Odometry Evaluation Pre-processing Tools

A collection of Python scripts and example data designed to pre-process odometry results—converting them between TUM and KITTI formats, applying frame transformations, and generating outputs suitable for subsequent evaluation with EVO or the RPG Trajectory Evaluator.

This work is from the weekly seminar in RPM Robotics Lab

[Presentation Slide] [Video (Not English)]

Repository Layout

├── Example/
│   ├── point_lio_helipr_lidar.txt
│   ├── Ouster_gt_time.txt
│
├── frame_changer/
│   └── drawer_est_lidarcoord.py
│
├── tum_kitti_converter/
│   ├── 1_kitti_to_tum.py
│   └── 2_tum_to_kitti.py
│
└── README.md

1. Example/

Contains sample odometry text files with timestamps, poses, etc. You can test the converter scripts on these.

2. frame_changer/

• drawer_est_lidarcoord.py
  Demonstrates:
  • Loading odometry text files into GTSAM’s Pose3.
  • Applying coordinate transformations or a calibration offset (e.g., changing the origin or frame).
  • Saving re-oriented results to a new file.
  • Visualizing in 3D using Open3D.

3. tum_kitti_converter/

• 1_kitti_to_tum.py and 2_tum_to_kitti.py
  Convert between KITTI and TUM pose files.
  • kitti → tum: each line of kitti_pose_file has [time r11 r12 r13 tx r21 r22 r23 ty r31 r32 r33 tz], and it outputs TUM format [time tx ty tz qx qy qz qw].
  • tum → kitti: the script takes [time tx ty tz qx qy qz qw] and writes a 3×4 transformation flattened to a single line plus the timestamp.

---

Installation & Dependencies

1. Python 3.7+
2. Core Python Libraries: NumPy, SciPy, CSV, etc.
   • pip install numpy scipy
3. Open3D for 3D visualization (optional, required by drawer_est_lidarcoord.py):
   • pip install open3d
4. GTSAM for advanced pose transformations (optional).
   • If you skip GTSAM, remove or comment the relevant lines in drawer_est_lidarcoord.py.
5. EVO for evaluation & plotting:
   • pip install evo --upgrade --no-binary evo

(Optional) The RPG evaluator can also be used if you prefer a different workflow or want additional metrics.

---

How to Use

A. Convert Between TUM & KITTI

1. KITTI → TUM

   cd tum_kitti_converter/
   python 1_kitti_to_tum.py --kitti_pose_file input_kitti.txt --tum_output_file output_tum.txt
   

2. TUM → KITTI

   cd tum_kitti_converter/
   python 2_tum_to_kitti.py --tum_file input_tum.txt --kitti_output_file output_kitti.txt
   

B. Change or Re-origin a Trajectory

cd frame_changer/
python drawer_est_lidarcoord.py

• Edits the input file path in traj_est_file.
• Adjust transformations in _source_to_target if you have a calibration offset or want to shift axes.
• Outputs a new “reorigin” text file.

---

Odometry Evaluation Tips (with EVO)

After you have an estimate (e.g., est.tum) and a reference (ground truth, e.g., reference.tum), you can run EVO to compute absolute or relative pose errors:

1. APE (Absolute Pose Error)

   evo_ape tum reference.txt estimate.txt --align --plot_mode xy --plot
   evo_ape tum reference.txt estimate.txt --align --plot_mode xy --plot --pose_relation angle_deg
   

   • --align tries to align the estimate to reference (SE(3) alignment).
   • --plot_mode xy plots in 2D overhead view.
   • --pose_relation angle_deg can measure orientation errors (in degrees).

2. RPE (Relative Pose Error)

   evo_rpe tum reference.txt estimate.txt --align --plot_mode xy --plot
   evo_rpe tum reference.txt estimate.txt --align --plot_mode xy --plot --pose_relation angle_deg
   

   • For relative/segment errors, you can specify distance or angle metrics.

3. Multiple Trajectories
   If you have multiple estimates:

   evo_traj tum est1.txt est2.txt est3.txt --ref reference.txt --align --plot_mode xy --plot
   

   • The --t_max_diff flag helps data association if your timestamps differ slightly.

Important Details

• Alignment: Even for translational error, ensure you align orientation first. If your sensor has a known roll/pitch offset or a calibration file, apply that before running EVO.
• Matplotlib Font: Some conferences (like IEEE) might disallow default fonts. For EVO’s plots, you may want to override the default font. One approach is editing your local matplotlibrc file:
  Find your matplotlibrc file:

  import matplotlib
  print(matplotlib.matplotlib_fname())

  Then open and edit that file. For example:

  font.family         : sans-serif
  font.sans-serif     : Arial
  

  Restart Python so the changes take effect.

Contributions

• Contributions are welcome—please open an issue or pull request! Maybe, there are many things we should do before the evaluation