2021

Ahmed Samiel Ahsan, University of Texas, Austin
Randy Machemehl, University of Texas, Austin

Street lane closures are necessary for transportation infrastructure rehabilitation and construction projects. Closures can be preplanned by allowing lanes to be closed when traffic demand is less than the remaining capacity of the streets. The objective of this study is to consider all available 24-hour traffic volume counts available for Austin streets, compute lane and street capacities, and determine the daily hours when traffic demand could be served with a lane closed. Moreover, with these potential closure hours as a basis, identify zones where specified streets will have similar lane closure times with as little variability as possible. Traffic Analysis Zones (TAZ’s) were chosen to depict the geographic areas and six overlays were constructed using travel direction and street type to form sub-classes to minimize lane closure time variability within each zone. The overlays were presented using ESRI ArcGIS. It was found that available traffic volume counts were sufficient to delineate zones with streets and street sections of similar lane closure times. However, many streets had no count data available, therefore, they were not attributed to any zone. Moreover, this methodology could be used to expand the overlays should traffic volume counts of more streets become available.

Yi Jing, Tongji University
Songhua Hu, University of Maryland
Hangfei Lin, Tongji University

A method combining sampling expansion with geographically weighted Poisson regression (GWPR) was proposed to estimate the road network traffic volume with limited observation values. Firstly, a sampling expansion method based on spatial similarity was introduced to correct the imbalanced sample. Then, the GWPR was employed to estimate the hourly traffic volume of the lane considering the influence of the geometric characteristics of the road and the built environment. Results show that compared with the traditional linear regression and the geographically weighted Gaussian regression, the proposed combination model demonstrates the best estimation performance. In addition, the local spatial heterogeneity of the relationship between independent variables and traffic volume is also well captured.

Huthaifa Ashqar (hiashqar@vt.edu), Booz Allen Hamilton, Inc.
Mohammed Elhenawy, Queensland University of Technology
Mahmoud Masoud, Queensland University of Technology
Mohammed Almannaa, King Saud University
Andry Rakotonirainy, Queensland University of Technology
Hesham Rakha, Virginia Polytechnic Institute and State University (Virginia Tech)

As the Autonomous Vehicle (AV) industry is rapidly advancing, classification of non-motorized (vulnerable) road users (VRUs) becomes essential to ensure their safety and to smooth operation of road applications. The typical practice of non-motorized road users’ classification usually takes numerous training time and ignores the temporal evolution and behavior of the signal. In this research effort, we attempt to detect VRUs with high accuracy be proposing a novel framework that includes using Deep Transfer Learning, which saves training time and cost, to classify images constructed from Recurrence Quantification Analysis (RQA) that reflect the temporal dynamics and behavior of the signal. Recurrence Plots (RPs) were constructed from low-power smartphone sensors without using GPS data. The resulted RPs were used as inputs for different pre-trained Convolutional Neural Network (CNN) classifiers including constructing 227×227 images to be used for AlexNet and SqueezeNet; and constructing 224×224 images to be used for VGG16 and VGG19. Results show that the classification accuracy of Convolutional Neural Network Transfer Learning (CNN-TL) reaches 98.70%, 98.62%, 98.71%, and 98.71% for AlexNet, SqueezeNet, VGG16, and VGG19, respectively. The results of the proposed framework outperform other results in the literature (to the best of our knowledge) and show that using CNN-TL is promising for VRUs classification. Because of its relative straightforwardness, ability to be generalized and transferred, and potential high accuracy, we anticipate that this framework might be able to solve various problems related to signal classification.

Keith Campbell, MITRE Corporation
Kathleen Berman, MITRE Corporation
Valerie Gawron, Zan Mitrev Clinic
Jeffrey Long, MITRE Corporation
Samir Yuha, Zan Mitrev Clinic

Logs of vehicle Controller Area Network (CAN) bus traffic supplemented by accelerometer and GPS data can provide valuable information about the use and operation of Advanced Driver Assistance Systems (ADAS) systems to the broader safety research community. Although CAN bus message codes are often manufacturer-specific, third-party libraries provide partial decoding of messages from many vehicle models, which can be augmented by reverse engineering additional signals.  This study explored the value of CAN bus, accelerometer and GPS data that were logged on a variety of light vehicle models with an emphasis on availability of Lane Keep Assist/Lane Departure Warning and Adaptive Cruise Control.  This study demonstrated that important in-vehicle ADAS variables such as system on/off status and whether the system is actively controlling the vehicle could be determined on a variety of vehicle types.  Associated control variables such as steering wheel angle, gas pedal state and brake pedal state could also be determined from most vehicles tested.  Taken together with roadway features identified via GPS location, CAN messages can provide a richer understanding in which situations ADAS best assists the driver.  Comparisons of message structure between models may also inform standardization efforts for Electronic Data Recorders and telematics.

Hessam Arefkhani (Arefkhani.hessam@gmail.com), Sharif University of Technology
Yousef Shafahi, Sharif University of Technology

In this study, the authors show that the optimal solution of the most recent Vehicle Identification Sensor (VIS) location model for full observability of routes flows may be non-unique. Afterwards, they propose a new model. The proposed model aims at selecting the most reliable solution from the set of possible solutions of the VIS location model considering sensors’ failure probability. The model is mathematically proven and three numerical examples consisting of a toy-network and two middle-scale networks (the Nguyen-Dupuis and the upper half of the Sioux-Falls networks) validate the model. Furthermore, a sensitivity analysis is performed on the values of the weighting coefficients of the reliability and cost terms of the objective function. The results are promising and confirming.

Yanqi Lian, Central South University
Jaeyoung Lee (jaeyoung@knights.ucf.edu), Central South University
Guoqing Zhang, Central South University
Helai Huang, Central South University

The era of Big Data has arrived. Recently, Big Data has been applied in various fields in transportation including traffic safety. In this study, we review the recent research studies that employed Big Data for analyzing traffic safety for intelligent transportation systems (ITS). The particular topics include crash detection or prediction, discovery of contributing factors to crashes, driving behavior analysis, and crash hotspot identification. From the reviewed studies, employing advanced analytics for Big Data have a great potential for understanding and enhancing traffic safety. Big Data application in traffic safety integrates and processes massive multi-source data, breaks through the limitations of the traditional data analytics, and discovers and solves the problems, which cannot be solved by the traditional safety analytics. Lastly, suggestions are provided for future Big Data safety analytics.

Yina Wu, University of Central Florida
Mohamed Abdel-Aty, University of Central Florida
Ou Zheng (ouzheng1993@GMAIL.COM), University of Central Florida
Jinghui Yuan, University of Central Florida

Near miss identification and analysis is prevalent in recent years. Video data that could provide high-resolution trajectories is one of the important data sources to identify near misses that happen on roadways. This study proposed a framework named “Near Miss Event Detection System (NMEDS)” for road safety diagnostics using video data collected from roadside video cameras. The proposed framework could obtain more accurate vehicles’ trajectory and better identify conflicts/near misses from the video images. Mask-RCNN and Channel and Spatial Reliability Tracking (CSRT) multi-object tracking are used to detect and track the vehicles in videos. Occlusion-Net is employed to detect and predict the key points of vehicles, including right-front headlight, left-front headlight, right-back taillight, and left-back taillight. The key points could provide more precise vehicle occupying locations compared with the bounding boxes generated based on Mask-RCNN. Moreover, the locations of the occluded key points are modified based on 2D bird views. Then, more precise post-encroachment time (PET) values could be calculated based on the modification. Near misses/conflicts could be identified by comparing the PET values and thresholds. A case study is presented at a typical intersection. The results indicate that the proposed framework has better performance for vehicle localization. Besides, more reasonable near misses could be obtained. It is expected that the proposed methods could help diagnose road safety problems using roadside video cameras.  Moreover, the proposed method could be incorporated with Connected Vehicle Systems and provide information to nearby drivers based on Infrastructure-to-Vehicle (I2V) technologies.

Nima Hoseinzadeh (nhoseinz@vols.utk.edu), University of Tennessee, Knoxville
Yangsong Gu, University of Tennessee, Knoxville
Hairuilong Zhang, University of Tennessee, Knoxville
Lee Han, University of Tennessee
Hyun Kim, University of Tennessee, Knoxville
Brad Freeze, Tennessee Department of Transportation

The year 2020 has marked the spread of a global pandemic, COVID-19, challenging many aspects of our daily lives. Different organizations have been involved in controlling this outbreak. The social distancing intervention is deemed to be the most effective policy in reducing face-to-face contacts and slowing down the rate of infections. The stay-at-home and shelter-in-place orders have been conducted in different states and cities, affecting daily traffic patterns. Social distancing interventions and fear of the disease result in a traffic decline in cities and counties. However, after stay-at-home orders end and some public places reopen, traffic gradually started reverting to pre-pandemic levels. It can be shown that counties have diverse patterns in the decline and recovery phases. This study analyzes county-level mobility change after the pandemic, explores the contributing factors, and identify possible spatial heterogeneity. To this end, 95 counties in Tennessee have been selected as the study area to perform geographically weighted regressions (GWR) models. The results show that density on non-freeway roads, median household income, percent of unemployment, population density, percent of people over age 65, percent of people under age 18, percent of work from home, and mean time to work are significantly correlated with vehicle mile traveled (VMT) change magnitude in both decline and recovery phases. Also, the GWR estimation captures the spatial heterogeneity and local variation in coefficients among counties. Finally, the results imply that recovery phase could be predicted depending on the identified spatial attributes.

Lingfeng Peng, Wuhan University
Nengchao Lyu (lnc@whut.edu.cn), Wuhan University
Chaozhong Wu, Wuhan University

An effective real-time crash prediction model (RTCPM) is crucial to active traffic safety management systems. RTCPMs use real-time traffic flow status to predict crashes in a critically brief time window. Traditional RTCPMs are usually based on actual crash records from specific locations and aggregated traffic data within a certain period. However, actual crash records are comparatively rare to normal traffic records, and aggregated traffic data does not reflect interaction between vehicles. This paper proposes a new feature parameter framework based on driving trajectories, surrogate safety measures (SSM), and machine learning to establish a real-time crash risk prediction model for use by traffic management to implement safety strategies. and then uses machine learning to establish a real-time crash risk prediction model for predicting conflicts. A microwave radar detector was used to obtain the real-time positions and speeds of vehicles in the detection area, and traffic trajectory data processing was used to extract the traffic flow parameters and several typical SSMs. Through the combination of time-to-collision (TTC)s value, rapid deceleration rates and manual observations, crash risk events were extracted. The feature parameters from 1-2 minutes before conflict events and normal traffic were used as the model input to construct a sample set. The random forest algorithm (RF) and synthetic minority oversampling technique (SMOTE) were used to process the sample set, filter the most important features and balance the number of positive and negative samples. Finally, a real-time crash risk prediction model based on the support vector machine (SVM) was established. The results show that the prediction model proposed herein detected conflict events with a 94.7% accuracy, 95.2% recall and 5.8% false-positive rate. As such, it can be reliably used for active traffic safety management.

Wei Wang, Chang'an University
Xinyao Tang, Chang'an University
Hua Cui, Chang'an University
Ying Li, Chang'an University

Accurate and rapid acquisition of vehicle spatial form is of great significance to the fields of intelligent transportation and autonomous driving. However, due to the limitations of projective geometry, it is difficult for monocular cameras to obtain the three-dimensional structure of the object. Therefore, the goal of this paper is to estimate the vehicle spatial form under monocular traffic cameras. To begin with, we establish a calibration model of the road scene, and jointly build the geometric constraint model of the vehicle spatial form by combining the vanishing point constraint. Furthermore, based on Mask-Rcnn, the contour and boundary constraints of the vehicle are obtained, and based on these constraints, an error constraint function to estimate the projection error of the vehicle spatial form is constructed. Finally, particle swarm algorithm is used to iteratively optimize the parameters in the constraint space to obtain accurate vehicle spatial form information. Experimental results show that the algorithm can effectively complete various types of vehicle spatial forms estimation with accuracy of 94% and real-time processing, which outperforms existing methods.

Biao YIN, Ecole des Ponts ParisTech
Fabien Leurent, Ecole des Ponts ParisTech

Understanding individual mobility patterns is significant for activity-based travel demand modeling. Data mining techniques can extract useful activity and travel information from large-scale data sources, such as mobile phone geolocation data.  This paper aims at exploring individual mobility patterns from the samples of mobile phone users using a two-week geolocation dataset in the Paris region, France. After filtering the dataset, we propose techniques to identify individual stays and activity places. The typical activity places, such as the primary anchor place and the secondary place are also detected. The daily activity-travel program (i.e. timeline) is reconstructed with the detected activity places and the trips in-between. Based on the user-day timelines, a three-stage clustering method is proposed. In the method framework, the activity types are firstly identified by clustering analysis. In the second stage, the daily mobility patterns are obtained after clustering the daily mobility features. The activity-travel topologies are investigated statistically, which help to interpret the daily mobility patterns. In the last stage, at individual level, we analyze statistically the individual mobility patterns for all simples over 14 days, measured by the mean number of days for all kinds of the daily mobility patterns. All individual samples are classified into several groups where people have the similar travel behaviors. A kmeans++ algorithm is applied to obtain the appropriate number of patterns in each stage. Finally, we interpret the individual mobility patterns with statistical descriptions and reveal the home-based differences in spatial distribution for the grouped individuals.

Behram Wali (bwali@vols.utk.edu), Urban Design 4 Health, Inc.
Sebastiano Milardo, Massachusetts Institute of Technology (MIT)
Umberto Fugiglando, Massachusetts Institute of Technology (MIT)
Paolo Santi, Massachusetts Institute of Technology (MIT)
Carlo Ratti, Massachusetts Institute of Technology (MIT)

Links between real-world driving stress and driving volatility are not well understood. The seemingly unstructured big data generated by computational and physical components of cyber-physical systems allows examination of the links between volatility in microscopic driving behaviors and real-world driving stress. A natural experiment is conducted to collect over 0.2 million geo-referenced temporal samples of real-world driving behavior and health biomarkers for 150 driving sessions/trips. The integrated kinematic and biometric data are linked with detailed roadway geometric data from OpenStreetMap. A rigorous data analytic methodology is proposed to measure driving volatility and stress (captured by health biomarkers) at the trip-level. Owing to the important methodological issues related to systematic and random heterogeneity, fixed parameter, random parameter, and random parameter models with extension to heterogeneity-in-means are estimated to examine the links between driving stress and driving volatility at the trip level. A sizable portion of drivers’ instantaneous heart rates does not fall in the normal range (60 to 100 heart beats per minute). Statistically significant and positive correlations are observed between driving volatility measures (based on speed, vehicular jerk) and driving stress. Motorways are associated with lower stress. Contrarily, larger standard deviation of number of lanes positively correlates with driving stress. Several of the heterogeneous associations captured by random parameters vary itself as a function of observed factors. The substantial heterogeneity in associations across the trips underscores the importance of accounting for observed and unobserved heterogeneity to avoid misleading inferences. Relevance of the results for proactive driver assistance systems is discussed.

Sepehr Ghader (sghader@umd.edu), University of Maryland, College Park
Jun Zhao
Minha Lee
Weiyi Zhou, University of Maryland, College Park
Guangchen Zhao, University of Maryland, College Park
Lei Zhang, University of Maryland, College Park

We utilized an integrated dataset, consisting of anonymized location data, COVID-19 case data, and population information, to study the impact of COVID-19 on human mobility. We revealed that statistics related to social distancing, namely trip rate, miles traveled per person, and percentage of population staying at home all showed a trend named “social distancing inertia”. As soon as COVID-19 cases were observed, the statistics started improving, regardless of government actions, suggesting that a portion of population who could and were willing to practice social distancing voluntarily did so. However, after about two weeks, the statistics stopped improving, despite the continuous rise in COVID-19 cases. The study suggests that there is a natural behavior inertia toward social distancing, putting a limit on the extent of improvement in the social-distancing-related statistics. The U.S. states showed a synchronized trend, regardless of the timeline of their statewide COVID-19 case spreads or government orders.

Maged Gouda (mgouda@ualberta.ca), University of Alberta
Amr Shalkamy, University of Alberta
Xinyi Li, Wuhan University
Karim El-Basyouny, University of Alberta

The placement of light poles within roadside clear zones might pose a safety concern as they are related to an increased risk of collision occurrence. Only a limited number of studies have explored the relationship between light poles and safety. This could possibly be attributed to the time-consuming and labor-intensive practices of collecting light pole assets data. With the widespread of light detection and ranging data, it provides a promising source for the automatic inventory of such data. There is limited research on the mapping of light poles using point cloud data on rural highways. This paper proposes an automated approach to light pole mapping. First, the scanning vehicle trajectory is extracted, smoothed, and then used to segment the point cloud data into smaller overlapped chunks of data. Several filters, based on Z-gradient values, density, and statistical outlier removal, are applied to extract pole-like objects from the data. The smaller data sections are reconnected back together, and a density-based clustering algorithm is used to group the remaining points into clusters. A geometric filter is finally applied to extract poles. The model is tested on 28 km of data on three rural highways in Alberta, Canada. The proposed algorithm is very accurate, with precision, recall, and F1 scores exceeding 94% for all test segments. Average scores were also high, exceeding 99% for all metrics.

Qian Gao, Tongji University
Chenglong Liu, Tongji University
Yishun Li, Tongji University
Yue Qiu, Tongji University
Yuchuan Du (wszwsz179@163.com), Tongji University
Han Zhang, Tongji University

Co-occurrence patterns among different subgrade distresses play a pivotal role in road maintenance, which draws much attention from road management departments. However, current work mainly focuses on the rapid detection and development evaluation of pavement distresses. Few shed light on the relationship among different subgrade distresses. In this paper, lots of field tests were conducted on 87 sections of the highways in China by the ground penetrating radar (GPR). Based on the processed GPR data, we applied the association rule mining algorithm (Apriori) to explore the co-occurrence pattern among thirteen types of subgrade distress. Results indicate that there is a significant correlation between light distresses. They have an average 53% probability of accompanying or inducing other distress, which are supposed to be prevented or maintained in time to prevent the road from further deterioration. Moreover, light distresses and severe distresses have a 36% probability of co-occurrence. Conversely, the relationship between severe distresses is proved to be weak. Compared with the external environment, the interaction between different distresses is also a considerable inducement for pavement performance deterioration. In conclusion, the paper proposes a new perspective focusing on co-occurrence patterns in subgrade distress analysis and provides insights into future maintenance scheduling.

Junaid Khan, Western Washington University
Kavyashree Bangalore, New York University
Abdullah Kurkcu, Ulteig
Kaan Ozbay, New York University

Trajectory data from Connected Vehicles (CVs) and other micro-mobility sources such as e-scooters, bikes and pedestrians is important for researchers, policy makers and other stakeholders to leverage the location, speed, heading along with other mobility data to improve safety and bolster technology development towards innovative location-based applications for citizens. Such raw data needs to be stored and accessed from a non-proprietary database while the obfuscation and encryption techniques on current cloud-based propriety solutions incurs data losses that are deemed inefficient for its accurate usage particularly in time-sensitive real-time operations. In this paper, we target the problem of scalably storing and retrieving potentially sensitive data generated by vehicles and propose \textit{TREAD}, a blockchain-based system comprising smart contracts to store this mobility data on a distributed ledger such that multiple peers can access and utilize it in different location-based applications while not revealing users' sensitive personal information. It is however challenging to scalably store large amount of constantly generated trajectories and to achieve scalability we leverage Interplanetary File System (IPFS), a scalable distributed P2P data storage system. To avoid users injecting malicious/fake trajectories to the ledger, we develop efficient consensus algorithms for the stakeholders to validate the storage and retrieval process in a distributed manner. We implemented TREAD on open-source Hyperledger Fabric blockchain platform using trajectory data generated for $700$ vehicles in a simulation environment well calibrated with vehicles trajectories from a real-world test-bed in New York City.  Results show that TREAD scalably stores trajectory data with lower delay and overhead.

Jiqian Dong, Purdue University
Sikai Chen (chen1670@purdue.edu), Purdue University
Paul (Young Joun) Ha, Purdue University
Yujie Li, Purdue University
Runjia Du, Purdue University
Aaron Steinfeld, Carnegie Mellon University
Samuel Labi, Purdue University

Sensing equipment that are on-board connected and autonomous vehicles (CAVs) have the capability to reasonably characterize the surrounding traffic environment. However, their performance is limited by their relatively short sensor range. There is opportunity to incorporate data obtained from a longer range through vehicle connectivity. A CAV can integrate short- and long-range information to comprehensively construct its surrounding environment, thereby allowing it to plan both short and long-term maneuvers. Such coalescence of local (short range) information and downstream (long range) information is critical for safe and effective driving decisions by the CAV. While literature is replete with CAV control approaches that use information sensed from the local traffic environment, the number of studies that fuse information from various temporal-spatial instances to facilitate CAV movements are limited. In this paper, we propose a Deep Reinforcement Learning (DRL) based approach that fuses information on the local and downstream environment obtained via sensing and connectivity, respectively, for maneuvering decision making in a specific context of CAV operations – lane changing. We adopt learning-based techniques to provide an integrated solution that incorporates the information fusion and movement-decision processor. We also determine the optimal connectivity range for each level of prevailing road conditions (traffic density). We anticipate that deployment of the proposed algorithm in a CAV will facilitate reliable proactive driving decisions and ultimately enhance the overall operational efficiency of CAVs in terms of safety and mobility.

Michalis Pouliasis, Aristotle University of Thessaloniki
Anastasia Nikolaidou, Aristotle University of Thessaloniki
Georgios Georgiadis, Aristotle University of Thessaloniki
Ioannis Politis, Aristotle University of Thessaloniki
Panagiotis Papaioannou, Aristotle University of Thessaloniki

The evolution of the World Wide Web, from a traditional passive platform where the information is just presented (Web 1.0), in today’s internet reality (Web 2.0), where the user interacts with the given information or produces new, led to the tremendous growth of the respective data. On the other hand, social networks pertain to be very famous, with millions of registered users, acting as unique data generation units. This paper examines knowledge discovery, through data mining processes, in social networks and more specifically knowledge derived – from the social media platform of Twitter. A total of 20 hashtags related to transport and mobility were examined as well as the hashtag related to the European Mobility Week. The collection, elaboration and analysis of data allows management and analysis of social networking data through graph theory (metric analysis of nodes) and the extraction of sentiment data (sentiment analysis). From the data analysis, useful conclusions can be drawn regarding the capacity of social networks as a source for the collection of important information that could be used at the level of operational planning by relevant policy makers at local, national and international level. Understanding users’ needs and perspectives can be used during transportation planning, management, and control, as well as for supporting the achievement of transport policy goals. Further investigation of data collected, either through geospatial analysis or through sentiment analysis could provide powerful tools for the exercise and evaluation of policies, by creating and enhancing the participatory process in citizens’ digital society.

Rezaur Rahman, University of Central Florida
Kazi Redwan Shabab, University of Central Florida
Mohamed H. Zaki, University of Central Florida
Samiul Hasan, University of Central Florida

The objective of this study is to evaluate the level of service of shared facilities through mining geotagged data from social media and analyzing the road-users' perception. In pursuit of this objective, we develop an algorithm adopting a text classification approach with contextual understanding to filter out relevant information related to user perception towards active mobility. Using a heuristic-based keyword matching approach produces about 75 % out-of-context tweets. Hence this approach is deemed as unsuitable for information extraction from Twitter. We implement six different text classification models and compare the performance of these models for tweet classification. We apply the model on real-world data to filter out relevant information and perform content analysis to check the distribution of keywords inside the filtered data. From the experiment, we find that the Term Frequency-Inverse Document Frequency (TF-IDF) vectorizer based logistic regression model (one of the six text classification models) performs best in classifying the tweets. To select the best model, we compared the performance of the models based on precision, recall, f1 score, and accuracy metrics. The findings from the analysis show that the proposed method can help produce more relevant information on walking and biking facilities as well as safety concerns.  By analyzing the sentiment of the filtered data, we can infer the existing condition of biking and walking facilities in the DC area. This method can be a critical part of the decision support system to understand the qualitative level of service of existing transportation facilities.

Zhiyong Cui, University of Washington
Meixin Zhu, University of Washington
Shuo Wang, University of Washington
Pengfei Wang, Southeast University
Yang Zhou, Tongji University
Qianxia Cao, Changsha University of Science and Technology
Cole Kopca, University of Washington
Yinhai Wang (yinhai@uw.edu), University of Washington

Measuring traffic performance is critical for public agencies who manage traffic and individuals who plan trips, especially when special events happen. The COVID-19 pandemic has significantly influenced almost every aspect of daily life, including urban traffic patterns. Thus, it is important to measure the impact of COVID-19 on transportation to further guide agencies and residents to properly respond to changes in traffic patterns. However, most existing traffic performance metrics incorporate only a single traffic parameter and measure only the performance of individual corridors. To overcome these challenges, in this study, a Traffic Performance Score (TPS) is proposed that incorporates multiple parameters for measuring network-wide traffic performance. An interactive web-based TPS platform that provides real-time and historical spatial-temporal traffic performance analysis is developed by the STAR Lab at the University of Washington. Based on data from this platform, this study analyzes the impact of COVID-19 on different road segments and the traffic network as a whole. Considering this pandemic has greatly reshaped social and economic operations, this study also evaluates how  COVID-19 is changing the urban mobility from both travel demand and driving behavior perspectives.

Qing Chang (qzc0010@tigermail.auburn.edu), Auburn University
Jiaxiang Ren, Auburn University
Huaguo Zhou, Auburn University
Yang Zhou, Auburn University
Yukun Song, Auburn University

Currently, transportation agencies have implemented different Wrong-Way Driving (WWD) detection systems based on loop detectors, radar detectors, or thermal cameras. Such systems are often deployed at fixed locations in urban areas or on toll roads. A majority of rural interchange terminals does not have real-time detection systems for  WWD incidents. Portable traffic cameras are used to temporarily monitor WWD activities at rural interchange terminals. However, it has always been a time-consuming task to manually review those videos to identify WWD incidents. The objective of this study is to develop an unsupervised trajectory-based method to automatically detect WWD incidents from regular traffic videos (not limited by mounting height and angle). The principle of the method includes three primary steps: vehicle recognition and trajectory generation, trajectory clustering, and outlier detection. This paper also developed a new sub-trajectory-based metric that makes the algorithm more adaptable regarding vehicle trajectory classification in different road scenarios. Finally, the algorithm has been tested through analyzing 357 hours of traffic videos from 14 partial cloverleaf interchange terminals in 7 states in the U.S. The results suggested that the method could identify all the WWD incidents with 80% accuracy rate. The method significantly reduced man-hours for reviewing the traffic videos. Additionally, The new method can also be applied in detecting and extracting other kinds of abnormal traffic activities, such as illegal U-turns.

Cheng Zhang (c.zhang@psu.edu), Pennsylvania State University, University Park
Shihui Shen, Pennsylvania State University, Altoona
Bin Shi, Tongji University
Hai Huang, Pennsylvania State University
Lijun SUN, Tongji University

Traffic information including vehicle load and speed are critical for pavement design, management, and health monitoring. Numerous in-pavement sensors such as Fiber Bragg Grating (FBG) sensor and piezoelectric sensor have been installed to collect the traffic volume and loading amplitude. However, limited attention has been paid on the algorithm of vehicle speed estimation, and the speeds of different vehicles are usually assumed at a normal speed. This research focuses on the estimation of the vehicle speed based on a cross-correlation method. A novel wireless Micro-Electromechanical Sensor (MEMS), RTS SmartRock, simply referred as SmartRock in this paper, is used to capture the triaxial acceleration, rotation, and stress data. The cross-correlation algorithms, i.e., normalized cross-correlation (NCC) algorithm, the Smoothed Coherence Transform (SCOT) algorithm, and the Phase Transform (PHAT) algorithm, are applied to estimate the loading speed of an accelerated pavement test (APT). The signal-noise-ratio (SNR) and the relative error (RE) are utilized to evaluate the stability and accuracy of the algorithms. The APT test data show that the noises of the APT data include the dependent and independent noises. In addition, the SCOT algorithm has a large SNR value of 12.98 and the lowest RE value of 0.0520 among the algorithms. It indicates that the main noises in the APT test are correlated noises that can be removed by the SCOT algorithm. Therefore, the cross-correlation method can be used to estimate the velocity of the vehicle in the APT setup. The quality of estimation in the field needs further evaluation and research.

Aldo D Fabregas, Florida Institute of Technology
Paul Crawford, Florida Institute of Technology
Rodrigo Mesa-Arango, Florida Institute of Technology
Agustina Calatayud, Inter-American Development Bank

Industrial Internet of Things or IIoT consists computing and communication at the operational edge of a value chain. Applications span manufacturing, energy, and transportation, among others. In the transportation context IIoT is translates to connected vehicle architecture and ITS applications. Current trends in communication allow more applications and devices to contribute to functions traditionally allocated to transportation ITS infrastructure. This poses challenges to design engineers to prototype emerging technologies in the existing transportation context and reduce integration, validation and time to market. Model-based systems engineering is crucial step for IIoT deployment since it allows to a feature-rich digital representation of the interaction of the system of interest and its context. This representation includes stakeholders, requirements, behaviors, flows, and numerical analyses associated with the solution space of the system of interest. This work presents an executable architecture developed in the System Modeling Language (SysML) in the domain of freight operations for trucks to and from maritime ports. The architecture was modeled in was that it can be customized based on presence/absence of services or devices in the deployment context (e.g. developing country, rural area, etc.). The model takes into consideration maritime port capabilities, truck communication capabilities, infrastructure capabilities and third-party actor’s capabilities (e.g. customs, freight forwarders, etc.). These integrations are typically part of larger initiatives such as supply chain integrations or port community systems. An example of parametric analysis and model-based metrics is presented.

Saeed Vasebi, Stevens Institute of Technology
Yeganeh Hayeri, Stevens Institute of Technology
Peter Jin, Rutgers University

Relatively recent increased computational power and extensive availability of traffic data have provided a unique opportunity to re-investigate drivers’ car-following (CF) behavior. Classic CF models assume drivers’ behavior is only influenced by their preceding vehicle. Recent studies have indicated that considering surrounding vehicles’ information (e.g., multiple preceding vehicles) could impact the performance of CF models. An in-depth investigation of surrounding vehicles’ contribution to CF modeling performance has not been reported in the literature. This study uses a deep-learning model with long short-term memory (LSTM) to investigate to what extent considering surrounding vehicles could improve the performance of CF models. This investigation helps to select the right inputs for traffic flow modeling. Five CF models are compared in this study (i.e., Classic, Multi-Anticipative, Adjacent-Lanes, Following-Vehicle, and All-Surrounding-Vehicles CF models). Performance of the CF models’ is compared in terms of accuracy, stability, and smoothness of traffic flow. The CF models are trained, validated, and tested by a large publicly available dataset. Results show insignificant performance differences between the Classic CF model and Multi-Anticipative model, or Adjacent-Lanes model in terms of accuracy, stability, or smoothness. The Following-Vehicle CF model shows similar performance to the Multi-Anticipative model. The All-Surrounding-Vehicles CF model has underperformed all the other models.

Xingzhou Guo, Purdue University
Yunfeng Chen, Purdue University
Jiansong Zhang, Purdue University

Infrastructure information modeling (IIM) can assist in life-cycle delivery and management of asset data, which requires the successful integration of all aspects of a business including process, technology, people, and information. However, issues in the current practice, such as unclear business process, isolated project phases, incompatible project software, and more importantly inaccurate, missing, or hard to find information required to effectively operate and manage infrastructure asset, have adversely affected the implementation of IIM. There is no shortage of theoretical discussion or solutions for life cycle integration of IIM. However, only limited aspects of IIM have been considered in previous studies. In addition, most research efforts lack sufficient empirical evidence for the identified challenges along a project’s life cycle. More significantly, no substantial theoretical justifications are provided for the potential solutions. Therefore, this study aims to explore the data need from operation and maintenance to data requirement for design and construction, and to explore process, technology, people, and information through a case study to outline the current workflow, identify specific gaps, and propose solutions. Main challenges identified are, for example, stakeholder’s reluctance to share digital files to avoid liability, blockages of data exchange caused by the use of different software and the lack of a data format that works across different software, etc. Correspondingly, new contract language has been proposed to improve current guidelines.

Haowen Xu, Oak Ridge National Laboratory
Chieh (Ross) Wang, Oak Ridge National Laboratory
Anne Berres, Oak Ridge National Laboratory
Tim LaClair, Oak Ridge National Laboratory

As traffic simulation software becomes more effective for realistically simulating and analyzing traffic dynamics and vehicle interactions on the mesoscopic and microscopic level, the management, dissemination, and collaborative visualization of traffic simulation results produced by individual transportation planners presents a significant challenge. Existing online content management systems, such as Dropbox and Google drive, can serve as repositories for storing and sharing output files from different traffic simulation software, but they have a very limited capability in allowing users to query specific traffic simulation scenarios and geospatially visualize simulation results through shareable and interactive web interfaces. In this paper, we present a web-based application for promoting the archiving, sharing, and visualization of large-scale traffic simulation outputs. The application is developed as an integral part of the \gls{rtmcs}, aims to enhance cyber-physical controls, communication, and public education for collaborative transportation planning.  The web application (a) allows users to upload their new traffic simulation scenarios (parameters and outputs), as well as search existing scenarios using easily-accessible interfaces; (b) optimizes simulation output files with heterogeneous data formats and geographic projections for web-based storage and management using a scalable and searchable data/metadata standard; and (c) provides shareable web visual interfaces to visualize multiscale traffic dynamics (e.g., regional mobility patterns and individual vehicle interactions). We present the conceptual design and implementation of our application, and demonstrate the application's performance for sharing, comparing, and visualizing simulation outputs from VISSIM and SUMO, two commonly used traffic simulation software.

Si Chen (sichen12@illinois.edu), University of Illinois, Urbana-Champaign
Brian Deal, University of Illinois, Urbana-Champaign
Shaowen Wang, University of Illinois, Urbana-Champaign
Julie Cidell, University of Illinois, Urbana-Champaign

The last mile problem describes the difficulty of public transit in getting people from transit stops to their homes. Previous studies on the last mile problem generally did not have an accessibility perspective, that is, they focus on analyzing access to the transit system but did not consider whether transit access helps people reach potential destinations. Problems arise when people can access to transit stops easily but the associated transit lines do not take them to their destinations. To address this issue, this paper develops a planning support system—the Last Mile Assessment System (LMAS)—to assess the last mile performance with an accessibility perspective that considers both the last-mile travel mode and the attractiveness of potential destinations. We used the city of Chicago to test and evaluate the performance of the LMAS. Google data was used to identify potential destinations and to evaluate their popularity. We developed an easy-to-navigate user interface to help visualize assessment scores and to collect public feedback for ranking and ordering. This engaging system brought in social sensed data and user interaction in destination attractiveness analysis, model parameter adjusting and result interpretation, to better support the application of LMAS and decrease model bias. Preliminary user feedback suggested that the model did reasonably well in assessing the last mile problems in Chicago. We suggest the LMAS is an effective planning and urban design support system that can help citizens, real estate agents, and planners to assess suitable living areas.

Hao Yang, University of Washington
Meixin Zhu, University of Washington
Ruimin Ke, University of Washington
Chenxi Liu, University of Washington
Yinhai Wang (yinhai@uw.edu), University of Washington

Traffic sensing devices are the eyes of the Intelligent Transportation Systems (ITS) nowadays, among which the surveillance camera system is one of the most widely deployed systems due to its information-rich data and intuitive format. However, the system has not been fully exploited: cameras are isolated and can only extract information from their own Field Of Views (FOVs). To extend studies from a single camera to network-level traffic parameters extraction and estimation, this study proposed a novel framework based on Multi-camera Object Tracking and Re-Identification (MOTRI). First, a cutting-edge cross-camera tracking and Re-ID framework based on temporal-attention deep neural networks are customized and improved, where a Spatial-temporal Camera Graph Inference Model (StCGIM) is designed to integrate the network level of features into the MOTRI. Finally, an Adaptative Accuracy Model (AAM) is designed to estimate the traffic information with various levels of cross-camera tracking accuracy for different camera links. The method was evaluated in a midsize USA city. Among all fifteen evaluated camera pairs, it was found that all the parameters estimation error is less than 8%, and the Kullback--Leibler (KL) distance between the estimated and real distributions is less than 3.42. This framework precisely models the network-scale traffic information utilizing the surveillance system rather than individual cameras, leading to a more advanced and efficient ITS.

Ruimin Ke, University of Washington
Zhiyong Cui, University of Washington
Yanlong Chen, University of Tokyo
Meixin Zhu, University of Washington
Yinhai Wang (yinhai@uw.edu), University of Washington

Our world is moving towards the goal of fully autonomous driving at a fast pace. While the latest automated vehicles (AVs) can handle most real-world scenarios they encounter, a major bottleneck for turning fully autonomous driving into reality is the lack of sufficient corner case data for training and testing AVs. Near-crash data, as a widely used surrogate data for traffic safety research, can also serve the purpose of AV testing if properly collected. To this end, this paper proposes an Internet-of-Things (IoT) system for real-time near-crash data collection. The system has several cool features. First, it is a low-cost and standalone system that is backward-compatible with any existing vehicles. People can fix the system to their dashboards for near-crash data collection and collision warning without the approval or help of vehicle manufacturers. Second, we propose a new near-crash detection method that models the target’s size changes and relative motions with the bounding boxes generated by deep-learning-based object detection and tracking. This near-miss detection method is fast, accurate, and reliable; particularly, it is insensitive to camera parameters, thereby having an excellent transferability to different dashboard cameras. We have conducted comprehensive experiments with 100 videos locally processed at Jetson, as well as real-world tests on cars and buses. Besides collecting corner cases, it can also serve as a white-box platform for testing innovative algorithms and evaluating other AV products. The system contributes to the real-world testing of AVs and has great potential to be brought into large-scale deployment.

Yi Ge, Rutgers University, New Brunswick
Peter Jin, Rutgers University
Tianya Zhang, Rutgers University, New Brunswick
Jonathan Martinez, Rutgers University

The video analytic systems for the CCTV (Closed Circuit Television) traffic videos can help alleviate the workload for traffic operators in traffic operations and management. Cloud computing technologies use an on-demand architecture to dynamic assign computer resources such as storage and computing powers without the need to actively managing a local computing center. Cloud computing had significant success in many areas such as video conferencing, video streaming, computer gaming, etc. This paper explores the cloud- versus server-based deployment scenarios of an enhanced computer vision platform for potential deployment on low-resolution 511 traffic video streams. An existing Spatial-Temporal Map (STMap) based computer vision algorithms designed for NGSIM (Next-Generation Simulation)-like high-angle traffic video is enhanced for the roadside CCTV traffic camera angles. Due to the lower visual angle, determining the directions, splitting vehicles from occlusions, and identifying lane-changes become difficult. A motion-flow-based direction determination method, a bi-section occlusion detection and splitting algorithm, and the lane-changing track method are proposed. The model evaluation is conducted by using videos from multiple cameras from the NJDOT (New Jersey DOT)’s 511 traffic video surveillance system. The results show promising performance in both accuracy and computational efficiency for potential large-scale cloud deployment. The cost analysis reveals that at the current pricing model of cloud computing, the cloud-based deployment is more convenient and cost-effective for an on-demand network assessment. In contrast, the dedicated server-based deployment is more economical for long-term traffic detection deployment.