一种基于视觉的车道线检测与跟踪算法

A Vision-based Lane Detecting and Tracking Algorithm

摘要:

本文提出了一种新颖的适用于自主驾驶系统的车道线检测与跟踪算法。该算法采用了广义曲线的道路参数模型，能同时适应弯道和直道的检测。该算法最突出的贡献在于，没有仅仅使用单一方法求解各个参数，而是根据不同参数的不同精度要求，分别使用自适应随机霍夫变换方法(ARHT)和Tabu Search算法计算道路模型中的各个参数，这样，既准确地确定了道路模型中的参数，同时也兼顾了车载系统的实时性要求。此外，为了提高算法的实时性，本文引入了多解析度的策略，以降低整个流程的时耗。最后，为了满足自主驾驶系统对稳定性的需求，本文还提出了一种基于粒子滤波器（particle filter）的跟踪算法。通过不同场景下进行的实验，充分表明本文提出的车道线算法检测与跟踪具有很好的鲁棒性和实时性，而且对不同的光照条件也有较好的适应性。

Abstract:

This paper presents a novel lane detection algorithm, which is suitable for automatic drive system. The algorithm chooses the common curve lane parameter model, which can describe both straight and curved lanes. The most prominent contribution of this paper is: instead of using one single method to calculate all the parameters in the lane model, both the Adaptive Random Hough Transformation (ARHT) algorithm and the Tabu Search algorithm are used to calculate the different parameters in the lane model, according to the different demands of accuracy for different parameters. Furthermore, in order to reduce the time-consume of the whole system, the strategy of multi-resolution is proposed. At last, this paper also presents a tracking algorithm based on the particle filter, which can make the system more stable. The algorithms presented in this paper are proved to be both robust and fast by a large amount of experiments in variable occasions, besides, the algorithms can extract the lanes accurately even in some bad illumination occasions.