一种基于双极系数法的斑马线检测改良方法

摘要

根据斑马线自身的颜色形状特点,对导盲系统中的斑马线识别问题进行研究.在双极系数法的基础上针对灰度化处理和线段检测两个方面进行改良,提出一种综合线性判别式公析LDA (Linear Discriminant Analysis)与EDLines(Edge Drawing Lines)算法的检测方法.利用LDA灰度化图像,使斑马线与周围路面、行人车辆有明显的区分;采用不均匀分块方式对图像求双极系数图,同时,进行区域优化筛选,再将优化后的双极系数图与二值图融合,得到候选区域;最后通过EDLines进行直线检测,对线段的长度、斜率、间隙和数量进行判断,确定斑马线位置.与传统的斑马线检测方法相比,LDA灰度化能更好地降低路面障碍物的干扰,而EDLines算法则能增强线段检测的准确性.实验结果表明识别率达到90.3％,对识别斑马线有良好的效果.%According to the characteristics of the zebra crossing's own color and shape,this paper studies the problem of zebra crossing recognition in blind guiding system.We improve the gray processing and line segment detection by bipolarity method,and put forward a detection method combining LDA (Linear Discriminant Analysis) and EDLines (Edge Drawing Lines) algorithm.Firstly,LDA was used to convert color images to gray images directly,which made the zebra crossing and the surrounding roads,pedestrians,vehicles were clearly distinguished.Secondly,the bipolarity image was calculated by non-uniform segmentation method.At the same time,the region was optimized and screened.Candidate regions were obtained by fusing the binary image and the bipolarity image that had been optimized.Finally,the line was detected by EDLines,and then the location of the zebra crossing was determined by selecting the length,slope,gap and number of line segments.Compared with the traditional detection method of the zebra crossing,the interference of the road obstacles is reduced by LDA,while the accuracy of line detection is enhanced by EDLines.Experimental results show that the recognition rate is 90.3 ％,which has a good effect on the zebra crossing.