中文摘要：

交通灯识别是智能车技术的关键，文章提出一种识别交通灯的新方法。首先基于多源信息，在摄像机像面上建立随机过程模型，确定车辆位置和像面概率分布的关系。然后选取输入图像中概率大于设定阈值的区域，在YCbCr颜色空间中基于颜色和亮度信息分割该区域得到候选区，对候选区域分别提取其颜色直方图和边缘方向直方图信息。最后采用支持向量机（SVM）对交通灯识别分类。实验结果表明，该方法能够实时准确地检测出交通灯。

英文摘要：

Traffic lights recognition is a key part of the intelligent vehicle technology. This paper proposed a new method to identify traffic lights. There are three steps. Firstly, based on multi-source information, this method builds a stochastic process model and determines the connection between vehicle position and probability distribution of image surface. Secondly, select the area whose probability is greater than the threshold, and segment the area as candidates based on color and luminance. Then extract the candidates＇ features by color histogram and edge orientation histogram. Finally, support vector machine （SVM） was used to classify traffic lights. The experimental results show that this method can detect traffic lights timely and accurately.