基于FPGA的高精度人脸检测加速  

High Accuracy Face Detection Acceleration Based on FPGA

作　　者：刘洪浩 李启海[2] 彭其飞 LIU Hong-hao;LI Qi-hai;PENG Qi-fei(No.91049 Army,Qingdao 266103;Changchun University of Science and Technology,Changchun 130022;China FAW Group Corporation Weishan Factory,Changchun 130012)

机构地区：[1]91049部队,青岛266103 [2]长春理工大学,长春130022 [3]中国第一汽车集团有限公司蔚山工厂,长春130012

出　　处：《长春理工大学学报（自然科学版）》2020年第3期90-94,101,共6页Journal of Changchun University of Science and Technology(Natural Science Edition)

摘　　要：针对人脸检测对高实时性的要求,系统测试了Viola-Jone算法的数据表示(定点/浮点)对检测假阳性和假阴性的影响,检测在逐步降低数据精度的基础上进行。最后在对定点检测评价的基础上提出了一种基于CPU和FPGA混合执行Viola-Jone算法的框架,充分利用FPGA的并行性,对定点表示的Viola-Jone算法进行显著加速处理,极大的提高了人脸检测速度。Aiming at the high real-time requirement of face detection,this paper systematically tests the influence of the data representation(fixed-point/floating-point)of Viola-Jone algorithm on the detection of false positive and false negative,and detects the face on the basis of gradually reducing the accuracy of the data.Based on the evaluation of fixed-point detection,a frame of Viola-Jone algorithm based on CPU and FPGA is proposed,which makes full use of the parallelism of FPGA to speed up the Viola-Jone algorithm.As a result,the speed of face detection is greatly improved.

关 键 词：人脸检测 Viola-Jone算法 FPGA 加速 

分 类 号：O482.31[理学—固体物理]