路面车辆实时检测与跟踪的视觉方法  

On Road Vehicles Real-Time Detection and Tracking Using Vision Based Approach

作　　者：沈峘[1] 李舜酩[1] 柏方超[1] 缪小冬[1] 李芳培[1] 

机构地区：[1]南京航空航天大学能源与动力学院,江苏南京210016

出　　处：《光学学报》2010年第4期1076-1083,共8页Acta Optica Sinica

基　　金：国家自然科学基金(50675099);江苏省自然科学基金(BK2007197);江苏省普通高校研究生科研创新计划(CX08B_044Z)资助课题

摘　　要：为向驾驶者提供有效的车辆位置信息,提高驾驶安全性,提出了一种融合多种目标特征的单目视觉车辆检测与跟踪方法。首先,利用车辆尾部的结构对称性提取出感兴趣区域,减少搜索范围。再利用车辆底部的阴影特征,在感兴趣区域中搜寻车辆可能出现的位置,找出假设目标。然后,利用亮度和轮廓信息对假设目标进行对称性验证,排除虚假目标。同时,融合颜色和梯度方向建立目标特征模型,利用均值平移算法在随后的图像序列中对目标进行快速跟踪定位。检测与跟踪联合工作在一种互动机制下,大幅改善了算法的有效性和实时性。实验结果显示,提出方法的正确识别率为96.34%,平均处理速度达24.27 frame/s,能够满足车辆驾驶安全性和实时性要求。A novel monocular camera based on road vehicle detection and trcking approach by fuse multi-cues of object is present to improve drive security by providing some effective on road vehicles position information for driver.First,the horizontal symmtery of vehicle rear view is utilized to achieve the region of interest（ROI） extract so as to reduce search area of following process.And then,the sign of underneath shadow is employed to generate hypothetical positions on which potantial vehicles maybe present.Following,both image intensity and figure information are combined to used to verify the vertical symmetry of the potential vehicle candidates.Meanwhile, mean shift procedure,based on the object feature model of combine color histogram and orientation histogram,is employ to fast search the potantial objects between two sequential image frames.More improtant,both detection and tracking cooperate work under a interactive mechanism which can dramatically improve both detection efficiency and real-time.Experimental results show that the propsed apporach can achieve 96.34%correct recognition rate and run on an average 24.27 frame/s,which validate the vehicle drive security and real-time requirements.

关 键 词：机器视觉 车辆检测 目标跟踪 智能车辆 智能交通系统 

分 类 号：U273.99[机械工程—车辆工程]