摘要
针对视频实时人脸识别存在服务器开销过大的问题,研究了适合人脸识别的嵌入式人脸检测摄像机,首先改进JDA算法,解决了服务器与摄像机传输无效数据问题;然后将改进后的算法嵌入到摄像机内,实现人脸检测与人脸识别无缝对接.在"商超"场景,测试了嵌入式人脸检测摄像机性能,结果表明:嵌入式摄像机与非嵌入式摄像机对比,服务器携带的摄像机由原来的2路提高到8路,部署成本降低了30%;人脸识别速度提高了近2个百分点.
Aim at solving the problems of too much cost in real-time face recognition server,this paper studies an embedded camera with face detection,which is suitable for face recognition. First of all,the JDA（Joint Cascade Face Detection and Alignment） algorithm is improved,and the problem of invalid data sent from server to camera solved; then the improved algorithm is embedded into video camera to realize the seamless docking face detection and face recognition. Performance of the embedded camera in ＂business super＂scene has been tested,the result shows that the server can carry cameras from 2 to 8 road,with deployment costs reduction by 30 %. Face recognition speed is increased by nearly 3 %.
出处
《江苏科技大学学报（自然科学版）》
CAS
北大核心
2017年第6期781-786,共6页
Journal of Jiangsu University of Science and Technology:Natural Science Edition
基金
国家自然科学基金资助项目(61771265
61340037)
江苏省现代教育技术研究课题(2017-R-54131)
南通先进通信技术研究院开放课题(KPKT2016B06)
南通大学自然科学基金资助项目(14Z008)
南通大学大学生创新训练项目(2017115)
关键词
摄像机
人脸检测
人脸识别
嵌入式
JDA
real-time face recognition server
embedded camera
face detection
face recognition
JDA（Joint Cascade Face Detection and Alignment） algorithm
invalid data
