基于生成对抗网络的驾驶行为风险监测方法  

A generative adversarial network⁃based risk monitoring approach for driving behavior

作　　者：赵栓峰[1] 王梦维 罗志健 ZHAO Shuanfeng;WANG Mengwei;LUO Zhijian(Faculty of Mechanical Engineering,Xi’an University of Science and Technology,Xi’an 710054,China)

机构地区：[1]西安科技大学机械工程学院,陕西西安710054

出　　处：《现代电子技术》2024年第18期133-138,共6页Modern Electronics Technique

基　　金：陕西省重点研发计划项目(2020ZDLGY04-06);陕西省秦创原“科学家+工程师”队伍建设项目(2023KXJ-249)。

摘　　要：针对真实驾驶场景下驾驶行为风险难以准确判断的问题,提出一种基于生成对抗网络的生成序列评估框架,旨在通过车辆行驶状态数据与驾驶场景信息匹配度评估来提高驾驶行为风险判断的准确性。构建真实驾驶场景数据匹配序列框架,利用生成对抗网络增加不同驾驶场景的数据量,提出一个驾驶员个性化情景感知样本数据集,对比分析真实数据和生成数据,并验证生成对抗网络检索与生成框架在匹配车辆行驶状态数据与驾驶场景信息方面的有效性和准确性。采用无监督学习生成评估框架,结合自回归模型中的监督训练控制策略来解决数据扩增问题,并通过迭代更新样本序列进行策略集成以保持模型鲁棒性。实验结果表明,相较于现有方法,所提研究框架在识别危险驾驶状态方面具有明显优势,为车辆行驶状态数据与驾驶场景信息的匹配评估提供了一种新的视角。In allusion to the problems that it is difficult to accurately judge the risk of driving behavior in real driving scenes,a generative sequence evaluation framework based on generative adversarial network is proposed,aiming to improve the accuracy of driving behavior risk judgment by evaluating the matching degree between vehicle driving state data and driving scenario information.A sequence framework for matching real driving scene data is constructed,the generative adversarial network is used to increase the data volume of different driving scenes,and a personalized scene perception sample dataset for drivers is proposed.The comparison analysis for real data and generated data is conducted,and the effectiveness and accuracy of the generative adversarial network retrieval and generation framework in matching vehicle driving status data with driving scenario information are verified.The problem of data augmentation is solved by means of the unsupervised learning generative evaluation framework and combined with supervised training control strategies in autoregressive model,and the policy integration is conducted by iteratively updating sample sequences to maintain the robustness of the model.The experimental results show that,in comparison with the existing methods,the proposed framework has obvious advantages in identifying dangerous driving states and can provide a new perspective for the matching evaluation of vehicle driving state data and driving scene information.

关 键 词：生成对抗网络 驾驶行为 风险监测 评估框架 数据匹配 无监督学习 车辆行驶状态 

分 类 号：TN911.73-34[电子电信—通信与信息系统] U491[电子电信—信息与通信工程]