基于特征点动态选择的三维人脸点云模型重建

针对典型的点云配准方法中伪特征点过多导致配准效率低和配准结果不精确的问题,提出一种基于特征点动态选择的三维人脸点云模型重建方法。该方法在粗配准阶段,采用动态特征矩阵求解法获取粗匹配特征变换矩阵以避免伪特征点的干扰。在精配准过程中,采用二次加权法向量垂直距离法在人脸流形表面选择更有效的特征点以减少伪特征点的数量,并采用基于特征融合与局部特征一致性的迭代最近点方法进行精配准。经过对比实验验证了算法的可行性,实验结果表明,提出算法能够实现高精度且快速的三维人脸点云模型重建,且均方根误差达到1.816 5 mm,相较于其他算法,在模型重建精度和效率方面都有所提升,具有良好的应用前景。

3D face point cloud model reconstruction based on dynamic selection of feature points

Aiming at the problems of excessive pseudo feature points can lead to low registration and imprecise registration results in typical point cloud registration methods, this paper proposed a 3D face point cloud model reconstruction method based on dynamic selection of feature points. This method used the dynamic feature matrix solving method to obtain the coarse matching feature transformation matrix during the coarse registration phase, which avoided the interference of false feature points. In the fine registration process, it used the second-order weighted normal vector perpendicular distance method to select more effective feature points on the face manifold surface to reduce the number of false feature points. And it also used the iterative closest point method based on feature fusion and local feature consistency for fine registration. Comparative experiments verified the feasibility of the algorithm. Experimental results show that the proposed method can achieve high-precision and fast reconstruction of three-dimensional face point cloud model. And the root mean square error reaches 1.816 5 mm. Compared with other algorithms, the proposed algorithm can improve the accuracy and efficiency of model reconstruction, and has good application prospects.