智能车中基于单目视觉的前车检测和跟踪

提出了一个改进的单目视觉方法，用于智能车在结构化公路环境下准确检测和跟踪前方车辆。该方法先利用图像灰度梯度检测前车，剔除可能的虚检测，建立新目标的二维模型；然后用卡尔曼滤波方法预测下一帧的目标位置，在预测位置附近用边缘投影方法定位目标；设计了一种新的四因素似然度函数，验证跟踪结果与检测结果的匹配度，当跟踪失败时，重新检测前车。利用长图像序列PETS2001进行实验，结果表明该方法可以有效的检测和跟踪本车车道前方视野中的车辆障碍物，为智能车的防撞预警和控制系统提供可靠信息。

Preceding car detection and tracking based on the monocular vision

An improved monocular vision method was proposed for intelligent vehicle to detect the preceding car in structural road environment. This method firstly detected the preceding car by identifying the edges of the object and then the false object was eliminated. The eligible object expressed as a 2-D model was acquired. Secondly the location of the object in the next frame was predicted by Kalman filter, and the object was detected near that location by means of edge projection. Finally a novel likelihood function was desired to verify the tracking results. If the likelihood was too low, the object was detected over again, or else the 2-D model would be updated. Experiment results of the image sequence from PETS2001 show that the proposed method can detect structural road automatically, rapidly and exactly and it can detect and track the preceding car automatically and exactly as well.