计算机应用

• 人工智能与仿真 •    下一篇

CCFAI2017 193 应用姿态估计人脸特征点的定位算法研究

张海艳   

  1. 淮阴工学院
  • 收稿日期:2017-07-10 发布日期:2017-07-10 出版日期:2017-11-10
  • 通讯作者: 张海艳

CCFAI2017 193 The Research on Facial Feature Points Localization Algorithm by Using Pose Estimation

  • Received:2017-07-10 Online:2017-07-10 Published:2017-11-10

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、成功率不足等已有问题,本文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。针对人脸特征点执行区域分块操作来进行形状约束条件实现;为提高算法性能,针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。本文实验结果表明本文算法针对复杂人脸及遮挡情况具备较高定位精度与算法鲁棒性,同时算法速度可达实时要求。

Abstract: According to the status of the existing robust regression algorithm of cascade lack of shape constraints and the low success rate in middle low positioning precision in the complex face and occlusion less than the existing problems, the novel positioning algorithm for attitude estimation of facial feature points was proposed to improve the precision and success rate. In order to improve the success rate and positioning accuracy, the facial feature points block operation has to implement regional shape constraints to achieve operation. To improve the performance of the algorithm, some feature points so as to reduce the operation to perform regression regression for scale and the introduction of shape index were sampled prior operation. The experimental results has shown that the proposed algorithm has higher localization accuracy and robustness for complex face and occlusion, and the computational speed of the algorithm can meet the real-time requirements.

中图分类号: