Project 2: Visual Odometry

ENPM 673, Robotics Perception
Project 2: Visual Odometry.

1 Visual Odometry - 100Pts
Visual Odometry is a crucial concept in Robotics Perception for estimating the trajectory of
the Robot (camera to be precise). The concepts involved in Visual Odometry are quite the
same for SLAM which needless to say is an integral part of Perception.
2 Details
You are given frames from a driving car, scripts to extract intrinsic parameters
and the output should be a plot for the trajectory of the camera - DATASET.
2.1 Pipeline
• The input images are in Bayer format from which you can recover the color images
using the demosaic function with GBRG alignment.
• Extract the camera parameters using ReadCameraModel.m as follows:
[fx, fy, cx, cy, G camera image, LUT] = ReadCameraModel(‘./stereo/centre’,
‘./model’); // top line of stereo narrow left.txt gives fx, fy, cx, cy
• Undistort the current frame and next frame using UndistortImage.m.
• Find point correspondences using ANY keypoint operator of your choice.
• Estimate the Fundamental matrix using these point correspondences.
• Recover the Rotation and Translation matrices from the Fundamental matrix.
• Plot the position of the camera center (for each frame) based on the rotation and
translation parameters between successive frames.
You should implement the functions to estimate Essential Matrix and also to recover
rotation/translation matrices and NOT use Matlab’s Computer Vision Toolbox or any
third party code.
1
3 Extra Credit - upto 40 Pts
Compare your result against the rotation/translation parameters recovered using relativeCameraPose,
estimateFundamentalMatrix from Matlab’s Computer Vision Toolbox. You can plot both
the trajectories and report the accumulated drift in trajectory from your implementation versus the one where you use Matlab’s in-built functions.
4 Submission Guidelines
Your submission SHOULD be a ZIP folder (no other file extensions) with the nam-ing
convention YourDirectoryID proj2.zip on to ELMS/Canvas. Additionally follow the
guidelines given below:
1. You will have a parent directory P2 Submission.
2. Under P2 Submission/VisualOdometry you will have three sub-folders code, input
and output.
3. You SHOULD also submit a report (Report.pdf) under P2 Submission/VisualOdometry
folder.
5 Useful Resources
• Check out relativeCameraPose from Matlab’s Computer Vision Toolbox.
• Watch this excellent video lecture to understant How to Estimate Essential Matrix
6 Acknowledgement
Dataset used is by courtesy of Oxford’s Robotics Institute.
7 Collaboration Policy
You are allowed to discuss the ideas with fellow students, but you need to give credits in the
report. But the code you turn-in should be your own and if you DO USE (try not to and
it is not permitted) other external codes - do cite them and you might get partial credits.
For other honor code refer to the University of Maryland Honor Pledge.
DISCLAIMER: You should take the effort to search online for any help regarding function
usage, however, any concept related queries can be discussed in TA Office Hours.
2