A Single Depth Sensor Based Human Activity Recognition via Convolutional Neural Network

Human activity recognition (HAR) has become an active research topic in the various fields. Depth sensor-based HAR recognizes human activities using features from depth human silhouettes via classifiers such as Hidden Markov Model (HMM), Conditional Random Fields Model etc. In this paper, we propose a new HAR system via Convolutional Neural Network (CNN), one of deep learning algorithms. We extract joint angles from multiple body joints changing in time and create a spatiotemporal feature matrix (i.e., multiple body joint angles in time). With these derived features, we train and test our CNN for HAR. In order to evaluate our system, we have compared the performance of our CNN-based HAR against the HMM- and Deep Belief Network (DBN)-based HAR using a database of Microsoft Research Cambridge-12 (MSRC-12). Our test results show that the proposed CNN-based HAR is able to recognize twelve human activities reliably and it outperforms the HMM- and DBN-based systems. We have achieved the average recognition accuracy of 98.59% for the activities. The results are 6.1% more accurate than that of the HMM-based HAR and 1.05% more accurate than that of the DBN-based HAR.