基于种子超像素生长的视觉道路检测  

VISION ROAD DETECTION BASED ON GROWTH OF SEED SUPERPIXEL

作　　者：杨波 Yang Bo(Chongqing College of International Business and Economics,Chongqing 401520,China)

机构地区：[1]重庆对外经贸学院,重庆401520

出　　处：《计算机应用与软件》2023年第9期199-204,共6页Computer Applications and Software

基　　金：国家自然科学基金青年科学基金项目(11901071);重庆市自然科学基金面上项目(cstc2019jcyj-msxmX0219)。

摘　　要：为了提高基于视觉的道路检测算法的准确性和鲁棒性,提出一种基于种子超像素生长的视觉道路检测算法。提出一种基于图像形态学开运算的方法去除道路车道线信息;将RGB颜色空间转换为更加符合人类视觉特性的HSV颜色空间,并对原图像进行超像素分割;提出一种道路种子超像素选取方法;计算各超像素块和种子像素块之间的归一化直方图距离以进行种子生长,并提出融合道路模型,完成基于视觉的道路检测。为验证所提算法的有效性,利用KITTI数据集进行道路检测实验。实验结果表明,该算法在不同场景下道路检测准确率均超过86.38%,召回率超过86.62%,相比其他算法在准确性和鲁棒性上有了显著的提高。In order to improve the accuracy and robustness of the vision-based road detection algorithms,a road detection algorithm based on growth of seed superpixel is proposed.We proposed a method based on morphological opening to remove road lane line information.The RGB color space was converted to HSV color space which was more consistent with human visual characteristics,and the original image was segmented to superpixels.A seed superpixel of road selection method was proposed.The distances of normalized histogram between each superpixel and the seed superpixels were calculated,and road model was fused to complete the task of vision-based road detection.In order to verify the effectiveness of the algorithm,road detection experiments were carried out by using KITTI datasets.The experimental results show that the road detection accuracy of the algorithm exceeds 86.38%and the recall is more than 86.62%in different scenes.Compared with other algorithms,the accuracy and robustness of the algorithm have been significantly improved.

关 键 词：形态学运算 超像素 归一化直方图 种子生长 道路检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]