利用三维网格模型的头部运动估计

Head motion tracking using 3D wire-frame model

头部三维运动跟踪是人机交互、虚拟现实及模型基编码等领域研究的重要内容。文中提出了一种借助合成模型估计头部三维运动的算法。首先由正交图像建立输入脸部模型 ,将其变换为初始图像人脸的姿态和大小 ;之后选取图像中 3个特征点和对应的模型点 ,利用块匹配确定图像中 3个特征点的运动 ,由 2 D到 3D的对齐算法 ,求出模型的三维运动参数 ;最后由运动后合成模型与图像的小范围相关检测 ,进一步确定最后的模型运动参数。实验表明该算法建模设备要求简单 ,速度较快 。

D head tracking is an important part of researches on man machine interaction, virtual reality, model based image coding and so on. An algorithm of head motion tracking based on 3D wire frame model is proposed in this paper. At first, the special facial model of the input face is built from orthogonal images and changed to the input face's scale and pose. Secondly, three feature points are chosen on image frame that their motion are determined by block matching, and the 3D model's motion is resumed by 2D 3D aligning algorithm through three model points' correspondence with the three feature points. Finally, the model's motion is refined by correlation matching of synthesis face and input face image. The result using this algorithm shows a good estimation of 3D motion with simple devices and a short time.