基于感兴趣区域AdaBoost分类器的视频车辆检测研究

近年来基于视频的车辆自动检测作为城市智能交通系统的一项重要技术一直受到关注.针对AdaBoost分类器目标检测所存在的漏检、误检和计算量过大等问题,提出一种基于混合高斯模型运动区域提取和Haar-like特征的AdaBoost级联分类器的交通视频车辆检测算法,首先通过建立混合高斯模型对运动目标的总体区域进行检测,进而提取基于车辆运动的感兴趣区域,再对其进行基于Haar-like特征的区域AdaBoost级联分类,实现对运动车辆的检测.由于采用了基于运动区域提取和分类相结合的检测模式,通过混合高斯背景模型较准确的提取出ROI作为车辆的候选区域,约束了每帧的搜索区域,使AdaBoost分类器的目标检测更具针对性,提高了检测的准确性,降低了漏检率;同时也减少了分类算法滑动窗口扫描所需要的时间,提高了检测速度.实验结果验证了所提出算法对复杂交通环境车辆检测的适应性和有效性.

Research on video vehicle detection based on AdaBoost classifiers of the ROI

Recently ,video-based automatic vehicle detection as a key technology of the urben intelli-gent transportation system has got more attention .As there are missing detections ,false detections , and large amount of calculations of AdaBoost classifier ,this paper proposes a detecting algorithm of video-based vehicle based on the motion region extraction of Gaussian mixture model and AdaBoost cascade classifier w hich has the haar-like features .Firstly ,the proposed paper detects the overall area of moving targets by Gaussian mixture model to extract the region of interest （ROI） of vehicle move-ment .Then ,it achieves the detection of the moving vehicles based on the AdaBoost cascade classifi-er .Because of the application of the detection mode ,which is based on the extraction and classifica-tion of the motion region ,the Gaussian background model can extract the ROI as the candidate re-gion for vehicles accurately .The proposed method restrains the search area of each frame ,which makes the target detection by AdaBoost classifier more specifically .In addition ,it improves the accu-racy and reduces the missing rate of detection .Besides ,the proposed method also reduces the scan-ning time that is required by the slide window of classification algorithm and improves the detecting rate .T he experimental results validate the adaptability and availability of the proposed method for the detection of vehicles in complicated traffic environment .