中文摘要：

驾驶员头部姿态跟踪是车辆辅助驾驶系统中的关键问题之一，文中提出了一种基于3D人脸模型的驾驶员头部姿态鲁棒跟踪算法；首先，将3D人脸模型映射到第一帧图像中，获取到脸部区域及估计出初始姿态；然后，在脸部区域中跟踪并检测特征点，并把匹配结果作为基于模型的光束法平差机制的输入来恢复出3D人脸模型的头部姿态；为提高检测精度，在每帧脸部区域内重新提取特征点用于跟踪；实验结果表明，该算法在部分遮挡及大幅转动时是有效的。

英文摘要：

This paper present a robust 3D model--based head pose tracking method in vehicle environments. First of all, the 3D face model is projected onto the first face image, then find out the face region and estimate the head pose. Then, after features are selected in the face region, a model--based bundle adjustment mechanism is implemented to automatic recovery head pose of the 3D face model： In order to enhance the detective accuracy, we reselect features in the face region of each frame. Experimental results show that the algorithm is capable of dealing with partial occlusion, large rotation.