一种结合Faceboxes与KCF的多尺度人脸检测算法  

A Multi⁃scale Face Detection Algorithm Combining Faceboxes and KCF

作　　者：邹景阳 ZOU Jingyang(School of Mechanical Engineering,Shenyang University,Shenyang Liaoning 110044,China)

机构地区：[1]沈阳大学机械工程学院,辽宁沈阳110044

出　　处：《信息记录材料》2024年第6期16-19,共4页Information Recording Materials

摘　　要：针对车间内员工面部多尺度变化导致现有人脸检测算法效果不佳的问题,本文提出了一种结合人脸检测模型Faceboxes与核相关滤波(kernel correlation fliter,KCF)的人脸检测融合算法。该算法首先对Faceboxes网络结构进行优化,并引入多尺度融合技术以提升多尺度人脸检测的精度;其次,通过方向梯度直方图(histogram of oriented gradients,HOG)与局部二值模式(local binary patterns,LBP)特征融合优化KCF,增强目标跟踪能力;最后,将二者整合,形成“检测-跟踪-检测”的循环系统。实验证明,本算法在静态图像和动态视频中的检测效果和速度均佳,能满足实际需求,展现出良好的实用性。Aiming at the problem of poor effect of existing face detection algorithms due to the multi-scale changes of employees'faces in the workshop,this paper proposes a face detection fusion algorithm that combines the face detection model Faceboxes with Kernel Correlation Fliter(KCF).The algorithm firstly optimises the Faceboxes network structure and introduces a multi-scale fusion technique to improve the accuracy of multi-scale face detection.Secondly,the fusion of Histogram of Oriented Gradients(HOG)and Local Binary Patterns(LBP)is used to optimise the KCF and enhance target tracking ability.Finally,integrate HOG and LBP to form a"detection-tracking-detection"loop system.Experiment results have shown that the algorithm has good detection effect and speed in both static images and dynamic videos,which can meet the practical needs and show good practicality.

关 键 词：Faceboxes KCF 人脸检测 多尺度融合 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]