基于GPU的并行人脸检测方法研究  

Research on Parallel Face Detection Method Based on GPU

作　　者：龚强[1] 谢凯[1] 贺建彪 阮宁君[1] 

机构地区：[1]长江大学电子信息学院,湖北荆州434023 [2]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《长江大学学报（自科版）（上旬）》2016年第9期31-35,4-5,共5页JOURNAL OF YANGTZE UNIVERSITY (NATURAL SCIENCE EDITION) SCI ＆ ENG

基　　金：国家自然科学基金项目(61272147);湖北省教育厅指导性项目(B2015446)

摘　　要：为了进一步提高人脸检测的精度和速度，提出了一种改进的Viola—Jones人脸检测算法，并在GPU平台上对改进后的算法加以优化。采用新型稀疏特征来进行人脸检测，利用计算机的图形处理单元GPU进行并行加速，并对优化前后的结果进行对比。结果表明，该算法把检测率提高到93．6％，与传统Viola-Jones人脸检测算法相比检测精度有了较大的提升；GPU并行加速实现了对OpenCV1．6～3．2倍左右的加速比，有效提高了人脸检测性能。In order to improve the accuracy and the speed of face detection,an improved Viola-Jones face detection algorithm is proposed,and the improved algorithm is optimized on the GPU platform.In this paper,a new type of sparse feature is used to detect the human face,and the GPU is used for parallel acceleration.The results before and after optimization are compared.Experiment results show that the detection rate of the improved algorithm increases to 93.6%,and has a greater improvement in detection accuracy compared with the traditional Viola-Jones face detection.GPU parallel acceleration is 1.6to3.2times faster than OpenCV,which effectively improves the performance of face detection.

关 键 词：人脸检测 Viola—Jones算法 GPU并行 OPENCV 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]