汽车工程

汽车工程 ›› 2019, Vol. 41 ›› Issue (7): 744-749.doi: 10.19562/j.chinasae.qcgc.2019.07.003

• • 上一篇    下一篇

基于图像识别的泊车车位检测算法研究

朱旺旺1, 黄宏成1, 马晋兴2   

  1. 1.上海交通大学,汽车电子控制技术国家工程实验室,上海 200240;
    2.上汽大众汽车有限公司,上海 201805
  • 出版日期:2019-07-25 发布日期:2019-07-30
  • 通讯作者: 朱旺旺,硕士,E-mail:zhuwangwang@sjtu.edu.cn。
  • 基金资助:
    国家自然科学基金(51677118)、国家重点研发计划重点专项(2017YFE0102000)和上海市政府间国际合作项目(16510711500)资助。

A Research on Parking Space Detection Algorithm Based on Image Recognition

Zhu Wangwang1, Huang Hongcheng1 ,Ma Jinxing2   

  1. 1.Shanghai Jiao Tong University, National Engineering Laboratory for Automotive Electronic Control Technology, Shanghai 200240;
    2.SAIC Volkswagen Automotive Company, Shanghai 201805
  • Online:2019-07-25 Published:2019-07-30

PDF (PC)

877

赞

摘要： 本文中为自动泊车提出了一种基于右侧单通道摄像头的车位检测算法。首先,通过顶帽变换融合直方图均衡化的方法降低了光照不均匀和光照强度变化对算法鲁棒性的影响;结合泊车工况,基于先验知识略去无价值域,压缩了Radon变换参数范围,获取车位线在Radon空间中坐标的同时,降低了Radon矩阵的维度;接着,通过阈值分割方法,优化了Radon矩阵的元素值,以在K-Means聚类分析的过程中,更准确高效地获取聚类中心;在聚类中心二维邻域内通过非极大值抑制实现了车位线的像素级定位。最后,设计了分层有限状态机,通过建立车位分类规则库实现了车位的分类。实验结果表明,该方法的计算效率高于基于Radon变换实现泊车的经典方法,且算法具有较强的鲁棒性。

关键词: 自动泊车, 顶帽变换融合直方图均衡化, K-均值聚类, Radon变换, 分层有限状态机

Abstract: A parking space detection algorithm based on right-side single-channel camera is proposed in this paper for automatic parking. Firstly, a scheme of top-hat transformation combined with histogram equalization is adopted to reduce the effects of uneven illumination and varying light intensity on the robustness of algorithm. Combined with parking condition, the worthless area is ignored based on prior knowledge, the parameter range of Radon transform is compressed, and while getting the coordinates in Radon space, the dimensions of Radon transform is reduced. Then the element values in Radon matrix are optimized by using threshold segmentation for more accurately and effectively getting cluster centers in K-means clustering, and non maximum suppression is used to achieve pixel-level location of parking lanes in the 2D neighborhood of cluster centers. Finally, hierarchical finite state machine is designed, and the classification of parking space is fulfilled by setting up the classification rule base for parking space. The results of experiments show that the computational efficiency of this method is higher than that of the classical method based on Radon transform, and the algorithm has stronger robustness

Key words: automatic parking, top-hat transform combined with histogram equalization, K-means clustering, Radon transform, hierarchical finite state machine