The idea is that the software will constantly analyze the raw data, looking for regions that have approximately the color that I'm interested in. (Later I will also analyze the shape of the detected regions.)
If it finds such a region, the camera will take a foto and let me know that it detected something.
(Either bleep, flash or something better than that.) Problem is of course that many objects that I'm not interested in will have approximately the color that triggers the detection.
To get a bit of inspiration I looked at the code of edge_detection and histogram, because these also take the raw data as input. However i gues that I need better performance, because I want to analyze 10 or more frames per second. To get started, I rewrote parts of edge_detection. It does not longer use the code in bit_vector, but uses a new overlay 'class' now that can draw itself very fast in the frame buffer. If the is overlay is finished, I will start working on the detection itself.