车辆行车实时目标区域特征提取及分类训练  

Target Region Feature Extraction and Classification Training of Drive Environment

作　　者：王全[1] 王长元[1] 穆静[1] 荆心 

机构地区：[1]西安工业大学计算机科学与工程学院,西安710021

出　　处：《西安工业大学学报》2015年第11期888-892,共5页Journal of Xi’an Technological University

基　　金：陕西省教育厅科研计划项目(14JK1347);西安工业大学校长基金(0852-302020615)

摘　　要：为了实现前方车辆智能实时检测,文中提出了一种基于梯度方向直方图特征和支持向量机的前方车辆检测的区域检测算法.采用一种改进Hough变换直线检测算法提取图像感兴趣区域;对图像感兴趣区域的梯度方向直方图(HOG)特征进行了提取,利用支持向量机(SVM)训练得到的分类器进行前方车辆的检测.实验结果表明:文中提出的区域检测算法能有效实时、准确检测前方车辆;车辆检测用时较梯度方向直方图和支持向量机直接检测算法降低了30%,识别率提高了1.6%.For the real-time forward intelligent vehicle detection,the paper presents a region detection algorithm based on the histogram of oriented gradient feature and the support vector machine. An improved Hough transform line detection method is used to extract the target area of the image. The histogram of oriented gradient feature in the region is extracted and the classifier obtained by support vector machine training is used to detect the vehicles ahead. The experimental results show that the region detection algorithm can detect vehicles ahead real-time and accurately. Compared with the direct HOG＋SVM detection algorithm,the vehicle detection time is reduced by 30 %,and the recognition rate is improved by 1.6 %.

关 键 词：车辆实时检测 梯度方向直方图(HOG) 特征提取 HOUGH变换 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]