ドライバ姿勢と動作のマルチタスク学習による高速かつ高精度なドライバ動作認識

抄録

Driver Action Recognition is a key component in driver monitoring systems, which is helpful for the safety management of commercial vehicles. Compared with traditional human action recognition tasks, driver action recognition is required to be fast and accurate on embedded systems. We propose a fast and accurate driver action recognition method that is composed of CNN based driver pose estimation and RNN based driver action recognition. We train our network model with multi-task learning includes localizing and detecting each body part of the driver, classifying state of each body part, and recognizing driver action at once. Our multi-task learning for the proposed model achieves a significant improvement compared to state-of-the-art human action recognition methods with limited computational resources. We also perform ablation study of our methods which composed of the driver pose localization, detection, and classification.