一种Haar-like和HOG特征结合的交通视频车辆识别方法研究  

Vehicle recognition method based on Haar-like and HOG feature combination in traffic video

作　　者：董天阳[1] 阮体洪 吴佳敏[1] 范菁[1] 

机构地区：[1]浙江工业大学计算机科学与技术学院,浙江杭州310023

出　　处：《浙江工业大学学报》2015年第5期503-507,共5页Journal of Zhejiang University of Technology

基　　金：浙江省重大科技专项项目(2013C01112;2012C01SA160034);杭州市重大科技创新专项(20132011A16)

摘　　要：由于前向和后向车辆的表观特征不同,单纯使用主流的HOG或者Haar-like特征来识别车辆会存在对某一方向行驶的车辆识别率低或者误识率高的问题.针对上述问题,提出了一种Haarlike和HOG特征结合的交通视频车辆识别方法.在训练阶段,对前后向车辆分别采用Haar-like和HOG特征来提取车辆特征,引入反馈式的AdaBoost算法训练车辆分类器,提高车辆识别的速度以及准确率;在识别阶段,根据车辆运行状态确定前后向车辆,再利用对应的车辆分类器进行多尺度遍历识别.在不同光照强度的高速公路视频中进行车辆识别实验,前后车辆的平均识别率达到93%,误识别为9%.Since apparent characteristics of forward and backward moving vehicle are different,only using HOG or Haar-like features to recognize vehicle will result in lower recognition rate or higher error rate in one direction.This paper presents a vehicle recognition method based on Haar-like and HOG feature combination in order to improve the recognition rate.During training phase,the Haar-like for the frontward vehicle and HOG for the backward vehicle are used to extract vehicle features respectively.The feedback AdaBoost algorithm is used to train vehicle classification.In recognition phase,the direction of vehicle is judged by the vehicle moving status and corresponding vehicle classification is used to traverse and identify vehicle in multiscale.The vehicle identification experiment is tested in the highway video at different light intensities..The average recognition rate reaches 93% and error rate is 9%.

关 键 词：特征结合 前后向车辆识别 HOG Haar-like ADABOOST 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]