基于视觉的自主车道路检测技术研究

针对结构化道路检测中基于单一特征的检测易受影响,非结构化道路检测算法对多种类型的非标准道路缺乏适应性的问题,分别提出了一种基于D-S证据理论的多视觉特征融合的车道线检测方法和一种基于增量模糊支持向量机(IFSVM)的非结构化道路在线学习检测方法。选取梯度幅度等检测算子分别设计基本概率分配函数,根据建立的分段线性道路模型进行求解,FSVM分类器通过从前先的检测结果中学习,在耗费少量计算时间与内存空间的情况下,不断再训练以增强分类器的性能。实验结果表明,该算法比单纯利用图像的边缘或颜色等特征进行道路检测具有更高的可靠性,对复杂环境下的道路检测具有较强的鲁棒性和较强的抗干扰能力。

Research on Vision-based Road Detection for Autonomous Land Vehicle

For structured road, a novel lane detection algorithm based on multi-visual-features fusion by using D-S evidence theory is introduced to improve the robustness. First, the detection operators are chosen to construct the evidence bodies, for which the basic probability assignment functions are designed respectively. Then, the parameters of piecewise linear lane model are calculated and KF is used for lane tracking. The experimental results show that this method can achieve higher reliability and adaptability for lane detection than the algorithm simply using the edge or color feature. For unstructured road,a novel online learning road detection method based on incremental fuzzy support vector machine (IFSVM) is introduced to improve the adaptability to environmental changes. FSVM classifier is updated by online learning, and the experimental results show that this method can achieve higher adaptability and robustness for road detection than the algorithm using traditional offline trained SVM.