中文摘要：

提出了一种基于颜色和形状神经网络的视觉可量测实景影像交通标志自动检测算法。该算法设计了2种类型的概率神经网络，一种可以将图像中像素分为黄色、红色、蓝色和其他颜色4类；另一种可以识别三角形、圆形、矩形和倒三角形4种形状。从而先利用颜色识别神经网络对影像进行颜色分割；然后，在分割后的二值图像上利用灰度投影确定交通标志的候选区域；最后，利用中心投影向量和形状识别神经网络，实现候选区域的形状判断和交通标志自动检测。使用车载三维数据采集系统拍摄的视觉可量测实景影像进行了实验，证明了该方法的有效性和鲁棒性。

英文摘要：

A new algorithm for automatic traffic sign detection in visual measurable image based on color and shape neural networks is proposed in this paper. In the algorithm two types of probabilistic neural networks are designed. One neural network can divide all the pixels of the image into four types of colors, i. e. , yellow, red, blue and other. The other neural network can recognize four types of shapes, i.e. , triangle, circle, rectangle and inverse-trian- gle. The color recognition neural network is firstly used to segment the image. Secondly, gray scale projection is used to determine the candidate regions of traffic signs on the segmented binary image. Lastly, central projected vector and shape recognition neural network are used to judge the shape type of each candidate region and realize the automatic identification of traffic signs. The new algorithm was applied to the visual measurable images obtained using the vehicle-borne 3D data acquisition system in different places and at different time. Experimental results prove the effectiveness and robustness of the proposed method.