高铁闸机智能监控系统中的行人检测算法研究

摘要

Aiming at high requirements of real-time on pedestrian detection system of turnstile' s smart surveillance in high-speed rail, we propose an improved pedestrian detection method which is based on histograms of oriented gradients ( HOG) features and AdaBoost classification. First, the features of HOG are extracted from sample images to generate a strong classifier with high detection rate after classifying and training them with Gentle AdaBoost; then the vertical edges of the testing images are pre-processed and a large number of windows without pedestrian will be eliminated according to the edges symmetry property features between the images with and without pedestrian; Finally the HOG features are extracted from the remnant windows and to be detected using the trained AdaBoost classifier to estimate whether the pedestrian can be found out from the windows. Experimental result shows that, the improved pedestrian detection algorithm has less computation load than the former algorithms. And it is faster in image detection as well with same accuracy rate as the former algorithms, therefore the real-time requirement on pedestrian detection system of turnstiles in high-speed rail is able to be satisfied.%针对高铁闸机智能监控行人检测系统中较高的实时性要求,提出一种改进的基于梯度直方图(HOG)特征与AdaBoost分类的行人检测算法.首先对图像样本提取HOG特征,进行Gentle AdaBoost分类训练,得到高检测率的强分类器；然后对待测图像进行垂直边缘预处理,根据行人图像与非行人图像的边缘对称性特征,排除大量非行人窗口；最后对剩余窗口提取HOG特征,依据训练出的AdaBoost分类器检测HOG特征向量,判断窗口是否含有行人.实验结果表明:改进的行人检测算法比原算法计算量少,能够在保证原有准确率的基础上,对图像进行更快的检测,满足高铁闸机行人检测系统的实时性要求.