Python package for distinguishing between the four colors of carpet in my office, from given images from a floor facing camera.
This is intended for use as a component in a carpet-based robot localisation system.
For an overview of the broader project, see the wiki.
Figure: Robot in target localisation environment, showing the four carpet colors to be classified.
Classification is performed by taking average HSV values for each input image, and using a gaussian mixture model (GMM) to distinguish the four color clusters in HSV space.
Figure: GMM clustering results distinguishing the four color clusters in HSV space.
For an overview on how the classifier is trained, see this notebook.
See test_carpet_color_classifier.py for an example demonstrating how to construct the classifier from a parameter file, and then use the classifier on given cv2 images.
This package includes a utility image_recorder for saving images from a webcam to disk (useful in creating training or testing datasets). Usage is as follows:
$ image_recorder -h
image_recorder.
Saves images from webcam to disk
Usage:
image_recorder <output_directory> [--device=<index>]
Options:
--device=<index> Index of video device for capture [default: 0]