基于可变形部件模型的驾驶员人脸检测

Driver Face Detection Based on Deformable Part-Based Model

交通监控中车辆驾驶室内环境较为复杂,如光线暗、遮挡、分辨率低等,现有的人脸检测方法效果不佳.提出了一种基于可变形部件模型的驾驶员人脸检测方法.通过提取聚合通道特征(局部二值模式和梯度方向直方图),得到候选人脸目标.基于监控图像中车牌与驾驶员人脸的相对位置存在比较固定的模式,将车牌与驾驶员人脸看作是可变形部件模型中的两个部件,用于验证车牌和候选目标相对位置关系的合理性,从而确定驾驶员人脸的位置.实验结果表明提出的方法提高了检测准确率和综合性能指标,有效地滤除了人脸虚警,且召回率影响较小.

To solve the problem of detecting driver faces from cabs images taken by traffic cameras,a driver face detection method was proposed based on deformable part-based model,overcoming the condition influence such as dim light,occlusion and low resolution in cabs.Firstly,extracting aggregate channel features（local binary pattern and histogram of oriented gradient）,the candidate faces were obtained.Then,considering the relative settled position between the license plate and driver face,the driver face and plate were taken as two deformable parts of a faceplate couple based on the concept of deformable part-based model,and the ubiety of two parts was used to determine the position of the candidate face.Experimental results show that the proposed method can improve the detection accuracy and overall performance,effectively filter out the false face alarm,and the recall rate is less affected.