im not sure if you'd be able to set up a fast implementation, because if im not wrong you'd need to copy it to a canvas, copy the canvas image with the image manipulator, and then get pixel colour using the image manipulator, just to get one pixel colour. im pretty sure most AR software posterizes the feed to deal with a smaller colour space to run calculations on, and most of the work is done by the gpu, but with construct the loop would be really slow is what im guessing.