Precise Segmentation and Position Estimation of Pedestrians By the Combination of the HOG Classifier and the S-T MRF Model

This paper presents a general algorithm for pedestrian detection by on-board monocular camera which can be applied to cameras of various view ranges. Under the assumption that motion of background can be nearly approximated as a linear function, the Spatio-Temporal MRF (S-T MRF) model segments foreground objects. Those foreground objects contain both pedestrian and non-pedestrian urban objects, verification by a cascaded classifier is conducted. However, segmentation result sometime contains error such as shrunk or inflated Region of Interest (ROI). The authors improved their system by implementing two types of feedback algorithm for ROI correction using the Kalman filter and by combining the results of motion classifier and HOG classifier. They confirmed that those ROI Corrections help the system decrease the false negative rate and extract highly accurate pedestrian trajectory. They expect that the trajectory could be used as a useful source for measuring the possibility of collision with pedestrian.

• Supplemental Notes:
  • This paper was sponsored by TRB committee ANF10 Pedestrians
• Corporate Authors:

  Transportation Research Board

  500 Fifth Street, NW
  Washington, DC  United States  20001
• Authors:
  • Kim, HyungKwan
  • Shibayama, Yuuki
  • Kamijo, Shunsuke
• Conference:
  • Transportation Research Board 91st Annual Meeting
  • Location: Washington DC, United States
  • Date: 2012-1-22 to 2012-1-26
• Date: 2012

Language

• English

Media Info

• Media Type: Digital/other
• Features: Figures; References; Tables;
• Pagination: 16p
• Monograph Title: TRB 91st Annual Meeting Compendium of Papers DVD

Subject/Index Terms

• TRT Terms: Algorithms; Cameras; Crash avoidance systems; Pedestrian detectors; Pedestrian safety; Proximity detectors
• Subject Areas: Highways; Pedestrians and Bicyclists; Safety and Human Factors; Vehicles and Equipment; I91: Vehicle Design and Safety;

Filing Info

• Accession Number: 01372466
• Record Type: Publication
• Report/Paper Numbers: 12-4526
• Files: TRIS, TRB
• Created Date: Jun 14 2012 10:55AM