masterThesis
Extraindo dados de tráfego a partir de vídeos em tempo real
Fecha
2017-07-31Registro en:
SILVA, Luiz Fernando Virginio da. Extraindo dados de tráfego a partir de vídeos em tempo real. 2017. 79f. Dissertação (Mestrado em Sistemas e Computação) - Centro de Ciências Exatas e da Terra, Universidade Federal do Rio Grande do Norte, Natal, 2017.
Autor
Silva, Luiz Fernando Virginio da
Resumen
Some of the major problems in large cities are related to urban mobility. Problems such
as traffic jams and vehicle accidents directly impact society in a negative way, and are
usually attributed to lack of urban planning from governments, the lack of public policies
or research projects aimed at solving this problems, even if partially. These researche
projects depend on data that must be collected in loco on the main avenues and streets
of the city, that are now performed manually through the observation of images captured
by CCTV cameras (Closed Circuit TV), the main means of traffic surveillance in the
city. Thus, there is a need for a solution that is able to automaticaly collect these data
in order to reduce costs with personnel, optimize the work and also reduce errors that
arise from this operation. In this way, we propose a method capable of collecting this data
automatically, in real time, using these video images to support the researche projects
and explore possible actions in traffic management. Our method consists of a continuous
flow of activities that use the collected images. First, it uses motion segmentation to
detect moving objects. Then, we apply, in each segmented object, an adaptation of the
Viola-Jones method to refine the search in the detection of vehicles, classifying them. In
this step, we deal with occlusion situations, a common phenomenon of objects overlapping
that directly interfere on results. Finally, we apply the Senior method to track each vehicle
in order to obtain relevant traffic data, initially direction, speed and intensity of flow. We
submit some videos collected on a large avenue to test our method. As a result, we
construct an efficient model with low computational cost capable of handling situations
of occlusion in distincts lighting conditions, which is the main contribution of this work.