A Framework for Vision Based Bearing Only 3D SLAM
Patric Jensfelt, Danica Kragic , John Folkesson and Mårten Björkman
Abstract:
In the last few years, a number of simultaneous localization and
mapping (SLAM) systems have been presented where visual sensors are
used for feature extraction. The examples include 2D and 3D maps
where either a single, stereo or a multi-camera system was used. In
many real applications, a single camera system is the most appealing
one due to the low cost for the hardware. However, single camera
SLAM suffers from the problem that features' depth cannot be
determined from a single image.
This paper presents a framework for 3D vision based bearing only SLAM
using a camera. The focus is on the management of the features so that
we can achieve real-time performance in extraction, matching and loop
detection. For matching image features to map landmarks a modified,
rotationally variant SIFT descriptor is used which is more appropriate
to the situation of a robot mounted camera. We also use a different
feature extractor than originally proposed for SIFT which is more
precisely localized in the image than the more diffuse
Difference-of-Gaussians features.
The richness of image data is beneficial to matching of features
between frames but also leads to a large number of poorly localized
features. The goal here is to have both robust matching from many
vision features and a few high quality 3D localized features for use
by the SLAM algorithm.
The framework has been combined with an EKF implementation for
SLAM. Experiments performed in indoor environments are
presented. These experiments demonstrate the validity and
effectiveness of the approach. In particular they show how the robot
is able to successfully match current image features to the map when
revisiting an area.
BibTeX Entry:
@InProceedings{Jensfelt06a,
author = {Patric Jensfelt and Danica Kragic and John Folkesson and M{\aa}rten Bj{\"o}rkman},
title = {A Framework for Vision Based Bearing Only {3D} {SLAM}},
booktitle = {Proc.~of the IEEE International Conference on Robotics and Automation (ICRA'06)},
year = 2006,
address = {Orlando, FL},
month = may
}
Download: pdf (996k)