The Driver Head Pose Estimation Method Based On Depth Image

In recent years, with the development of economic, the quantity of vehicles grows rapidly. Besides the transportation convenience for our daily life, the frequency of traffic accidents is also increasing. According to the research, besides objective factors of force majeure, the incidence of traffic accidents is mainly due to a lack of focus on driving. If some kinds of traffic auxiliary system can be applied to monitor the driver’s behavior, some accidents caused by the lack of focus can be reduced. The head pose estimation technique plays a critical role in investigating the driver’s observing orientation, testing the driver’s attention focusing or not, and improving the driver’s safety, which is a popular subject of extensive research in the world.Nowadays, some existing methods on driver head pose estimation have already achieved certain result. However, during the driving, there are some unavoidable factors, such as light, shelter and so on, in which case the simple 2D RGB image cannot meet the actual needs.To improve the accuracy of diver head pose estimation, a new estimation algorithm is proposed in this paper, which combines the 2D RGB image with the depth image. The face location is determined by the active appearance model. Then the part of face is transferred to the rigid head point cloud and the head pose is analyzed roughly using the ICP algorithm. Through the self-learning algorithm, a template of nine head pose zones is set regarding the driver’s gaze zone in the front of the car as the referenced area. Finally, the gaze zone and its neighborhood area in the template are applied to estimate the driver’s current head pose with particle filter tracking algorithm. More accurate estimation of driver’s head pose is realized.The experiment result shows that the algorithm in the paper can estimate the driver’s head pose well, and the accuracy is more than 85%. The data result all accords with the standard of driving behavior. So this method can meet the need of the safety driving assist system and safety driving behavior.