poorly defined: how to accurately track if an image is color or not? This is causing a bug in the @with_frame decorator in GameModel.
There's also an antipattern of "cvimage.image = cvimage.mod_function()", to perform in-place modifications to images. This is the place for an ImagePipeline object.
TODO: standardize the chainable outputs from CVImage. Will we return a new CVImage or a numpy array? I'm leaning towards the second one, so CVImage's return values can be used as numpy/cv2 inputs, and the ImagePipeline class will handle multiple modification calls.
TODO: remove all image manipulation code from GameModel class. Moving all the OpenCV image manipulation actions to the CVImage class would make the GameModel procedures more legible.
TODO: abstract multi-step processes in GameModel class as an ImagePipeline. The purpose is to improve testability by making each manipulation action in CVImage to result in a series of images and data structures.
Rocks, lives, and missiles are identified by their rectangles from cv2.matchTemplate.
Clusters from SIFT are checked against these detected rectangles.
The remaining objects (like the rotated ship) are classified as "mysteries". These will be target for further analysis (like ship angle determination), or we can just shoot at them :)
Set up a SIFT method and massively upgraded the debug view to look at the image processing outputs in color.
TODO: the template matching used for unrotated ship icons and asteroids does not work on missiles.
The data returned by the gamemodel function is overly verbose and not interpreted well yet. I'd like to return a (pt, wh, label) tuple like the find_asteroids function does. This would make it compatible with the existing display_results function.
Changed os.name to platform.system() because platform.system() is much more reliable, and is easy to check for different system.
Also added a check for linux.
John McCardle
Martysh12
Error_404