融合表观与曲率特征的三维点云眉毛提取

3D Point Cloud Eyebrow Extraction Based on Apparent and Curvature Features

针对三维点云数据中眉毛区域提取研究较少的问题,提出了一种融合表观与曲率特征的方法。首先在3D-2D投影的基础上提出新的3D-2D-3D框架,借助二维关键点实现三维眉毛关键点定位;然后利用眉毛关键点构建感兴趣的区域ROI(region of interest),以曲率说明眉毛的形状,并基于表观特征提取三维眉毛区域。在DMCSv1数据库的三维点云数据集上,针对不同表情、不同眉毛形状的眉毛进行提取,实验结果验证了本文所提方法的可行性。

There are few studies on the extraction of eyebrow area in 3D point cloud data,This paper proposed a method to fuse the apparent and curvature characteristics.Firstly,a new 3D-2D-3D framework was proposed on the basis of 3D-2D projection,and 3D eyebrow landmarks locating is realized by using 2D key points.Then ROI was built by using landmarks of eyebrows,the shape of eyebrows was explained by curvature,and 3D eyebrow regions were extracted based on the appearance characteristics.On the 3D point cloud data set of DMCSv1 database,eyebrows with different expressions and different eyebrows shapes were extracted,and the experimental results verified the feasibility of the method proposed in this paper.