计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (14): 210-218.DOI: 10.3778/j.issn.1002-8331.2011-0462

• 图形图像处理 • 上一篇    下一篇

基于SLIC和改进区域生长的非结构化道路识别

谢习华,王刚,辛涛,赵喻明   

  1. 1.中南大学 高性能复杂制造国家重点实验室,长沙 410083
    2.山河智能装备股份有限公司,长沙 410100
    3.中国人民解放军32181部队
    4.清华大学 精密仪器系,北京 00084
  • 出版日期:2022-07-15 发布日期:2022-07-15

Unstructured Road Detection Based on SLIC and Improved Region Growing

XIE Xihua, WANG Gang, XIN Tao, ZHAO Yuming   

  1. 1.State Key Laboratory of High Performance Complex Manufacturing, Central South University, Changsha 410083, China
    2.Sunward Intelligent Equipment Co. Ltd., Changsha 410100, China
    3.Unit 32181 of PLA, China
    4.Department of Precision Instruments, Tsinghua University, Beijing 100084, China
  • Online:2022-07-15 Published:2022-07-15

摘要: 非结构化道路一般没有车道标识线且道路边界模糊,区分道路区域与背景区域难度较大。针对现有非结构化道路识别方法存在全像素域计算分类处理实时性差、易受噪声数据干扰等问题,提出一种基于SLIC(simple linear iterative clustering)超像素分割和改进区域生长算法的非结构化道路识别方法。利用均匀化初始聚类中心的SLIC算法生成低分辨率超像素特征图。在此基础上,利用聚类算法与邻域搜索算法自适应选择种子点,并引入CIEDE2000色差理论作为区域生长法生长准则,初步确定道路区域。根据道路连续一致特点,优化超像素级生长图并映射轮廓区域至原图,获得道路最终区域。基于数据集及真实场景的实验结果表明,该方法具有较高的识别率和抗干扰能力。

关键词: 非结构化道路, 简单线性迭代聚类(SLIC), 超像素特征图, CIEDE2000, 区域生长算法

Abstract: For unstructured roads, there are generally no lane markings and no clear road boundaries, so it is difficult to distinguish the road area from the background area. In view of the shortcomings of the existing unstructured road recognition methods, such as the poor real-time performance caused by the calculation and classification in full pixel domain and the vulnerability to the interference of noise data, a new unstructured road recognition method based on SLIC(simple linear iterative clustering) superpixel segmentation and improved regional growth algorithm is proposed. Firstly, the SLIC algorithm that homogenizes the initial clustering center is used to generate a low-resolution superpixel feature map. On this basis, the seed point is selected adaptively by clustering algorithm and neighborhood search algorithm, and the CIEDE2000 color difference theory is introduced as the growth criterion for the region growing to preliminarily determine the road area. Finally, according to the continuous characteristics of the road, it optimizes the superpixel-level growth map and maps the contour area to the original image to obtain the final area of the road. The experimental results based on data sets and real scenes show that the method has a high recognition rate and anti-interference ability.

Key words: unstructured road, simple linear iterative clustering(SLIC), superpixel feature map, CIEDE2000, region growing algorithm