2
$\begingroup$

I have tested my SLAM algorithm on EuRoC, TUM and KITTI Datasets. But I would like to test on my own custom dataset. Moreover, I already rectify how to make my own dataset in TUM dataset format, but I didn't know how to calculate the ground truth of my own data set so that I calculate the accuracy of my SLAM algorithm.

Would kindly provide me any suggestions and recommendation?

Thank you.

Improve this question
$\endgroup$

2 Answers 2

Reset to default
1
$\begingroup$

This depends on how you acquire your data. You can either manually label ground truth or use sensors such as GPS. I have used GPS data with RTK corrections (sub cm accuracy) as ground truth in the past. You could also place markers such as AR/april tags at known locations in your surroundings and use their measurements in your dataset as ground truth.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Thank you for the reply. for example I placed a black thick line on the white ground. I calculate it's distance and thickness with the help of tap. what you think it will help? $\endgroup$
    – Qamar Ul Islam
    Commented Mar 31, 2021 at 17:53
  • $\begingroup$ How do you deal with imprecise initial orientation? Slight errors in initial orientation well destroy evaluating the accuracy of your system $\endgroup$
    – FourierFlux
    Commented Jan 5 at 19:00
0
$\begingroup$

If you want to calculate real-world ground truth poses for evaluating your SLAM algorithm in indoor environments, there are several methods you might consider:

1. Motion Capture (MoCap) or Laser Tracker Systems: MoCap systems use multiple high-speed cameras to track reflective markers, offering nearly millimeter-level pose accuracy. However, they are very costly, require complex setup, and are limited to small, controlled areas.

2. Fiducial Markers (e.g., ArUco markers): These provide a cost-effective solution by detecting known markers placed around the environment. They involve manual setup and calibration, and their accuracy can be affected by lighting and marker placement.

3. Using a Prior Reference Map and SLAM2REF: Creating a high-resolution reference 3D map with a terrestrial laser scanner or a reliable mapping system, like those from NavVis, can be effective. This approach covers larger areas and offers high accuracy. Once you have a prior map, SLAM2REF provides a solution for accurate, real-world ground truth calculation in indoor and outdoor environments. This project, which I developed, aims to assist researchers in automatically aligning and correcting their LiDAR-based SLAM data with a reference map or across multiple sessions, providing precise 6-DoF poses with an accuracy of up to 3 cm. For more information on SLAM2REF and how it works, you can check out its GitHub repository.

Improve this answer
New contributor
Miguel Vega is a new contributor to this site. Take care in asking for clarification, commenting, and answering. Check out our Code of Conduct.
$\endgroup$
4
  • $\begingroup$ This seems like spam for your Github repo. You should disclose that you are the author. Also, this is similar to your other answer $\endgroup$
    – Greenonline
    Commented yesterday
  • $\begingroup$ Thank you for pointing that out. I apologize if it seemed like spam. It was not my intention. I am indeed the author of the project mentioned. I intend to share information about my work and contribute to the discussion and development of better solutions. Regarding the authorship, I think you are right, thanks for the hint! I edited the post, and I’ll make sure to also disclose authorship clearly in future posts. If you have any questions or need further details about the project, feel free to ask! $\endgroup$
    – Miguel Vega
    Commented yesterday
  • 1
    $\begingroup$ Welcome to Robotics Miguel Vega. The way this was written it looks like you may be associated with the product being promoted, so please read What kind of behavior is expected of users? and How not to be a spammer. The community here tends to vote down overt self-promotion and flag it as spam, but if you post good, relevant answers then it's okay if some (but not all) of the content happens to be about your product/website/company. But you must disclose any affiliation in your answers. $\endgroup$
    – Chuck ♦
    Commented yesterday
  • $\begingroup$ Oh, I see. Yes, you are right. I think I overpromoted my project. Sorry about that. I edited the post. Is it better now? $\endgroup$
    – Miguel Vega
    Commented yesterday

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.