Would it be possible to implement a face detection algorithm as part of the motion detection?
Face detection algorithms are commodity these days and are very reliable. Most can recognise the same face.
Having Canary able to notify only when it detected a person (as an option) would be a great way of refining notifications and reducing false positives.
If it was deemed computationally expensive it could just be run as a second pass filter on detected motion video.
If you were feeling more advanced you could even enable the tagging of people to be able to not notify on recognised faces, e.g. the cleaner.