Short-Time Activity Recognition With Wearable Sensors Using Convolutional Neural Network

Human activity recognition is still a challenging problem in particular environment. In this paper, we propose a novel method based on wearable sensors to effectively recognize the short-time human activity. Our proposed method is based on two stages: First stage, constructing an over-complete pattern library which includes different patterns of short-time human activity. This library is produced by segmenting a long-time activity with sliding window method. Second stage, extracting robust features from an over-completed pattern library and establishing an off-line classification model through convolutional neural network (CNN). Consequently, an outstanding classification result on benchmark database WARD1.0 is successfully achieved based on the previous idea. Experimental results indicate that the proposed method is able to recognize the short-time human activity and at the same time satisfy the requirement of online recognition.