Customização e aplicação de ferramenta para coleta automatizada de dados de travessia de pedestres em interseções semaforizadas

• Ano de defesa: 2023
• Autor(a) principal: Trez, Juliana de Abreu e
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Circulação de pedestres; Headways; Visão por computador; Pedestrian circulation; Computer vision; CNPQ::ENGENHARIAS::ENGENHARIA DE TRANSPORTES
• Link de acesso: http://repositorio.ufc.br/handle/riufc/80517

Pedestrian crossings on red are an issue that still have gaps to investigate, due to the complexity of the variables involved and their relations, mostly in big urbam centers in Brazil. Automated collection tools can be an important aid on modelling these studies. This M.S. thesis has as main objective to customize and apply an automated collection tool to gather important variables for pedestrian crossings’ studies on signalized intersections. To reach this objective, three specific goals were established: 1) definition of the variables for the comprehension of the headways’ acceptance phenomenon; 2) calibration and validation of an automated collection tool of variables related to the headway acceptance process; and 3) collection and exploration of results related to variables that are part of the headway acceptance process on signalized intersections. A video of the intersection between Antonio Sales avenue and Desembargador Moreira avenue, in Fortaleza, was chosen. The tools used to collect the variables defined (headway, pedestrian delay, vehicular speed, vehicle type and crossing instant by lane) were YOLOv7 and StrongSORT. A training was conducted, resulting in a validation mAP of 94% and a test map of 90%. The class ‘Truck’ showed false positives, then a new trainning is recommended to improve the precision of this class. In total, 9427 vehicles and 723 pedestrians were tracked; the headways presented large range, the average speed of the vehicles was 28 km/h and the average pedestrian delay was 18 s. A validation step was made with a manual collection tool (RUBA), and results showed no significative difference between the collection with both methods of vehicle passage times and headways; the mean of differences for the vehicular speeds was -0,9 s and the mean of differences for pedestrian variables was below 0,2 s. Finally, exploratory analyses of four types of headways : accepted, minimum accepted, accepted on red and minimum accepted on red by the pedestrian of both origin zones (A and B), by vehicle type and vehicular lane. The point was made that, for all the four analyses, the mean accepted headways and the mean vehicular speeds were similar for the pedestrians of both origins, most headways from bicycles were accepted, and the pedestrians of origin B had the shortest delay.