应用姿态估计的人脸特征点定位算法

摘要

针对已有鲁棒级联姿势回归算法缺少形状约束条件的现状,同时在复杂人脸及遮挡情况中定位精度较低、成功率不高等问题,提出应用姿态估计人脸特征点的新型定位算法来提高定位精度和成功率.对人脸特征点执行区域分块操作来实现形状约束条件;为提高算法性能,对部分特征点位置执行回归操作从而降低回归器规模,并引入形状索引特征进行采样先验操作.实验结果表明,所提算法针对复杂人脸及遮挡情况具备较高定位精度与鲁棒性,同时算法速度可达实时要求.%Aiming at the problem that the existing robust cascade postural regression algorithm lacks shape constraint,and has low localization accuracy and unsatisfactory success rate in complex face and occlusion situations,a novel positioning algorithm for pose estimation of facial feature points was proposed to improve the accuracy and success rate.A regional block operation was performed on face feature points to implement shape constraint.To improve the algorithm performance,a regression operation was performed on partial feature point positions to reduce the scale of regression,and the shape index feature was introduced to sampling prior operation.The experimental results show that the proposed algorithm has higher localization accuracy and robustness for complex face and occlusion,and meets the realtime requirement.