Development of real-time vision based traffic flow information estimation systems for intersection and highways

In this study, a real-time traffic flow monitoring system that can be used in the planning of urban traffic and can generate statistical traffic flow data was developed using image processing methods. Intersection/highway traffic images were obtained with cameras. Normally, the highly congested places in cities are intersections, roads or highway areas where vehicles have to share. Hence, this study focused on intersections/highways. Online traffic images obtained separately from intersections and highways were processed with AI techniques, and instant information about the traffic of selected areas of the city was calculated. For example, the classification of vehicles passing through city roads/intersections (such as cars, trucks, buses) and number ​​(for example, frequency information) of each vehicle class, the direction and speed information of the vehicles were estimated. Deep/machine learning techniques were utilized for the online software system developed in the study. The image processing module; object detection, tracking and association (extraction of time-dependent vehicle trajectory) and traffic flow data such as speed, counts (directional and total vehicle counts), entry and exit points of vehicles to the specified area, and the time period between these points. YOLOv3 architecture based on convolutional neural networks (CNN) was used for the object detection task. By retraining the YOLOV3 algorithm, case-specific weight models were created. Based on the trained new weight models, vehicle detection (localization and classification) tasks were performed from the traffic images obtained from case study intersections and highways. Bounding box information, which is the output of the object detection phase, was used as the input of the object tracking and trajectory extraction modules. Traffic flow information was estimated from the vehicle trajectory data. City traffic can be analyzed from the real-time outputs of this developed software system or the obtained output data can be used as input for the traffic analysis simulation software tools (for example, PTV Vissim). The simulation tools can give accurate results only with the right statistical data, so this study is important for robust city traffic planning. The studies were carried out within the scope of the project and the software tool developed was presented to Sakarya Metropolitan Municipality (SBB). SBB, which was selected among the cities to implement the smart city vision, interested in the study and decided to support it within the scope of traffic and signaling projects. The outputs of the project will directly contribute to the traffic planning of the city, and indirectly, it is anticipated that it will contribute to increasing the quality of city living space, reducing emissions and traffic noise. In this context, it is also aimed to increase the cooperation between Sakarya Metropolitan Municipality and Sakarya University by using up-to-date technologies.