On the Use of Bayesian Networks for Real-Time Urban Traffic Measurements: a Case Study with Low-Cost Devices

  • PDF / 1,415,583 Bytes
  • 12 Pages / 595.276 x 790.866 pts Page_size
  • 21 Downloads / 162 Views

DOWNLOAD

REPORT


On the Use of Bayesian Networks for Real-Time Urban Traffic Measurements: a Case Study with Low-Cost Devices Ginés Doménech-Asensi 1

&

María-Dolores Cano 2 & Víctor Morales-Esteras 2

Received: 1 October 2019 / Revised: 29 August 2020 / Accepted: 25 September 2020 # Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract This paper describes a low cost computer vision system able to obtain traffic metrics at urban intersections. The proposed system is based on a Bayesian network based reasoning model. It employs the data extracted from background subtraction and contrast analysis techniques applied to predefined regions of interest of the video sequences, to evaluate different traffic metrics. The system has been designed to be able to work with already installed urban cameras, in order to reduce installation costs. So, it can be configured to work with different types of image sizes and video frame rates, as well as to process images taken from different distances and perspectives. The validity of the proposed system has been proved using a Raspberry Pi platform and tested using two real surveillance video cameras managed by the local authority of Cartagena (Spain) during different environmental light conditions. Using this hardware the system is able to process VGA grayscale images at a rate of 8 frames per second. Keywords Traffic signaling . Intelligent traffic lights . Image processing . Intelligent transportation, Bayesian networks

1 Introduction Nowadays, different types of surveillance systems for transportation are being deployed in different environments: urban, inter-urban, or indoor. According to the extensive review presented in [1], hierarchical and networked vehicle surveillance in intelligent transportation systems (ITS) are constructed from four layers: image acquisition, extraction of attributes, behavior understanding and, finally, ITS services. The first layer is based on digital video cameras, to capture color or, more likely, grayscale video sequences, which can come in a variety of formats at different frame rates. The image processing itself begins in the second layer. This layer uses a wide variety of computer vision techniques to robustly detect a vehicle in an image and identify both dynamic attributes, like speed or direction, and static ones (e.g., type of vehicle and license plate number). Above this layer, the behavior understanding is evaluated in the third layer. Examples of such behavior are the location of the vehicle or changes in its speed. * Ginés Doménech-Asensi [email protected] 1

Dpto. de Electrónica, Tecnología de Computadoras y Proyectos, Universidad Politécnica de Cartagena, Cartagena, Spain

2

Dpto. de Tecnología de la Información y las Comunicaciones, Universidad Politécnica de Cartagena, Cartagena, Spain

Typical actions that can be recognized are: running, turning or stopping. This is a complex task that requires training and matching methods based on a wide variety of clustering and modelling techniques. Finally, the fourth layer provid