depthmap-toolkit 0.1.0

Code for loading and processing depthmaps collected by CGM app

Depthmap toolkit

Depthmap toolkit is an utility to convert and visualise the data captured by cgm-scanner.

Overview

CGM-Scanner currently captures the data as depthmaps. Depthmap is our own format developed for high compressed data.

Tools

Visualisation of depthmaps

• The tool accepts only the data captured by cgm-scanner. The data could be captured by any ARCore/AREngine device supporting ToF sensor. Tool could be opened by following command:

python toolkit.py depthmap_dir calibration_file

• The depthmap_dir folder has to contain subfolder depth containing one or more depthmap files.
• By arrows "<<" and ">>" you can switch to next or previous depthmap in the folder.
• Export pointcloud will export the data into PLY file in export folder.
• Export textured mesh will triangulate and texturize the data and export in into OBJ file in export folder.
• Export poisson mesh will triangulate and extrapolate the data and export in into OBJ file in export folder.
• All exported data data will be reoriented using depthmap pose (if available)
• calibration_file is the txt file with calibration for the device.

Visualisation types

The tool generates 5 different visualisations (from left-to-right order):

• Depth image - this is a most raw visualisation of depthmap
• World-oriented normals - green value indicates a horizontal surface, blue and red a vertical surface (supported from scan type v1.0)
• Metrical segmentation visualisation - repeating pattern mapped on surfaces based on world-oriented normals, the pattern repeats every 10 centimeters (this might help ML models to calibrate measures captured by different devices, supported from scan type v1.0), blue is the detected floor, yellow is the detected child/object
• Confidence map - amount of IR light reflected into ToF receiver, this information might vary a lot (every sensor uses a different amount of IR light) and it is not recommended to use it for ML training
• RGB photo - captured photo (if available)