Finally, different methods of estimating traffic volumes from the counts were implemented and assessed. The processing system pulls static images at high-frequency intervals from the traffic camera APIs and generates real-time counts of the detected vehicles (car, bus, motorcycle, etc.). Object detection was implemented using the open source You Only Look Once, Version 3 object-detection model that was trained on the Common Objects in Context dataset. Different classes of vehicles were detected from these images. Then, a prototype system was developed to collect imagery from three traffic camera jurisdictions through use of their Application Programming Interfaces (APIs). The process started with research on available traffic camera programs in Canada. This paper presents a computer vision-based system to periodically extract vehicle counts from Canadian traffic camera imagery. Traffic monitoring in large urban areas remains a challenge for both practical and technical reasons. The following symbols are used in this paper: N d,c Daily vehicle counts for road segment N t The number of vehicles in the image at different instances N c Average daily vehicle count N d,est Estimated traffic volume N AADT Historical annual average daily traffic (AADT) T pass,i The time it takes different vehicles to pass across the camera’s range of view V P Traffic volume over a period P N P,avg Average number of vehicles per image V Ti Ground-truth traffic volume for a sample period T i V T Estimated traffic over a time duration T Summary The authors would also like to thank Statistics Canada’s Research and Development Board for funding this project. Alessandro Alasia for his valuable input and feedback. For intersection control type selection, the gray areas were identified and visualized.The authors of this paper would like to thank Rastin Rassoli for his assistance with parts of this work, and Dr. Lastly, gray areas in a decision-making process based on quantifiable measures were addressed to cope with uncertainties in modeling outputs. The algorithm was evaluated by using various real congested traffic flow data. This algorithm was designed to detect dynamic queueing conditions in a spatio-temporal domain rather than detect a queue and congestion directly from traffic flow variables. Third, a real-time traffic queue detection algorithm was developed based on traffic fundamentals combined with a statistical pattern recognition procedure. The proposed algorithm consists of a data dimension reduction module and a nonparametric multivariate time-series analysis module. Second, a short-term traffic speed prediction algorithm was proposed that provides accurate prediction results and is scalable for a large road network analysis in real time. Different types of kriging methods were evaluated to utilize the spatio-temporal characteristic of traffic data with respect to two factors, including missing patterns and use of secondary data. #ONTIME TRAFFIC SERIES#To this end, this dissertation presents a series of studies on real-time traffic operation for highway facilities focusing on detection and prediction.First, a spatio-temporal traffic data imputation approach was studied to exploit multi-source data. There is still room to improve the traffic detection and prediction capabilities through studies on the emerging resources. During the last decade, great attention has been paid to the increases in the number of traffic data sources, the amount of data, and the data-driven analysis methods. This characteristic has been widely studied and various applications have been developed and enhanced. This spatio-temporal dependency of traffic flow should be considered and used to enhance the performance of real-time traffic detection and prediction capabilities.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |