Urban Traffic State Evaluation Methods Based On Automatic Number Plate Recognition Data

Modern cities continue to encounter serious traffic problems as urbanization expands,impeding city construction and economic development.In order to solve traffic congestion,traffic managers need a real-time,reliable,and scientific urban traffic state evaluation system.The most accessible traffic data for traffic managers is Automatic Number Plate Recognition(ANPR)data.It offers a high sample rate,a large coverage area,and a high level of reliability.We use machine learning approaches to build a traffic state evaluation system based on ANPR data and road network topology data from the perspectives of saturation,space and time equilibrium,and congestion degree.The system provides a panoramic traffic state monitoring.Free-flow travel time is the benchmark parameter in traffic state evaluation.Firstly,we propose an urban road free-flow travel time estimation method.The travel time distribution function is derived using the traffic wave model and the assumption of uniform arrival.To perform function fitting,we convert the interrupted traffic flow data to data that satisfy the uniform arrival assumption using the data sampling method.The free-flow travel time could be obtained from the fitted function.We transform the freeflow travel time estimation problem into a function fitting problem and verify the method’s practicality and accuracy using real data.The free-flow travel time is the basis of saturation degree calculation.The travel time reliability will greatly affect the experience of travelers,indicating the traffic state from the perspective of time equilibrium.To evaluate travel time reliability,we propose a Copula based travel time distribution model of urban road segments and paths considering the correlation of traffic movements.We further use the Dip test to quantify the travel time reliability.The necessity of considering interactions among movements and the superiority of the proposed model is demonstrated by analyzing the correlation between movements,testing and establishing best-fitting models and comparing with traditional models.We further evaluate the spatial equilibrium of traffic states.Road network traffic states are often heterogeneous in space due to unbalanced traffic demand and planning.We propose a graph theory based traffic cluster identification method to describe the spatial equilibrium.The traffic state index of each segment is calculated using ANPR data,road network topology information,and free-flow travel time.We modify the RatioCut algorithm and turn hyper-parameters automatically to identify traffic clusters in the road graph,which is based on graph theory.A large-scale urban network in Hangzhou,China,is used to test the proposed approach.The visualization and comparison results demonstrate that the proposed approach can reasonably classify connected segments with similar states into the same traffic cluster.The results are used to quantify the spatial equilibrium of traffic states.The congestion level,in addition to saturation and equilibrium,is an important aspect of the traffic state evaluation.We classify segments into four congestion levels based on multi-dimensional traffic state indexes,ranging from clear to extremely congested.We augment multi-dimensional traffic state indexes through feature learning from the ANPR data and map them to the 2 dimensions hidden layer using a multi-layer autoencoder.The congestion level is obtained using the Gaussian mixture clustering method in the hidden layer.The results can be introduced to congestion level indicators.We build an urban traffic state evaluation system based on the aforementioned study that evaluates and estimates traffic states of road segments,clusters,and networks from different viewpoints,including saturation,space and time equilibrium,and congestion level.The system is tested in Hangzhou,where real-world traffic data is used to validate its applicability and accuracy.The research adds to the knowledge of urban traffic flow,provides theoretical support for traffic managers,and provides references for traffic participants.