Robust estimation of stereo Disparity in bad weather

Introduction

Robust estimation of stereo disparity in bad weather

  • Existing stereo matching algorithms have difficulty to calculating disparity in various environment
  • Developed robust matching algorithm using geometrical features about surface of road and dynamic modeling

Development of stereo matching algorithm

  • Accuracy : 95.28%
  • Dense Registration : 100%

Development of stereo matching algorithm in bad weather

  • 2.9sec(12 it) ※ Baseline : 12cm, CPU Intel i7-based

Contents of Research

Development of robust and dense stereo matching method

Vertical-Horizontal Stixel aggregation

  • Obstacle detection on the ground using disparity => separating ground, obstacle, background => removal information of ground and background
  • Vertical stixel(connecting disparity label with same-value) occurs, when specific object exists
  • Vertical stixel(horizontal disparity label with same-value) appears on the ground part
  • Reducing error with stixel correction using relaxation labeling

Superior our algorithms

  • Offer various context recognition service from captured real-time depth information around vehicle
  • Dynamic obstacle detection from depth image
  • Detect safety region from experimental result

Experimental Results

Publication

  1. Kwang Hee Won, Soon Ki Jung, Billboard sweep stereo for obstacle detection in road scenes, Electronics Letters, Vol.48, No.24, pp.1528-1530, ISSN.00135194, 2012 (2012.11.22)
  2. Kwang Hee Won, Soon Ki Jung, hSGM: Hierarchical Pyramid Based Stereo Matching Algorithm, Lecture Notes in Computer Science, Vol.6915, pp.693-701, ISSN.03029743, 2011 (2011.08.22)
  3. Kwang Hee Won, Jongwoo Son, Soon Ki Jung, Stixels Estimation through Stereo Matching of Road Scenes, International Conference on Research in Adaptive and Convergent System (ACM RACS 2014), pp.116-120, 2014 (2014.10.05 ~ 2014.10.08)
  4. 정근호, 박정필, 원광희, 정순기, 스틱셀을 활용한 강인한 시차맵 추정, 한국 컴퓨터 그래픽스 학회 2015 학술대회, pp.53-54, 2015 (2015.07.13 ~ 2015.07.16)