Overview

Compared to traditional traffic monitoring, UAVs (also known as drones) are advantageous in many ways. UAVs can cover a continuous stretch of roadway or even an entire traffic network that would normally necessitate a great many surveillance cameras. Camera calibration and estimation of real-world distances from an UAV video are easier from its viewing angle. As a flexible platform with high mobility, UAV can also provide emergency response and assessment. STAR Lab researchers developed frameworks to address issues surrounding the use of UAVs in traffic data collection applications such as irregular ego-motion, low estimation accuracy of flow parameters in dense traffic situations, and high computational complexity. We have also developed benchmarking datasets for various computer vision applications.

▪ Ke et al., Real-Time Traffic Flow Parameter Estimation From UAV Video Based on Ensemble Classifier and Optical Flow, 2019.

▪ Ke et al., An Advanced Framework for Microscopic and Lane-level Macroscopic Traffic Parameters Estimation from UAV Video, 2020.

The Digital Roadway Interactive Visualization and Evaluation Network (DRIVE Net) is a representative engineering project showcasing the STAR Lab’s work with state and city transportation agencies. The system provides users with the capability to store, access, and manipulate transportation data via a platform that can be run from anywhere on a web browser. The goal of this platform is to remove existing barriers in the current storage and processing of datasets by agencies, and to achieve the integration and visualization of information needed for decision support.

DRIVE Net archives Washington State Department of Transportation (WSDOT) loop detector data, weather data, incident data, and probe vehicle data, among many other data sets. It also provides functions to calculate level of service (LOS), estimate traffic-related emissions, predict crashes, and analyze/visualize many other types of transportation data.

With an enthusiasm for studying artificial intelligence, machine learning, and their applications in transportation data science, the STAR Lab has gained abundant experience with the research and application of traffic modeling and prediction in a deep learning context. We have been a big data analytics pioneer in the transportation community and consider sharing our experience helpful to our colleagues.

AI Net is a new open-source platform for precisely the purpose of knowledge sharing on AI applications in transportation. Our researchers have collected representative traffic network datasets, designed interactive coding environments, and set up scripts implementing various AI methodologies for transportation analysis tasks. On this platform, we have enabled any visitor to test established or uploaded algorithms and to study maintained or self-uploaded datasets. Transportation researchers interested in deep learning and traffic network data can use our platform as a testbed and we encourage those unfamiliar with the area to come and learn more via our tutorials!

▪ Ma et al., DRIVE Net: E-science transportation platform for data sharing, visualization, modeling, and analysis, 2011.

▪ Xiao et al., Data-driven Geospatial-Enabled Transportation Platform for Freeway Performance Analysis, 2015.

▪ Cui et al., An Artificial Intelligence Platform for Network-Wide Congestion Detection and Prediction using Multi-Source Data, 2019.

Traffic state estimation and short-term prediction is vital in all transportation services. Bus arrival time prediction, Uber travel time estimation, and Google Map routing are good examples of these tasks. Conventionally, in our field, methods have focused on the application of point-based data (e.g., loop detectors) analyzing time series collected at different locations. However, these methods do not consider the topology of neighboring transportation networks and therefore are insufficient to capture the complex network-wide influencing factors, some of which are latent.

The STAR Lab researchers understand the importance of network analysis and have thus applied machine learning methods to traffic networks as graphs. Our efforts culminated in the building of graph-based deep learning methods on the basis of Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs). These methods are built with transportation expertise and are designed to make short-term predictions even in the context of missing data.

▪ Cui et al., Deep bidirectional and unidirectional LSTM recurrent neural network for network-wide traffic speed prediction, 2018.

▪ Cui et al., Traffic graph convolutional recurrent neural network: A deep learning framework for network-scale traffic learning and forecasting, 2019.

▪ Cui et al., Learning Traffic as a Graph: A Gated Graph Wavelet Recurrent Neural Network for Network-scale Traffic Prediction, 2020.

▪ Ke et al., Two-Stream Multi-Channel Convolutional Neural Network for Multi-Lane Traffic Speed Prediction Considering Traffic Volume Impact, 2020.

To meet the aforementioned demand of traffic sensing, the supporting data infrastructure must continue to scale efficiently to ensure sufficient network coverage can be balanced with efficient computation. As such, the STAR Lab has devoted significant attention to the research of novel sensors to meet the objectives. Unlike most transportation engineering groups, we have students from a diverse background with degrees in areas beyond civil engineering, such as electrical engineering and computer science that allow us to develop hardware and software in-house. We introduce prototypes by the STAR Lab applied to parking, icy highways and road navigation.

Urban computing in transportation looks at mobility within city ranges. The acquisition, integration, and analysis of data from various sources provide us with different angles to understand who, how, when, why, and where people move. These sources of information are rich and essential for future smart city traffic operations.

By studying these large-scale multi-source data (e.g., smart card data, probe vehicle data, taxi fleet data, etc.), we, as transportation specialists and smart city dreamers, have worked to understand demand and capacity for urban transportation systems on various time scales (from hours to years). We then can make predictions and plan the available resources accordingly.

Smart card data embeds abundant information of origin-destination pairs, time of travel, fare payment, mode of travel, etc. Analysis of smart card datasets sheds light on people’s travel behavior, which serves as ground truth demand for regional transportation system planning and decision making. STAR Lab researchers analyzed datasets in China and discussed travel patterns and passenger origin distributions.

In response to the COVID-19 pandemic, the STAR Lab dedicated group work to quantifying the overall traffic condition in the Greater Seattle area. Based on WSDOT data collected from more than 44,800 inductive loop detectors deployed on freeways in the Seattle area, we calculated Traffic Performance Score (TPS) and Vehicle Miles of Travel (VMT). Furthermore, we built a open-source website (tps.uwstarlab.org) visualizing:

▪ Temporal dynamic of network-wide TPS of different types of lanes with various time resolutions, ranging from 5 minutes to one day;

▪ Varying Spatial distribution of segment-based TPS on interactive maps;

▪ Traffic changes in response to COVID-19 reflected by the TPS;

▪ Other traffic performance metrics;

Notably, VMT in the Seattle area dropped by 64% (the worst point) compared to Jan and Feb 2020 baselines, during Washington State's “Stay Home, Stay Healthy” order.

In this project, the STAR Lab will work closely with the City of Bellevue in managing the city’s curb spaces. Employing state-of-the-art traffic sensing techniques, the Lab will install MUST sensors at test fields and build a management system to achieve (1) multi-source traffic sensing, (2) edge computing and sensor-fusion, (3) on-street parking space monitoring, and (4) crowd density estimation. Chosen test fields include a transit station, a commercial parking area, and a parking space for mixed usage. (shuttle, app-based taxi, etc.)

▪ Ma et al., Transit smart card data mining for passenger origin information extraction, 2012.

▪ Ma et al., Mining smart card data for transit riders’ travel patterns, 2013.

▪ Ma et al., Development of a data-driven platform for transit performance measures using smart card and GPS data, 2014.

▪ Tang et al., Uncovering urban human mobility from large scale taxi GPS data, 2015.

▪ Pu et al., Evaluation of spatial heterogeneity in the sensitivity of on-street parking occupancy to price change, 2017.

▪ Chen et al., Examining regional mobility patterns of public transit and automobile users based on the smart card and mobile Internet data: a case study of Chengdu, China, 2019.

Moving traffic safely and efficiently are the primary goal of transportation systems. It is always a challenge with the complex inputs: wide range of users (cars, buses, pedestrians & bikers), time-varying traffic demand, random incidents, occasional adverse weather conditions and different-by-point roadway geometries. To understand the dynamics of transportation systems, STAR Lab researchers build models at all scales - micro, meso and macro. To improve system performances, we then optimize targeting specific metrics, e.g., traffic delays and queue lengths. At STAR Lab, our expertise of traffic operations research lies in the following areas.

Intelligent Transportation System (ITS) is an application that manages multiple modes of transportation via innovative services and facilitates safe, coordinated and “smart” use of the network. To improve safety, mobility and environment, innovations bond tightly with advances in algorithms, sensing and smart infrastructure. Typical applications vary from commonly seen loop detectors and red light cameras, to more recent floating car data, V2X communications and traffic signal coordiantion.

Given the benefits of ITS (ITS JPO, USDOT), STAR Lab researchers have devoted a significant amount of efforts to ITS research, including vehicle re-identification through multiple traffic surveillance cameras, optimizing ramp metering rates to improve freeway traffic operations, and vehicle automation technologies.

Traffic signal control aims to optimize the operations of traffic intersections by determining which traffic movements (e.g., left turn at south bound approach) should have the right of way at a given certain time. Here at STAR Lab, we use traffic operations data to model how signal timing will affect intersection delays. With one or multiple intersections modeled and simulated, we can apply more sophisticated knowledge, e.g. in control theory and reinforcement learning, to optimize signal timing plans for independent intersections or a coordinated system.

▪ Wang et al., Criteria for the Selection and Application of Advanced Traffic Signal Control Systems, 2012.

▪ Zhang et al., Optimizing Coordinated Ramp Metering: A Preemptive Hierarchical Control Approach, 2013.

▪ Zou et al., Application of finite mixture models for analysing freeway incident clearance time, 2016.

▪ Zhu et al., Capacity Modeling and Control Optimization for a Two-Lane Highway Lane-Closure Work Zone, 2017.

▪ Zou et al., Developing a Clustering-Based Empirical Bayes Analysis Method for Hotspot Identification, 2017.

▪ Zou et al., Quantile analysis of factors influencing the time taken to clear road traffic incidents, 2017.

▪ Zou et al., Jointly analyzing freeway traffic incident clearance and response time using a copula-based approach, 2018.

▪ Ash et al., Comparison of confidence and prediction intervals for different mixed-Poisson regression models, 2019.

▪ Huang et al., Multi-view vehicle re-identification using temporal attention model and metadata re-ranking, 2019.

▪ Pu et al., Full Bayesian Before-After Analysis of Safety Effects of Variable Speed Limit System, 2020.

▪ Wang et al., Optimizing Signal Timing Control for Large Urban Traffic Networks Using an Adaptive Linear Quadratic Regulator Control Strategy, 2020.

America lost 620k citizens total as a result of all wars since 1775. That said, more than 3 million were lost on the nation’s highways during the last century alone as a result of traffic crashes. Thus, traffic safety has consistently been the primary research priority in transportation engineering. The main goal of planners and engineers is to provide a safe travel experience for all road users. Researchers seek to deliver on this goal by studying how to better understand, detect, and prevent traffic crashes.

Highway safety research has consistently been a priority in the STAR Lab. Cooperating with government agencies at various jurisdictions, as well as private companies, we have amassed large sets of real-world data (e.g., crash data, traffic flow data, weather data, etc.) for use in data-driven safety research projects. These projects range from studying how to improve crash frequency and severity modeling paradigms, to novel applications such as real-time crash prediction modeling among many others.

Everyone is a pedestrian sometime in a day. Unfortunately, US pedestrians fatalities remain high: 6,283 killed in traffic crashes 2018, a 3% increase from previous year (NHTSA). People often ask how walkable a community is and try to avoid dangerous neighborhoods. But walkability scores on websites do not tell us how and why the accidents took place, which is, in fact, fundamental to mitigate such crashes. The STAR Lab therefore dedicates to investigating pedestrian-involved crashes, comparing rural vs urban walker safeties, understanding the triggers and finally raising the awareness of all parties, e.g. pedestrians, drivers and agencies.

Traffic crashes often result in significant injury as well as property damages; how to mitigate these impacts is a top focus for transportation agencies. Understanding the factors that contribute to crashes is therefore a prerequisite for numerous planning and engineering decisions (i.e., in design and construction).

Researchers from the STAR Lab specifically developed nonlinear models in predicting traffic crash rates on Washington State highways.

▪ Zeng et al., A generalized nonlinear model-based mixed multinomial logit approach for crash data analysis, 2017.

▪ Pu et al., Evaluating the Non-Linear Correlation between Vertical Curve Features and Crash Frequency on Highways using Random Forests, 2020.