Ensuring safety in both autonomous driving and advanced driver-assistance systems (ADAS) depends critically on the efficient deployment of traffic sign recognition technology. While current methods show effectiveness, they often compromise between speed and accuracy. To address this issue, we present a novel real-time and efficient road sign detection network, YOLO-TS. This network significantly improves performance by optimizing the receptive fields of multi-scale feature maps to align more closely with the size distribution of traffic signs in various datasets. Moreover, our innovative feature-fusion strategy, leveraging the flexibility of Anchor-Free methods, allows for multi-scale object detection on a high-resolution feature map abundant in contextual information, achieving remarkable enhancements in both accuracy and speed. To mitigate the adverse effects of the grid pattern caused by dilated convolutions on the detection of smaller objects, we have devised a unique module that not only mitigates this grid effect but also widens the receptive field to encompass an extensive range of spatial contextual information, thus boosting the efficiency of information usage. Moreover, to address the scarcity of traffic sign datasets, especially under adverse weather conditions, we introduce two novel datasets: Generated-TT100K-weather and CAWTSSS. Extensive evaluations conducted on challenging public benchmarks—including TT100K, CCTSDB2021, and GTSDB—as well as on our proposed datasets, demonstrate that YOLO-TS surpasses current state-of-the-art methods in both accuracy and inference speed. The code, datasets and weights are available athttps://github.com/Heqiang-Huang/YOLO-TS

The cut-in maneuver is a common high-interaction behavior between vehicles, where improper execution may lead to driving risks and is closely associated with the vehicle’s inherent driving style. Existing driving risk prediction studies lack targeted modeling for this typical maneuver, while current driving style modeling approaches often fail to capture stable and inherent behavioral traits. This study proposes a classification method for inherent driving styles and develops a proactive prediction framework for cut-in risk, which is validated and analyzed using wide-area trajectory data. The results indicate that: (a) the indicator system constructed based on car-following, lane-changing, and interaction characteristics effectively captures inherent driving traits, and the recognition model performs well when the number of style clusters is set to three; (b) the cut-in risk prediction model developed using LightGBM achieves optimal predictive performance, with a fixed observation window and a 2-second lead time offering the most practical feasibility for proactive warning applications; (c) incorporating inherent driving style into the model reduces the prediction error by 3.9% and supports more targeted decision-making. The proposed framework enables proactive recognition of cut-in risks from surrounding vehicles by identifying and sharing inherent driving style information in a connected environment, thereby supporting ego vehicle intervention and decision-making to actively adjust its behavior and reduce driving risk.

To optimize a vehicle platoon system in terms of car-following behavior, a decentralized model predictive control (MPC) strategy for longitudinal velocity control was established (namely, CF-MPC). Firstly, considering the influence of car-following behavior on vehicle states, a longitudinal velocity control model for platoons of connected and automated vehicles (CAV) was designed. Based on that model, an upper-level MPC controller was built to obtain the desired acceleration of the vehicles. Secondly, a lower-level controller received the desired acceleration signal and converted it into the expected throttle opening/braking pressure, to control acceleration/deceleration. Then, the Lyapunov stability method was used to detect the stability conditions that the model should satisfy. Finally, three simulation procedures—constant speed, acceleration, and deceleration were tested, and the validity of the CF-MPC method was verified from the perspectives of a model strategy and a control strategy. The simulation results show that with the proposed CF-MPC method, CAV platoons quickly completed velocity tracking and maintained a safe distance, thereby improving traffic efficiency, fuel economy, driving safety, and transportation capacity.

Modern complexities associated with an arterial traffic makes existing safety prediction methods insufficient to meet desired standards required by recent developmental needs. This paper proposes an enhanced active safety prediction method based on big-data approach and Stacked AutoEncoder-Gated Recurrent Unit. Firstly, the big-data technology is used to construct a dynamic identification model to recognize real-time operation state and risk state. Secondly, the Stacked AutoEncoder-Gated Recurrent Unit is used to predict a level of safety based on associated recognition results. This paper uses data from working days of Sunset Boulevard, California, from January$1^{\mathrm{st}}$, 2020, to February$28^{\mathrm{th}}$, 2020. The results of analysis show that the accuracy of the proposed dynamic recognition model reaches 98.92%, which is better than existing models such as random forest, K-nearest neighbor, and naïve Bayes models. In addition, it is found that the Stacked AutoEncoder-Gated Recurrent Unit can achieve a prediction accuracy of 95.157% and has significant advantages in terms of efficiency. The proposed methods will provide feasible solutions for actively monitoring safety levels.

Side collisions caused by sudden vehicle cut-ins comprise a significant proportion of traffic accidents. Due to the complex and dynamic nature of traffic environments, the warning algorithms in advanced driving assistant systems (ADAS) often misjudge and misdiagnose risk and omit necessary warnings, because they rely solely on the sensing information of the single vehicle equipped with ADAS and have limited insights from and communication with the surrounding vehicles and traffic environment. To improve the effectiveness of ADAS in cut-in scenarios, this study established a collision warning model in a vehicle-to-vehicle (V2V) communication environment. Firstly, based on the support vector machine-recursive feature elimination (SVM-RFE) lane-change intent-recognition model, the lane-change feasibility and the change rate of the lateral offset, the logical “and” was used to establish a lane-change behavior prediction model, and a trajectory prediction model was established based on the long short-term memory (LSTM). Then, based on the proposed comprehensive prediction model for lane-change behavior, the driving trajectory prediction model, and the oriented bounding box (OBB) detection algorithm, a collision warning model was established for a V2V environment. Finally, based on a driving simulation platform and a real-world vehicle test, a cut-in experiment in a V2V environment was designed and implemented. By comparing the warning confusion matrix and warning time, it was found that the proposed cut-in collision warning model is superior to the traditional collision warning model. The results of this study can provide new modeling ideas and a theoretical basis for ADAS to further optimize for a cut-in scenario.

Vehicle behavior learning can be used in video surveillance systems to identify normal and abnormal vehicle motion patterns for the management of traffic operations, public services, and law enforcement. The purpose of this paper is to develop a novel adaptive sparse reconstruction method for vehicle behavior learning based on video surveillance systems. First, the ℓ0 minimization problem of sparse reconstruction is relaxed to the ℓp minimization problem (0 <; p <; 1). A hybrid algorithm orthogonal matching pursuit-quasi-Newton is proposed to effectively find the sparse solutions. Then, a sparse reconstruction and similarity-based trajectory classifier is developed to learn vehicle behavior based on the sparse solutions and the trajectory similarity. In order to validate the performance and the effectiveness of the proposed method, four datasets, including CROSS, i-LIDA, Stop Sign, and I5 are used in the experiments. The results show that the classification and the anomaly detection accuracies of the proposed method are superior to the representative methods, including the Naïve Bayes classifier, k nearest neighbor, support vector machine, and traditional sparse reconstruction-based trajectory learning methods.