A Visual Measurement Algorithm of Approaching Vehicle Speed Based on Deep Learning

DOI
  • Yurong Zhu
    School of Electrical and Electronic Engineering, Shanghai Institute of Technology
  • Huailin Zhao
    School of Electrical and Electronic Engineering, Shanghai Institute of Technology
  • Junjie Liu
    School of Electrical and Electronic Engineering, Shanghai Institute of Technology
  • Jinping Zhang
    School of Electrical and Electronic Engineering, Shanghai Institute of Technology
  • Xiaojun Ji
    Shanghai Gazecloud Software Technology Co.Ltd. Shanghai Institute of Technology

説明

With the urbanizational process expediting and the national economy developing rapidly and healthily, the amount of private cars is on the rise, and traffic accidents occur frequently due to speeding and other reasons, and the difficulty of traffic supervision has also increased. This topic will use semantic segmentation and feature extraction and matching. Based on the video data of the traffic surveillance camera, an algorithm is designed to quickly calculate the matching of feature points in adjacent frames with low computing power to achieve the calculation. The same vehicle moves within the two frames of the target, so as to calculate the speed of the vehicle. Firstly, performing semantic segmentation based on deep learning, we choose a fully convolutional network to achieve semantic segmentation of depth maps, and distinguish the picture's principal part. After that, we can realize features extraction and mapping. The HOG algorithm is used on the matching step, the target's relative movement is calculated based on these matched point pairs to measure the moving speed of the vehicle. The experiment and the test prove that the system can realize the efficient speed measurement of moving vehicles.

収録刊行物

キーワード

詳細情報 詳細情報について

  • CRID
    1390010292571631488
  • DOI
    10.5954/icarob.2022.os33-1
  • ISSN
    21887829
  • 本文言語コード
    en
  • データソース種別
    • JaLC
    • Crossref
  • 抄録ライセンスフラグ
    使用不可

問題の指摘

ページトップへ