主办：陕西省汽车工程学会
ISSN 1671-7988  CN 61-1394/TH
创刊：1976年
汽车实用技术

汽车实用技术 ›› 2021, Vol. 46 ›› Issue (24): 64-67.DOI: 10.16638/j.cnki.1671-7988.2021.024.014

• 设计研究 • 上一篇    

基于深度学习的前车视觉测距研究

毛旭伟   

  1. 长安大学 汽车学院
  • 发布日期:2022-01-21
  • 通讯作者: 毛旭伟
  • 作者简介:毛旭伟（1997—），硕士研究生，就读于长安大学汽 车学院，主要研究方向：车辆安全。

Research on Vision Distance Measurement of Front Vehicle Based on Deep Learning

MAO Xuwei   

  1. School of Automobile, Chang'an University
  • Published:2022-01-21
  • Contact: MAO Xuwei

PDF

摘要： 针对测量前方车辆纵向距离的问题，文章提出一种基于深度学习的前方车辆检测和视觉测 距方法。该方法首先通过 YOLOv3 算法实现车牌位置检测并利用 OpenCV 进行剪裁，接着计算前 车车牌字符区域在相机成像区域所占像素长度，最后利用小孔成像原理对前方车辆的纵向距离进 行测量。试验表明：在 50 米范围内，该方法最大误差在 8%以内，平均误差约为 3.077%，可为前 车测距提供一种新的思路，在智能车辆感知方面有一定的应用前景。

关键词: 车辆测距；深度学习；车牌检测；小孔成像

Abstract: Aiming at the problem of measuring the longitudinal distance of vehicles ahead, this paper puts forward a vehicle distance measurement method with the detection of vehicles ahead and visual measurement distance. Firstly, this method realizes license plate position detection through YOLOv3 algorithm and use OpenCV to crop license plate images. Then, the pixel length of the license plate character area on the camera imaging area was calculated. Finally, the longitudinal distance of the car was measured by the pinhole imaging principle. The test shows that the maximum error of this method is less than 8% and the average error is about 3.077% in the range of 50 meters. It can provide a new idea for the distance measurement of the preceding vehicle, and has certain application prospects in intelligent vehicle perception.

Key words: Vehicle distance measurement; Deep learning; License plate detection; Pinhole imaging