Avaliação da qualidade da estimação de atributos de fluxo veicular a partir do processamento de imagens

• Ano de defesa: 2018
• Autor(a) principal: Lira, Johny Alves
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Transportes; Trajetórias - Veículos; Processamento de imagens; Quality of tracking; Vehicular trajectories; Images processing
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/36459

The increasing increase of the vehicular fleet and of the congestion of vehicles and traffic accidents generates need of better mechanisms of traffic management. The possibility of analyzing the trajectory of large-scale vehicles with greater autonomy is very useful for traffic management, and can subsidize decisions that involve the cost generated by the loss of time of users or money spent with accident victims. The method of extraction of vehicular trajectories may contain errors resulting from the adjustment of parameters of the extraction algorithm or the models used in it. The evaluation of its quality is important in that the traffic indicators are generated from the data exported by the algorithm. The estimation of the precision of the indicators calculated from the exported trajectories allows to inform the reliability that can be had in the estimates of the indicators. This work aims to evaluate the quality of vehicle trajectory extraction through image processing. For this, a vehicle tracking algorithm was consolidated and adjusted based on VISSIM microsimulator footage, in order to evaluate the quality of the tracing through error indicators exported by these. Twelve scenarios were generated for different camera heights, traffic operation regime and vehicular flow. Position and time information of each vehicle were obtained from the results of the algorithm and microsimulation in the VISSIM by which, visually, they presented a good overlap. The clustering model for feature based using a storage system proved to be efficient in dealing with interrupted flow regime. The results of this research presented a mean of the differences between -0.70 m and 0.67 m for the coordinate estimates, between -0.85 m and 0.11 m for the vehicular length, between -3.4% and 5.44% for the vehicle counts and below 0.95km / h and 1.10km / h for medium and instantaneous speeds, respectively. It was observed that the average velocities of the current estimated by the algorithm and by VISSIM were very close, that the standard of the average speed errors decreased with the increase of the height of the camera and that the use of VISSIM allowed a good evaluation of the quality of the tracking. Excluding external interferences, such as variation of luminosity, camera balance, obstacles and climate change, it was possible to evaluate traffic indicator errors alone, informing us of the origin and size of the errors.