基于全局与局部模型相互制约及具有模型不确定性评估的车道线检测方法

Lane markings detection approach based on interaction of global and local models and with uncertainty evaluation

针对无人驾驶车辆,提出了一种新的基于全局与局部模型相互制约及具有模型不确定性评估的车道线检测方法.该方法基于车道线的全局与局部模型,通过它们之间的相互制约与影响,有效地实现了视觉测量噪声的过滤与全局信息的获取.进一步利用对检测结果或全局模型的不确定性程度进行评估,不仅可稳定可靠地检测到车道线,而且还可为多源异构传感器之间的信息融合提供基础支撑.面向高速公路进行了大量的真实道路实验.实验结果表明,利用所提方法获得的预瞄点,其平均侧向偏差小于0.3 m,平均角度偏差小于0.03°,能够满足无人驾驶车辆在高速公路上进行自主行驶的需要.

This paper proposes a novel lane markings detection approach for unmanned ground vehicle（UGV）.The approach is based on the interaction of global and local models w ith uncertainty evaluation.The elimination of visual measurement noise w ith lane markings can be effectively accomplished and the global information can be reliably acquired through constrains and interaction betw een the local model of lane markings and the global one.Moreover,the uncertainty evaluation of global lane markings model can be used to improve the stability of lane markings detection and to further provide the basic support for the information fusion of heterogeneous multi-sensors.Many experiments on real roads under highw ay environments are carried out.The experimental results show that the average lateral deviation at the preview point achieved by using the proposed approach is less than 0.3 m and the average azimuth is not greater than 0.03°.This can fulfill the need of the UGV driving on the express w ay by itself.