Abstract
Object detection in a dynamic backgroundis a challenging task in many computer vision applica-tions. In some situations, the motion of objects can bepredicted thanks to its regularity (e.g. vehicle motion,pedestrian motion). In this article, we propose to modelsuch motion knowledge and to use it as additional infor-mation to help in foreground detection. The inclusionof object motion information provides a measure fordistinguishing moving objects from a background thathas similar sizes and brightness levels. This informationis obtained by applying statistical methods on data ob-tained during the training period.When available, priorknowledge can be incorporated into the foreground de-tection process to improve robustness and to decreasefalse detection. We apply this framework to moving ob-ject detection in rivers, one of the situations in whichclassic background subtraction algorithms fail. Our ex-periments show that the incorporation of prior motiondata into background subtraction improves object de-tection.