Synthetic Data Generation of Car Driver's Head and Hands Motions Based on Human Tracking and Virtual Reality Technology

Aiming at safety car-driving support based on grasping of human driver's motion by tracking techniques such as particle filtering, ideal video data with annotation are not easy to obtain by manually annotating on the raw data due to human cost. To circumvent this difficulty, synthetic data generation method for car driver's head and hands motions has been proposed in this article. It utilizes tracking results of human driver's head and hands motions as input to a virtual humanoid, synthetic computer-graphics videos of driving scene in a car cabin have been generated based on virtual reality technology where virtual cameras capture face region and steering wheel region in the cabin. The pairs of the video and the input to the virtual humanoid constitute the annotated video data of car driver in the cabin. Demonstrative results of synthetic video data have been shown in the experimental section of this article along with technical details on the implementation.