Jump to Content
  1. Queen's
  2. Smith Eng.
  3. ECE

Estimation, Search, and Planning (ESP) Research Group

SE(3) multimotion estimation through occlusion

Authors
  1. Kevin M. Judd
  2. Jonathan D. Gammell
Publication Date
Abstract

Visual motion estimation is an integral and well- studied challenge in autonomous navigation. It is significantly more challenging in highly dynamic environments with multiple moving objects. This paper introduces an approach to this multimotion estimation problem capable of estimating the full SE(3) trajectory of every motion in a scene, even when motions become occluded. The Multimotion Visual Odometry (MVO) pipeline employs multilabeling techniques and continuous motion models to estimate all motions simultaneously, including the camera egomotion. Motion closure is used to recognize when trajectories become unoccluded, and the motion models are used to interpolate the occluded estimates. The estimation performance of the pipeline is demonstrated on real-world trajectory data from the Oxford Multimotion Dataset.

Publication Details
Type
Abstract-Refereed Conference Paper
Conference
Long-Term Human Motion Prediction (LHMP) Workshop, IEEE International Conference on Robotics and Automation (ICRA)
Location
Montréal, QC, Canada
Manuscript
Google ScholarGoogle Scholar
Google Scholar
BibTeX Entry
@inproceedings{judd_lhmp19,
author = {Kevin M Judd and Jonathan D Gammell},
title = {{SE(3)} multimotion estimation through occlusion},
booktitle = {Proceedings of the Long-Term Human Motion Prediction ({LHMP}) Workshop, {IEEE} International Conference on Robotics and Automation ({ICRA})},
year = {2019},
address = {Montréal, QC, Canada},
month = {24 } # may,
}