Traffic flow estimation and vehicle‐type classification using vision‐based spatial–temporal profile analysis

This paper provides a review of a 3-year research program on the feasibility of using airborne LiDAR (Light Detection and Ranging) and imagery collected simultaneously over transportation corridors for estimation of traffic flow parameters such as: (1) vehicle counts, (2) vehicle classification, (3) velocity per vehicle category, and (4) intersection movement patterns. This work is conducted by The National Consortium for Remote Sensing in Transportation-Flows (NCRST-F), led by The Ohio State University, supported by the U.S. Department of Transportation and the National Aeronautics and Space Administration (NASA). The major focus is on improving the efficiency of transportation systems by integration of remotely sensed data with traditional ground data to monitor and manage traffic flows.

Real-time citywide reconstruction of traffic flow from moving cameras on lightweight edge devices

Nowadays, traffic flow estimation is the one of the most important topics in intelligent transportation systems (ITS). Accordingly, the authors of this article propose a traffic flow estimation method using time-series analysis and geometric correlation. Firstly, they define a 3D heat-map to present the traffic state and spatial and temporal adjacent traffic condition. Thereafter, they model the dependency heat-map using spatiotemporal Markov Random Field and estimate the probability using logistic regression. To evaluate the performance of the proposed method, it was tested using data collected from expressway traffic that were provided by the Korean Expressway Corporation, and its performance was compared with those of other existing approaches. The results showed that the proposed method has a superior accuracy to others method, which has the accuracy of 85%.

This paper proposes a estimation model based on three-layer stacking model to estimate traffic flow. Compared with estimation models of single machine learning and a two-layer stacking model, the three-layer stacking model can achieve higher estimation performance. The traffic flow data set we apply are accumulated from 2010 to 2018 in Wales, UK. The three-layer stacking model has a three-layer structure. The first layer is made up of two estimation models and trained by 5-fold cross-validation from the original traffic flow data set to generate a new data set. The second layer takes the new data set as the input data to train the new estimation model. In the third layer, the original traffic flow data set are trained as the input of a new estimation model. Then the weights of the two generalizers are measured according to the accurate estimation results. Finally, the weighted average is done to get the final estimation results. Experimental results show that in comparison to some single machine learning model and a model using only two-layer of stacking model, the three-layer stacking model can achieve higher traffic flow estimation accuracy and reduce the variance of the model.

<p>A comprehensive traffic monitoring system can assist authorities in identifying parts of a road transportation network that exhibit poor performance. In addition to monitoring, it is essential to develop a localized and efficient analytical transportation model that reflects various network scenarios and conditions. A comprehensive transportation model must consider various components such as vehicles and their different mechanical characteristics, human and their diverse behaviours, urban layouts and structures, and communication and transportation infrastructure and their limitations. Development of such a system requires a bringing together of ideas, tools, and techniques from multiple overlapping disciplines such as traffic and computer engineers, statistics, urban planning, and behavioural modelling. In addition to modelling of the urban traffic for a typical day, development of a large-scale emergency evacuation modelling is a critical task for an urban area as this assists traffic operation teams and local authorities to identify the limitations of traffic infrastructure during an evacuation process through examining various parameters such as evacuation time. In an evacuation, there may be severe and unpredictable damage to the infrastructure of a city such as the loss of power, telecommunications and transportation links. Traffic modelling of a large-scale evacuation is more challenging than modelling the traffic for a typical day as historical data is usually available for typical days, whereas each disaster and evacuation are typically one-off or rare events. Damage due to a disaster, combined with a sudden increase of demand due to the evacuation of people will likely result in increased pressure on the remaining, potentially fragmented, infrastructure. The lessons learnt from evacuation modelling can assist traffic operation teams and local authorities to provide safer and more efficient planning. The development of pervasive personal digital devices such as phones, watches, and headphones which can be interconnected with technologies such as Bluetooth, has led to a disruptive change in the ways in which local governments can monitor traffic flows within their cities. Moreover, modern vehicles and navigation systems can interconnect to the personal devices of drivers and passengers primarily via Bluetooth technologies. By continuously monitoring such devices when they are discoverable and in range, traffic patterns can be estimated based on, not only the volume of detection, but also other characteristics of the devices that can be used to give more refined estimates of the real underlying traffic flows. This thesis examines Bluetooth traffic data collected from Bluetooth Traffic Monitoring Systems (BTMS) for modelling and monitoring the urban traffic. BTMS can monitor and track individual detected vehicles through a city. Installation, processing, data transmission, and maintenance of BTMS are easier, quicker and cheaper than existing standard monitoring systems such as CCTV cameras and inductive loops. Inductive loops are typically point-wise traffic monitoring systems that are installed in the roads and can measure the traffic flow. However, the use of BTMS devices presents several challenges: not every vehicle has a detectable device, some have many, and there are devices carried by pedestrians and non-motor vehicles as well as stationary devices. This thesis enumerates and investigates these challenges through statistical modelling, various protocols for cleaning and data preparation, dynamic estimation of the detection rate, and simulation through the case study of the city of Wellington, New Zealand. The city of Wellington experienced damage from the 2016 Kaikoura earthquake (a magnitude 7.8 earthquake), which led to road closures and other infrastructure damage. As part of modelling, performance evaluation, and identifying impacted routes by the 2016 Kaikoura earthquake, this thesis analyses three weeks of BTMS data from the periods before and after the earthquake. Furthermore, this thesis proposes a multi-disciplinary dynamic traffic modelling (TFDA2M) framework and evaluates the performance of TFDA2M on various large-scale evacuation scenarios. These scenarios cover a wide range of real-world use cases which may occur during a disaster such as power failure, an abrupt increase in demand, and damage to the main transportation infrastructure. The findings of this thesis highlight an immediate need for preparations of a large-scale evacuation planning for Wellington to mitigate the consequences of a large-scale evacuation due to a future disaster. Moreover, TFDA2M can assist traffic operation managers and authorities in making smarter decisions (both quantitative and spatially) through the simulation process. Since TFDA2M has a flexible schema, it can be set to monitor, assess, and manage the traffic flow on a daily basis and disaster occasions.</p>

Large welfare costs related to deaths and injuries in road accidents have led, in the last few years, to a great number of studies and research projects focusing on highway safety. Road accidents have been identified as the greatest cause of death in Europe. The paper presents a simulation procedure for the assessment of impacts of traffic management measures on urban road safety. The procedure consists of two main modules: one, aiming at the estimation of traffic flow on links and at junctions, and the other estimating by means of performance functions of urban road safety, the accident frequency on each element of the network given the average daily traffic flows. Presented are the specification and calibration results of road safety functions based on data gathered in the urban area of Rome, Italy; also discussed are preliminary applications of the procedure to the case study.

Estimation of traffic flow with reasonable accuracy is essential for successful implementation of an Intelligent Transportation System (ITS). Crossroads are important part of urban traffic system, whose flow prediction on each direction is one of the most extraordinary key functions in the urban ITS. Some forecasting models have been developed, but these methods' precision usually can't meet with practical requirement. In this article, a neural network model is presented for foreccasting crossroads traffic flow using Backpropagation (BP) Neural Network. Through forecasting traffic flow at Hongqi crossroad in Ganzhou City, the result shows that this model has a considerable accuracy, which provides a new reliable and effective way of forecasting short term traffic flow of crossroads in urban road network.

With the advancement of modern UAV technology, UAVs have become integral to creating traffic management monitoring systems. Additionally, UAV-based traffic monitoring systems can predict traffic flow by integrating machine learning methods. Specifically, traffic flow data contains both spatial and temporal information, which can be effectively processed by graph neural networks (GNNs). However, GNNs often face the challenge of oversmoothing, which hinders their ability to capture complex structures in the data. The Spatio-Temporal Graph Ordinary Differential Equations (STGODE) model addresses this issue by introducing Neural Ordinary Differential Equations (NODEs) to construct deeper GNNs. Despite this, STGODE relies on initially predefined semantic neighborhood matrices, which do not adapt well to the dynamic nature of traffic information. To overcome this limitation, we propose an evolutionary graph neural network for traffic prediction, capable of continuously updating the semantic adjacency matrix throughout the training process. This dynamic evolution of the semantic adjacency matrix allows it to adapt to the features and semantic relations of the current data, enhancing its ability to capture the complexity and variability of traffic patterns. We validate our approach through experiments on several real-world datasets, demonstrating that our method outperforms state-of-the-art benchmarks.

An intermodal freight transport system for optimal supply chain logistics

Estimating traffic flow condition is a tough but beneficial task. In Intelligent Transportation System (ITS), many applications have been done to collect and analyze traffic data. However, the surveillance video data are still only used for engineer's manual check. To better utilize this data source, traffic flow estimation from surveillance camera should be explored. This study uses Faster Regional Convolutional Neural Network (Faster R-CNN) with ResNet 101 as the backbone to achieve multi-object detection. Then a tracking algorithm based on histogram comparison is applied to link objects across frames. Finally, this study uses warping method to convert vehicle speeds from the pixel domain to the real world. The results show that estimating vehicle speed at intersection is more challenging than in uninterrupted flow.

A traffic situation in studied in which some counters are placed in the streets to count some combination of traffic flows in various directions. Based on these observations, individual traffic flows are estimated using recent results on filtering with counting observation. Simulation studies are given to demonstrate the workability of the approximate estimation equations obtained from theoretical consideration.

Traffic flows are key parameters for design, channelization, and signal control of the intersection. Thus, it is important to study how to correct the output of the assignment model so it can be used for the intersection design. This paper uses the assigned and observed traffic flows of some intersections as the input/output factors to train the neural network to obtain an intersection traffic forecast model. In order to train the neural network, the number and location of the sampled intersections are determined, and traffic flows in the intersection are collected. Then, the surveyed and assigned data neural network is trained. Finally, the accuracy of the trained model is examined.

Improving traffic management through consideration of uncertainty and stochastics in traffic flow

In metropolitan areas, common traffic issues include traffic congestion, traffic accidents, air pollution, and energy consumption occur. To resolve this issues, Intelligent Transportation Systems (ITS) have been evolved by many researchers. One of the important sub-systems in the development of ITS is a Traffic Management System (TMS) which attempts to reduce a traffic congestion. In fact, TMS itself relies on the estimation of traffic flow, therefore providing such an accurate traffic flow prediction is needed. For this reason, we aim to provide an accurate traffic flow prediction to facilitate this system. In this works, a Supervised Deep Learning Based Traffic Flow Prediction (SDLTFP) was proposed which is a type of fully-connected deep neural network (FC-DNN). Timely prediction is also a major issue in guaranteeing reliable traffic flow prediction. However, training a deep network could be time-consuming, and overfitting is might be happening, especially when feeding small data into the deep architecture. The network is learned perfectly during the training, but in testing with the new data, it could fail to generalize the model. We adopt the Batch Normalization (BN) and Dropout techniques to help the network training. SGD and momentum are carried out to update the weight. We then take advantage of open data as historical traffic data which are then used to predict future traffic flow with the proposed method and model above. Experiments show that the Mean Absolute Percentage Error (MAPE) for our traffic flow prediction is within 5 % using sample data and between 15% to 20% using out of the sample data. Training a deep network faster with BN and Dropout reduces the overfitting.

Towards improving quality of video-based vehicle counting method for traffic flow estimation