基于行人与车辆关系模型的行人检测

摘要

For the pedestrians in traffic scenarios,and considering that the vehicles and pedestrians coexist in majority of traffic scenes, we put forward a method that detects vehicles on background image while carries outs preliminary pedestrian detection,and build a pedestrians and vehicles relationship model.We use the location of vehicle as the basis of auxiliary detection,introduce the true and false positive tests to exclude the pedestrians appearing in impossible region,and present the specific method.This method first defines pedestrians,vehicles,and the characteristics of pedestrians and vehicles relationship,and models them to form the correlated function relation.Then this is deduced to obtain the standard form suitable for support vector machine.Finally the method uses the support vector machine regression method to train the classifier for classification and recognition.Site measurement results show that this method greatly reduces the rate of error detection,and has good recognition effect on pedestrians on the pictures with different resolutions.%针对在交通场景下的行人，考虑到绝大多数交通场景中车辆与行人同时存在，提出一种在对背景图像进行初步行人检测的同时对车辆进行检测的方法，建立一种行人与车辆关系模型。以车辆位置作为辅助检测基础，引入真假阳性检验用以排除出现在不可能区域的行人并介绍了具体方法。该方法首先对行人、车辆、行人与车辆关系特征进行定义并建模形成与其有关的函数关系，然后推导得到适用于支持向量机的标准形式，最后通过支持向量机回归法训练分类器进行分类识别。现场实测结果表明，此种方法大大降低了误检率，对不同分辨率图片中的行人均有较好的识别效果。