Fall Detection Using a Multistage Deep Convolutional Network Architecture

摘要

Fall detection is a major challenge in the field of public healthcare, especially for the elderly. And reliable surveillance is critical to mitigate the incidence rate of falls. In this paper, we propose a multistage architecture to obtain the human pose estimation. The proposed network architecture contains two branches. The first branch is the confidence maps of joint points; the second branch proposes a bi-directional graph structure information model (BGSIM) to encode the rich contextual information. Then we define a linear function to determine whether the people (especially the elderly) fall or have tendency to fall. We test the system in a simulated environment, such as a bathroom, a kitchen, and a hallway. Meanwhile, we also give some prediction results from real scenes.