智能辅助驾驶技术广泛应用，其决策系统的潜在不稳定性对城市交通安全构成了重大挑战。本文旨在深入研究智能辅助驾驶车辆驾驶人在复杂城市交通场景下的驾驶行为，通过分析各种城市交通场景中的车辆行驶数据及其对驾驶人驾驶决策产生影响的关键环境要素，有效监测和评估驾驶行为的风险程度，为辅助驾驶安全性提供实证数据支持。主要研究内容如下：

(1) 影响驾驶行为的场景要素提取方法设计。深入探讨了智能驾驶技术在城市复杂驾驶环境中识别关键场景元素的策略，通过设计关键静态和动态场景要素的识别方法，实现对驾驶环境的高效且精准监测。利用分布式多任务联合检测方法，整合关键静态和动态场景要素检测方法，对真实行车环境下采集的交通场景图像进行深入分析和识别，为有效处理车辆、行人、交通标志检测任务及驾驶场景数据处理提供了重要的技术支撑，规避了传统佩戴式数据采集设备可能对驾驶人造成的干扰，使研究成果更贴近实际应用需求。

(2) 基于特征生成网络的行驶状态与驾驶场景数据检索方法。利用特征生成网络对真实行车场景下的交通场景数据及相应时刻车辆行驶状态数据进行数据扩增，有效地匹配提取的驾驶员周围核心目标要素信息与相应的驾驶场景数据，在真实驾驶环境中评估驾驶员的安全驾驶阈值，验证特征生成网络在检索和生成驾驶员感知景观信息方面的有效性。以更接近实际驾驶状况变化的方式进行模型构建，研究利用时间序列变化的车辆动态特征数据映射出驾驶员在驾驶过程中的周边场景信息，为驾驶安全性评估、交通流优化、自动化技术发展以及人机交互改进提供了理论和实践依据。

(3) 驾驶人行为风险监测及预警方法。建立驾驶人行为风险监测阈值，利用负感知注意力数据匹配方法对驾驶场景信息与行驶状态数据进行匹配度分析，使驾驶行为风险监测的分析更加直观。在真实驾驶环境中评估驾驶员的安全驾驶阈值，有助于准确分析驾驶人的操作决策过程，为智能辅助驾驶系统的决策策略提供更个性化的修正支持。对识别和预防驾驶人的分心驾驶、注意力分散和视觉偏离道路等不安全行为提供了有效的技术支持。

(4) 驾驶人行为风险度监测实验验证。通过智能驾驶技术提取复杂城市场景中的关键场景要素，验证了本文提出的分布式多任务联合检测技术在图像识别方面的准确性和实验结果的可靠性。此外，验证了本论文基于时间动态的特征生成网络对驾驶场景和行车状态数据增强结果的可靠性，通过负感知注意力数据匹配方法进行驾驶人行为风险性分析，证明了该方法在模拟实际驾驶条件下优于已有方法的风险监测效果。最后，深入模拟仿真平台的危险驾驶行为数据与正常驾驶数据的低匹配度情况，进一步验证了风险监测与预警方法的实用性和有效性。

本文提出的城市复杂场景的驾驶行为风险监测方法为智能驾驶逻辑场景中的风险域识别及智能驾驶系统安全性的全面评估提供了有效的方法论和实践指南，对智能辅助驾驶技术的安全优化和城市交通安全管理具有重要的理论和实践意义。

Intelligent driving technology is widely deployed and the potential instability of its decision-making system poses a major challenge to urban road safety. The purpose of this paper is to study in depth the driving behavior of intelligent vehicle drivers in complex urban traffic scenarios, to effectively monitor and assess the risk level of driving behavior by analysing the vehicle driving data in several urban traffic scenarios and its key environmental elements that affect the driver's driving decision, and to provide empirical data support for the safety of assisted driving. The main research content is as follows:

(1) Design of a scene element extraction method influencing driving behavior. This study extensively explores the intelligent driving technology strategy for identifying key scene elements in complex urban driving environments. By introducing an identification method for both static and dynamic scene elements, efficient and accurate monitoring of the driving environment is achieved. Through the utilization of a distributed multi-task joint detection approach and integration of key static and dynamic scene element detection methods, thorough analysis and recognition of traffic scene images collected in real-world driving environments are conducted. These findings provide crucial technical support for effective processing tasks such as vehicle, pedestrian, traffic sign, and lane detection, as well as driving scene data processing. Moreover, this approach avoids potential interference caused by traditional wearable data acquisition devices on drivers while ensuring that research results align closely with practical application requirements.

(2) A method for retrieving driving state and driving scene data based on feature generation networks. The traffic scene data and vehicle driving state data in real driving scenarios are expanded using feature generation networks to approach the actual driving condition changes in a more realistic way. The dynamic vehicle feature data are mapped to the surrounding scene information of the driver during driving, providing theoretical and practical basis for driving safety evaluation, traffic flow optimization, automation technology development, and human-machine interaction improvement.

(3) A method for monitoring and warning driver behavioral risks. Behavioral risk monitoring thresholds are established, and the negative perceptual attention data matching method is used to analyze the matching degree between driving scene information and driving state data, making the analysis of driver behavioral risk monitoring more intuitive. Evaluating the safe driving threshold of drivers in real driving environments helps accurately analyze the driver's operational decision process and provide more personalized correction support for the decision strategy of intelligent assisted driving systems. It provides effective technical support for identifying and preventing driver distractions, attention diversion, and visual deviation from the road, etc.

(4) Experimental verification of driver behavior risk degree monitoring. The key scene elements in complex urban scenes are extracted by intelligent driving technology, and the accuracy of distributed multi-task joint detection technology in image recognition and the reliability of experimental results are verified. In addition, the reliability of the feature generation network for driving scene and driving state data enhancement results is verified, and the risk analysis of driver behavior is carried out by the negative perception attention data matching method, which proves that the risk monitoring effect of the proposed method is better than that of existing methods under simulated actual driving conditions. Finally, the low matching degree between dangerous driving behavior data and normal driving data of the simulation platform is deeply simulated, which further verifies the practicability and effectiveness of the risk monitoring and early warning method.

The risk monitoring method of driving behavior in complex urban scenes proposed in this paper provides an effective methodology and practical guide for the identification of risk domains in intelligent driving logic scenes and the comprehensive assessment of the safety of intelligent driving systems, which has important theoretical and practical significance for the safety optimization of intelligent assistant driving technology and urban traffic safety management.

[1]Muhammad K, Ullah A, Lloret J, et al. Deep learning for safe autonomous driving: Current challenges and future directions[J]. IEEE Transactions on Intelligent Transportation Systems, 2020, 22(7): 4316-4336.

[2]Yang K, Tang X, Qiu S, et al. Towards robust decision-making for autonomous driving on highway[J]. IEEE Transactions on Intelligent Transportation Systems, 2023, 72(9):11251-11263.

[3]Tang X, Yang K, Wang H, et al. Driving environment uncertainty-aware motion planning for autonomous vehicles[J]. 2022, 35(120):1-14

[4]Yang K, Tang X, Li J, et al. Uncertainties in onboard algorithms for autonomous vehicles: Challenges, mitigation, and perspectives[J]. IEEE Transactions on Intelligent Trans-portation Systems, 2023, 24(9):8963-8987.

[5]Ma S, Xu S C, Song J F, et al. Study on driver’s active emergency response in dangerous traffic scenes based on driving simulator[J]. Traffic Injury Prevention, 2024, 25(2):116-121.

[6]李佳恺.驾驶人弯道行车生理特性研究进展[J].汽车实用技术.2024,(4):184-188.

[7]陈林,曲延羽,刘宇凡,等.基于驾驶行为习惯的智能驾驶人机交互系统[J].时代汽车. 2021,(14):197-198.

[8]李佳芯.深度调查除隐患护平安[J].道路交通管理,2024(02):30-31.

[9]Ferdowsi A, Challita U, Saad W. Deep learning for reliable mobile edge analytics in intelligent transportation systems[J]. IEEE Vehicular Technology Magazine, 2019, 14:62–70.

[10]Lefèvre S, Vasquez D, Laugier C. A survey on motion prediction and risk assessment for intelligent vehicles[J]. ROBOMECH journal, 2014, 1: 1-14.

[11]Chen T, Shi X, Wong Y D, et al. Predicting lane-changing risk level based on vehicles’ space-series features: A pre-emptive learning approach[J]. Transportation research part C: emerging technologies, 2020, 116: 102646.

[12]江昱山. 基于机器学习的换道意图识别与换道轨迹预测研究[D]. 长安大学, 2021.

[13]CHEN X Y, XU J J, ZHOU R, et al. TrajVAE: A Variational AutoEncoder model for trajectory generation[J/OL]. Neurocomputing, 2021, 428:332-339.

[14]熊璐,吴建峰,冯天悦等.面向自动驾驶逻辑场景评价的危险域识别方法[J/OL].中国公路学报:1-18[2024-03-19].

[15]周士谦,高永强,齐龙等.基于LSTM的驾驶行为风险评估研究[J].公路交通技术, 2024, 40(01):163-170.

[16]Gao Y, Huang Y, Meng S. Evaluation and interpretation of driving risks: Automobile claim frequency modeling with telematics data[J]. Statistical Analysis and Data Mining: The ASA Data Science Journal, 2023, 16(2): 97-119.

[17]Sheikh M S, Peng Y. Modeling collision risk for unsafe lane-changing behavior: A lane-changing risk index approach[J]. Alexandria Engineering Journal, 2024, 88: 164-181.

[18]孙宫昊,荣建,常鑫,等.基于驾驶行为模式转移的驾驶行为风险评估方法[J].汽车技术,2021(11):22-29.

[19]南彦洲,柯辉,朱才华,等.面向驾驶员心电数据的交叉口驾驶风险评估方法[J].中国安全科学学报,2023,33(02):185-193.

[20]高翔,陈龙,王歆叶,等.基于轨迹预测的智能汽车行驶风险评估方法[J].汽车工程,2023,45(04):588-597.

[21]彭浩楠,唐明环,查奇文,等.自动驾驶汽车双车道换道最优轨迹规划方法[J].吉林大学学报(工学版),2022,52(12):2852-2863.

[22]刘玉民,张雨虹.基于迁移学习的交通标志识别系统设计[J].唐山学院学报, 2023,36(06):1-4.

[23]景荣荣,吴兰,张坤鹏.基于生成对抗网络的自动驾驶交通场景生成[J].电子测量技术,2023,46(14):146-154.

[24]LiuY, Li M Y, Shhen C Y, et al. Discriminative Transfer Learning for Driving Pattern Recognition in Unlabeled Scenes[J]. IEEE Transactions on Cybernetics, 2022, 52(3):1429-1442.

[25]Gan S, Li Q, Wang Q, W. Chen, et al. Constructing personalized situation awareness dataset for hazard perception, comprehension, projection, and action of drivers[C]. Indianapolis, IN, USA:2021 IEEE International Intelligent Transportation Systems Conference, 2021.

[26]熊坚, 施锦浩, 万华森. 人车路综合风险场模型构建及驾驶风格评估[J]. 交通运输系统工程与信息, 2021, 21(6): 105-114.

[27]Wang C, Sun Q, Guo Y, et al. Improving the user acceptability of advanced driver assistance systems based on different driving styles: A case study of lane change warning systems[J]. IEEE Transactions on Intelligent Transportation Systems, 2019, 21(10): 4196-4208.

[28]Park H, Oh C, Moon J, et al. Development of a lane change risk index using vehicle trajectory data[J]. Accident Analysis & Prevention, 2018, 110: 1-8.

[29]Huang H, Wang J, Fei C, et al. A probabilistic risk assessment framework considering lane-changing behavior interaction[J]. Science China Information Sciences, 2020, 63: 1-15.

[30]Chen Q, Huang H, Li Y, et al. Modeling accident risks in different lane-changing behavioral patterns[J]. Analytic methods in accident research, 2021, 30: 100159.

[31]Mao H, Guo F, Deng X, et al. Decision-adjusted driver risk predictive models using kinematics information[J]. Accident Analysis & Prevention, 2021, 156: 106088.

[32]Chen S, Xue Q, Zhao X, et al. Risky driving behavior recognition based on vehicle trajectory[J]. International journal of environmental research and public health, 2021, 18(23): 12373.

[33]Seo H, Shin J, Kim K H, et al. Driving risk assessment using non-negative matrix factorization with driving behavior records[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(11): 20398-20412.

[34]刘通, 付锐,马勇,等. 考虑驾驶人风格的跟车预警规则研究[J].中国公路学报, 2020, 33(2): 170-180.

[35]安玉,焦朋朋,白紫秀.考虑多因素的驾驶行为安全评价与风险等级预测[J].系统仿真学报,2021,33(01):118-126.

[36]戢晓峰,谢世坤,覃文文,等.基于轨迹数据的山区危险性弯道路段交通事故风险动态预测[J]. 中国公路学报, 2022, 35(04): 277-285.

[37]张晖,刘永杰,吴超仲,等.考虑行车安全事件严重程度和个体差异的驾驶行为风险评估[J].中国安全科学学报,2023,33(07):24-31.

[38]张瑞宾. 融合时空状态与场景特征的车辆运行风险评估及预警研究[D].长安大学,2023.

[39]王常态. 考虑驾驶工况的驾驶习性辨识策略研究[D].吉林大学,2024.

[40]Shangguan Q, Fu T, Wang J, et al. A proactive lane-changing risk prediction framework considering driving intention recognition and different lane-changing patterns[J]. Accident Analysis & Prevention, 2022, 164: 106500.

[41]冯忠祥, 李靖宇, 张卫华, 等. 面向人机共驾车辆的驾驶人风险感知研究综述[J]. 交通信息与安全, 2022, 40(2): 1-10.

[42]孙宫昊,常鑫,高亚聪,等.基于信息熵与高风险行为的驾驶行为风险评估方法[J].交通工程,2023,23(04):22-28.

[43]陈志发,余贵珍,张传莹,等.露天矿区无人驾驶行车风险评估及防控策略仿真研究[J].煤炭学报,2023,48(04):1782-1797.

[44]刘平,沈跃,杨明亮,等.跟车工况下基于风险评估的人机共驾策略[J/OL].汽车工程学报:1-10[2024-04-09].

[45]Liu M, Li Z, Li Y, et al. A fast and accurate method of power line intelligent inspection based on edge computing[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1-12.

[46]Jiang T, Li C, Yang M, et al. An improved YOLOv5s algorithm for object detection with an attention mechanism[J]. Electronics, 2022, 11(16): 2494.

[47]Li Y, Fan Q, Huang H, et al. A modified YOLOv8 detection network for UAV aerial image recognition[J]. Drones, 2023, 7(5): 304.

[48]Zhu X, Lyu S, Wang X, et al. TPH-YOLOv5: Improved YOLOv5 based on transformer prediction head for object detection on drone-captured scenarios[C]//Proceedings of the IEEE/CVF international conference on computer vision. 2021: 2778-2788.

[49]Wang C Y, Bochkovskiy A, Liao H Y M. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023: 7464-7475.

[50]马燕婷, 赵红东, 阎超, 等. 改进 YOLOv5 网络的带钢表面缺陷检测方法[J]. 电子测量与仪器学报, 2022, 36(8): 150-157.

[51]Uladzislau S, Feng X. Modified omni-scale net architecture for cattle identification on their muzzle point image pattern characteristics[C]//International Conference on Computer, Artificial Intelligence, and Control Engineering (CAICE 2023). SPIE, 2023, 12645: 489-494.

[52]Creswell A, White T, Dumoulin V, et al. Generative adversarial networks: An overview[J]. IEEE signal processing magazine, 2018, 35(1): 53-65.

[53]Elakkiya R, Vijayakumar P, Kumar N. An optimized generative adversarial network based continuous sign language classification[J]. Expert Systems with Applications, 2021, 182: 115276.

[54]陈泓佑, 陈帆, 和红杰, 等. 基于样本特征解码约束的 GANs[J]. 自动化学报, 2022, 48(9): 2288-2300.

[55]张文博. 基于多源数据融合的异构车辆数据匹配算法研究及应用[D].吉林大学,2019.

[56]林瑀,陈日成,金涛.面向复杂信息系统的多源异构数据融合技术[J].中国测试,2020,46(7):1-7.

[57]Kessler L, Rempe F, Bogenberger K. Multi-Sensor Data Fusion for Accurate Traffic Speed and Travel Time Reconstruction[J]. Frontiers in Future Transportation, 2021.

[58]Haule HJ, Ali M S, Alluri P, et al. Evaluating the effect of ramp metering on freeway safety using real-time traffic data[J]. Accident Analysis & Prevention, 2021, 157(5): 106181.

[59]Diao H, Zhang Y, Ma L, et al. Similarity reasoning and filtration for image-text matching[C]. Proceedings of the AAAI conference on artificial intelligence. 2021, 35(2): 1218-1226.

[60]Huang Y, Wu Q, Song C, et al. Learning semantic concepts and order for image and sentence matching[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 6163-6171.