Road congestion levels currently seen in the main cities of the world, have generated problems associated with environmental air quality, road safety and excessive trip times. Intelligent Transport Systems (ITS) emerge as an important alternative to assist urban planning by monitoring the different urban road corridors. Traditionally, ITS use fixed camera systems which are sometimes insufficient to monitor and extend the traceability of the vehicles with extended coverage. Video capture mechanisms supported by Unmanned Aerial Vehicles (UAVs) e.g. drones, have greater continuous visibility of road segments, better visibility and allow more precise analysis on the density and behaviour of vehicle flows. This chapter presents detection, classification and tracking analysis for three categories of vehicles: Light (automobiles), Heavy (trucks and buses) and Motorcycles corresponding to 95% of urban traffic in emerging countries. For this purpose, 3 datasets with more than 2.5 million annotated objects are introduced using video sequences captured from UAVs from different views. An analysis of detection ability is carried out by considering traditional methods (prior to deep learning detectors) and using state-of-the-art detectors such as YOLO v3 and Faster R-CNN, obtaining results up to 98.5% AP (Average Precision) in all the analysed categories. Finally, multiple object tracking techniques such as Deep Sort are investigated, achieving results of up to 82% MOTP (Multiple Object Tracking Precision) in light vehicles.
Classification and Tracking of Vehicles Using Videos Captured by Unmanned Aerial Vehicles
Advances in Science, Technology & Innovation
Machine Learning Techniques for Smart City Applications: Trends and Solutions ; Kapitel : 6 ; 59-73
2022-09-20
15 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Urban vehicle detection , Urban vehicle tracking , Yolo , Faster R-CNN , Region-based detector , CNN , Deep learning , Multiple object tracking Computer Science , Machine Learning , Artificial Intelligence , Health Policy , Transportation Technology and Traffic Engineering , Logistics , Earth and Environmental Science
Tracking from Unmanned Aerial Vehicles
Trans Tech Publications | 2015
|Tracking from Unmanned Aerial Vehicles
British Library Conference Proceedings | 2015
|Tracking unmanned aerial vehicles using reflected light
Europäisches Patentamt | 2017
|Airspace activity tracking using unmanned aerial vehicles
Europäisches Patentamt | 2018
|