基于DM642的多姿态人脸检测实时系统  

Multi-Posture Face Detection Real Time System Based on DM642

作　　者：王平[1] 邹暘[1] 熊殷[1] 

机构地区：[1]南昌大学信息工程学院,江西南昌330031

出　　处：《电视技术》2013年第5期179-182,共4页Video Engineering

摘　　要：为满足在嵌入式系统领域上进行人脸检测的需要,采用Gentle Adaboost人脸检测算法设计基于Haar特征的级联型人脸分类器。提出并设计了基于DM642的人脸检测实时系统,以及工程与代码上的优化方案。优化后的系统进行人脸检测的时间为67 ms,在系统上达到了13 f/s(帧/秒)的速度,并保证了较高的检测率,验证了在DSP系统上进行人脸检测的可行性。To meet the requirement of the face detection in the field of embedded system, the cascaded boosted classifiers based on Haar feature is learned by gentle Adaboost algorithm. The real time multi-posture face detection system is proposed and designed based on DM642 and the prioritization scheme on the project and codes. The time of face detection on optimal system is 67 ms , resulting in 13 frames per second, and the high detection rate is guaranteed, both of which verify the feasibility of face detection implementing on DSP system.

关 键 词：DSP 人脸检测 GENTLE ADABOOST HAAR特征 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]