Automatic vehicle detection and tracking in aerial surveillance using DBN and Graph cut model

The robust tracking system for aerial surveillance is one of the important challenge in computer vision due to its large uncertainty. We tackle this problem by providing an approach for automatic vehicle detection and tracking for aerial surveillance. The proposed system includes feature extraction, color transformation, Dynamic Bayesian Network (DBN) and post processing. Support Vector machine is constructed for the classification purpose. Finally the results of detection after post processing are provided as input to the tracking system. To effectively handle the, tracking process the Graph cuts and the 2D parametric motion models. From the results obtained the performance of vehicle tracking on the detected vehicles gets increased on the challenging dataset with images taken at different camera angles and different heights.