Trajectory Dataset Research Articles

City-scale Vehicle Trajectory Data from Traffic Camera Videos

Vehicle trajectory data underpins various applications in intelligent transportation systems, such as traffic surveillance, traffic prediction, and traffic control. Traditional vehicle trajectory datasets, recorded by GPS devices or single cameras, are often biased towards specific vehicles (e.g., taxis) or incomplete (typically < 1 km), limiting their reliability for downstream applications. With the widespread deployment of traffic cameras across the city road network, we have the opportunity to capture all vehicles passing by. By collecting city-scale traffic camera video data, we apply a trajectory recovery framework that identifies vehicles across all cameras and reconstructs their paths in between. Leveraging this approach, we are the first to release a comprehensive vehicle trajectory dataset that covers almost full-amount of city vehicle trajectories, with approximately 5 million trajectories recovered from over 3000 traffic cameras in two metropolises. To assess the quality and quantity of this dataset, we evaluate the recovery methods, visualize specific cases, and compare the results with external road speed and flow statistics. The results demonstrate the consistency and reliability of the released trajectories. This dataset holds great promise for research in areas such as unveiling traffic dynamics, traffic network resilience assessment, and traffic network planning.

Colossal Trajectory Mining: A unifying approach to mine behavioral mobility patterns

Spatio-temporal mobility patterns are at the core of strategic applications such as urban planning and monitoring. Depending on the strength of spatio-temporal constraints, different mobility patterns can be defined. While existing approaches work well in the extraction of groups of objects sharing fine-grained paths, the huge volume of large-scale data asks for coarse-grained solutions. In this paper, we introduce Colossal Trajectory Mining (CTM) to efficiently extract heterogeneous mobility patterns out of a multidimensional space that, along with space and time dimensions, can consider additional trajectory features (e.g., means of transport or activity) to characterize behavioral mobility patterns. The algorithm is natively designed in a distributed fashion, and the experimental evaluation shows its scalability with respect to the involved features and the cardinality of the trajectory dataset.

PPDF-FedTMI: A Federated Learning-based Transport Mode Inference Model with Privacy-Preserving Data Fusion

Mobile Crowd-Sensing (MCS) represents a distributed data fusion system that aggregates diverse sources to enhance inference capabilities beyond any single input. As an essential application in MCS, intelligent transportation system deploys a large number of Transport Mode Inference (TMI) models. However, robust TMI models necessitates large trajectories datasets, engendering significant privacy risks. Existing solutions fail to adequately preserve trajectory privacy in collaborative MCS. To enable secure distributed modeling while safeguarding user anonymity, a Federated Learning-based Transport Mode Inference Model with Privacy-Preserving Data Fusion (PPDF-FedTMI) is proposed in this paper. Firstly, federated learning is incorporated into the TMI architecture to mitigate isolated data silos in MCS. Subsequently, a fuzzy fusion algorithm leverages fuzzy clustering of labels to facilitate stable and effective training under non-independent and non-identically distributed decentralized data. Furthermore, a local differential privacy-based joint training technique is devised to enhance gradient privacy, preventing exposure of original user data through model updates. Finally, the experiments demonstrate PPDF-FedTMI provides superior inference accuracy and effective privacy-preserving compared to current models.

CATS: Conditional Adversarial Trajectory Synthesis for privacy-preserving trajectory data publication using deep learning approaches

The prevalence of ubiquitous location-aware devices and mobile Internet enables us to collect massive individual-level trajectory dataset from users. Such trajectory big data bring new opportunities to human mobility research but also raise public concerns with regard to location privacy. In this work, we present the Conditional Adversarial Trajectory Synthesis (CATS), a deep-learning-based GeoAI methodological framework for privacy-preserving trajectory data generation and publication. CATS applies K-anonymity to the underlying spatiotemporal distributions of human movements, which provides a distributional-level strong privacy guarantee. By leveraging conditional adversarial training on K-anonymized human mobility matrices, trajectory global context learning using the attention-based mechanism, and recurrent bipartite graph matching of adjacent trajectory points, CATS is able to reconstruct trajectory topology from conditionally sampled locations and generate high-quality individual-level synthetic trajectory data, which can serve as supplements or alternatives to raw data for privacy-preserving trajectory data publication. The experiment results on over 90k GPS trajectories show that our method has a better performance in privacy preservation, spatiotemporal characteristic preservation, and downstream utility compared with baseline methods, which brings new insights into privacy-preserving human mobility research using generative AI techniques and explores data ethics issues in GIScience.

Do looks matter for an academic career in economics?

We show that physical appearance plays a role in the success of economics PhD graduates and investigate the underlying mechanisms driving this relationship. Leveraging a unique dataset of career and research productivity trajectories of PhD graduates from leading economics departments in the United States, we provide robust evidence that appearance is a predictive factor for both research productivity and job placement. Our analysis goes beyond establishing the association between attractiveness and success within the profession. By jointly examining appearance, job outcome, and research productivity, as well as the longitudinal development of the latter two over time, we show that the effect of appearance can be partially, but not fully, attributed to its role as a predictor of research productivity, with the remainder of the effect reflecting an intrinsic demand for attractiveness.

Predicting Urban Region Heat via Learning Arrive-Stay-Leave Behaviors of Private Cars

Urban region heat refers to the extent of which people congregate in various regions when they travel to and stay in a specified place. Predicting urban region heat facilitates broad applications ranging from location-based services to intelligent transportation management. The region heat is essentially characterized by the ‘arrive-stay-leave (ASL)’ behaviors, while it is a challenging task to well capture the spatial-temporal evolution of region heat since the following issues remain: i) ASL behaviors of private cars is usually heterogeneous resulting in a hierarchical distribution of region heat. ii) Urban region heat contains complex spatial-temporal correlations hidden in ASL behaviors and how to collaboratively integrate them is challenging. To address these challenges, we propose a Hierarchical Spatial-Temporal Network (HierSTNet) to forecast urban region heat, which contains two representations, namely, grid region from micro perspective and node region from macro perspective. For the grids, three-dimension spatial and temporal convolutional network (3D-STCNN) is proposed to model multi-scale properties in temporal dimension of ASL behaviors. For the nodes, multi-head graph attention networks are utilized to model the periodicity and spatial heterogeneity among macro region. Hierarchical structures are designed for multi-view modeling spatial-temporal distribution of ASL behaviors, by which they capture small-scale features in micro regions and embeds the global representation into graph propagation. Finally, we design an interaction decoder layer to integrate the external factors and aggregate spatial-temporal information across hierarchical structures. Extensive experiments based on real-world private car trajectory dataset demonstrate the superiority and effectiveness of proposed framework.

VNAGT: Variational Non-Autoregressive Graph Transformer Network for Multi-Agent Trajectory Prediction

Accurately predicting the trajectory of road agents in complex traffic scenarios is challenging because the movement patterns of agents are complex and stochastic, not only depending on their own past trajectories but also being closely related to the social interaction with other surrounding agents. Besides accuracy, efficient prediction with low inference latency is also a highly desirable feature for the practical application of trajectory prediction. To address these issues, we propose the VNAGT model, a variational non-autoregressive graph transformer that adopts the framework of conditional variational autoencoder and incorporates the non-autoregressive approach such that diverse trajectories and low prediction latency can be achieved simultaneously. In order to capture the social and temporal interaction, we put forward a unified graph attention-based module that is applicable for homogeneous and heterogeneous multi-agents such that the class information can be seamlessly integrated when it is available. Non-autoregressive decoding is combined with variational learning to produce multiple plausible predictions with low latency. We train and validate the model on two real-world homogeneous and heterogeneous trajectory datasets. The experimental results demonstrate the superior performance of our model in comparison with the state-of-the-art methods.

Mobility trajectory generation: a survey

Mobility trajectory data is of great significance for mobility pattern study, urban computing, and city science. Self-driving, traffic prediction, environment estimation, and many other applications require large-scale mobility trajectory datasets. However, mobility trajectory data acquisition is challenging due to privacy concerns, commercial considerations, missing values, and expensive deployment costs. Nowadays, mobility trajectory data generation has become an emerging trend in reducing the difficulty of mobility trajectory data acquisition by generating principled data. Despite the popularity of mobility trajectory data generation, literature surveys on this topic are rare. In this paper, we present a survey for mobility trajectory generation by artificial intelligence from knowledge-driven and data-driven views. Specifically, we will give a taxonomy of the literature of mobility trajectory data generation, examine mainstream theories and techniques as well as application scenarios for generating mobility trajectory data, and discuss some critical challenges facing this area.

Identifying dynamic interaction patterns in mandatory and discretionary lane changes using graph structure

AbstractA quantitative understanding of dynamic lane‐changing interaction patterns is indispensable for improving the decision‐making of autonomous vehicles (AVs), especially in mixed traffic with human‐driven vehicles. This paper develops a novel framework combining the hidden Markov model (HMM) and graph structure to identify the difference in dynamic interaction patterns between mandatory lane changes (MLC) and discretionary lane changes (DLC). An HMM is developed to separate the interaction patterns considering heterogeneity in lane‐changing processes and reveal the temporal properties of these patterns. Conditional mutual information is used to quantify the interaction intensity, and the graph structure is used to characterize the relationship between vehicles. Finally, a case study is conducted to demonstrate the practical value of the proposed framework and validate its effectiveness in predicting lane‐changing trajectories. Based on the lane‐changing events extracted from a real‐world trajectory dataset, the proposed analytical framework is applied to model MLC and DLC under congested traffic with levels of service E and F. The results show that there could be multiple heterogeneous dynamic interaction patterns in a lane‐changing process. A comparison of MLC and DLC demonstrates that MLC involves more intense interactions and more frequent transitions of the interaction network structure, while the evolution rules of interaction patterns in DLC do not exhibit a clear trend. The findings in this study are useful for understanding the connectivity structure between vehicles in lane‐changing interactions and for designing safe and smooth driving decision‐making models for AVs and advanced driver‐assistance systems.

A Gated-Recurrent-Unit-Based Interacting Multiple Model Method for Small Bird Tracking on Lidar System.

Lidar presents a promising solution for bird surveillance in airport environments. However, the low observation refresh rate of Lidar poses challenges for tracking bird targets. To address this problem, we propose a gated recurrent unit (GRU)-based interacting multiple model (IMM) approach for tracking bird targets at low sampling frequencies. The proposed method constructs various GRU-based motion models to extract different motion patterns and to give different predictions of target trajectory in place of traditional target moving models and uses an interacting multiple model mechanism to dynamically select the most suitable GRU-based motion model for trajectory prediction and tracking. In order to fuse the GRU-based motion model and IMM, the approximation state transfer matrix method is proposed to transform the prediction of GRU-based network into an explicit state transfer model, which enables the calculation of the models' probability. The simulation carried out on an open bird trajectory dataset proves that our method outperforms classical tracking methods at low refresh rates with at least 26% improvement in tracking error. The results show that the proposed method is effective for tracking small bird targets based on Lidar systems, as well as for other low-refresh-rate tracking systems.

Automatic Clustering of Excited-State Trajectories: Application to Photoexcited Dynamics.

We introduce automatic clustering as a computationally efficient tool for classifying and interpreting trajectories from simulations of photo-excited dynamics. Trajectories are treated as time-series data, with the features for clustering selected by variance mapping of normalized data. The L2-norm and dynamic time warping are proposed as suitable similarity measures for calculating the distance matrices, and these are clustered using the unsupervised density-based DBSCAN algorithm. The silhouette coefficient and the number of trajectories classified as noise are used as quality measures for the clustering. The ability of clustering to provide rapid overview of large and complex trajectory data sets, and its utility for extracting chemical and physical insight, is demonstrated on trajectories corresponding to the photochemical ring-opening reaction of 1,3-cyclohexadiene, noting that the clustering can be used to generate reduced dimensionality representations in an unbiased manner.

An integrated approach of machine learning and Bayesian spatial Poisson model for large-scale real-time traffic conflict prediction

The use of traffic conflicts in road safety evaluation is gaining considerable popularity as it plays a vital role in developing a proactive safety management strategy and allowing for real-time safety analysis. This study proposes an integrated approach that combines a machine learning (ML) algorithm and a Bayesian spatial Poisson (BSP) model to conduct large-scale real-time traffic conflict prediction by considering traffic states as the explanatory variables. Traffic conflicts are measured by two indicators, the Time to Collision (TTC) and the Post-Encroachment Time (PET). Based on both TTC and PET, traffic conflict severity is classified into five categories. For each conflict severity category, a binary variable (conflict occurrence) and a count variable (conflict frequency) are developed, respectively. In addition to conflict variables, traffic state parameters are extracted from a large-scale high-resolution trajectory dataset. The traffic parameters include volume, density, speed, and the corresponding space-based and space–time-based measures within a 30-second interval. Eight ML-based classifiers are applied to predict conflict occurrence, and the best classifier is selected. A binary logistic regression is developed to explore the potential linkages between traffic states and conflict occurrence. As well, a resampling technique Borderline-SMOTE is used to mitigate the sparsity caused by the predefined short interval. The BSP model is utilized to predict the specific number of conflicts. Further, the BSP model can also explain the relationship between traffic states and conflict frequency, and thus the significant influencing traffic states are identified. The results show that random forest outperforms the other MLs in terms of conflict occurrence prediction accuracy. Further, the proposed integrated approach achieves a high performance of conflict frequency prediction with RMSE values of 0.1384 ∼ 0.1699, MAPE values of 9.25% ∼ 36.99%, and MAE values of 0.0087 ∼ 0.6398. The finding emphasizes the need for separately predicting the occurrence and frequency of conflicts with different severities.

I-24 MOTION: An instrument for freeway traffic science

The Interstate-24 MObility Technology Interstate Observation Network (I-24 MOTION) is a new instrument for traffic science located near Nashville, Tennessee. I-24 MOTION consists of 276 pole-mounted high-resolution traffic cameras that provide seamless coverage of approximately 4.2 miles I-24, a 4-5 lane (each direction) freeway with frequently observed congestion. The cameras are connected via fiber optic network to a compute facility where vehicle trajectories are extracted from the video imagery using computer vision techniques. Approximately 230 million vehicle miles of travel occur within I-24 MOTION annually. The main output of the instrument is vehicle trajectory datasets that contain the position of each vehicle on the freeway, as well as other supplementary information such as vehicle dimensions and class. This article describes the design and creation of the instrument, and provides the first publicly available datasets generated from the instrument. The datasets published with this article contains at least 4 hours of vehicle trajectory data for each of 10 days. As the system continues to mature, all trajectory data will be made publicly available at i24motion.org.

Treating Noise and Anomalies in Vehicle Trajectories From an Experiment With a Swarm of Drones

Unmanned aerial systems, known as “drones”, are relatively new in collecting traffic data. Data from drone videography can have potential applications for traffic research. Drones can record the vehicles from their aerial point-of-view and provide their naturalistic driving behavior. Processing raw data from drones to remove noise and anomalies is crucial to ensure that the data are fit for subsequent applications, e.g., the development of traffic flow or crash risk models. This study uses a part of the pNEUMA dataset, a large dataset with almost half a million trajectories captured by a swarm of drones over Athens, Greece. This novel dataset offers an opportunity to analyze the data attributes and treat the noise and outliers in the data. We use a combination of smoothing filters and Extreme Gradient Boosting with adaptive regularization to process the speed and acceleration profiles of the vehicle trajectories in the dataset. Our approach can help prospective data users treat this or similar trajectory datasets alternatively to applying manual thresholds and assist in accelerating research in microscopic traffic analysis.

CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling

Large-scale simulations with complex electron interactions remain one of the greatest challenges for atomistic modelling. Although classical force fields often fail to describe the coupling between electronic states and ionic rearrangements, the more accurate ab initio molecular dynamics suffers from computational complexity that prevents long-time and large-scale simulations, which are essential to study technologically relevant phenomena. Here we present the Crystal Hamiltonian Graph Neural Network (CHGNet), a graph neural network-based machine-learning interatomic potential (MLIP) that models the universal potential energy surface. CHGNet is pretrained on the energies, forces, stresses and magnetic moments from the Materials Project Trajectory Dataset, which consists of over 10 years of density functional theory calculations of more than 1.5 million inorganic structures. The explicit inclusion of magnetic moments enables CHGNet to learn and accurately represent the orbital occupancy of electrons, enhancing its capability to describe both atomic and electronic degrees of freedom. We demonstrate several applications of CHGNet in solid-state materials, including charge-informed molecular dynamics in LixMnO2, the finite temperature phase diagram for LixFePO4 and Li diffusion in garnet conductors. We highlight the significance of charge information for capturing appropriate chemistry and provide insights into ionic systems with additional electronic degrees of freedom that cannot be observed by previous MLIPs.

Servo-Level Human–Machine Shared Control Flexible Strategy Based on Driving Ability, Status and Regionalized Environmental Risk

The design of interaction strategy between the human driver and Automated Driving System (ADS) brings great challenges to driving automation. A servo-level human-machine shared control flexible strategy (SHSC-FS) is proposed to solve the problem of how to reasonably allocate Driving Control Authority (DCA—the proportional distribution of outputs from the driver and ADS when they complete the same dynamic driving task at the same time together). The strategy can realize DCA allocation by evaluating whether the driver’s long-term ability and status can cope with the real-time environmental risk from different directions around. Through the verification test and the analysis of the natural vehicle trajectory dataset, it is found that the regionalized environmental risk faced by human drivers during driving obeys normal distribution. The real-time driving abilities (which is defined as driving status here) of human drivers also obey normal distribution. In addition, the verification tests show that SHSC-FS proposed tends to give higher DCA to the drivers with strong driving ability when they are facing low environmental risk from regions where they are sensitive and good at coping; When dealing with the high environmental risk from regions where drivers are insensitive and not good at dealing with, DCA given to the drivers with weak driving ability is lower.

Cooperative Task Offloading for Mobile Edge Computing Based on Multi-Agent Deep Reinforcement Learning

Driven by the prevalence of the computation-intensive and delay-intensive mobile applications, Mobile Edge Computing (MEC) is emerging as a promising solution. Traditional task offloading methods usually rely on centralized decision making, which inevitably involves a high computational complexity and a large state space. However, the MEC is a typical distributed system, where the edge servers are geographically separated, and independently perform the computing tasks. This fact inspires us to conceive a distributed cooperative task offloading system, where each edge server makes its own decision on how to allocate local computing resources and how to migrate tasks among the edge servers. To characterize diverse task requirements, we divide the arrival tasks into different priorities according to the tolerance time, which enables to dynamically schedule the local computing resources for reducing the task timeout. In order to coordinate the independent decision makings of geographically separate edge servers, we propose a priority driven cooperative task offloading algorithm based on multi-agent deep reinforcement learning, where the decision making of each edge server not only depends on its own state but also on the shared global information. We further develop a Variational Recurrent Neural Network (VRNN) based global state sharing model which significantly reduces the communication overhead among edge servers. The performance evaluation conducted on a movement trajectories dataset of mobile devices verifies that the proposed algorithm can reduce the task consumption time and improve the edge computing resources utilization.

Using data to develop cost-effective battery-swap networks for electric vehicles

Electric vehicle (EV) development faces the challenges of range anxiety and the high construction and operating costs of charging and battery-swap infrastructure. To address these issues, this paper describes a proposed battery-swap network (BSN) comprising battery-swapping van service points and battery-swap stations to provide local, fast and cost-effective battery-swapping services. A data-driven approach was developed to analyse and model vehicle trajectory data to obtain realistic battery-swap demand distributions. An optimisation model based on location cover was then developed to minimise the cost of a BSN using a greedy algorithm. Real traffic network driving distance was adopted to improve authenticity and accuracy. In a case study, a BSN used a 1 week trajectory data set of taxis in Beijing, China and tested the effects in eight scenarios. The results showed that the mesh size of service points, service radius and coverage rate all had an impact on service quality and cost. It is concluded that a BSN could be cost-effectively deployed by adjusting these parameters to suit the needs of EVs at different development stages. This paper provides a useful reference for decision makers and operators to formulate policies and strategies.

Efficient Large-Scale GPS Trajectory Compression on Spark: A Pipeline-Based Approach

Every day, hundreds of thousands of vehicles, including buses, taxis, and ride-hailing cars, continuously generate GPS positioning records. Simultaneously, the traffic big data platform of urban transportation systems has already collected a large amount of GPS trajectory datasets. These incremental and historical GPS datasets require more and more storage space, placing unprecedented cost pressure on the big data platform. Therefore, it is imperative to efficiently compress these large-scale GPS trajectory datasets, saving storage cost and subsequent computing cost. However, a set of classical trajectory compression algorithms can only be executed in a single-threaded manner and are limited to running in a single-node environment. Therefore, these trajectory compression algorithms are insufficient to compress this incremental data, which often amounts to hundreds of gigabytes, within an acceptable time frame. This paper utilizes Spark, a popular big data processing engine, to parallelize a set of classical trajectory compression algorithms. These algorithms consist of the DP (Douglas–Peucker), the TD-TR (Top-Down Time-Ratio), the SW (Sliding Window), SQUISH (Spatial Quality Simplification Heuristic), and the V-DP (Velocity-Aware Douglas–Peucker). We systematically evaluate these parallelized algorithms on a very large GPS trajectory dataset, which contains 117.5 GB of data produced by 20,000 taxis. The experimental results show that: (1) It takes only 438 s to compress this dataset in a Spark cluster with 14 nodes; (2) These parallelized algorithms can save an average of 26% on storage cost, and up to 40%. In addition, we design and implement a pipeline-based solution that automatically performs preprocessing and compression for continuous GPS trajectories on the Spark platform.

Inferring vehicle spacing in urban traffic from trajectory data

This study presents a new method to infer the average two-dimensional (2D) spacing between interacting vehicles in urban traffic from trajectory data. In this context, 2D spacing reflects the amount of road space consumed by pairs of interacting vehicles, and is related to 2D density at the macroscopic level. Due to complex interaction and conflicts in urban traffic, the inherent assumptions in traditional traffic flow models, such as unidirectional flow and homogeneity, are often violated. Such violation challenges direct measurement of urban vehicle spacing. The proposed method addresses this challenge by focusing on the relative movement between interacting vehicles and aggregating the accumulated presence of vehicles in similar scenarios. We apply the method to a large-scale urban trajectory dataset called pNEUMA, and validate the consistency of the method through bootstrapping. By applying the method we obtain a new empirical relation between the average 2D spacing and the relative speeds between interacting vehicles. There are similarities between this empirical relation with the classical Fundamental Diagram of traffic flow in terms of shape and interpretation, and so we term it the “interaction Fundamental Diagram” (iFD). However, there are also key differences. The iFD does not represent steady-state (homogeneous and stationary) longitudinal behaviour; it describes the average amount of road space needed for vehicle interactions at different relative speeds. We believe these iFD relations contribute to understanding vehicle interaction in urban traffic, and can offer new insights for designing safer and more efficient urban intersections.