Kinematic Model based Visual Odometry for Differential Drive Vehicles

Abstract

This work presents KMVO, a ground plane based visual odometry that utilizes the vehicle’s kinematic model to improve accuracy and robustness. Instead of solving a generic image alignment problem, the motion parameters of a differential drive vehicle can be directly estimated from RGB- D image data. In addition, a method for outlier rejection is presented that can deal with large percentages of outliers. The system is designed to run in real time on a single thread of a mobile CPU. The results of the proposed method are compared to other publicly available visual odometry and SLAM methods on a set of nine real world image sequences of different indoor environments.

BibTeX

@inproceedings{JordanECMR2017,
  title = {Kinematic Model based Visual Odometry for Differential Drive Vehicles},
  author = {Jordan, Julian and Zell, Andreas},
  booktitle = {IEEE 8th European Conference on Mobile Robots (ECMR)},
  year = {2017},
  address = {ENSTA ParisTech, Université Paris-Saclay, France},
  month = sept,
  abstract = {This work presents KMVO, a ground plane based
visual odometry that utilizes the vehicle’s kinematic model
to improve accuracy and robustness. Instead of solving a
generic image alignment problem, the motion parameters of a
differential drive vehicle can be directly estimated from RGB-
D image data. In addition, a method for outlier rejection is
presented that can deal with large percentages of outliers. The
system is designed to run in real time on a single thread of a
mobile CPU.
The results of the proposed method are compared to other
publicly available visual odometry and SLAM methods on a
set of nine real world image sequences of different indoor
environments.},
  days = {6-8},
  note = {},
}