基于改进PBAS算法的级联特征行车检测

Vehicle detection using cascaded feature based on improved PBAS algorithm

随着车辆迅速增加,智能交通系统中的监控系统需要在复杂环境中快速、准确地检测车辆,在现有研究的基础上提出一种高效的车辆检测方案。首先选取像素自适应分割算法对其背景模型作线性优化,减少运算复杂度,提取前景斑点为定义区域;然后通过设定阈值确定感兴趣区域;在感兴趣区域里,选取哈尔(Haar-like)特征和方向梯度直方图特征,输入到优化后的AdaBoost+支持向量机(support vector machine,SVM)级联分类器中进行车辆检测。大量的实验证明了线性化像素自适应分割算法的优越性、AdaBoost+SVM级联分类器的快速性、整体车辆检测算法在检测车辆时的实时性和光照鲁棒性。

With the increasing number of the vehicles,the intelligent transportation system requires accurately and quickly detect vehicles in videos under complex conditions.Hence,this paper proposed an efficient vehicles detection scheme based on the existing works.Firstly,it selected a pixel-based adaptive segmentation algorithm to linearly optimize its background model,which could reduce the compute complexity and extract the foreground spot as defined range approach.Then,it used threshold determination to determine the ROI.In the ROI,it selected the Haar-like features and HOG features and used them as input of the optimized AdaBoost+SVM cascade classifier for vehicles detection.The substantial experiments demonstrate the superiority of the linearized pixel-based adaptive segmentation,the rapidity of the AdaBoost+SVM cascaded classifier,and the real-time processing ability and the illumination robustness of the overall vehicle detection algorithm in detecting vehicles.