Trajectory estimation based on globally consistent homography

Siwook Nam, Hanjoo Kim, Jaihie Kim

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

We propose a method for estimating trajectories of objects moving on a world plane. Motivation of this work is to estimate the field trajectories of players and the ball from uncalibrated monocular soccer image sequences. In order to find mappings between images and the plane, four feature points, no three of them are collinear, should exist in each image. However, many soccer images do not satisfy that condition. In that case, the object positions in the given image are mapped to those in the reference image of the sequence, and then mapped again to those in the soccer field. Conventional mapping between given image and the reference image is given by concatenation of homographies between consecutive image pairs. However, small correspondence error is accumulated in the concatenation of homographies over long image sequence. To overcome this problem, we compute globally consistent homographies for all the feature-sufficient images by solving a sparse linear system of equations which consists of consecutive and non-consecutive homographies of feature-sufficient image pairs. Experimental results with real and synthetic soccer data show that the proposed method is more accurate than existing method.

Original languageEnglish
Pages (from-to)214-221
Number of pages8
JournalLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume2756
Publication statusPublished - 2003 Dec 1

Fingerprint

Homography
Trajectories
Trajectory
Linear systems
Concatenation
Image Sequence
Consecutive
Sufficient
Sparse Linear Systems
Linear system of equations
Collinear
Feature Point
Synthetic Data
Moving Objects
Ball
Correspondence

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

@article{c957c9625357407db5a8329369f4dc0a,
title = "Trajectory estimation based on globally consistent homography",
abstract = "We propose a method for estimating trajectories of objects moving on a world plane. Motivation of this work is to estimate the field trajectories of players and the ball from uncalibrated monocular soccer image sequences. In order to find mappings between images and the plane, four feature points, no three of them are collinear, should exist in each image. However, many soccer images do not satisfy that condition. In that case, the object positions in the given image are mapped to those in the reference image of the sequence, and then mapped again to those in the soccer field. Conventional mapping between given image and the reference image is given by concatenation of homographies between consecutive image pairs. However, small correspondence error is accumulated in the concatenation of homographies over long image sequence. To overcome this problem, we compute globally consistent homographies for all the feature-sufficient images by solving a sparse linear system of equations which consists of consecutive and non-consecutive homographies of feature-sufficient image pairs. Experimental results with real and synthetic soccer data show that the proposed method is more accurate than existing method.",
author = "Siwook Nam and Hanjoo Kim and Jaihie Kim",
year = "2003",
month = "12",
day = "1",
language = "English",
volume = "2756",
pages = "214--221",
journal = "Lecture Notes in Computer Science",
issn = "0302-9743",
publisher = "Springer Verlag",

}

TY - JOUR

T1 - Trajectory estimation based on globally consistent homography

AU - Nam, Siwook

AU - Kim, Hanjoo

AU - Kim, Jaihie

PY - 2003/12/1

Y1 - 2003/12/1

N2 - We propose a method for estimating trajectories of objects moving on a world plane. Motivation of this work is to estimate the field trajectories of players and the ball from uncalibrated monocular soccer image sequences. In order to find mappings between images and the plane, four feature points, no three of them are collinear, should exist in each image. However, many soccer images do not satisfy that condition. In that case, the object positions in the given image are mapped to those in the reference image of the sequence, and then mapped again to those in the soccer field. Conventional mapping between given image and the reference image is given by concatenation of homographies between consecutive image pairs. However, small correspondence error is accumulated in the concatenation of homographies over long image sequence. To overcome this problem, we compute globally consistent homographies for all the feature-sufficient images by solving a sparse linear system of equations which consists of consecutive and non-consecutive homographies of feature-sufficient image pairs. Experimental results with real and synthetic soccer data show that the proposed method is more accurate than existing method.

AB - We propose a method for estimating trajectories of objects moving on a world plane. Motivation of this work is to estimate the field trajectories of players and the ball from uncalibrated monocular soccer image sequences. In order to find mappings between images and the plane, four feature points, no three of them are collinear, should exist in each image. However, many soccer images do not satisfy that condition. In that case, the object positions in the given image are mapped to those in the reference image of the sequence, and then mapped again to those in the soccer field. Conventional mapping between given image and the reference image is given by concatenation of homographies between consecutive image pairs. However, small correspondence error is accumulated in the concatenation of homographies over long image sequence. To overcome this problem, we compute globally consistent homographies for all the feature-sufficient images by solving a sparse linear system of equations which consists of consecutive and non-consecutive homographies of feature-sufficient image pairs. Experimental results with real and synthetic soccer data show that the proposed method is more accurate than existing method.

UR - http://www.scopus.com/inward/record.url?scp=35248898784&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=35248898784&partnerID=8YFLogxK

M3 - Article

VL - 2756

SP - 214

EP - 221

JO - Lecture Notes in Computer Science

JF - Lecture Notes in Computer Science

SN - 0302-9743

ER -