Image Processing Framework For Extracting High Resolution Vehicle Trajectories From UAV Videos

Recently,with the development of the capacity and processing power of the computers,the big data is increasingly being applied in the academic field.For Transportation Engineering,works based on vehicle trajectories are getting more and more popular due to the reliability of the obtained results.The present Master degree thesis was developed aiming at applying and improving an existing framework that extracts high resolution vehicle trajectory database from Unmanned Aerial Vehicle(UAV)videos.The complete extraction methodology is summarized as: vehicle detection through Canny edge detector method,vehicle tracking by Kernelized Correlation Filter(KCF)and data validation.Along this work we also introduced an image stabilization proposal for the vehicle detection process in which it was applied a methodology based on the Sum of Absolute Transformed Differences(SATD).The existing detection method was able to detect 368(59.3%)of correct detections and,after applying the proposed image stabilization method,the accuracy turned into 451(72.6%)correct detections.The tracking method obtained 368(69.6%)of the trajectories that were considered low quality.The bad detections and trajectories were manually obtained or corrected in order to complete the database and make the data validation.After the data validation,the final database contained 621 vehicles in total.