论文导读 | IEEE T-ITS 2024年第25期核心论文导读（三）

点击蓝字 关注我们

核心论文

41. CAV-Enabled Active Resolving of Temporary Mainline Congestion Caused by Gap Creation for On-Ramp Merging Vehicles

通过自动驾驶车辆解决由间隙创建导致的主线临时拥堵

作者：Y. Zhou, J. Chen, E. Chung, and K. Ozbay

摘要：This article proposes a simple and novel method to actively dissipate temporary congestion caused by freeway mainline vehicles’ maneuvers for creating gaps for on-ramp merging vehicles. Under the framework of the kinematic wave theory with a triangular fundamental diagram, the authors first show that when the prevailing mainline traffic is under-critical, the total delay of mainline vehicles due to the gap creation does not depend on the choice of the speed by which the gap is created, which is consistent with previous observations of Aimsun simulations. Then, exploiting the connection between Newell’s simplified car-following model and the kinematic wave theory with a triangular fundamental diagram, they pro- pose a cooperative longitudinal movement strategy of CAVs to actively resolve the congestion. Simulations are conducted to verify the effectiveness of the proposed method.

关键词：Connected automated vehicles, on-ramp merging, kinematic wave model, Newell’s simplified car-following model, triangular fundamental diagram, congestion dissipation.

42. Promoting Collaborative Dispatching in the Ride- Sourcing Market With a Third-Party Integrator

在第三方整合器的帮助下促进网约车市场中的协同调度

作者：Y. Wang, J. Wu, H. Sun, Y. Lv, and J. Zhang

摘要：A novel two-stage dispatching framework using the multi- graph hierarchical multi-head attention-deep deterministic pol- icy gradient (MGHMHA-DDPG) algorithm is introduced for ride-sourcing services. This approach, addressing competition- induced inefficiencies in the market, employs a partially observable Markov decision process to optimize driver- passenger pairings. Validated with Beijing data, it outperforms benchmarks in market revenues and response rates, enhancing overall efficiency in urban transportation.

关键词：Ride-sourcing, order-dispatching, third-party integrator, collaborative dispatching, reinforcement learning.

43. Joint Optimization of Crack Segmentation With an Adaptive Dynamic Threshold Module

基于自适应动态阈值模块的裂缝分割联合优化

作者：Q. Lei, J. Zhong, and C. Wang

摘要：In this article, the authors redefine crack segmentation as a joint optimization problem, focusing on both the segmentation model’s objective function and the binarization process. They introduce the adaptive dynamic thresholding module (ADTM), which leverages spatial features in the segmentation network for optimal thresholding of each crack image. ADTM, a plug- and-play component, requires minimal additional memory for deployment and significantly enhances inference accuracy. Experiments on four diverse datasets confirm ADTM’s effectiveness in improving crack segmentation performance, indicating its potential applicability in other binary classification image segmentation tasks.

关键词：Crack semantic segmentation, crack instance segmentation, adaptive dynamical threshold, joint optimization.

44. Generalizable Journey Mode Detection Using Unsupervised Representation Learning

使用无监督表示学习的通用出行模式检测

作者：S. Bandyopadhyay, A. Datta, R. K. Ramakrishnan, and A. Pal

摘要：A generalizable journey mode identification method is pro- posed exploiting unsupervised representation learning, embed- ding the commonalities and diversities across various users’ different modes of journeys. A multistage unsupervised learning mechanism is exploited to form clusters of the different modes of transport on the learned latent representation with a choice of best distance measure. A limited amount of labeled data is used to map the formed clusters to the different transport modes during inference which is performed in a user- specific manner where the user can be of the same or different city. The mobile-phone-based accelerometer sensor instead of the GPS sensor is used, without any placement constraint of the device. Extensive experimental analysis is performed using publicly available transportation datasets along with in-house data collected from different Indian cities.

关键词：Representation learning, transport mode detection, distance measure, domain generalization.

45. Autonomous Driving Decision Algorithm for Complex Multi-Vehicle Interactions: An Efficient Approach Based on Global Sorting and Local Gaming

面对复杂多车交互的自动驾驶决策算法：基于全局排序和局部博弈的高效方法

作者：D. Li, J. Zhang, and G. Liu

摘要：Aiming to solve the complex multi-vehicle interaction problem at unsignalized intersections, a global sorting and local gaming decision algorithm is proposed to mimic human expert knowledge in complex driving interactions. The global sorting, working as a guide to the local game solution, helps greatly reduce the computational complexities of the multi-player game. Both simulations and human-in-the-loop experiments of corner cases show that the algorithm can comprehensively consider both the overall traffic efficiency and the interaction between vehicles. Compared to a traditional approach that decomposes a multi-vehicle game into multiple two-vehicle games, the proposed algorithm can improve both safety and traffic efficiency in intensively interactive driving scenarios.

关键词：Autonomous driving, decision-making, driving safety, traffic efficiency, game theory, interactive driving, unsignalized intersection.

46. Video-Based Driver Drowsiness Detection With Optimised Utilization of Key Facial Features

基于视频的驾驶员疲劳检测，通过优化利用关键面部特征

作者：L. Yang, H. Yang, H. Wei, Z. Hu, and C. Lv

摘要：Driver drowsiness detection is of great significance in improving driving safety and has been widely studied in recent years. However, some existing methods have not fully utilized the drowsiness-related information, and some methods are susceptible to interference from the redundant information of input data. To address these issues, a video-based driver drowsiness detection method according to the key facial features including facial landmarks and local facial areas (VBFLLFA) is proposed in this article. In order to fully utilize the key facial features related to drowsiness and exclude the interference of redundant information, the head movement information is obtained through facial landmark analysis, and the movement information of eyes and mouth is acquired from the local facial areas. The spatial filtering based on the common spatial pattern (CSP) algorithm is introduced to improve the discrimination of different classes of samples. To adequately extract the temporal and spatial features, a two-branch multi-head attention (TB-MHA) module is designed in this article. Furthermore, the center loss with center vector distance penalty is introduced to further improve the discrimination of different classes of samples in the feature space. In addition to two public datasets, the authors specifically create a novel video-based driver drowsiness detection (VBDDD) dataset to evaluate the effectiveness of the method. The experimental results verify that the method can achieve very excellent performance in driver drowsiness detection tasks.

关键词：Driver drowsiness detection, multi-head attention, facial landmarks, local facial areas.

47. E²CoPre: Energy Efficient and Cooperative Collision Avoidance for UAV Swarms With Trajectory Prediction

E2CoPre：具有轨迹预测的能效与合作碰撞避免，用于无人机群作者：S. Huang, H. Zhang, and Z. Huang

摘要：This article presents a novel solution to address the challenges in achieving energy efficiency and cooperation for collision avoidance in UAV swarms. The proposed method enhances environmental awareness and fosters implicit coordination among UAVs, ensuring collision avoidance through cooperative and energy-efficient means. Furthermore, the incorporation of trajectory prediction enables proactive actions to better resolve the collisions between UAVs. Extensive experiments are conducted to evaluate the effectiveness and efficiency of the method in various situations.

关键词：UAV, swarm, collision avoidance, PSO, APF.

48. Automatic Background Filtering for Cooperative Perception Using Roadside LiDAR

使用路边激光雷达的协同感知自动背景过滤

作者：J. Liu, J. Zhao, J. Guo, C. Zou, X. Yin, X. Cheng, and F. Khan

摘要：This article proposes an automatic background filtering method with innovative frame selection and background matrix extraction modules to improve accuracy and real-time performance in vehicle-road cooperative perception. The method first utilizes a space division approach to divide roadside LiDAR point clouds, to reduce the impact of LiDAR vibrations. Then, the terminal-edge-cloud framework is introduced to balance delay-constrained and computation-intensive tasks and improve real-time performance. Finally, a variance-based frame selection strategy with a sliding window and a back- ground matrix extraction method is proposed to reduce the impact of moving foreground objects and increase the accuracy of background filtering. The experimental results demonstrate that the proposed method has a lower error rate and higher integrity rate, with frame filtering processing time within 10 ms and almost no network delay.

关键词：Automatic background filtering, frame selection, background matrix extraction, space division.

49. FHSI and QRCPE-Based Low-Light Enhancement With Application to Night Traffic Monitoring Images

基于FHSI和QRCPE的低光增强技术，应用于夜间交通监控图像

作者：C. Hu, W. Yi, K. Hu, Y. Guo, X. Jing, and P. Liu

摘要：This article proposes a fast hue, saturation, and intensity (HSI) color space and an orthogonal triangular with column pivoting (QRCP) enhancement model to tackle the large-size red, green, and blue (RGB) night traffic monitoring (NTM) image. First, 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. Second, 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.

关键词：Low-light image enhancement, night traffic monitoring image, fast HSI color space, orthogonal triangular with column pivoting enhancement.

50. YOLOv7-RDD: A Lightweight Efficient Pavement Dis- tress Detection Model

YOLOv7-RDD：一种轻量高效的路面病害检测模型

作者：Z. Ning, H. Wang, S. Li, and Z. Xu

摘要：An effective detection method was proposed for 12 pavement distress types of urban roads based on low-cost front- view video data. The model achieved the best precision and efficiency with mAP@0.5 reaching 89.5% and FPS reaching 145. The improved spatial feature pyramid (SPPCSPD) structure has been proven to reduce the running time by nearly 54% compared to the spatial pyramid pooling structure in YOLOv7 while losing little precision. Except for linear cracks, the AP values for the other types of distress detection are all greater than 85%. Moreover, the excellent performance on deformation type distresses of this method forms a relative advantage over professional top-view detection.

关键词：Front-view video data, multi-objective detection, pavement distress, pavement maintenance, urban road.

51. TFAC-Net: A Temporal-Frequential Attentional Convolutional Network for Driver Drowsiness Recognition With Single-Channel EEG

TFAC-Net：用于单通道EEG的驾驶员疲劳识别的时频注意卷积网络

作者：P. Gong, P. Wang, Y. Zhou, X. Wen, and D. Zhang

摘要：This article proposes a novel temporal-frequential attentional convolutional neural network to take full advantage of spectral–temporal features for single-channel EEG driver drowsiness recognition. The model considers the temporal- frequential domain features of the single-channel EEG signals to effectively utilize the complementary information. Moreover, critical spectral–temporal regions related to the driver’s mental state can be learned in the temporal-frequential attention module, while task-relevant feature channels can be further filtered and aggregated by the adaptive feature fuse module. Extensive experiments on the public EEG dataset demonstrate the extraordinary performance of the approach compared to other competitive models.

关键词：Driver drowsiness recognition, EEG, convolutional neural networks, attention mechanism.

52. DCOR: Dynamic Channel-Wise Outlier Removal to De-Noise LiDAR Data Corrupted by Snow

DCOR：用于去除被雪污染的激光雷达数据噪声的动态信道异常值去除

作者：S. Zhou, H. Xu, G. Zhang, T. Ma, and Y. Yang

摘要：A neighborhood-based noise removal methodology is pro- posed to eliminate snow noises from LiDAR data. It identifies a point of interest from a specific laser channel as an outlier, if the number of neighboring points in the same channel within a dynamic search radius is fewer than a threshold. Unlike existing methods that filter the entire LiDAR point cloud, the proposed methodology processes LiDAR data channel- by-channel, which helps reduce the data dimensionality and decouple the snow effects along the vertical axis of the 3D point cloud, leading to more effective and efficient outlier detection. Furthermore, by dynamically changing the search radius based on the point-to-sensor distance rather than adopting a fixed search radius, the proposed methodology can account for the reduced point density at far distances caused by the non-uniformity of LiDAR data.

关键词：LiDAR data, adverse weather, snow noise, channel-wise outlier removal, dynamic search radius.

53. Enhancing Perception for Intelligent Vehicles via Electromagnetic Leakage

通过电磁泄漏增强智能车辆的感知

作者：Q. Zhang, F. Zeng, J. Hu, Z. Xiao, J. Fang, K. Lei, and H. Jiang

摘要：Accurate perception is critical for safe autonomous vehicles, but current methods struggle when obstacles block sensors. Collaborative perception presents privacy/trust issues. This study presents EIV, a cost-effective and comprehensive perception system. It utilizes fluctuating memory currents in vehicles’ system-on-chips, which emit electromagnetic radiation (EMR). EIV has specialized antenna arrays to scan EMR signals and identify vehicles via micro-Doppler signatures. A database is constructed and a multi-antenna algorithm estimates target vehicles’ positions, distances, and directions. Experiments show that EIV enhances timeliness, robustness, and accuracy. Key advantages are utilizing untapped EMR signals and avoiding complex sensor fusion or V2X communication. Challenges include handling noise, encryption, and integrating with legacy systems. Overall EIV pioneers an innovative paradigm for robust, precise, and affordable vehicle perception without compromising privacy.

关键词：Autonomous vehicles, vehicle perception, memory EMR.

54. Intersec2vec-TSC: Intersection Representation Learning for Large-Scale Traffic Signal Control

Intersec2vec-TSC：大规模交通信号控制的交叉口表示学习

作者：H. Huang, Z. Hu, Y. Wang, Z. Lu, and X. Wen

摘要：Using car trajectory data and intersection attribute information, an Intersec2vec model based on network representation learning is proposed to achieve accurate representations of intersection nodes. Furthermore, the Intersec2vec-TSC model based on sub-area division is proposed to optimize the common cycle length of the sub-area and the green time for each phase. Experiments are conducted using Chengdu online car- hailing data to prove the effectiveness of the proposed model.

关键词：Intersection network, network representation learning (NRL), attribute information, deep reinforcement learning (DRL), traffic signal control (TSC).

55. Spatiotemporal Pricing and Fleet Management of Autonomous Mobility-on-Demand Networks: A Decom- position and Dynamic Programming Approach With Bounded Optimality Gap

自动驾驶需求网络的空间-时间定价和车队管理：一种具有有界最优性间隙的分解和动态规划方法

作者：Z. Lai and S. Li

摘要：This article studies the spatiotemporal pricing and fleet management of an AMoD system, which includes pricing, relocating, and fleet sizing strategies. A network flow model that accounts for passengers’ elasticity to prices and service quality is formulated to characterize the system dynamics, and the platform’s profit maximization problem is formulated as a non-convex optimization over multiple stages. An algorithm is proposed to compute the candidate solution, and an integrated

relaxation, decomposition, and dynamic programming method is developed to provide a theoretical upper bound on the gap between the candidate solution and the unknown globally optimal solution. The proposed mathematical model and solution approach are validated with case studies for Manhattan.

关键词：Spatiotemporal pricing, fleet management, elastic demand, autonomous mobility-on-demand.

56. Detecting Critical Mismatched Driver Visual Attention During Lane Change: An Embedding Kernel Algorithm

在换道过程中检测驾驶员视觉注意力不匹配：嵌入核算法

作者：J. Xu, C. Qian, S. Han, and F. Guo

摘要：An embedding kernel algorithm is proposed to identify the critical moment at which the driver’s eye-glance pattern differs significantly between safety-critical events and normal driving maneuvers. The proposed method can incorporate the driver’s visual behavior on the timing and location of the driver’s visual behaviors and identify the safety-critical moment during lane change events. A case study is performed on lane change events from the second strategic highway research program naturalistic driving study. The analysis revealed that the duration and number of safety-critical moments could vary depending on factors such as the driver’s age, the time of day, traffic conditions, driving speed, and lane change direction.

关键词：Driver visual behavior, driving safety, embedding-based kernel test, lane change, inattention, distraction.

57. Fooling Decision-Based Black-Box Automotive Vision Perception Systems in Physical World

在物理世界中欺骗基于决策的黑盒汽车视觉感知系统

作者：W. Jia, Z. Lu, R. Yu, L. Li, H. Zhang, Z. Liu, and G. Qu

摘要：This article proposes PRAD, an end-to-end framework that transfers the existing decision-based black-box adversarial attack algorithms (as the backbone of PRAD) targeting the digital domain to the physical world for the first time. Specifically, T is first introduced to simulate the real environment changes, e.g., angle, distance, and illumination. Then and crucially, PRAD bridges the non-differentiable black-box attack and the differentiable T by the L_1 loss function. The physical world experimental results (average 90% in target attacks and nearly 100% in non-target attacks) demonstrate the robustness and generalization of PRAD.

关键词：Vision perception, black-box attacks, physical world attacks, deep neural networks, autonomous vehicles.

58. Multi-Step Regression Network With Attention Fusion for Airport Delay Prediction

带有注意力融合的多步回归网络，用于机场延误预测

作者：Z. Wei, S. Zhu, Z. Lyu, Y. Qiao, X. Yuan, Y. Zhao, and H. Zhang

摘要：This article addresses the critical issue of airport delay prediction by introducing a novel multi-step regression pre- diction method, DA-BILSTM. Recognizing the complexity of factors influencing airport delays, the authors propose an attention fusion network that adaptively learns sequence and condition correlation features, thereby portraying the influence mechanisms of different factors on airport delays from different dimensions. Moreover, the inclusion of a Bayesian optimization algorithm enhances DA-BILSTM’s performance by optimizing hyperparameters objectively. The method is applied to two real flight datasets, demonstrating superior predictive accuracy compared to many state-of-the-art methods. Overall, this research significantly advances the field by offering a com- prehensive approach to airport delay prediction, combining the strengths of multi-step forecasting and attention-based feature extraction.

关键词：Airport delay prediction, attention fusion network, bidirectional long short-term memory network, sequence to sequence.

59. Energy-Efficient Timetable Optimization Empowered by Time-Energy Pareto Solution Under Actual Line Conditions

在实际线路条件下通过时间-能量帕累托解决方案优化能效时间表

作者：H. Zhang, L. Jia, L. Wang, X. Xu, and F. Dou

摘要：This article introduces a two-level method for optimizing railway timetables to enhance energy efficiency. At the train level, it presents an efficient model that accounts for complex actual line conditions like slopes and curves, aiming to balance travel time and energy consumption by using a multi-objective optimization approach. Utilizing an improved non-dominated sorting genetic algorithm II (INSGA-II) combined with differential evolution (DE) and a novel crowding distance operator, the method identifies optimal speed profiles for individual trains. At the timetable level, an integer linear programming (ILP) model incorporates these profiles to create an energy- efficient schedule, ensuring minimal headway between trains. This is achieved through a new branch-and-bound method based on operation time, which expedites the search for the best control strategies. A case study on the Beijing–Shanghai high-speed railway demonstrates significant energy savings of up to 18.69% with punctual trains and 4.69%, 9.93%, 8.05% in various delay scenarios, highlighting the effectiveness of the proposed optimization approach.

关键词：Energy-efficient driving, energy-efficient timetable, multi-objective optimization, Pareto solution.

60. A Hybrid Long Short-Term Memory and Kalman Filter Model for Train Trajectory Prediction

一种用于列车轨迹预测的混合长短期记忆和卡尔曼滤波模型

作者：E. Ahmad, Y. He, Z. Luo, and J. Lv

摘要：A novel hybrid model is presented for predicting train trajectories, integrating long–short-term memory (LSTM) and Kalman filter (KF). The KF component incorporates train dynamics mechanisms to extract local features from train operation data, thereby improving the smoothness of the trajectory predicted by the LSTM model. An innovative on- the-fly algorithm is developed for efficient implementation of the proposed LSTM-KF model. The experimental results show a notable improvement in the accuracy of long-term train trajectory prediction with the hybrid LSTM-KF model.

关键词：Trajectory prediction, long short-term memory, Kalman filter, train separation.

创立宗旨：交通邦旨在构建泛交通领域最大的交流分享平台，促进产学研融合，最小化信息不对称。欢迎各位学生/教授/专家/企业在学术成果、招生计划、书籍推介、内推岗位、比赛冠名、会议合作等与交通邦合作！

联系方式：添加交通邦小助手微信（jiaotongbang），备注“姓名-学校-合作内容”。