stereo camera visual odometry

A very lightweight visual odometry with local bundle adjustment (BA).

• Test running on kitti dataset. Processing each frames with 20 ms.

• Test running on DJI M300 front stereo camera, the video is recorded around Concordia University, Montreal, CA.

• Mapping with octomap.

NOTE:

• Green frame indicates current camera pose, while the red and black point clouds are the active and inactive landmarks respectively.

Features

• Stereo Camera based visual odometry.
• LK optical flow is used to track the keypoints.
• Lost tracking then relocate.
• Insert keyframes according to the keypoints number.
• Local bundle adjustment.

Dependencies

• g2o

  optimization in local bundle adjustment and frontend camera pose estimation.

• pangolin

  visualization

• Eigen & Sophus

  basic matrix operations and Lie Algebra operations.

• OpenCV 4

  general image processing methods and basic data structures.

Future Work

Integrate with forest fire detection system to locate in GPS denied environments.

Acknowledgement

"14 Lectures on Visual SLAM: From Theory to Practice" Xiang Gao