Make-ORB-SLAM2-RGBD-dataset

This repository helps you make Custom Dataset (RGB-D) used in ORB_SLAM2.

When you excute ORB_SLAM2 RGB-D mode with your own dataset, you need to create

• rgb.txt : the timestamps and RGB images list
• depth.txt : the timestamps and depth images list
• association.txt : the list of matching timestamp, rgb filename, timestamp, and depth filename.
• rgb/ : the folder of RGB image sequence

- rgb /  
    |---- rgb00002.png
    |---- rgb00003.png
    |---- rgb00004.png
    |---- rgb00005.png ..

• depths/ : the folder of depth image sequence

- depths /  
    |---- depth00002.png
    |---- depth00003.png
    |---- depth00004.png
    |---- depth00005.png ..

You can make these things with Real Sense Camera! 😊

I tested this repository with Real Sense D435 & L515!

1. Prerequisites

• pyrealsense2
• numpy
• opencv-python
• python

2. Make RGB and depth sequence

With rs-get-Sequence.py you could get RGB and depth sequence.

First you would make empty folders which name rgb/ and depths/.
After you connect your real sense camera, just run the python rs-get-Sequence.py in terminal.

When you run this code, press 's' button and 'enter' to start.
You could visualize RGB image and Depth colorized image.
And save images and rgb.txt and depth.txt!

3. Make a little modification about rgb.txt and depth.txt

When you get rgb.txt and depth.txt, It looks like this.
For example rgb.txt,

0.00585412979126 rgb/rgb00002.png
0.00706601142883 rgb/rgb00003.png
00.0154299736023 rgb/rgb00004.png
00.0148220062256 rgb/rgb00005.png
00.0163168907166 rgb/rgb00006.png
00.0162618160248 rgb/rgb00007.png
00.0166661739349 rgb/rgb00008.png
00.0166158676147 rgb/rgb00009.png
00.0157341957092 rgb/rgb00010.png
00.0166420936584 rgb/rgb00011.png
00.0167319774628 rgb/rgb00012.png ...

You should fix this txt file to follow ORB_SLAM2 rgb.txt rules!

Just add these 3 lines to fit the format.

# rgb maps
# file: 'example of rgb sequence'
# timestamp filename

You could get this result.

# rgb maps
# file: 'example of rgb sequence'
# timestamp filename
0.00585412979126 rgb/rgb00002.png
0.00706601142883 rgb/rgb00003.png
00.0154299736023 rgb/rgb00004.png
00.0148220062256 rgb/rgb00005.png
00.0163168907166 rgb/rgb00006.png
00.0162618160248 rgb/rgb00007.png
00.0166661739349 rgb/rgb00008.png
00.0166158676147 rgb/rgb00009.png
00.0157341957092 rgb/rgb00010.png
00.0166420936584 rgb/rgb00011.png
00.0167319774628 rgb/rgb00012.png ...

4. Make associations.txt

You could make associations.txt with associate.py.
When you run associate.py, you need to provide 3 argments.

• rgb_path : Path of rgb.txt
• depth_path : Path of depth.txt
• output : Output name of txt file

You run the code like this.

• python associate.py --rgb_path [rgb path] --depth_path [depth path] --output [Name of txt file].
• (For example) python associate.py --rgb_path rgb.txt --depth_path depth.txt --output associations.txt.

It's Done! Just excute ORB_SLAM2 RGB-D Mode 😆