The CMU Landmark-Based Pose Dataset

Robert T. Collins, Robotics Institute, Carnegie Mellon University

This is a 3D landmark-based pose estimation dataset from Carnegie Mellon University. The dataset contains measured 3D landmark points, and a set of pan-tilt observations of these landmarks from 5 cameras mounted on campus. This is enough information to compute the pose (location and orientation) of the cameras. My code for doing the pose computation is included. It is described in Section 4 of the paper
  R.Collins and Y.Tsin, "Calibration of an Outdoor Active Camera
  System," IEEE Computer Vision and Pattern Recognition (CVPR'99),
  Ft.Collins, CO, June 23-25, 1999.

In order to get the dataset, download CMUpose.tar and untar to get the following directories
In the data/ directory you will find
which is a file of GPS landmark data. Read the header to decipher the format. The position of the landmarks on campus is depicted in the gif file campuspts.gif. You will also find the following set of pan-tilt measurement files for each of 5 cameras on campus (measured by sighting a subset of the landmarks in the landmarks.dat file)
Again read the header to understand the format. In the src/ directory you will find my pose calibration code. It is perhaps the best example of how to read and use the data in the above files. Go to the src/ directory and type "make". I've compiled it under SGI IRIX and RedHat Linux without problems. It will make an executable called posecalib. Run "posecalib" with no arguments to see what the input arguments should be. To try a test run, type
  posecalib ../data/landmarks.dat ../data/prb_calib.dat 1
and compare the results with the file prbpose.out. If the results deviate significantly there is a problem. You are the alpha test site! Good luck.
  -Bob Collins 
   Robotics Institute
   Carnegie Mellon University
   5000 Forbes Ave
   Pittsburgh, PA 15213

Click here to go back to my home page.