Highway variable toll collection is an important means of traffic governance in modern society. A reasonable variable tolling scheme can adjust the traffic flow through price, and further develop the bearing capacity of highway on the basis of existing infrastructure to improve its traffic efficiency and service level. In the past, the formulation of variable tolling schemes is usually based on various assumed models, which belongs to the knowledge-driven method. In this paper, a data-driven variable tolling scheme is proposed, which uses the traffic flow prediction model integrating attention mechanism, recurrent neural network and graph neural network to mine the historical traffic information data on the road. In this paper, the traffic flow information on the future road is forecasted, and the information is used to replace some assumptions of bottleneck theory which are not completely in line with the actual situation. Taking the forecast information of traffic flow as the new constraint condition of dynamic congestion pricing in bottleneck theory can help us develop a more scientific and reasonable variable tolling scheme.
With the development of intelligent transportation, traffic sign detection has become a very important task. Traffic sign detection requires the positioning and classification of traffic signs in the road environment. Due to the complexity and diversity of road environments, traffic signs can only take up a small proportion of videos or pictures, and common algorithms still show high false detection rate and high time and calculation overhead in practical application. So far, traffic sign detection is still a difficult task. At the same time, most traffic sign data sets used by algorithms at the present stage have unreasonable data structures or incomplete database types. Blind application of unstructured data sets easily lead to the difficulty of obtaining good results in model training. Therefore, this paper decided to carry out the traffic sign detection task based on the lightweight YOLOv5 (You Only Look Once) neural network model. In order to achieve excellent target detection effect, this paper adopts optimized Tsinghua-Tencent-100K (TT100K) data set to train the model. The experimental results show that the trained models mAP@0.5 and mAP@0.5:0.95 reach 59% and 40% respectively, which basically meet the requirements of application.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.