Abstract

Vision based intelligent transport system applications are extensively utilized and researched in recent years. Several applications with tracking, classification and counting functionalities are used for automatization of traffic management. Work in this thesis aims to provide an accurate vehicle detection method for improving performance of these tasks. Vehicle detection starts with detection of moving objects, using a background subtraction algorithm. Then, accuracy of the foreground mask is improvedusingashadowdetectionalgorithm. Occlusionsaredetectedfrombothgeometrical properties of blobs in binary mask, and associations between objects from consecutive observations. A segmentation method based on the assumptions on objectgeometryunderocclusionisproposedandimplemented,todetectvehiclesunder occlusion correctly. Proposed solution is tested on several videos collected from different intersections. Results indicate a significant improvement in performance compared to the existing methods in literature. M.S. - Master of Science