一种基于光照无关图的车道检测方法

针对无人车辆道路检测时,特征提取易受光照变化和阴影影响的问题,提出一种基于光照无关图的车道检测方法.使用最小熵值法离线求取图像的光照无关角,对图像每个像素点在对数色度空间向光照无关方向投影得到光照无关灰度图;利用Canny算子提取光照无关图边缘,在边缘图上寻找线段基元以减弱细小边缘的干扰;使用改进投票空间的Hough变换检测直线,采用分段线性道路模型描述道路边界.实验表明,该算法能够有效减弱阴影和光照变化造成的影响,准确识别道路边界,满足实时性要求.

Lane Detection Based on Illumination Invariant Image

This paper aims at overcoming the defect caused by illumination variations and shadows in the feature extraction of lane detection. Illumination invariant angle is obtained with entropy-minimization method and it leads to a 1D, gray-scale image representation which is illumination invariant at each image pixel in log-chromaticity space. By finding line elements in Canny edge map, trivial edges in shadow sections are eliminated. Then, an improved voting scheme Hough transform is adopted to detect lines and the lane boundaries is represented with piecewise linear road model. The experimental results show the efficiency of proposed method in terms of invariance illumination, shadow removal, reliability and adaptability detection and real-time navigation.