A fast local descriptor for dense matching - Robot Vision Group

Vision-Based Odometry and SLAM for
Medium and High Altitude Flying UAVs
J Intell Robot Syst 2009
F.Caballero L.Merino J.Ferruz A.Ollero
University of Seville

Odometry
odometry is composed from the Greek words hodos (meaning
"travel", "journey") and metron (meaning "measure").
Odometry is the use of data from the movement of actuators to
estimate change in position over time.

Introduction
Describing a monocular visual odometer which could be used as a
backup system when the accuracy of GPS is reduced to critical
levels.
Integrating the visual odometer into a SLAM scheme in order to
reduce the impact of cumulative errors in odometry-based position
estimation approaches.

Homography-Based Visual Odometry for UAVs
Robust Homography Estimation
The algorithm for homography computation basically consists of a
point-feature tracker that obtains matches between images, and a
combination of least median of squares and M-Estimator for outlier
rejection and accurate homography estimation from these matches.
Two Problems:
non-uniform distribution of the features along the images
the parallax effect will increase, and the planarity assumption will
not hold

Homography-Based Visual Odometry for UAVs
Robust Homography Estimation
Complete homography. Least median of squares (LMedS) is used for
outlier rejection and a M-Estimator to compute the final result. This
model is used if more than the 65% of the matches are successfully
tracked.
Affine homography. If the percentage of success in the tracking step
is between 40% and 65%, then the LMedS is not used, given the
reduction in the number of matches. A relaxed M-Estimator (soft
penalization) is carried out to compute the model.
Euclidean homography. If the percentage is below 40%, the set of
data is too noisy and small to apply non-linear minimizations. The
model is computed using least-squares.

Homography-Based Visual Odometry for UAVs
Geometry of Two Views of the same Plane

Homography-Based Visual Odometry for UAVs
Motion Estimation from Homography

Homography-Based Visual Odometry for UAVs
Motion Estimation from Homography
where

Homography-Based Visual Odometry for UAVs
Correct Solution Disambiguation
1. The distance among and the rest normal in S n is computed.
2. is set to an initial value.
3. For the current value ,check if there exist an unique solution.
4. If no solution is found, increase the value of and try again with
the step 3. If multiple solutions were found decrease and try
again with step 3. If an unique solution was found, then finish.

Homography-Based Visual Odometry for UAVs
An Estimation of the Uncertainties

Homography-Based Visual Odometry for UAVs
Experimental Results
Sensors:
DGPS
IMU
Camera

Homography-Based Visual Odometry for UAVs
Experimental Results
Vision-based technique VS DGPS

Homography-Based Visual Odometry for UAVs
Experimental Results
Vision-based technique VS DGPS

Homography-Based Visual Odometry for UAVs
Experimental Results
Vision-based technique VS IMU

Homography-Based Visual Odometry for UAVs
Experimental Results
Vision-based technique VS IMU

Application of Homography-Based Odometry to the
SLAM Problem
The State Vector
Measurement Vector

Application of Homography-Based Odometry to the
SLAM Problem
Prediction Stage
Updating Stage

Application of Homography-Based Odometry to the
SLAM Problem
Filter and Landmarks Initialization

Application of Homography-Based Odometry to the
SLAM Problem
Experimental Results
SLAM VS DGPS

Application of Homography-Based Odometry to the
SLAM Problem
Experimental Results
SLAM VS DGPS

Application of Homography-Based Odometry to the
SLAM Problem
Experimental Results
SLAM VS DGPS SLAM&IMU VS DGPS