中文摘要：

为提高前方车辆检测的实时性和准确率,提出了一种新型的分层HOG对称特征,并利用假设检验的方法对视频图像中的前方车辆进行检测。在假设阶段,利用前方车辆车底阴影的特点,提取满足特点的图像子区域,并将此区域作为假设车辆图像。在验证阶段,首先利用多层低维HOG特征来代替高维HOG特征,并对HOG特征进行对称化处理,得到分层HOG对称特征;然后将训练样本的分层HOG对称特征用于ELM（Extreme Learning Machine）分类器的训练;最后,用ELM分类器对假设车辆图像进行验证,得到检测结果。实验结果表明：该方法平均处理速度为25.5帧/s,验证准确率平均值达到了99.59%,比原始的HOG特征提高了4.01%。

英文摘要：

In order to improve the real-time and accuracy of preceding vehicles detection,a Layered HOG Symmetrical Feature（ LHSF） was proposed,and the method of hypothesis and verification to detect the vehicles in front of the camera was used. In the hypothesis stage,considering the characteristics of the shadows below the preceding vehicles,the image sub-region,which would be the hypothetic sub-images of the vehicles,was extracted. And in the verification phase,first,instead of the higher dimensional HOG,multilayer lower dimensional HOG was used to compute the HOG symmetrical vector,the layered HOG vector and the symmetrical vector together as a LHSF vector was mixed. Then,the LHSF vectors of samples were used for ELM（ Extreme Learning Machine） classifier training. At last,the sub-images with the ELM classifier to get the test results were verified. The experimental results show that the proposed method can handle 25. 5frame / s on an average,and the mean accuracy rate is99. 59%,which have a raise of 4. 01% than the rate of HOG.