Human Learning Behavior Recognition in Seated Posture Using Millimeter-Wave Radar

The collection and analysis of student behavior data play a crucial role in providing educators with insights into the learning process, enabling them to optimize teaching strategies. Currently, visual sensors are predominantly used for student behavior recognition. However, these sensors pose privacy risks and may not be suitable for home environments due to concerns over personal data leakage. In this paper, we propose a learning behaviors recognition method based on local and global features in seated posture using mmWave radar. The behaviors identified by our system include reading, writing, playing mobile games, sleeping, and using the computer. We firstly perform local-limbs classification on the point cloud data. Then we extract global features by utilizing the relative motion relationships between limbs, and local features from the left and right hands. Finally, we perform an LSTM network with the extracted global-local features to classify the behaviors. Experimental results demonstrate that our proposed method achieves an average accuracy of 90.29 % across five types of behaviors.