Vehicle tracking using a human-vision-based model of visual similarity

In this paper we propose an automatic vehicle tracking method for monitoring traffic intersections. The method uses a weighted combination of low-level features and low-level human-visual-system (HVS) modeling. Given an input video, moving vehicles are first detected from the scene and low-level features are extracted from the detected vehicles. Next, each detected region in the current video frame is compared with each detected region in the next frame by using an HVS-based similarity model. Finally, tracking is performed by locating the vehicle with the closest matching low-level features and greatest visual similarity. We demonstrate that combining low-level features with an HVS-based model can be an effective strategy for vehicle tracking.