基于BING似物性检测的行人快速检测算法*

针对传统滑动窗行人检测速度慢、实时性差的问题,提出了一种基于似物性的行人快速检测算法。首先,算法通过提取正负训练样本的规范化二进制梯度特征,训练级联SVM分类器得到行人似物检测模型。然后利用尺寸调节和聚类算法对初始候选区域进行聚类融合,进一步优化行人候选窗口区域。最后,提取各候选区域的HOG特征并利用SVM分类器对其进行进一步行人检测。实验结果表明：本算法在保证行人检测率的同时在检测实时性上有明显提高。

Fast Human Detection Algorithm Based on BING Objectness

In order to solve the poor real-time performance of traditional pederstrian detection algorithm,A new fast human detection method based on binarized normed gradient objectness was proposed in this paper.First, Binarized normed gradients features were extracted from the training image sets to train a cascade SVM classifier,which was used to extimate the objectness value of some areas of an image.Secondly,sizes of the original candidate boxes produced by the first step were corrected and a rectangle cluster algorithm was applied to those boxes to improve the quality of the initial candidate boxes.Finally,HOG features were extracted from those candidate areas and SVM classifier were used to check there was pedestrian in these boxes or not. Experimental results demonstrate that the proposed algorithm outperforms the traditional detection models in detection speed while maintains the comparable detection rate to the traditional algorithm.