Bayesian object-level change detection in grayscale imagery

We present a change detection algorithm formulated in a Bayesian framework that uses the output of an object detector to reason about change at a higher level than comparing pixels. The object detector mitigates pixel-level noise, and presents objects to the change detection framework. This in turn ties the objects across images and determines change. The Bayesian framework allows us to easily add domain knowledge into the change detection process to improve detection. We show that our approach can successfully detect changes across grayscale images with significantly greater variance in imaging conditions (such as viewpoint, resolution, and illumination) than those handled by traditional methods.