汽车智能辅助驾驶系统中的单目视觉导航技术

在智能交通系统中，自动驾驶系统或许是最难于实现的技术之一．许多学者努力 想找到解决自动驾驶这个难题的方法．本文介绍了一种新的思想和一些新的算法，它立足于 单目视觉技术来实现车道保持和进行障碍物检测，进而力图解决视觉导航问题．首先运用多 阈值分割技术提取高速公路上当前车道的分道线，接着对分道线建立了圆锥曲线模型进行二 维重建．在道路模型的约束下，笔者提出方向分形维的算法确定前方车辆的位置，进而运用 旋转，平移和缩放中心不变矩（RTM）判断其真伪，最后采用有色噪声的卡尔曼滤波对真实 障碍物进行跟踪．本系统不但能根据分道线模型获得保持车道所需的方向信息，也能检测本 车道前方障碍物而防止碰撞．系统在四川和重庆的高速公路上以每小时100公里的速度进行 了试验，能圆满完成车道保持和障碍物检测的任务．

MONOCULAR VISION NAVIGATION IN VEHICLE INTELLIGENT AUXILIARY DRIVE SYSTEM

In the intelligent transportation systems, maybe the automatic navigation system is the most difficult problem. Many researchers try to find a way to carry out cruise assistance. This paper describes some new thoughts and algorithms for lane keeping and obstacle detection with a single view for cruise assistance. The system uses multi-threshold to extract the lane line. Then, the conic model is used to reconstruct them. Based on the image direction-fractal dimension and rotation, translation and magnification invariable central moment (RTM), it detects the obstacle in front in the same lane. At last, it uses Kalman filtering to track the obstacle. We have experimented the system with the algorithms on highway at 100km/h in Sichuan province and Chongqing city in China. The result shows that the algorithms can work perfectly.