基于视频的人脸检测算法研究

Research on Face Detection Algorithm Based on Video

针对传统AdaBoost算法在视频中检测人脸误检率较高的问题,文中提出了一种结合运动分析和肤色检验的改进型人脸检测算法。该方法通过运动检测来提取运动前景,并选择肤色模型对人脸肤色进行相似度求取,利用几何特征进一步缩小检测范围;采用增加新Haar特征和改进权重更新方式的改进型AdaBoost算法对人脸候选区域进行实时检测。实验结果表明,与传统AdaBoost方法和增加肤色检验的AdaBoost方法相比,该方法的误检率分别降低了18.68%和8.79%,检测时间则分别缩短约800 ms和250 ms。

Aiming at the shortcomings of the traditional AdaBoost algorithm in detecting high false detection rate in video,an improved face detection algorithm combining motion analysis and skin color detection was proposed.The proposed method extracted motion prospects through motion detection,and selected the skin color model to obtain the similarity of the face skin color,and further narrowed the detection range by using the geometric features.The face candidate region was detected in real time detection the improved AdaBoost algorithm which was added with new Haarfeatures and improved weight update mode.The experimental results showed that compared with the traditional AdaBoost method and the AdaBoost method with skin color detection,the false detection rate of this method was reduced by 18.68%and 8.79%respectively,and the detection time was shortened by about 800 ms and 250 ms,respectively.