Search

Traffic oscillation on freeways results in abrupt and frequent vehicle deceleration, which remarkably increases crash risks. Artificial Intelligence based traffic control provides a new opportunity for addressing the safety issues. This study aims at proposing a Reinforcement Learning (RL)-based variable speed limits (VSL) control algorithm to reduce crash risks associated with oscillations. The state, action and reward, which are the critical components in the RL, were designed carefully for safety improvement. The RL was trained to learn the optimal speed limit for various traffic states to achieve a goal of safety optimization. A rear-end crash risk model was applied to assess crash risks associated with oscillations near freeway bottlenecks. The cell transmission model was modified as the simulation platform for evaluating the control effects. The results showed that after the training process, the proposed RL-based VSL control successfully reduced the crash risks by 19.4%. A continuous online learning function was developed in RL to enhance the robustness of our strategy. The results showed that with continuous learning, the RL-based VSL control performed reasonably well under lower driver compliance situations.

Road surface friction significantly impacts traffic safety and mobility. A precise road surface friction prediction model can help to alleviate the influence of inclement road conditions on traffic safety, Level of Service, traffic mobility, fuel efficiency, and sustained economic productivity. Most related previous studies are laboratory-based methods that are difficult for practical implementation. Moreover, in other data-driven methods, the demonstrated time-series features of road surface conditions have not been considered. This study employed a Long-Short Term Memory (LSTM) neural network to develop a data-driven road surface friction prediction model based on historical data. The proposed prediction model outperformed the other baseline models in terms of the lowest value of predictive performance measurements. The influence of the number of time-lags and the predicting time interval on predictive accuracy was analyzed. In addition, the influence of adding road surface water thickness, road surface temperature and air temperature on predictive accuracy also were investigated. The findings of this study can support road maintenance strategy development and decision making, thus mitigating the impact of inclement road conditions on traffic mobility and safety. Future work includes a modified LSTM-based prediction model development by accommodating flexible time intervals between time-lags., Comment: arXiv admin note: text overlap with arXiv:1911.00605

An accurate road surface friction forecasting algorithm can allow travelers and managers to schedule trips and maintenance activities based on the road weather condition to enhance traffic safety and efficiency in advance. Previously, scholars developed multiple forecasting models to predict road surface conditions using historical data. However, historical dataset used for model training may have missing values caused by multiple issues, e.g. the data collected by on-vehicle sensors may be influenced when vehicles cannot travel due to high economic and labor cost or weather-related issues.The missing values in the road surface condition dataset can damage the effectiveness and accuracy of the existing prediction methods. This study proposed a road surface friction forecasting algorithm by employing a time-aware Gated Recurrent Unit (GRU-D) networks that integrate a decay mechanism as extra gates of the GRU to handle the missing values and forecast the road surface friction in future periods simultaneously. The evaluation results present that the proposed GRU-D networks outperform all selected baseline algorithms. The impacts of missing rate on predictive accuracy, learning efficiency, and learned decay rates are investigated as well. The findings can help improve the forecasting accuracy and efficiency of road surface friction prediction using historical data with missing values, therefore mitigating the negative impact of wet or icy road conditions on traffic safety and efficiency.

In public areas, crowd stampedes and incidents generate huge negative impacts on public security. Accurate and efficient crowd density estimation is critical to monitor crowd status for developing evacuation strategies. The existing crowd density estimation methods are established based on complex deep-learning algorithms which are usually more accurate, but, on the other side, they require much more computational resources. Consequently, cloud-computing is the only option for deploying crowd density estimation algorithms, which needs tremendous resources for real-time video data transmission and the malfunction and delay of internet service may cause wrong and delayed estimation results. Edge computing is a novel concept of accomplishing computing tasks only relying on the computational resources of edge devices. Estimating crowd density on edge devices rather than conveying images to cloud server for further analysis has several advantages, including 1) reducing network bandwidth pressure from remote transmission; 2) avoiding risks of leaking privacy on images; and 3) improving computation efficient. This study designs an edge computing-enabled crowd density estimation model based on the residual bottleneck block and dilated convolution. The experiments are designed and conducted on public crowd data sets to verify accuracy, storage cost and computation efficiency of our model. According to the experimental results, the proposed model achieves a considerable improvement in operational efficiency, while keep the accuracy at the same level with the complex deep-learning algorithms. Furthermore, the proposed model is implemented on a real edge device to detect real-world crowd density in a Beijing subway station.

As the automobile market gradually develops towards intelligence, networking, and information-orientated, intelligent identification based on connected vehicle data becomes a key technology. Specifically, real-time crash identification using vehicle operation data can enable automotive companies to obtain timely information on the safety of user vehicle usage so that timely customer service and roadside rescue can be provided. In this paper, an accurate vehicle crash identification algorithm is developed based on machine learning techniques using electric vehicles' operation data provided by SAIC-GM-Wuling. The point of battery disconnection is identified as a potential crash event. Data before and after the battery disconnection is retrieved for feature extraction. Two different feature extraction methods are used: one directly extracts the descriptive statistical features of various variables, and the other directly unfolds the multivariate time series data. The AdaBoost algorithm is used to classify whether a potential crash event is a real crash using the constructed features. Models trained with the two different features are fused for the final outputs. The results show that the final model is simple, effective, and has a fast inference speed. The model has an F1 score of 0.98 on testing data for crash classification, and the identified crash times are all within 10 s around the true crash times. All data and code are available at https://github.com/MeixinZhu/vehicle-crash-identification.

To simulate pedestrian evacuation processes on a metro station platform in a case of fire, a specific evacuation model is proposed, using an integrated cellular automata (CA) approach, in which the impacts from exits, other evacuees, and fire and smoke are included to measure the probability of the evacuee getting to each neighboring cell. The evacuation is firstly identified as a two-stage process including the motion on the platform and on the treads. Then the evacuation space is drawn to be a three-dimensional grid space, in which the cell size is defined by the stair structure and human body size. Based on that, this study proposes two CA models to simulate the evacuees’ movement and the smoke diffusion separately. Moreover, to describe the evacuation process in detail, the evacuation model is modified in three ways. First, transition rules in the evacuees’ movement model are embedded by social force theory to measure the impacts from the environment. Second, the smoke diffusion process is modified by considering the smoke control measures on the metro platform. Third, impact from smoke is quantified by the proportion of smoke in the centroid cell of evacuees. Finally, results from simulation experiments show that this model is able to recognize the arching and stagnation phenomenon at the foot of staircases, and the relations between the evacuation time and the crowd density for different parameters are also analyzed. The proposed method of simulating the pedestrian evacuation process can be useful in providing guiding principles for the software design of evacuation in metro systems.

A robust, integrated and flexible charging network is essential for the growth and deployment of electric vehicles (EVs). The State Grid of China has developed a Smart Internet of Electric Vehicle Charging Network (SIEN). At present, there are three main ways to attack SIEN maliciously: distributed data tampering; distributed denial of service (DDoS); and forged command attacks. Network attacks are random and continuous, closely related to time. By contrast, when analyzing the alarm in malicious attacks, the traditional Markov chain based model ignores the association relationship in the time series between states of alarm, so that the analysis and prediction of alarms are not suitable for real situations. This paper analyzes the characteristics of the three types of attack and proposes an association state analysis method on the time series. This method firstly analyzes alarm logs at different locations, different levels, and different types, and then establishes the temporal association of scattered and isolated alarm information. Secondly, it tracks the transition trend of abnormal events in the SIEN’s main station layer, the channel layer, and the sub-station layer. It also identifies the real attack behavior. This method not only provides a prediction of security risks, but, more importantly, it can also accurately analyze the trend of SIEN security risks. Compared with the ordinary Markov chain model, this method can better smooth the fluctuation of processing values, with higher real-time performance, stronger robustness, and higher precision. This method has been applied to the State Grid of China.

With the development of mobile internet technology, on-demand ridesourcing services have rapidly spread across the world and have caused debates in the transportation industry. While many researchers have begun to study the characteristics and impacts of ridesourcing, there are few published studies specifically on ridesplitting, a ridesourcing service that matches riders with similar origins or destinations to the same ridesourcing driver and vehicle in real time. This paper aims to explore the characteristics and effects of ridesplitting using observed ridesourcing data provided by DiDi Chuxing that contain complete datasets of the ridesourcing trajectories and orders in the city of Chengdu, China. First, a ridesplitting trip identification (RTI) algorithm is developed to separate the shared rides from the single rides (non-ridesplitting orders) and derive ridesplitting scales. Second, a ridesplitting trajectory reconstruction (RTR) algorithm is proposed to estimate the ridesplitting effects on delays and detours. Then, we analyze and compare the scales, spatiotemporal patterns and travel characteristics between shared rides and single rides, which are very different. The results show that the current percentage of ridesplitting in ridesourcing is still low (6–7%), which may be explained by the extra delay (about 10 min on average), detour (about 1.55 km on average), and degraded travel time reliability caused by ridesplitting. In addition, built environment factors, such as density, diversity, and development, are also correlated with ridesplitting demand and delay. The findings of this study can help better understand the features of ridesplitting and develop strategies for improving its use in emerging ridesourcing services.

The effectiveness of the variable speed limit (VSL) control is affected by the deployment locations of VSL signs. In this paper, a procedure was proposed to help determine the deployment location of VSL signs to reduce collision risks at freeway recurrent bottlenecks. The procedure is started from determining the hazardous recurrent bottleneck section by constructing the profile of the collision risks. The length of the hazardous section was determined according to the pre-selected threshold of crash risks. Various scenarios were considered for different lengths of VSL controlled sections with various densities of VSL signs. A modified cell transmission model (CTM) was used for modeling the traffic flow at the freeway bottlenecks under the VSL control. The VSL control factors were optimized by using the genetic algorithm. The safety effects of VSL were greatly affected by the placement of VSL signs. In general, the scenario with a longer VSL controlled section and more speed limit signs was more effective in reducing rear-end collision risks. The cost-benefit analyses showed that the placement of VSL signs in scenario C2 with the controlled section length of 11.7 mi and average VSL density of 1.0 mi had the best benefit/cost effect. Using 12 VSL signs, the collision risks were reduced by 69.16% and the total travel time was increased slightly by 7.73%. The procedure can help determine the optimal deployment of VSL signs on freeways, considering both the safety benefits and the cost.

Although various theories have been adopted to develop reliable pedestrian walking models, limited effort has been made to shed light on the case at signalized crosswalk, especially during the pede...

Vertical curve features on interstate highways greatly affect traffic operations and vehicle performance and, thus, could have an impact on the occurrence of traffic crashes. Most studies t...

Traditional travel modes have been slowly losing their market share to e-hailing services, especially among younger generations. This paper examines characteristics of young travelers’ trav...

Short-term traffic forecasting based on deep learning methods, especially recurrent neural networks (RNN), has received much attention in recent years. However, the potential of RNN-based models in traffic forecasting has not yet been fully exploited in terms of the predictive power of spatial–temporal data and the capability of handling missing data. In this paper, we focus on RNN-based models and attempt to reformulate the way to incorporate RNN and its variants into traffic prediction models. A stacked bidirectional and unidirectional LSTM network architecture (SBU-LSTM) is proposed to assist the design of neural network structures for traffic state forecasting. As a key component of the architecture, the bidirectional LSTM (BDLSM) is exploited to capture the forward and backward temporal dependencies in spatiotemporal data. To deal with missing values in spatial–temporal data, we also propose a data imputation mechanism in the LSTM structure (LSTM-I) by designing an imputation unit to infer missing values and assist traffic prediction. The bidirectional version of LSTM-I is incorporated in the SBU-LSTM architecture. Two real-world network-wide traffic state datasets are used to conduct experiments and published to facilitate further traffic prediction research. The prediction performance of multiple types of multi-layer LSTM or BDLSTM models is evaluated. Experimental results indicate that the proposed SBU-LSTM architecture, especially the two-layer BDLSTM network, can achieve superior performance for the network-wide traffic prediction in both accuracy and robustness. Further, comprehensive comparison results show that the proposed data imputation mechanism in the RNN-based models can achieve outstanding prediction performance when the model’s input data contains different patterns of missing values.

Cloud computing has been a main-stream computing service for years. Recently, with the rapid development in urbanization, massive video surveillance data are produced at an unprecedented speed. A traditional solution to deal with the big data would require a large amount of computing and storage resources. With the advances in Internet of things (IoT), artificial intelligence, and communication technologies, edge computing offers a new solution to the problem by processing all or part of the data locally at the edge of a surveillance system. In this study, we investigate the feasibility of using edge computing for smart parking surveillance tasks, specifically, parking occupancy detection using the real-time video feed. The system processing pipeline is carefully designed with the consideration of flexibility, online surveillance, data transmission, detection accuracy, and system reliability. It enables artificial intelligence at the edge by implementing an enhanced single shot multibox detector (SSD). A few more algorithms are developed either locally at the edge of the system or on the centralized data server targeting optimal system efficiency and accuracy. Thorough field tests were conducted in the Angle Lake parking garage for three months. The experimental results are promising that the final detection method achieves over 95% accuracy in real-world scenarios with high efficiency and reliability. The proposed smart parking surveillance system is a critical component of smart cities and can be a solid foundation for future applications in intelligent transportation systems.

In recent years, with the advancement in traffic sensing, data storage, and communication technologies, the availability and diversity of transportation data have increased substantially. When the volume and variety of traffic data increase dramatically, integrating multisource traffic data to conduct traffic analysis is becoming a challenging task. The heterogeneous spatiotemporal resolutions of traffic data and the lack of standard geospatial representations of multisource data are the main hurdles for solving the traffic data-integration problem. In this study, to overcome these challenges, a transportation data-integration framework based on a uniform geospatial roadway referencing layer is proposed. In the framework, on the basis of traffic sensors’ locations and sensing areas, transportation-related data are classified into four categories, including on-road segment-based data, off-road segment-based data, on-road point-based data, and off-road point-based data. Four data-integration solutions are proposed accordingly. An iterative map conflation algorithm as a core component of the framework is proposed for integrating the on-road segment-based data. The overall integration performance of the four types of data and the efficiency of the iterative map conflation algorithm in terms of percentage of integrated roadway segments and computation time are analyzed. To produce efficient transportation analytics, the proposed framework is implemented on an interactive data-driven transportation analytics platform. Based on the implemented framework, several case studies of real-world transportation data analytics are presented and discussed.

Same-day delivery and on-demand delivery with driverless delivery robots (DDRs) are becoming new attractive options for more customers looking for grocery or medication delivery, as these delivery methods can customize time demand and meet consumers’ safety expectations. However, meeting these requirements for instant shipping necessarily increases the need for more vans and DDRs for last-mile delivery, thus increasing the economic and ecological costs. To optimize the economic costs and environmental effects of the delivery network, and also to meet customer satisfaction simultaneously, an effective model considering the new constraints of the van-DDR system and an efficient algorithm are needed to obtain the solutions. Therefore, the goals of this study are to establish a model and develop an algorithm for a multi-objective multi-depot two-tier location routing problem with parcel transshipment (MOMD-2T-LRP-PT), where vans and DDRs serve the two tiers, respectively. In this study, we split the MOMD-2T-LRP-PT model into two subproblems: the location-allocation problem and the vehicle routing problem. The two problems are solved sequentially and iteratively with a “k-prototype cluster” and a hybrid artificial immune algorithm (HAIA). We firstly illustrate the effectiveness of the MOMD-2T-LRP-PT model with the ∊-constraint method on a small-scale data set. Then the proposed HAIA algorithm is compared with a nondominated sorting genetic algorithm II (NSGA-II) using different data sets including a real case test. Both the analytic results and the real case application show that the ∊-constraint method can produce the best solution with up to six customers, and the HAIA algorithm produces better-optimized results than NSGA-II in real-life applications. These results imply that the MOMD-2T-LRP-PT model and the proposed HAIA algorithm are promising and effective in optimizing practical E-grocery delivery that can achieve optimization and balance among economic costs, environmental effects, and customer satisfaction.

Ridespitting, which enables riders with similar routes to share a ridesourcing trip, is a promising transportation technology to reduce traffic congestions and air pollutions. This study aims to explore how ridesplitting reduces emissions from ridesourcing based on GPS trajectory data from Didi Chuxing in Chengdu, China. First, this study quantifies the emission factors of both regular ridesourcing and ridesplitting trips to evaluate the emission reductions per ride-km from ridesplitting. The results show that the average emission reduction rates of CO2, CO, NOx, and HC are 28.7%, 32.5%, 27.7%, and 31.2%, respectively. Then, a spatiotemporal analysis of the emission reductions indicates that ridesplitting generally reduces more emissions around the expressways and during peak hours. Finally, a spatial error model is adopted to analyze the effects of travel-related and built environment variables on emission reductions from ridesplitting. The trajectory overlapping rate of shared rides turns out to be the most important determinant for expanding the environmental benefits of ridesplitting.

Adjustment of parking price has long been considered an effective way to control parking demand and demand has often been shown to be affected by spatial factors. The primary objective of this study is to investigate the spatial heterogeneity in the sensitivity of parking occupancy to price change using data obtained in downtown San Francisco between 2011 and 2014. The performance-based pricing implemented in the study area allows parking rate to increase, decrease or remain unchanged in neighborhoods with parking occupancy levels higher than, lower than, or within a desired range. As such, the relationship between change in occupancy and change in parking rate is explored. The geographically weighted regression (GWR) method was used to capture the spatial heterogeneity in sensitivity in different blocks and modeling results showed that there is a significant negative correlation between occupancy change and parking rate change. Thus, sensitivity of on-street parking occupancy to price change has an obvious trend of spatial variation. By capturing the spatial heterogeneity in the dataset, the GWR model achieved higher prediction accuracy than a global model. Variables including time of day, block-level features, and socio-demographic characteristics were also found to be correlated with occupancy change. Based on the GWR outputs, a generalized linear model was estimated to further identify how various factors affect sensitivity in different block areas. Findings of this study can be used to help parking authorities with tasks such as identifying which blocks are suitable for balancing parking demand and supply by adjusting price and designing optimal parking rate schemes to achieve desired on-street parking occupancy levels.

Nowadays phone distraction has started to become an increasingly recognized phenomenon. This article aims to examine the influences of phone use on pedestrian crossing behavior at signalized intersections in China. Using video recording and manual counting, pedestrian crossing behavior, age, gender, phone use, and waiting time are obtained at four signalized intersections. Totally, 4196 pedestrians are observed in four peak hours. Among them, 328 pedestrians (7.82%) are using their mobile phones, including 162 male pedestrians and 166 female pedestrians. The average phone use rate in different age groups are presented as follows: children (4.49%), youth (10.69%), middle-aged (6.87%), and elderly (1.15%). In terms of the phone using behavior on the crosswalk, age is a significant factor, while gender is not. For the type of violation behavior, the results show that pedestrians who use mobile phones are most likely to be late starters. In addition, some other important results are found: mobile phone use can raise the violation of pedestrian crossing behavior; pedestrians using their phones are more likely to cross on red; and pedestrians using their phones while crossing walk more slowly. Finally, the application significance of this study and some recommendations are provided to improve pedestrian safety.

Old town fringe area refers to a surrounding zone of an old town. The fringe undertakes the important function of evacuating traffic volume between suburb districts and central city. Traffic flow and travel behavior of the old town fringe are complex and different from other areas. Thus, recognition of fringe area is of importance for researchers to better understand the unique travel features and propose proper policies for fringe renewal. This study took the city of Nanjing as an example and fused the residents’ travel survey data and the point-of-interest data for fringe recognition. The study estimated the travel intensity of each travel analysis zone per day. The mutation point of travel intensity was decided to divide the fringe. The point-of-interest data was used for validating the boundaries of the fringe. The fusion of the two data sets jointly decided the core old town fringe area. The travel behavior characteristics of the fringe area, including the fringe internal trips, crossing fringe trips, and those with only origin or destination in the fringe, were then evaluated and policy suggestions were provided. The findings of the study will benefit the urban space planning and coordinated transportation system development in the fringe area of old towns.

A model used for velocity control during car following was proposed based on deep reinforcement learning (RL). To fulfil the multi-objectives of car following, a reward function reflecting driving safety, efficiency, and comfort was constructed. With the reward function, the RL agent learns to control vehicle speed in a fashion that maximizes cumulative rewards, through trials and errors in the simulation environment. A total of 1,341 car-following events extracted from the Next Generation Simulation (NGSIM) dataset were used to train the model. Car-following behavior produced by the model were compared with that observed in the empirical NGSIM data, to demonstrate the model's ability to follow a lead vehicle safely, efficiently, and comfortably. Results show that the model demonstrates the capability of safe, efficient, and comfortable velocity control in that it 1) has small percentages (8\%) of dangerous minimum time to collision values (\textless\ 5s) than human drivers in the NGSIM data (35\%); 2) can maintain efficient and safe headways in the range of 1s to 2s; and 3) can follow the lead vehicle comfortably with smooth acceleration. The results indicate that reinforcement learning methods could contribute to the development of autonomous driving systems., Under the first-round revision for transportation research part c

Network-wide traffic forecasting is a critical component of modern intelligent transportation systems for urban traffic management and control. With the rise of artificial intelligence, many recent studies attempted to use deep neural networks to extract comprehensive features from traffic networks to enhance prediction performance, given the volume and variety of traffic data has been greatly increased. Considering that traffic status on a road segment is highly influenced by the upstream/downstream segments and nearby bottlenecks in the traffic network, extracting well-localized features from these neighboring segments is essential for a traffic prediction model. Although the convolution neural network or graph convolution neural network has been adopted to learn localized features from the complex geometric or topological structure of traffic networks, the lack of flexibility in the local-feature extraction process is still a big issue. Classical wavelet transform can detect sudden changes and peaks in temporal signals. Analogously, when extending to the graph/spectral domain, graph wavelet can concentrate more on key vertices in the graph and discriminatively extract localized features. In this study, to capture the complex spatial-temporal dependencies in network-wide traffic data, we learn the traffic network as a graph and propose a graph wavelet gated recurrent (GWGR) neural network. The graph wavelet is incorporated as a key component for extracting spatial features in the proposed model. A gated recurrent structure is employed to learn temporal dependencies in the sequence data. Comparing to baseline models, the proposed model can achieve state-of-the-art prediction performance and training efficiency on two real-world datasets. In addition, experiments show that the sparsity of graph wavelet weight matrices greatly increases the interpretability of GWGR.

As the amount of traffic congestion continues to grow, pinpointing freeway bottleneck locations and quantifying their impacts are crucial activities for traffic management and control. Amon...

Surveillance video cameras have been increasingly deployed on roadway networks providing important support for roadway management. While the information-rich video images are a valuable source of traffic data, these surveillance video cameras are typically designed for manual observation of roadway conditions and are not for automatic traffic data collection. The benefits of turning these surveillance cameras into data collection cameras are obvious, but collecting traffic data would normally require the development of a cost-effective method to efficiently and accurately calibrate surveillance video cameras. This paper proposes such a robust and efficient method that calibrates surveillance video cameras using standard shipping container as the reference object. The traditional camera calibration model can be simplified and camera parameters can be recovered with precise mathematical derivation. After solving for all the camera parameters, the 3D object world coordinates can be reconstructed from 2D image coordinates, thus enabling the collection of a variety of traffic data using surveillance video camera data.

Many urban streets are designed with on-street bike lanes to provide right-of-way for bicycle traffic. However, when bicycle flow is large, extensive passing maneuvers could occupy vehicle lanes and thus cause interferences to vehicle traffic. The primary objective of this study is to evaluate how bicycle traffic affects vehicle operation on urban streets with bike lanes. Data were collected on six street segments in Nanjing, China. The cumulative curves were constructed to extract traffic flow information including individual bicycle and vehicle speeds and aggregated traffic parameters such as flow and density. The results showed that as bicycle density on bike lanes continuously increases faster bicycles may run into vehicle lanes causing considerable reductions in vehicle speeds. A generalized linear model was estimated to predict the vehicle delay. Results showed that vehicle delay increases as bicycle flow and vehicle flow increase. Number of vehicle lanes and width of bike lane also have significant impact on vehicle delay. Findings of the study are helpful to regions around the world in bike infrastructure design in order to improve operations of both bicycles and vehicles.

The impact of cell phone use on driving performance has been widely known yet is carried out to varying degrees under different driving conditions. 40 participants whose ages ranged from 23 to 25 took part in experiments on both highway and urban roads using a driving simulator. The experiment was carried out under five different conditions: 1) no call; 2) different phone types (hand-held or hands-free); 3) different conversation contents (casual or cognitive); 4) dialing; 5) texting messages. Driving performance indices such as speed maintenance, lane keeping, pedal force and displacement were monitored and analyzed. A reliability model of cell phone use under different driving conditions was established for the comprehensive analysis of different factors on driving performance. Results indicated that driving reliability decreased significantly despite the phone type and conversation content as the time of cell phone use increased, and frequent short-time cell phone conversation was more dangerous.