Compensating for angular image references
Elliot Geno Oct 13, 2017
From what I can tell, the original image reference needs to be fairly square to the camera for it to start the AR experience. I tried uploading various angles of a surface to enhance the recognition, but from what I can tell, there seems to be no way to mask out areas that should not be included as feature points. This leads to some issues concerning what background the marker plane is sitting on.
It would be great if we as developers could provide transparent PNGs that would ignore or mask the feature points of the image that are transparent. This could facilitate better recognition at odd angles. If this were possible, we could more easily detect a planar marker at extreme angles on various backgrounds. Perhaps this could even be an optional automated task in the CMS?
Another way to do this is to allow users to upload various angular shots and manually outline the plane to specify 'good' feature point areas. Then the software would just need to ignore everything outside of it.
Additionally the API could try to compensate the source image based on the accelerometer and/or identified in-scene planar lines? These could be used as hints to find markers at weird angles.