Combining Kalman filtering and vision-based trajectory estimation

Fuan Tsai, Huan Chang, Yih Shyih Chiou, Yao Tsung Lin, Shin Hui Li

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

The objective of this research is using video frames or highly overlapped images to obtain the camera orientation and translation parameters for trajectory estimation. One form of measurements comes from the computer vision community where successive frames from a camera approximately looking at the ground can be used to compute the translation between frames. In order to deal with the corner effect and registration problems, normalized cross correlation (NCC) is used to recognize the landmarks as the control points. The developed algorithms consist of four steps for camera trajectory estimation: (1) feature points detection and matching; (2) homography calculation; (3) control points detection and registration; (4) motion estimation. The first step is data decimation in order to reduce data amount and increase computation efficiency. Then, corner detector is employed to extract the feature points and match them between the frames using sum of absolute differences (SAD). The metric part of homography can provide the camera orientation and translation parameters according to the conjugate points between each frames. After that, this research uses RANSAC to remove the outlier of the previous step. NCC is then used to check if the camera pass thought the control points or not. This study compared the results with on-site measurement and with or without Kalman filter. Examples of applying the developed algorithm to tracking applications demonstrate the effectiveness of the methods. The example in outdoor environment indicates that the developed method for determining camera orientation and translation parameters can be used in providing initial conditions to real-time positioning and tracking in indoor or outdoor environments.

Original languageEnglish
Title of host publication33rd Asian Conference on Remote Sensing 2012, ACRS 2012
Pages2304-2311
Number of pages8
StatePublished - 2012
Event33rd Asian Conference on Remote Sensing 2012, ACRS 2012 - Pattaya, Thailand
Duration: 26 Nov 201230 Nov 2012

Publication series

Name33rd Asian Conference on Remote Sensing 2012, ACRS 2012
Volume3

Conference

Conference33rd Asian Conference on Remote Sensing 2012, ACRS 2012
Country/TerritoryThailand
CityPattaya
Period26/11/1230/11/12

Keywords

  • Computer vision
  • Homography
  • Motion estimation
  • Normalized cross correlation
  • Visual navigation

Fingerprint

Dive into the research topics of 'Combining Kalman filtering and vision-based trajectory estimation'. Together they form a unique fingerprint.

Cite this