In mixed-traffic of human-driven vehicles (HDVs) and Connected and Automated Vehicles (CAVs), the trajectory planning of CAV is a critical issue. This study proposes a Hierarchical Dynamic Trajectory Planning Framework for CAVs. The framework consists of two layers. At the path planning layer, an ID* Lite algorithm is proposed with four enhancement modules: a risk-aware target update module, a spatiotemporal search module, a look-ahead safety inspection module, and a path smoothing module. At the trajectory tracking layer, a Model Predictive Control (MPC) algorithm refines the planned path by imposing vehicle dynamics and input constraints. Simulation experiments were conducted on a MATLAB-based upstream lane scenario of a signalized intersection. Results show that the proposed ID* Lite significantly outperforms state-of-the-art baselines. Specifically, the planning success rate, average path length, node expansion, and computation time are improved by 50%, 12.16%, 23.29%, and 20%, respectively. The hierarchical dynamic safe trajectory planning framework enhances motion control, reducing steering angle changes by 12.06%, endpoint error by 52.42%, and trajectory length by 0.07%. The results highlight the proposed framework’s effectiveness in achieving safe, real-time, smooth, and stable trajectory planning in mixed-traffic environments, suggesting its potential for autonomous driving applications.

Vehicle trajectory data serves as a crucial foundation for vehicle-level control in mixed traffic environments incorporating connected and automated vehicles (CAVs). However, due to the limited sensing range of CAVs, perception-blind areas inevitably emerge, resulting in partially unobservable vehicle trajectories. To reconstruct unobservable trajectories between CAVs, this paper proposes a novel physics-informed neural network for platoon trajectory reconstruction under partially observed conditions, termed PINN-PTR. PINN-PTR integrates a physics-uninformed neural network (PUNN) with physics-based computational graphs derived from car-following models, in which each vehicle is assigned a distinct parameter set to capture heterogeneity in driving behavior. The proposed model leverages the advantages of both physics-based models, which are data-efficient and interpretable, and deep learning-based models, which are generalizable. Moreover, a bidirectional Newell computational graph (BNCG) is introduced to comprehensively capture the bidirectional temporal and spatial correlations of the ego vehicle with both its leading and following vehicles, based on the propagation of kinematic waves. Two types of PINN-PTR models are studied: one designed solely for trajectory reconstruction and another that jointly reconstructs trajectories and calibrates parameters within physics-based computational graphs. Experimental results on both real-world and synthetic datasets demonstrate that the proposed approach outperforms baseline methods in terms of trajectory reconstruction accuracy and car-following model calibration. Ablation studies further confirm the contributions of the BNCG and heterogeneous modeling to improving reconstruction performance. Finally, vehicle emission estimation is applied as a practical case study to evaluate the effectiveness of the proposed model in real-world scenarios.

The effective control of connected and automated vehicles (CAVs) in mixed platoons offers fresh opportunities to optimize the emerging mixed traffic flow environment in the future. The goals of existing studies are modeling accuracy and control effectiveness. As a continued work of such a pursuit, this article develops a data-driven robust predictive control framework for the mixed platoon composed of CAVs and human driven vehicles (HDVs). A deep variational Koopman network (DVKoN) was proposed to learn the HDVs’ uncertainty-aware dynamics driving behavior based on the HighD dataset. A DVKoN-based robust predictive control framework (DVKoRPC) was designed for optimizing the mixed vehicle platoon. The DVKoRPC has two components, i.e., a nominal system and an error system. The nominal system, which integrates multiple DVKoNs based on the platoon formation, is used as the state predictive model for the mixed vehicle platoon. The error system is used to compensate for deviations and uncertainties within platoon operation. Moreover, the asymptotic stability of the mixed vehicle platoon with the DVKoRPC was proved using Lyapunov theory. The experimental was conducted to verify the proposed DVKoN and DVKoRPC. The results show that DVKoN can accurately predict the uncertain car-following behavior of HDVs in the mixed vehicle platoon. The proposed DVKoRPC can effectively alleviate traffic oscillations, improve traffic efficiency, and reduce fuel consumption.

With the advancement of self-driving vehicles, autonomous driving systems depend on multimodal data to achieve a dynamic perception of the surrounding environment. Synthetic aperture radar (SAR) techniques can enhance azimuth resolution by utilizing the relative motion between the vehicle and targets, requiring a precise trajectory of the vehicle, normally without the assistance of automotive-grade navigation systems. In this case, data-driven autofocus-based algorithms are typically used to implement compensation for non-systematic motion errors. Despite demonstrating robust autofocus capabilities in numerous scenarios, their potential for application in automotive scenarios still needs to be exploited. This paper aims to provide a comprehensive automotive SAR imaging with autofocus workflow and to analyze the performance of autofocus algorithms based on phase gradient autofocus (PGA) in typical automotive scenarios. We rigorously derive the Omega-$\boldsymbol {K}$algorithm based on the system-grade waveform of frequency modulated continuous wave (FMCW) signals. Based on the analysis of motion error and phase error characteristics, a sub-aperture contrast-based weighted PGA (SAC-WPGA) method, a contrast-based selection strategy (CBSS), and a contrast-based WPGA kernel are proposed to improve the robustness of autofocus for automotive scenarios. In addition, we theoretically discuss the impact of the selection strategy, the PGA kernel, and the selection threshold in detail, highlighting the validity of the proposed method. Finally, we showcase the superiority of the proposed technique by employing experimental data in two typical automotive scenarios, i.e., a simple scenario with isolated dominant points and a complex scenario with strong clutter.

This paper proposes a novel normalizing flow learning based method IIAG-CoFlow for low-light image enhancement (LLIE), which consists of an inter-and intra-channel attention Transformer based conditional generator (IIAG) and a complete flow (CoFlow). On the one hand, IIAG is designed as a U-shape network, whose down-sampling and up-sampling layers are constructed by IIZAT (i.e., inter-and intra-channel and zero-map attention Transformer) and IIAT (i.e., inter-and intra-channel attention Transformer) respectively. IIAT is designed to calculate inter-channel attention and intra-channel attention independently. Based on IIAT, IIZAT is designed to perform parallel fusion of zero-map attention and intra-channel attention. On the other hand, based on existing normalizing flow, we bring in unconditional affine coupling layer and design 3 invertible linear transformation layers, to develop CoFlow. The height and width axes based cross attention network (HWCAN) is proposed to learn affine/linear transformation parameters for conditional feature-driven layers of CoFlow. Experiments show that IIAG-CoFlow outperforms existing SOTA LLIE methods on several benchmark low-light datasets, and real NTM images. The source codes and pre-trained models are available athttps://github.com/NJUPT-IPR-ChenTS/IIAG-CoFlow.

We observe that the low-light RGB images, as well as night traffic monitoring (NTM) images, contain lots of color pixels with zeros caused by the low-light, which means that the low-light images suffer both information weakness and information loss of zero-element pixels. In this paper, we propose a novel flow-based generative method JTE-CFlow for low-light image enhancement, which consists of a joint-attention transformer based conditional encoder (JTE) and a map-wise cross affine coupling flow (CFlow). Specifically, JTE executes short-range and long-range operations by RRDBs (i.e., residual-in-residual dense blocks) and JATs (i.e., joint-attention transformer blocks) in series connection. JAT achieves weak information amplification and information loss restoration of zero-element pixels by the integration of self-attention and specific-attention with sharing the same value vectors, where the query and key vectors of specific-attention are from the zero-map feature of the low-light image. On the other hand, CFlow develops a map-wise cross affine coupling (MCAC) layer to perform cross learning for the flow feature, and a multiplication coupling network (MCN) to learn the transformation parameters of MCAC. JTE-CFlow learns to map the subtraction of outputs of CFlow and JTE (i.e., the residual code) into a standard normal distribution, and the inverse network of CFlow takes the latent feature of the low-light image as its input to infer the enhanced image. Experiments show that JTE-CFlow outperforms most SOTA methods on 7 mainstream low-light datasets with the same architecture, and can be applied to enhance NTM images. The source code and pre-trained models are available athttps://github.com/NJUPT-IPR-HuYin/JTE-CFlow.

Vehicle trajectory data is essential for analyzing and modeling complex traffic behaviors. Although extraction of vehicle trajectory from aerial video data is not a new problem, obtaining trajectories with heading information across various road types, such as intersections or long road segments, requires further research. In this paper, we propose OpenVTER, a generalized Open-source Vehicle Trajectory Extraction framework based on Rotated bounding boxes (RBBs). This framework includes several key components: video stabilization, image division, vehicle detection, vehicle tracking, and data post-processing. Specifically, the rotated vehicle detection model, named YOLOX-R, is applied to detect the small and rotated vehicles using RBBs that provide vehicle heading information. A base-frame video stabilization method is proposed to reduce error accumulation in the transformation matrix and improve the computational efficiency. The rotated vehicle tracking model, named SORT-R, is proposed to enable real-time tracking of RBBs. The performance of YOLOX-R is evaluated on two datasets, showing that vehicle detection challenges are well addressed. Ablation experiments were also conducted to analyze the effectiveness of different modules. Subsequently, we evaluate the completeness of the extracted trajectories under various road types and lighting conditions. The extracted trajectories are also compared with the NGSIM dataset, focusing on internal and platoon consistency. These evaluations demonstrate both the effectiveness and practicality of the proposed framework. Additionally, the visualization analyses of different road types demonstrate the advantages of the trajectories extracted by OpenVTER in various road scenarios for traffic research. The code and dataset are available online for non-commercial research purposes.

Detecting spatiotemporal anomalies is imperative for addressing critical societal and engineering challenges, including public safety assurance, environmental hazard identification, epidemic surveillance, and transportation system optimization. Existing methodologies, however, face persistent limitations due to sparse labeled datasets and the inherent complexity of dynamic spatiotemporal systems. In order to bridge this gap, we present unsupervised-semi-supervised stacking (USemiS), a novel framework that synergizes semi-supervised learning with ensemble meta-learning. USemiS introduces three core innovations: 1) unsupervised component learners that extract low-level representations of heterogeneous anomalies, 2) a consensus-based tuning mechanism that dynamically weights robust learners via stability metrics, and 3) spatiotemporal MixUp (ST-MixUp), a tailored augmentation strategy that interpolates anomalies across spatial and temporal dimensions to enhance decision boundaries. By integrating these components, USemiS effectively disentangles latent anomaly patterns while mitigating label scarcity. Evaluated on large-scale traffic anomaly and crowd fall detection datasets, USemiS achieves state-of-the-art performance, outperforming existing methods by 1.3% and 2.1% in AUC under extreme low-label regimes (0.4% and 0.8% labeled data, respectively). These results underscore USemiS's capacity to generalize across diverse spatiotemporal contexts, offering a scalable and robust solution for real-world applications where labeled anomalies are scarce yet critical.

This paper proposes a fast HSI (hue, saturation, intensity) color space and an orthogonal triangular with column pivoting (QRCP) enhancement model to tackle the large size RGB (red, green, blue) night traffic monitoring (NTM) image. Firstly, the fast HSI (FHSI) is proposed to decompose the light and color information of the RGB image, whose hue is defined as the cosine value of the included angle, instead of the included angle in HSI. The saturation of FHSI is defined as the ratio of the projection vector length and the side length of the projection equilateral triangle, and a saturation correction model is further proposed to correct color distortion of the low-light image by adjusting the saturation of FHSI. FHSI is more concise and faster than HSI. Secondly, a novel QRCP enhancement (QRCPE) model is proposed to improve the light of the low-light image by enhancing the intensity of FHSI, which first strengthens diagonal elements of QRCP, and followed by controlling the normalization of strengthened diagonal elements of QRCP. Finally, the FHSI-QRCPE based RGB image can be obtained by transforming the processed FHSI to RGB. The experimental results on NTM, SICE, ExDark, and BDD 100K databases, indicate that the proposed FHSI-QRCPE is fast and efficient to tackle low-light image enhancement.

The real traffic monitoring driver face (TMDF) images are with complex multiple degradations, which decline face recognition accuracy in real intelligent transportation systems (ITS). This paper is the first to propose joint image-to-image (I2I) translation to enhance TMDF images of ITS. First, as TMDF images are without corresponding clear ones, identity preserving is critical for TMDF images under unpaired I2I translation. This paper proposes a fast diagonal symmetry pattern (FDSP) to preserve identity structure under unpaired I2I translation. Second, FDSP is introduced into CycleGAN to form FDSP-CG, which aims to learn the degradation mapping (i.e., FDSP-CG-d) from the clarity domain to the degradation domain. FDSP-CG-d can generate massive degradation/clarity image pairs for paired I2I translation training. Third, this paper proposes the dual residual block (DRB) to strengthen Pix2pix for rich face detail features learning (i.e., DRB-P2P), which learns the enhancement mapping from the degradation image to its clear version under paired I2I translation. Finally, the experiments on TMDF (i.e., the brevity name of the face database collected from real ITS) and Chinese famous face (CFF) databases, as well as CelebA and MegaFace databases, indicate that the proposed method can efficiently enhance TMDF images whose degradation variations are learned by FDSP-CG.

This paper proposes HSV (hue, saturation, value) with three sectors (HSV-3S) and two-dimensional gradient descent algorithm (2D-GDA) for high-saturation low-light image enhancement in night traffic monitoring (NTM). The saturation of HSV-3S is defined as the ratio of the projection vector length and twice length of the sector start vector, which results in that the saturation of HSV-3S is smaller than that of HSV, and a saturation weakening model is proposed to further decrease the saturation of HSV-3S. The hue of HSV-3S is defined as the cosine value of the included angle between the projection vector and the sector start vector in each of three sectors. HSV-3S is more concise and faster than HSV. Then, 2D-GDA extends the gradient descent algorithm to 2D image domain. 2D-GDA employs the iteration matrix with variable step values (i.e., the step values of the dark regions are less than those of the bright regions), which can improve the pixel distribution of the 2D-GDA enhanced image. Finally, the HSV-3S+2D- GDA based RGB image can be obtained by performing 2D-GDA on the value of HSV-3S with transforming the processed HSV-3S to RGB. The experimental results on NTM (i.e., the brevity name of the database collected from real ITS), LOL, ExDark and SICE databases, indicate that HSV-3S+2D-GDA is fast and efficient for high-saturation low-light image enhancement.

To estimate the accurate fundamental relationship in traffic flow, this paper proposes a novel framework that extends classical fundamental diagram (FD) models to incorporate more dimensions of traffic state variables and allow for the impact of the supply-side factors of roads. The proposed framework is suitable for real-time traffic management, especially in urban areas, due to its reliance on minimal assumptions, its flexibility in adapting to various data sources, and its scalability to higher-dimensional data. The Gaussian process (GP) model is adopted as the base model for learning the optimal mapping from these input features to traffic volume. To enhance the GP model, an in-depth analysis of the properties of its kernel and likelihood function is provided. To cope with the hyperparameter optimisation of the GP, a modified Newton method for GP-based traffic flow model is also designed, which can jump over regions with small gradients. Experiments based on simulation data demonstrate the ability of the proposed framework to capture complex relationships between traffic state variables and supply-side factors, and show its value for estimating dynamic road capacity.

The study aims to evaluate the performance of the transfer learning algorithm to enhance the transferability of a deep reinforcement learning-based variable speed limits (VSL) control. The Double Deep Q Network (DDQN)-based VSL control strategy is proposed for reducing total time spent (TTS) on freeways. A real merging bottleneck is developed in the simulation and considered for the VSL control as the source scenario. Three types of target scenarios are considered, including the overspeed scenarios, adverse weather scenarios, and diverse capacity drop scenarios. A stable testing demand and a fluctuating testing demand are adopted to evaluate the effects of VSL control. The results show that by updating the neural networks, the transfer learning in the DDQN-based VSL control agent successfully transfers knowledge learned in the source scenario to other target scenarios. With the transfer learning, the entire training process is shortened by 32.3% to 69.8%, while keeping a similar maximum reward level, as compared to the VSL control with full learning from scratch. With the transferred DDQN-based VSL strategy, the TTS is reduced by 26.02% to 67.37% with the stable testing demand and 21.31% to 69.98% with the fluctuating testing demand in various scenarios, respectively. The results also show that when the task similarity between the source scenario and target scenario is relatively low, the transfer learning could lead to local optimum and may not achieve the global optimal control effects.

This paper models the driver face recognition problem under the intelligent traffic monitoring systems as severe illumination variation face recognition with single sample problem. Firstly, in the point of view of numerical value sign, the current illumination invariant unit is derived from the subtraction of two pixels in the face local region, which may be positive or negative, we propose a generalized illumination robust (GIR) model based on positive and negative illumination invariant units to tackle severe illumination variations. Then, the GIR model can be used to generate several GIR images based on the local edge-region or the local block-region, which results in the edge-region based GIR (EGIR) image or the block-region based GIR (BGIR) image. For single GIR image based classification, the GIR image utilizes the saturation function and the nearest neighbor classifier, which can develop EGIR-face and BGIR-face. For multi GIR images based classification, the GIR images employ the extended sparse representation classification (ESRC) as the classifier that can form the EGIR image based classification (GIRC) and the BGIR image based classification (BGIRC). Further, the GIR model is integrated with the pre-trained deep learning (PDL) model to construct the GIR-PDL model. Finally, the performances of the proposed methods are verified on the Extended Yale B, CMU PIE, AR, self-built Driver and VGGFace2 face databases. The experimental results indicate that the proposed methods are efficient to tackle severe illumination variations.

In this paper, we present a novel high-frequency facial feature and a high-frequency based sparse representation classification to tackle single sample face recognition (SSFR) under varying illumination. Firstly, we propose the assumption that QRCP bases can represent intrinsic face surface features with different frequencies, and their corresponding energy coefficients describe illumination intensities. Based on this assumption, we take QRCP bases with corresponding weighting coefficients (i.e. the major components of energy coefficients) to develop the high-frequency facial feature of the face image, which is named as QRCP-face. The normalized QRCP-face (NQRCPface) is constructed to further constraint illumination effects by normalizing the weighting coefficients of QRCP-face. Moreover, we propose the adaptive QRCP-face (AQRCP-face) that assigns a special parameter to NQRCP-face via the illumination level estimated by the weighting coefficients. Secondly, we consider that the differences of pixel images cannot model the intraclass variations of generic faces with illumination variations, and the specific identification information of the generic face is redundant for the current SSFR with generic learning. To tackle above two issues, we develop a general high-frequency based sparse representation (GHSP) model. Two practical approaches separated high-frequency based sparse representation (SHSP) and unified high-frequency based sparse representation (UHSP) are developed. Finally, the performances of the proposed methods are verified on the Extended Yale B, CMU PIE, AR, LFW and our self-built Driver face databases. The experimental results indicate that the proposed methods outperform previous approaches for SSFR under varying illumination.

In this paper, a deep learning (DL) framework was proposed to predict the taxi-passenger demand while the spatial, the temporal, and external dependencies were considered simultaneously. The proposed DL framework combined a modified density-based spatial clustering algorithm with noise (DBSCAN) and a conditional generative adversarial network (CGAN) model. More specifically, the modified DBSCAN model was applied to produce a number of sub-networks considering the spatial correlation of taxi pick-up events in the road network. And the CGAN model, fed with the historical taxi passenger demand and other conditional information, was capable to predict the taxi-passenger demands. The proposed CGAN model was made up with two long short-term memory (LSTM) neural networks, which are termed as the generative network G and the discriminative network D, respectively. Adversarial training process was conducted to the two LSTMs. In the numerical experiment, different model layouts were compared. It was found that different network layouts provided reasonable accuracy. With limited training data, more LSTM layers in the generator network resulted in not only higher accuracy, but also more difficulties in training. Comparisons were also conducted between the proposed prediction model and four typical approaches, including the moving average method, the autoregressive integrated moving method, the neural network model, and the LSTM neural network model. The comparison results showed that the proposed model outperformed all the other methods. And the repeated experiment indicated that the proposed CGAN model provided significant better predictions than the LSTM model did. Future research was recommended to include more datasets for testing the model and more information for improving predictive performance.

The primary objective of this paper was to incorporate the reinforcement learning technique in variable speed limit (VSL) control strategies to reduce system travel time at freeway bottlenecks. A Q-learning (QL)-based VSL control strategy was proposed. The controller included two components: a QL-based offline agent and an online VSL controller. The VSL controller was trained to learn the optimal speed limits for various traffic states to achieve a long-term goal of system optimization. The control effects of the VSL were evaluated using a modified cell transmission model for a freeway recurrent bottleneck. A new parameter was introduced in the cell transmission model to account for the overspeed of drivers in unsaturated traffic conditions. Two scenarios that considered both stable and fluctuating traffic demands were evaluated. The effects of the proposed strategy were compared with those of the feedback-based VSL strategy. The results showed that the proposed QL-based VSL strategy outperformed the feedback-based VSL strategy. More specifically, the proposed VSL control strategy reduced the system travel time by 49.34% in the stable demand scenario and 21.84% in the fluctuating demand scenario.

The primary objective of this paper was to develop a control strategy of variable speed limits (VSLs) to reduce rear-end collision risks near freeway recurrent bottlenecks. The risks of rear-end collisions were estimated using a crash risk prediction model that is specifically developed for rear-end collisions in freeway bottleneck areas. The effects of the VSL control strategy were evaluated using a cell transmission model. Several control factors were tested, including the start-up threshold of the collision likelihood, the target speed limit, the speed change rate, and the speed difference between adjacent links. A genetic algorithm was used to optimize critical control factors. For the high demand scenario, the proposed control strategy used 25% of the maximum collision likelihood for the start-up threshold, 35 mi/h for the target speed limit, 10 mi/h per 30 s for the speed change rate, and 10 mi/h for the speed difference between different links. For the moderate demand scenario, the strategy used 20% of the maximum collision likelihood for the start-up threshold, 40 mi/h for the target speed limit, 15 mi/h per 30 s for the speed change rate, and 10 mi/h for the speed difference between different links. The results of comparative analyses suggested that the proposed control strategy outperformed other strategies in reducing the rear-end collision risks near freeway recurrent bottlenecks. With the proposed control strategy, the VSL control reduced the rear-end crash potential by 69.84% for the high demand scenario and by 81.81% for the moderate demand scenario.

This paper aimed at evaluating the application of the genetic programming (GP) model for real-time crash prediction on freeways. Traffic, weather, and crash data used in this paper were obtained from the I-880N freeway in California, United States. The random forest (RF) technique was conducted to select the variables that affect crash risk under uncongested and congested traffic conditions. The GP model was developed for each traffic state based on the candidate variables that were selected by the RF technique. The traffic flow characteristics that contribute to crash risk were found to be quite different between congested and uncongested traffic conditions. This paper applied the receiver operating characteristic (ROC) curve to evaluate the prediction performance of the developed GP model for each traffic state. The validation results showed that the prediction performance of the GP models were satisfactory. The binary logit model was also developed for each traffic state using the same training data set. The authors compared the ROC curve of the GP model and the binary logit model for each traffic state. The GP model produced better prediction performance than did the binary logit model for each traffic state. The GP model was found to increase the crash prediction accuracy under uncongested traffic conditions by an average of 8.2% and to increase the crash prediction accuracy under congested traffic conditions by an average of 4.9%.