可扩展梯度直方图人体检测算法研究与实现

Research and Implementation of Human Detection Based on Extended Histograms of Oriented Gradients

人体检测已经成为机器视觉研究的一个热门课题,针对以梯度直方图作为人体特征描述的人体检测算法存在密集人群检测率较低这一问题。本文根据人体特征差异性,提出一种可扩展梯度直方图人体检测算法,使用非统一的区域方式提取图片梯度直方图描述算子,有效改善传统梯度直方图算法在密集人群检测中漏检率过高的情况。

Human detection has become a hot topic in the field of machine vision.Traditional histograms of oriented gradients（HOG） for human detection have a lower detection rate in dense population.To solve this problem,an extended HOG for human detection is proposed based on differences in human characterisics,which can extract the HOG descriptor of Image with different blocks.Experimental results show this method gives a better performance in dense population than the traditional method.