laomeng0703/k4a_mkv2image

Tool to convert mkv files to image files for Azure Kinect

k4a_mkv2image

This is convert tool that works on cross-platform (Windows, Linux) for Azure Kinect.
This tool converts all data of each stream types (Color, Depth, Infrared) that contained in mkv file to image file.

Modifications

The project has been adapted to work with ETH3D/badslam and the exported datasets should now be compatible with the ETH SLAM Datasets format. The directory for color images is called "rgb" instead of "color" and text files for rgb, depth and calibration have been added. To prepare datasets for badslam use the following commands:

k4arecorder -c 720p -r 30 my_file.mkv												# starts recording until ctrl+c is pressed
k4a_mkv2image -i="my_file.mkv" -t=true -s=false -d=true                         	# creates a directory "my_file" with depth/rgb files.
python associate.py my_file/rgb.txt my_file/depth.txt > my_file/associated.txt		# associates rgb and depth frames. The python script can be found in the badslam docs

Then run badslam on the dataset. You might have to tune some settings like: "Maximum depth to use" -> 5, "Use photometric residuals" -> off (?), "Keyframe interval" -> lower a bit (but not too low!),

Sample

Command

k4a_mkv2image -i="./file.mkv" -s=true -d=true

Output

Image files are output in the following folder structure.
Image file name is "<file_index>_<device_timestamp>.jpg".
The file index is sequencial number for output images. The unit of device timestamp is microseconds.

k4a_mkv2image.exe
file.mkv
file
  |-rgb
  |   |-000000_00000000000.jpg
  |   |-000001_00000000033.jpg
  |   |-000002_00000000066.jpg
  |
  |-depth
  |   |-000000_00000000000.png
  |   |-000001_00000000033.png
  |   |-000002_00000000066.png
  |
  |-infrared
  |   |-000000_00000000000.jpg
  |   |-000001_00000000033.jpg
  |   |-000002_00000000066.jpg
  |-rgb.txt
  |-depth.txt
  |-calibration.txt

Option

option	description
-i	input mkv file path. (requered)
-s	enable depth scaling. false is raw 16bit image. (bool)
-t	enable transform depth to color camera. false is not transform. (bool)
-q	jpeg encoding quality for infrared. [0-100]
-d	display each images on window. false is not display. (bool)

Environment

• Visual Studio 2017/2019 / GCC 7.4 / Clang 6.0 (require <filesystem> supported)
• Azure Kinect Sensor SDK v1.3.0 (or later)
• OpenCV 4.1.2 (or later)
• CMake 3.15.4 (latest release is preferred)
• Intel Thread Building Blocks (latest release is preferred)

^{* This tool requires Intel TBB only on Linux.}

License

Copyright © 2019 Tsukasa SUGIURA
Distributed under the MIT License.

Contact

• Tsukasa Sugiura
  • [email protected]
  • http://unanancyowen.com