汽车工程 ›› 2021, Vol. 43 ›› Issue (5): 650-656.doi: 10.19562/j.chinasae.qcgc.2021.05.002

• • 上一篇    下一篇

基于双目逆投影的停车位占位快速检测方法

肖汉彪1,2,胡钊政1,2(),周哲2,3,伍锦祥1,2   

  1. 1.武汉理工大学信息工程学院,武汉 430070
    2.武汉理工大学智能交通系统研究中心,武汉 430063
    3.武汉理工大学重庆研究院,重庆 401120
  • 收稿日期:2020-10-19 修回日期:2020-12-11 出版日期:2021-05-25 发布日期:2021-05-18
  • 通讯作者: 胡钊政 E-mail:zzhu@whut.edu.cn
  • 基金资助:
    国家重点研发计划(2018YFB1600801);重庆市自然科学基金(cstc2020jcyj-msxmX0978);武汉市科技局(2020010601012165)

Fast Detection Method of Parking Space Occupation Based on Inverse Perspective Mapping with Binocular Cameras

Hanbiao Xiao1,2,Zhaozheng Hu1,2(),Zhe Zhou2,3,Jinxiang Wu1,2   

  1. 1.School of Information Engineering,Wuhan University of Technology,Wuhan 430070
    2.ITS Research Center,Wuhan University of Technology,Wuhan 430063
    3.Chongqing Research Institute of Wuhan University of Technology,Chongqing 401120
  • Received:2020-10-19 Revised:2020-12-11 Online:2021-05-25 Published:2021-05-18
  • Contact: Zhaozheng Hu E-mail:zzhu@whut.edu.cn

摘要:

本文中提出了一种基于双目逆投影变换的停车位快速检测与定位方法。首先,通过逆投影变换将双目相机获取的两张图像投影到参考平面,生成两个新视图,即双目逆投影图,并利用其差分图来区分平面上与平面外的目标;随后,通过对差分图进行阈值化与滤波来实现障碍物的快速检测;同时,障碍物的位置也可从参考平面获取。与现有方法相比,该方法不需要明确的目标检测或三维重建,仅利用基本的图像变换实现停车位的快速检测与障碍物定位。在地下停车场与室外停车场环境下,利用智能车搭载的双目摄像机采集数据对本文算法进行验证。试验结果表明,此方法能快速有效地检测和定位车位附近的动静态障碍物,例如路障、行人、车辆和车位锁等,检测速度达到18帧/s,识别准确率高于95.0%,在算法效率、准确率和定位精度方面均优于现有的检测方法。

关键词: 自动驾驶, 停车位占位检测, 逆投影, 双目相机

Abstract:

This paper proposes a method for fast parking space detection and location based on inverse perspective mapping with binocular cameras. Firstly, the two images captured by binocular camera are projected to the reference plane through back projection to generate two new views, which are called Binocular Inverse Perspective Mapping (BIPM) views. The difference map of the two DBP (BIPM Difference) views can thus be utilized to distinguish objects on the plane or out of the plane. Then, obstacle fast detection is realized by thresholding and filtering the difference map. Meanwhile, the obstacle is localized readily from the reference plane. Compared to the existing methods, the proposed method requires no explicit object detection or 3D reconstruction. It can achieve fast parking space detection and obstacle localization by basic image warping. The proposed method has been verified with actual data collected by the binocular camera of the intelligent vehicle in indoor and outdoor parking lot. The experimental results show that this method can quickly and effectively detect and localize both dynamic and static obstacle objects near the parking space, such as roadblocks, pedestrians, vehicles, ground lockers, etc. The detection speed reaches 18 frames per second (fps), and the recognition accuracy rate is above 95.0%, outperforming existing methods in terms of algorithm efficiency, accuracy and localization precison.

Key words: autonomous driving, parking space occupation detection, inverse perspective mapping, binocular cameras