基于单目视觉的车辆碰撞模型  

Vehicle collision model based on mono-vision

作　　者：王超[1] 赵春霞[1] 任明武[1] 王欢[1] 

机构地区：[1]南京理工大学计算机科学与工程学院,江苏南京210094

出　　处：《南京理工大学学报》2014年第6期739-744,共6页Journal of Nanjing University of Science and Technology

基　　金：国家"973"计划资助项目(30920130122004);国家自然科学基金(91220301);南京理工大学重点实验室基金(30920130122005/6)

摘　　要：为了解决主动安全研究中车辆在行驶过程中与前车的碰撞危险判定问题,该文提出了一种车辆碰撞模型。基于针孔成像原理,分析图像中目标车辆与世界坐标系中实际车辆的映射关系。检测图像中路面消失点与车辆底部的位置,并以其差值作为车辆尺寸特征。分析多帧图像中车辆目标尺寸特征的变化规律,从而分析出车辆行进趋势,并估算出前车同本车的相对碰撞时间。该碰撞模型既为驾驶员反馈了碰撞时间信息,又通过分析加速度避免虚警。与已有模型相比较,该文模型在车辆距离大于30 m时效果不稳定,在距离小于30 m时误差低于5%。实验结果表明该模型具备较强的实用性与准确性。In order to solve the problem of collision safety analysis of moving vehicles in active safety field, a new vehicle collision model is proposed here. The corresponding relations between target vehicles in images and real vehicles in world coordinate system are analyzed based on keyhole imaging principle. The vanishing point of road and the location of vehicle bottom in images are detected, and their difference is treated as the scale of vehicles. The changing law of the scale of vehicles in multi- frame images is analyzed, and the moving tendency of the vehicles is calculated and the time of collision between the front vehicle and the self-vehicle is evaluated. The vehicle collision model proposed here provides exact warning time and avoids needless alarms by analyzing acceleration. Compared with other models,the result of this model is astable when the distance between vehicles is longer than 30 m,and the error of this model is below 5% when the distance between vehicles is shorter than 30 m. A real vehicle experiment verifies the practicability and accuracy of this model.

关 键 词：单目视觉 车辆 碰撞模型 主动安全 目标车辆 世界坐标系 实际车辆 消失点检测 碰撞时间 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]