Abstract: Traffic lights recognition is important for the intelligent vehicles research. A real-time traffic lights recognition method was presented. Firstly, the image was pre-processed by morphological operations. And then the candidate areas were found according to the samples of three kinds of traffic light got from Lab color space, which describes red, green, yellow color more effectively. Then, the candidate areas were filtered by traffic lights' shape characteristics. After that, three kinds of traffic lights templates, which were surrounded by a black rectangular box, were designed and used to confirm the candidate areas. At last, the results were validated by statistical methods. The experimental results show that the method is effective.

Key words: computer vision, traffic lights recognition, morphology operation, Lab color space, template match

摘要： 交通灯的识别是无人驾驶汽车研究的一个重要方面，提出一种实时交通灯识别算法。首先对图像进行形态学预处理，然后对三种色彩的交通灯取样，利用交通灯在色彩方面的特征，在对红、黄、绿色彩描述更为清晰的Lab色彩空间，查找交通灯的候选区域。根据交通灯在结构上被一个黑色矩形框包围的特征，设计出三种交通灯模板，使用模板匹配对候选区域进行确认。最后使用统计方法对结果进行验证。实验结果表明，该算法能实时准确地识别出交通灯。

关键词: 计算机视觉, 交通灯识别, 形态学操作, Lab色彩空间, 模板匹配