Visual SLAM on a Real World Dataset - Kapernikov/tech-session-visual-odometry GitHub Wiki

KITTI Vision Benchmark Suite is a project by Karlsruhe Institute of Technology and Toyota Technological Institute of Chicago and consists of datasets recorded from a moving car equipped with various sensors (IMU, GPS, LIDAR, Stereo Camera). We will use a reduced dataset converted into a convenient rosbag format with:

  • Left and right camera images and calibrations
  • GPS data
  • IMU data
  • Coordinate transforms (car coordinate frames and pose ground truth)

Instructions

  1. Download a KITTI dataset rosbag and put it in /path/to/tech-session-visual-odometry/datasets/kitti/
  2. Open a terminal and execute the following command to launch RTAB-Map for the KITTI dataset you just downloaded:
$ roslaunch visual_odometry kitti_dataset_vslam.launch

kitti.png

As you can see the estimated trajectory is not identical to the ground truth.

As part of the last exercise you get to try to improve the results by tuning the various parameters in /path/to/tech-session-visual-odometry/visual_odometry/config/rtabmap_kitti.ini