Context Enhanced Traffic Segmentation: traffic jam and road surface segmentation from aerial image

Yubo Wang, Zhao Wang, Yuusuke Nakano, Ken Nishimatsu, Katsuya Hasegawa, Jun Ohya

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Traffic jam detection and density estimation of aerial images have been widely utilized in various scenarios, such as vehicle routing and city management. Rather than directly detecting traffic jams or estimating density, traffic condition analysis based on traffic jam segmentation could yield more accurate results. Therefore, we propose a Context Enhanced Traffic Segmentation Model to simultaneously segment the traffic jam parts and road surface. However, there are two critical issues for traffic jam segmentation in aerial images: one is the scale variation problem and the other is the difficulty of accurately segmenting ambiguous traffic jam boundaries. Thus, we design a traffic estimation module to handle the scale variation problem and present a context attention module to enhance the boundary of traffic jam segmentation. Experimental results demonstrate the superiority of our proposed method.

Original languageEnglish
Title of host publicationIVMSP 2022 - 2022 IEEE 14th Image, Video, and Multidimensional Signal Processing Workshop
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781665478229
DOIs
Publication statusPublished - 2022
Event14th IEEE Image, Video, and Multidimensional Signal Processing Workshop, IVMSP 2022 - Nafplio, Greece
Duration: 2022 Jun 262022 Jun 29

Publication series

NameIVMSP 2022 - 2022 IEEE 14th Image, Video, and Multidimensional Signal Processing Workshop

Conference

Conference14th IEEE Image, Video, and Multidimensional Signal Processing Workshop, IVMSP 2022
Country/TerritoryGreece
CityNafplio
Period22/6/2622/6/29

Keywords

  • aerial image segmentation
  • self-attention mechanism
  • traffic jam segmentaion

ASJC Scopus subject areas

  • Computer Vision and Pattern Recognition
  • Signal Processing
  • Media Technology

Fingerprint

Dive into the research topics of 'Context Enhanced Traffic Segmentation: traffic jam and road surface segmentation from aerial image'. Together they form a unique fingerprint.

Cite this