Graph view not generated with RGB monocular video input (Missing some information)

[osamakhamees92-stack]

Hi @matlabbe

System

• OS: Ubuntu 22.04 (running inside VirtualBox)
• RTAB-Map version: 0.23.3

Input

• RGB monocular video (AVI / MP4)
• Recorded using phone camera
• Not stereo
• Not RGB-D
• No depth information
• Camera calibration YAML file created and loaded

Configuration

• Source type: RGB → Video file
• Odometry: Enabled
• Odometry → Visual Registration: 2D → 2D
• ICP: Disabled
• Stereo: Disabled
• RGB-D: Disabled
• LiDAR: Disabled
• Camera viewer shows the video correctly

Expected behavior

When playing the RGB video, Graph View should display nodes and edges generated from visual odometry.

Actual behavior

The video plays correctly in the camera viewer, but no nodes are created in Graph View.
The following error is repeatedly printed in the console: Console error:
[ERROR] OdometryThread.cpp::addData() Missing some information (images empty or missing calibration)!

[osamakhamees92-stack]

Thank you for the clarification.

So to confirm:
Using RGB monocular video input (without depth, stereo, or external odometry),
RTAB-Map is not able to generate odometry and thus no graph view can be created.
The video plays correctly, but no nodes/edges are added to the graph.

This matches the observed behavior.

[matlabbe]

RTAB-Map cannot do pure monocular VSLAM. It can do some at some extent (only back-end) if external odometry is provided (for example on a phone, ARKit or ARCore poses can be used) and Mem/StereoFromMotion is enabled. That is what we are using at 8:45 in this video https://www.youtube.com/watch?v=rVpIcrgD5c0 so that we can do VSLAM (detecting loop closures) on any phone (without depth camera). Note that 3D reconstruction or loop closure detection won't be as accurate than with a depth/stereo camera.

[osamakhamees92-stack]

thank you ... but i want just the graph view , not 3D map

[matlabbe]

Without metric data, we cannot generate a graph in the Graph View. For appearance-based loop closure detection, we can however show a graph like this:

To do it in RTAB-Map UI in appearance-based mode, open Preferences, uncheck this:

Play your video, then do this to generate the *.dot file:

See end of https://github.com/introlab/rtabmap/wiki/Cplusplus-Loop-Closure-Detection to export that dot file into a picture/pdf.

[osamakhamees92-stack]

Just a question about chessboard calibration... Should the calibration camera be the same camera used for video recording (you can't use a laptop camera for calibration and a phone camera for video recording)?... Should the video resolution be 1080x1920?... And should the x, y, and size values ​​be greater than 70%?

[matlabbe]

The same camera should be used. The highest resolution that the camera support is ideal (for best calibration), afterwards we can re-scale the calibration parameters depending on the resolution chosen on runtime.