基于PHOG特征的行人检测算法研究

Pedestrian detection algorithm based on PHOG feature

摘要:

HOG特征对行人轮廓有很好的描述能力，但基于HOG特征的行人检测存在检测速度慢、漏检率较高的问题，使得该算法的实践应用范围受限。本文针对检测速度慢、漏检率较高的问题，提出了一种基于PHOG特征的行人检测算法。首先，提出了PHOG特征，该特征对cell内的梯度特征进行强化，增大了目标与背景的梯度分布区别，从而使目标更容易被分类器学习和识别。然后提出了构建特征金字塔的方法，并对PHOG特征进行有效地降维，大幅度减少了检测时间。试验结果表明，本文提出的PHOG-PCA特征将漏检率从35%降到了22%，检测速度也比一些流行算法快。

Abstract:

The HOG feature has a good description of the pedestrian profile, but the pedestrian detection based on the HOG feature has the problems of slow detection speed and high missed detection rate, which makes the practical application of the algorithm limited. In this paper, aiming at the problem of slow detection rate and high missed detection rate, a pedestrian detection algorithm based on PHOG features is proposed. Firstly, the PHOG feature is proposed. This feature enhances the gradient features in the cell and increases the gradient distribution difference between the target and the background so that the target can be easily learned and identified by the classifier. Then, a method of constructing characteristic pyramid is proposed, and the PHOG features are effectively reduced in dimension, which greatly reduces the detection time. The experimental results show that the proposed PHOG-PCA feature reduces the missed detection rate from 35% to 22% and the detection speed is faster than some popular algorithms.