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 ›› 2022, Vol. 48 ›› Issue (4): 23-30.DOI: 10.16638/j.cnki.1671-7988.2023.04.006

• 智能网联汽车 • 上一篇    

基于深度学习的多目标车辆检测及追踪方法

王 锋   

  1. 长安大学 汽车学院
  • 出版日期:2023-02-28 发布日期:2023-02-28
  • 通讯作者: 王 锋
  • 作者简介:王锋（1997—），男，硕士研究生，研究方向为智能网联车及无人驾驶，E-mail:1612868433@qq.com。

Method of Multi-target Vehicle Detection and Tracking Based on Deep Learning

WANG Feng   

  1. School of Automobile, Chang'an University
  • Online:2023-02-28 Published:2023-02-28
  • Contact: WANG Feng

PDF

摘要： 文章是以 MTALAB 软件为主要平台，基于深度学习建立一种多目标车辆检测及追踪的 方法。首先建立一个基于深度学习的模型用于训练的不同场景的车辆数据集，并对所采集的 数据集进行标注和格式归一化处理。然后使用 K-means 聚类算法进行锚框，建立以 YOLOv3 SPP 算法为主的神经网络框架，采用非极大值拟制（NMS）算法得到最终的预测框。最终训 练神经网络模型，再对该模型进行测试和评定。经实验可以得出该模型能够准确地检测及追 踪多目标车辆。

关键词: 深度学习；多目标车辆；K-means 聚类算法；YOLOv3 SPP 算法；非极大值拟制算法

Abstract: This paper uses MTALAB software as the main platform, and establishes a multi-target vehicle detection and tracking method based on deep learning. Firstly, a deep learning-based model is established for training vehicle datasets of different scenarios, and the datasets are labeled and format normalized. Then uses the K-means clustering algorithm to anchor the frame, establishes a neural network framework based on the YOLOv3 SPP algorithm, and uses the non-maximum suppression (NMS) algorithm to obtain the final prediction frame. At last, the neural network model is trained. Meanwhile, it is tested and evaluated. The study shows that this model can accurately detect and track multi-target vehicles.

Key words: Deep learning; Multi-target vehicle; K-means clustering algorithm; YOLOv3 SPP algorithm; Non-maximum simulation algorithm