Vehicle tracking using the k-shortest paths algorithm and dual graphs

• Autor(a) principal: Lima Azevedo, C.
• Data de Publicação: 2014
• Outros Autores: Cardoso, J. L., Ben-Akiva, M.
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://repositorio.lnec.pt:8080/jspui/handle/123456789/1006425
• topic: Vehicle trajectories; Image processing; Driver behaviour; Remote sensing

Vehicle trajectory descriptions are required for the development of driving behaviour models and in the calibration of several traffic simulation applications. In recent years, the progress in aerial sensing technologies and image processing algorithms allowed for easier collection of such detailed traffic datasets and multiple-object tracking based on constrained flow optimization has been shown to produce very satisfactory results, even in high density traffic situations. This method uses individual image features collected for each candidate vehicle as criteria in the optimization process. When dealing with poor image quality or low ground sampling distances, feature-based optimization may produce unreal trajectories. In this paper we extend the application of the k-shortest paths algorithm for multiple-object tracking to the motion-based optimization. A graph of possible connections between successive candidate positions was built using a first level criteria based on speeds. Dual graphs were built to account for acceleration-based and acceleration variation-based criteria. With this framework both longitudinal and lateral motion-based criteria are contemplated in the optimization process. The k-shortest disjoints paths algorithm was then used to determine the optimal set of trajectories (paths) on the constructed graph. The proposed algorithm was successfully applied to a vehicle positions dataset, collected through aerial remote sensing on a Portuguese suburban motorway. Besides the importance of a new trajectory dataset that will allow for the estimation of new behavioural models and the validation of existing ones, the motion-based multiple-vehicle tracking algorithm allowed for a fast and effective processing using a simple optimization formulation.