who live alone. Currently, fall detection has attracted a lot of research attention and machine learning (ML) has shown promising performance in this task due to their strengths in person recognition. However, many existing methods using RGB images as the training data, resulting in the main information to be lost, or do not appropriately consider the effect of light, resulting in weak generalizability of the fall detection. Moreover, traditional methods pose a risk of leakage of personal privacy. This paper proposes a fall detection IoT system based on depth camera and fast Fourier transform (FFT) to overcome these problems. We first use depth camera to get the skeleton images of a person who is standing or falling down. We then get the characteristic quantity of these images and train them by ML to get the training model. Finally, we use FFT to encrypt images and detect the fall. We constructe a training database that includes 1131 images, and the experimental evaluation of the images demonstrates that our algorithm is effective for detecting falls and maintain privacy.