汽车工程 ›› 2023, Vol. 45 ›› Issue (6): 989-996.doi: 10.19562/j.chinasae.qcgc.2023.06.009

所属专题: 智能网联汽车技术专题-感知&HMI&测评2023年

• • 上一篇    下一篇

基于Retinex的低光照车位图像增强

苗作华1,2(),朱良建1,赵成诚1,刘代文1,李诒雯3,陈澳光1   

  1. 1.武汉科技大学资源与环境工程学院,武汉  430081
    2.冶金矿产资源高效利用与造块湖北省重点实验室,武汉  430081
    3.武汉光庭信息技术股份有限公司,武汉  430000
  • 收稿日期:2022-12-01 修回日期:2022-12-27 出版日期:2023-06-25 发布日期:2023-06-16
  • 通讯作者: 苗作华 E-mail:270226741@qq.com
  • 基金资助:
    国家自然科学基金(41071242);教育部产学合作协同育人项目(202102136008)

Image Enhancement of Low-Light Parking Space Based on Retinex

Zuohua Miao1,2(),Liangjian Zhu1,Chengcheng Zhao1,Daiwen Liu1,Yiwen Li3,Aoguang Chen1   

  1. 1.School of Resources and Environmental Engineering,Wuhan University of Science and Technology,Wuhan  430081
    2.Hubei Key Laboratory of Efficient Utilization and Blocking of Metallurgical Mineral Resources,Wuhan  430081
    3.Wuhan Kotei Informatics Co. ,Ltd. ,Wuhan  430000
  • Received:2022-12-01 Revised:2022-12-27 Online:2023-06-25 Published:2023-06-16
  • Contact: Zuohua Miao E-mail:270226741@qq.com

摘要:

针对自动泊车过程中车载鱼眼相机拍摄的车位图像因低光照环境导致图像整体偏暗、车位信息模糊而无法检测出车位的问题,本文中设计了一种融合限制对比度自适应直方图均衡化(CLAHE)和改进单尺度Retinex(SSR)的车位线图像增强算法。首先将鱼眼相机图像畸变矫正后转换生成鸟瞰图;然后使用优化映射函数计算过程的CLAHE算法对鸟瞰图预处理;进而使用基于迭代方框滤波估计入射分量的单尺度Retinex算法增强图像;经过滤波和形态学处理,最后基于亮通道先验将图像灰度化,得到最终的增强结果。本文采集多组低光照场景下单侧鱼眼相机摄取的实际泊车过程视频,截取驶过停车位过程中的单帧图像作为数据输入,并使用一种基础车位检测算法对增强结果进行检测,试验结果表明,经过增强后可被检测出车位的视频帧数量超过90%,且单帧图像增强算法处理时间仅38 ms。

关键词: 自动泊车, 低光照, 车位线增强, 车位检测

Abstract:

In the process of automatic parking, the parking space image taken by the fisheye camera on the vehicle is dark due to the low light environment and the parking space information is fuzzy, which makes it impossible to detect the parking space. In this paper, an image enhancement algorithm of parking space line is designed, which combines limited contrast adaptive histogram equalization (CLAHE) and improved single-scale Retinex (SSR). Firstly, the image distortion of fisheye camera is rectified and converted to generate a bird's eye view. Then, the bird's-eye view is preprocessed by the CLAHE algorithm, which optimizes the calculation process of mapping function. Then, a single-scale Retinex algorithm based on iterative block filtering to estimate incident components is used to enhance the image. After filtering and morphological processing, the image is grayed based on the bright channel prior, and the final enhancement result is obtained. In this paper, multiple sets of videos of actual parking process captured by one-sided fisheye camera in low-light scenes are collected, and single-frame image in the process of passing through parking space is intercepted as data input, and a basic parking space detection algorithm is used to detect the enhancement result. The experiment shows that the number of video frames that can detect parking spaces after enhancement exceeds 90%, with only 38 ms of the processing time of single-frame image enhancement algorithm.

Key words: automatic parking, low illumination, parking space line enhancement, parking space detection