华中科技大学学报（自然科学版）2025，Vol.53Issue(5)：58-64,7.DOI:10.13245/j.hust.250046
基于姿态约束特征增强的人脸关键点检测算法
Facial landmark detection algorithm based on pose constraint and feature enhancement
摘要
Abstract
To solve the large pose problem in facial landmark detection and achieve a balance between accuracy and computational cost,a three-stage lightweight network based on pose constraint and feature enhancement was proposed.A lightweight backbone network was adopted in the first stage.A feature pyramid based on spatial-channel aware module was proposed in the second stage to help the model perceive facial geometric relationships.A feature enhancement network was used in the third stage to enhance features and maintain high-resolution representation,which was beneficial for improving the accuracy of heatmap regression.The detection head combined coordinate regression and heatmap regression to simultaneously predict heatmap scores and coordinate offsets,helping to improve the accuracy of the model.Subsequently,a loss function integrating the data-balanced pose factor and the yaw factor was proposed for pose constraint.For the data-balanced pose factor,the samples were first projected into one dimension through procrustes analysis and principal component analysis to obtain pose coefficients,and then the interval proportion and pose coefficients were used to derive the factor.For the yaw factor,a yaw degree calculation formula was designed for computation.Finally,experimental results show that the proposed algorithm achieves a balance between accuracy and computational cost,and effectively improves the detection accuracy under large poses.关键词
人脸关键点检测/轻量级网络/空间通道感知模块/特征增强/姿态约束Key words
facial landmark detection/lightweight network/spatial-channel aware module/feature enhancement/pose constraint分类
信息技术与安全科学引用本文复制引用
谢佳,邹腊梅,钟胜..基于姿态约束特征增强的人脸关键点检测算法[J].华中科技大学学报（自然科学版）,2025,53(5):58-64,7.基金项目
国家自然科学基金面上项目(62176100). （62176100）
