基于监控视频的稳定实时人脸跟踪系统构建  

Robust and real-time face tracking system construction based on surveillance video

作　　者：李明夏 陈斌[1,2] 

机构地区：[1]中国科学院成都计算机应用研究所,成都610041 [2]中国科学院大学,北京100049

出　　处：《计算机应用》2015年第A02期250-253,257,共5页journal of Computer Applications

摘　　要：针对基于可变形模板的人脸跟踪算法容易出现的目标丢失问题,提出了基于等间隔的矫正算法和基于变化程度的矫正算法,在特定的时机利用人脸检测算法重新定位人脸,对跟踪过程加以矫正,提升了跟踪的鲁棒性。为了保障跟踪的实时性,人脸检测采用了Haar-like特征的Adaboost算法。在跟踪质量方面,以跟踪结果中目标人脸的可辨识情况来评判算法的准确度。实验结果表明,矫正后的算法在跟踪的稳定性和跟踪结果的质量上都比原有跟踪算法优越。Using the deformable template technique to track human faces may lose target frequently. To solve this problem,two rectifying methods based on equidistant frame and change rate of aspect ratio of tracking frame were proposed.The robustness was improved after rectifying through relocating human face at specified time. To ensure efficiency of the tracking system,Adaboost algorithm based on Haar-like features was used to detect human faces. The accuracy of these algorithms was evaluated by ratio of the number of faces that can be recognized from the tracking result to total number of result. The experiment result shows that the proposed method is superior to the unrectified method in both robustness and accuracy.

关 键 词：可变形模板 人脸跟踪 反馈调节 视频监控 HAAR-LIKE特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]