交通信号灯的检测是复杂场景下交通灯识别的重点。采用了色彩分割与关联滤波方案进行交通灯的检测。首先，建立交通信号灯的高斯模型，利用高斯向量与多色彩空间结合的方法进行图像分割。然后，采用基于区域增长与相似性判定的关联滤波，对色彩分割后的图像进行处理。最后，使用基于canny算子的边缘提取算法获取方向指示灯轮廓特征，并使用基于改进hu不变矩和马氏距离对方向指示信号灯进行分类。-The detection of traffic lights is the focus of traffic lights recognition in complex scenes. Using color segmentation and the associated filtering scheme, the detection of traffic lights. First, the Gaussian model to establish the traffic lights, the use of Gaussian vector with multi-color space for image segmentation. Then, using the associated filter, based on regional growth and the similarity determination processing of the image color segmentation. Finally, the canny operator edge extraction algorithm to obtain the direction indicator profile feature, and use the classification based on improved hu moment invariants and the Mahalanobis distance directional signal lights.