If there was some extensibility point in the motion detection pipeline, such as the ability to load an external library, which would allow us to plug in other motion detection algorithms, it would be fantastic.
Then folks like myself in the community could work on more advanced motion detection, such as facial recognition, or simply filtering our 'bugs' and 'rain', which have fairly consistent characteristics.
I don't know much about how the motion detection is implemented, but I can at least assume that a) it has some number of previous frames available, and b) it needs to be fast.
I think the solution that would get the most community support would by a way to implement python based plugins, which would give people the ability to execute c/c++ code as well as use something like OpenCV for advanced image processing easily. For example, you could expose a public method for a python script to register a callback, and the callback would get pointers to the video data on each frame.
An even easier solution at your end would be simply dynamic loading a shared library to handle the motion detection, then anyone could implement the symbol/contract and drop their binary in the right folder to use it. This is likely very little work and would be perfectly suitable.
Thanks for considering! If you add support I'll happily contribute time to implement some interesting motion detection algorithms.
Another vote here. This way the community could write software to interface with IP cameras for near-hardware level motion detection or using motion sensors like burglar alarms. This is a big issue for us right now. 5x Megapixel IP cameras and we can't run motion detection on any of them (due to another flaw in the BC software). Having an "external" trigger would be amazing.