Fall Detection Based on Colorization Coded MHI Combining with Convolutional Neural Network

Abstract

Fall detection is a matter of great concern in the field of public healthcare system especially for elderly. Video-based methods have been publicly acceptable methods, where frame sequences are used to extract feature for fall detection. Recently, deep learning models have been applied in the field of visual fall detection to avoid the impact of environmental constraints brought by hand-crafted features. However, most deep learning-based methods have simple inputs into the deep networks, which may make networks ignore detail information and weak features for detection. In this paper, we propose a fall detection method based on colorization coded motion history image (cc-MHI) combining with VGG16 neural network. First, a colorizaion coding method is applied to obtain cc-MHI, which not only contains the motion and shape information in the traditional MHI, but also enhances perceptual information and improves SNR to distinguish fall events from fall-like events more effectively. Then, the VGG16 neural network is employed to extract features from improved frame sequences with fall event automatically. Experimental results have shown the superiority of our method based on colorization coding technology and the fall detection accuracy has arrived at 92.34%.