低照度下运动车辆的检测方法  

Moving vehicle detection in low light conditions

作　　者：王海晖[1,2] 刘洋[1] 陶玲[1] 

机构地区：[1]武汉工程大学计算机科学与工程学院,湖北武汉430205 [2]武汉工程大学智能机器人湖北省重点实验室,湖北武汉430205

出　　处：《武汉工程大学学报》2013年第10期41-45,共5页Journal of Wuhan Institute of Technology

基　　金：湖北省教育厅科学研究项目计划重点项目(D20111509);武汉工程大学研究生教育创新基金项目(CX201273)

摘　　要：针对一般车辆检测系统在夜间等低照度情况下检测能力急剧下降的问题,提出一种低照度下有效的基于Adaboost算法的检测方法.该方法通过提取训练样本的梯度方向直方图(HOG)特征,训练一个可用于分类的二类分类器,并采用Gamma校正法对当前帧作图像亮度处理,来降低低照度对特征提取的影响;再通过载入分类器在校正后的当前帧内作多尺度检测,最后用矩形框标识出检测出来的目标,统计车辆数目.实验表明,该方法对交通路口的车辆有较好的实时检测效果,在夜间、雾天及其他能见度低的天气等低照度状况下,也能保持较高的检测率.As the general vehicle detection systems＇ detection ability declines sharply in the night and low light conditions,this paper presents a detection system based on Adaboost algorithm.A two class cascaded classifier was trained by using Histogram of Oriented Gradients（HOG） features of the vehicle samples.The current frame was adjusted by the Gamma correction to reduce the effect of low light intensity on feature extraction.Then the classifier was loaded to do a multi-scale detection in the adjusted current frame.Eventually the vehicles were counted by marking the detection results with rectangles.The results of experiments show that this system is capable of detecting the vehicles in effective area of the traffic intersection,and can keep a high detection rate in dark scene such as nights,foggy days and other low visibility weather.

关 键 词：车辆检测 低照度 ADABOOST算法 梯度方向直方图特征 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]