基于轮廓重合度分析的道路标记鲁棒识别

Robust recognition of road markings using contour coincidence analysis

基于道路标线识别是自动驾驶中的一个关键问题,提出一种消除道路场景中大量非目标干扰的道路标线识别方法。该方法将基于椭圆傅里叶描述子的支持向量机分类结果和对道路标线轮廓分析结果进行融合。在此算法中,为了减小非目标区域的干扰,对轮廓图像与Canny边缘图像重合度进行分析以滤除非目标区域。研究结果表明:该道路标记识别算法的分类准确率高达98.69%,召回率高达94.02%,同时误报率较低,为0.61%;精确率和召回率的调和平均数F1高达96.30%;整个道路标记识别算法平均运行时间为34.79 ms,能够实时地检测并分类8种常见道路标记,表明该方法的识别效果较好。

Considering that the road marking recognition is a key problem in automatic driving, a method of road marking recognition was proposed to eliminate a lot of non-target interference in road scenes. This method fused the classification results of support vector machine based on elliptic Fourier descriptor with the results of contour analysis of road markers. In this algorithm, in order to reduce the interference of non-target region, the coincidence degree analysis method of contour image and Canny edge image was adopted to filter the non-target region. The results show that the classification accuracy is up to 98.69%, the recall rate is up to 94.02%, and the false alarm rate is 0.61%. The harmonic average F1 for classification accuracy and the recall rate is 96.30%, the average running time of the algorithm is 34.79 ms, and the algorithm can detect and classify 8 common road markings in real time. This method has good recognition effects.