实时鲁棒的矩形交通标识牌倾斜校正方法

ReaL-time Robust Correction Method for Rectangular Traffic Signs

交通标识牌识别是无人驾驶智能汽车视觉环境感知的重要组成,为了解决交通牌倾斜造成的识别率下降的问题,本文提出了一种实时鲁柔的矩形交通标识牌倾斜校正方法。首先,对图像进行灰度化、Canny边缘检测等预处理,再通过Hough直线变换检测交通标识牌的边缘线段,根据这些边缘线段计算矩形的四个角点坐标,然后再根据逆透视变换(IPM)原理得到矫正矩阵,最后利用IPM将倾斜的交通标识牌图像校正。实验结果表明,该算法可以有效地矫正不同程度倾斜的矩形交通标识牌图像,具有很好的实时性、鲁柔性,适用于无人驾驶系统中的交通标识牌识别预处理技术。

In order to solve the problem of the decrease of recognition rate caused by traffic card inclination,this paper presents a real-time robust rectangular traffic sign correction method.Firstly,the image is preprocessed,such as grayscale,edge detection.Then the edge line of the traffic sign is detected by Hough transformation.The coordinates of the four corners of the rectangle are calculated according to the edge lines,and the correction matrix is obtained according to the principle of inverse perspective transformation.Finally,Correction of the tilted traffic signboard image via IPM.Experiments show that the proposed method has robust and real-time performance and It is suitable for traffic sign recognition and preprocessing technology in unmanned vehicle.