Skip to Main Content
Detecting moving objects from an image sequence is challenging, especially when the camera is moving and the background varies significantly in every frame. In addition, classifying moving objects using only their appearances creates ambiguities in complex scenes. In this sense a Markov random field (MRF) approach is proposed incorporating a stereo vision-based structure-from-motion scheme in order to robustly detect the moving objects from image sequences. In this MRF formulation, the new energy terms of a high-order likelihood and a temporal pairwise potential are added to improve the detection performance further. The performance of the proposed method is demonstrated from publicly available datasets.
Date of Publication: August 2 2012