计算机与现代化

• 图像处理 • 上一篇    下一篇

一种改进YOLOv3-Tiny的行车检测算法

  

  1. (1.南京航空航天大学电子信息工程学院,江苏南京210016;2.南京航空航天大学无人机研究院,江苏南京210016)
  • 收稿日期:2019-04-24 出版日期:2020-03-24 发布日期:2020-03-30
  • 作者简介:刘力冉(1995-),女,安徽滁州人,硕士研究生,研究方向:无人机测控,图像处理,目标检测,E-mail: 56415618@qq.com。
  • 基金资助:
    国家重点研发计划资助项目(2017YFC0822404)

An Improved YOLOv3-Tiny Traffic Detection Algorithm

  1. (1. College of Electronic and Information Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China;
    2. UAV Research Institute, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China)
  • Received:2019-04-24 Online:2020-03-24 Published:2020-03-30

摘要: YOLO系列算法的简化版本YOLOv3-Tiny具有较为简单的网络框架,对GPU显存要求较低,该算法虽然实时性较高,却存在精度较低的问题,在识别行车目标方面不能得到精确的结果。对此,本文首先改变输入图片的大小,目的是获取图片更多的横向信息,使得网络更容易学习行车的信息,其次改进算法的网络结构提高算法的精度,最终得出改进的YOLOv3-Tiny算法。实验结果表明,改进之后的算法在保证实时性的情况下,提高了精确性。

关键词: 深度学习, 行车检测, YOLOv3-Tiny, 聚类

Abstract: The simplified version YOLOv3-Tiny of YOLO series algorithm has a relatively simple network framework and low requirement for GPU display and memory. Although the algorithm has high real-time performance but accuracy is low, it can not get accurate results in identifying driving targets. This paper first changes the size of the input pictures in order to obtain more lateral information of the pictures so that the network can easily learn the driving information. Secondly, the network structure of the algorithm is improved so as to improve the accuracy of the algorithm. Finally, the improved YOLOv3-Tiny algorithm is obtained. The experimental results show that the improved algorithm improves the accuracy while guaranteeing real-time performance.

Key words: deep learning, vehicle detection, YOLOv3-Tiny, clustering

中图分类号: