Analysis of truck drivers’ unsafe driving behaviors using four machine learning methods

Unsafe driving behaviors are the leading causes of truck crashes. Therefore, an enhanced understanding of truck drivers' unsafe driving behaviors is of considerable significance for preventing truck crashes. However, previous studies have rarely encompassed proactive factors such as safety management. Therefore, a classification framework for truck drivers' unsafe driving behaviors was established according to a survey of 2000 truck drivers using four machine learning models (CART, RT, AdaBoost, and GBDT). The classification framework included six first-level input dimensions, 51 s-level input indicators, covering both objective and proactive factors. Nine types of unsafe driving behaviors were determined as outputs. Unique risk factors associated with each of nine unsafe driving behaviors were identified. The results showed that the model's predictive performance varies with different driving behaviors (Classification Accuracy ranges from 0.64 to 0.95, F1-score ranges from 0.52 to 0.72), which was caused by different formation mechanisms of different driving behaviors. Similarly, the results related to factor importance for different driving behaviors were also significantly different, regardless of the first-level and second-level factors. Furthermore, the correlation analysis and OR value strengthened the interpretability of the factor importance, revealing possible reasons for the differences between various driving behaviors.     

SF-CNN在驾驶行为识别中的应用研究

为了减少不良驾驶行为的潜在危险,通过智能手机内置传感器对驾驶行为进行实时监测,辅助驾驶者安全驾驶,提出了一种优化特征分布的无监督特征学习算法模型——稀疏滤波-卷积神经网络模型（Sparse Filter-Convolutional Neural Network,SF-CNN）。该方法利用移动终端在车辆行驶中采集的三轴加速度数据,通过稀疏滤波进行范数联合约束,得到紧凑的初级特征表达,将该表达矩阵作为卷积神经网络首层的输入,进行非线性分类来识别驾驶行为。实验结果表明,稀疏滤波-神经网络的识别模型对驾驶行为具有更高的识别率和鲁棒性,优于传统神经网络模型,对辅助驾驶系统的效能评价有重要的理论意义。     

驾驶行为的初步研究

疲劳驾驶对交通安全的危害日益严重，通过给驾驶行为建模来研究疲劳行为，主要介绍了给驾驶行为建模的主要流程，信号采集，对方向信号进行傅立叶级数展开来提取特征值，确定特征值的个数等。通过对驾驶员正常的驾驶行为进行建模，当驾驶员处于疲劳状况下驾车偏离正常模型时则能够有效的区别开。     

A smartphone based technique to monitor driving behavior using DTW and crowdsensing

Safety issues while driving in smart cities are considered to be top-notch priority in contrast to traveling. Today’s fast paced society, often leads to accidents. In order to reduce the road accidents, one key area of research is monitoring the driving behavior of drivers. Understanding the driver behavior is an essential component in Intelligent Driver Assistance Systems. One of potential cause of traffic fatalities is aggressive driving behavior. However, drivers are not fully aware of their aggressive actions. So, in order to increase awareness and to promote driver safety, a novel system has been proposed. In this work, we focus on DTW based event detection technique, which have not been researched in motion sensors based time series data to a great extent. Our motivation is to improve the classification accuracy to detect sudden braking and aggressive driving behaviors using sensory data collected from smartphone. A very significant feature of DTW is to be able to automatically cope with time deformations and different speeds associated with time-dependent data which makes it suitable for our chosen application where data might get affected due to factors such as: high variability in road and vehicle conditions, heterogeneous smartphone sensors, etc. Our technique is novel as it uses fusion of sensors to enhance detection accuracy. The experimental results show that proposed algorithm outperforms the existing machine learning and threshold-based techniques with 100% detection rate of braking events and 97% & 86.67% detection rate of normal left & right turns and aggressive left & right turns respectively.     

基于移动群智感知的极端驾驶行为感知方法

为提高驾驶安全性，减少交通事故的发生，本文提出一种基于移动群智感知的极端驾驶行为识别方法。对收集到的用户智能手机相关传感器的数据进行预处理，进而利用动态步数检测和随机森林等方法来识别乘客的用户情境信息。针对不同的极端驾驶行为，选取不同位置乘客的智能手机来进行数据的收集，综合考虑乘客的手机放置位置因素所造成的相关影响，实现多特征融合的极端驾驶行为感知。针对不同位置的乘客所感知的结果不一致问题，提出采用贝叶斯投票模型来解决。通过真实数据实验，结果表明本文方法能够有效地识别出司机的极端驾驶行为。     

虚拟驾驶环境中车辆智能体的驾驶行为模型

建立虚拟交通环境的多智能体结构,分析车辆智能体的驾驶行为分层模型以及感知、决策和操作等过程。采用模糊专家系统建立车辆智能体的驾驶行为模型。为模拟现实中的驾驶员行为特性,加入驾驶员因子,使驾驶模拟器的虚拟交通环境更符合现实。运用OpenGVS产生和显示实时交互的虚拟驾驶场景。结果表明该模型能体现实际驾驶行为的多样性、随机性和模糊性。该模型通用有效,它使驾驶模拟器的虚拟交通场景更真实满意。     

基于深度学习的驾驶员分心行为识别

分心驾驶行为识别是提高驾驶安全的主要方法之一。针对分心驾驶行为识别精度低的问题,本文提出一种基于深度学习的驾驶员分心行为识别算法,由目标检测网络和行为精确识别网络级联构成。基于State Farm公开数据集,第一级利用目标检测算法SSD(Single Shot Multibox Detector)对数据集中的驾驶员原始图像进行局部信息提取,确定行为识别候选区域;第二级分别利用迁移学习VGG19、Res Net50和MobileNetV2模型对候选区域内的行为信息进行精确识别;最后,实验对比级联架构与单模型架构对分心驾驶行为的识别精度。结果表明,提出的级联网络模型相较于主流单模型检测方法,驾驶员行为识别的准确率总体上提升4～7%个百分点。该算法不仅减少噪声和其他背景区域对模型的影响,提高分心行为识别准确率,还可以有效识别更多的行为类别以避免动作的误分类。     

Driving Activity: How to Improve Knowledge of the Environment

This paper concerns the problem of driving assistance and, in particular, how to improve the perception of the surrounding environment to make this assistance really helpful. The main aims of a Driving Assistance System are to improve the security of the driver, passengers, and other road users. Driving is a complex activity, where the interactions between the driver, the vehicle, and the environment are continuous and numerous. The vehicle moves in a dynamic environment, so the Driving Assistance System, for its diagnosis, needs a map that represents as well as possible the actual situation of this environment. This paper presents a multi-sensor fusion module embedded in a real vehicle. The problem considered here is the dynamic reconstruction of the environment of the vehicle, based on measurements of a set of sensors.     

驾驶行为智能分析的研究与发展

越来越多的研究者认识到: 深入地理解驾驶员的驾驶行为将有助于制定更为合理的交通法规和设计更加有效的智能驾驶导航系统, 从而达到减少交通事故提高交通效率的目的. 本文综述了已有的尝试, 较为完整地阐述了目前驾驶行为智能分析研究的四个主要方向: 纵向驾驶行为分析和避撞, 横向驾驶行为分析和道路偏离预警, 复杂驾驶行为学习以及驾驶员状态(疲劳、分心等)分析, 并指出了今后该领域(特别是国内)的可能发展方向.     

Specifying comfortable driving postures for ergonomic design and evaluation of the driver workspace using digital human models

Specifying comfortable driving postures is essential for ergonomic design and evaluation of a driver workspace. The present study sought to enhance and expand upon several existing recommendations for such postures. Participants (n = 38) were involved in six driving sessions that differed by vehicle class (sedan and SUV), driving venue (laboratory-based and field) or seat (from vehicles ranked high and low by vehicle comfort). Sixteen joint angles were measured in preferred postures to more completely describe driving postures, as were corresponding perceptual responses. Driving postures were found to be bilaterally asymmetric and distinct between vehicle classes, venues, age groups and gender. A subset of preferred postural ranges was identified using a filtering mechanism that ensured desired levels of perceptual responses. Accurate ranges of joint angles for comfortable driving postures, and careful consideration of vehicle and driver factors, will facilitate ergonomic design and evaluation of a driver workspace, particularly when embedded in digital human models.     

驾驶员多源信息融合协同仿真算法研究

交通仿真是交通控制与管理方案评价和优化的重要实验研究手段。传统的微观交通仿真模型,特别是刻画驾驶员行为的车辆跟驰模型,未能综合考虑交通环境中信息刺激的多源性和驾驶员任务集聚、协调反应的行为过程。文章利用Bayes方法描述驾驶员在复杂行驶环境中多源信息的融合过程,确定驾驶员任务集聚后对车辆应采取的驾驶行为。模型验证表明:交通仿真过程中,在车辆跟驰模型实施之前,利用Bayes算法模型化驾驶员在多源信息刺激下任务集聚、协同反应的过程是行之有效的。     

基于MEMS和GPS的驾驶行为和车辆状态监测系统设计

为了适应智能车辆辅助驾驶系统对驾驶和车辆状态监测的要求,利用MEMS惯性传感器自主设计了微惯性测量单元,并结合GPS设计了一种驾驶行为和车辆状态监测系统,实现对驾驶员操纵动作的感知、汽车6自由度运动状态参数和汽车运行车速的实时监测。介绍了MEMS传感器的选型,设计,安装和布置。实车道路实验结果表明:系统对驾驶员踩踏刹车踏板、离合器踏板和变换档位的操纵动作的感知效果较好,侧向加速度和方向盘转角的理论识别曲线与实车实验曲线在趋势上比较吻合。该系统为开发驾驶人员操纵动作自动识别系统提供理论基础和技术支持,也可为提高汽车行驶性和安全性提供重要的理论依据和工程应用指导。     

视觉协同的违规驾驶行为分析方法

本文针对危险驾驶识别中主流行为检测算法可靠性差的问题,提出了一种快速、可靠的视觉协同分析方法。对手机、水杯、香烟等敏感物体进行目标检测,提出的LW(low weight)-Yolov4(You only look once v4)通过去除CSPDarknet53(cross stage partial Darknet53)卷积层中不重要的要素通道提升了检测速度,并L₁正则化产生稀疏权值矩阵,添加到BN(batch normalization)层的梯度中,实现优化网络模型的目的;提出姿态检测算法对驾驶员指关节关键点进行检测,经过仿射逆变换得到原始帧中的坐标;通过视觉协同分析对比敏感物品的检测框位置与驾驶员手部坐标是否重合,判定驾驶员是否出现违规驾驶行为及类别。实验结果表明,该方法在识别精度与检测速度方面均优于主流的算法,能够满足实时性和可靠性的检测要求。     

A Review.of: “Man and Motor Cars: An Ergonomic Study.” By S. Black. (Seeker and Warburg, 1966.) [Pp. 373.] 63s

This study was concerned with examining driver patterns for three different sized vehicles (small cars, family cars, trucks and buses) through an intersection controlled by a traffic signal. Observations were made morning, noon and afternoon, when there was precipitation and no precipitation. Records were also kept as to whether the drivers wore alone or with passengers and whether the drivers were males or fomales. Results were found showing that family car drivers were the most cautious and that small ear, truck and bus drivers were approximately similar to each other, being less cautious. Drivers with passengers were more cautious and did not violate the law by passing a red light as frequently as did drivers without passengers. Approximately similar driving patterns were found for males and fomales. Different driving patterns were obtained for the three-sized vehicles during morning, noon and afternoon, when there was precipitation and no precipitation.     

虚拟驾驶员视觉感知模型研究

视觉是驾驶员获得驾驶信息的主要通道,虚拟驾驶员的视觉感知模型是驾驶行为建模与仿真的重要内容,直接影响驾驶行为仿真的逼真度。介绍了四种视觉感知建模方法,根据视觉感知系统的反馈性和选择性,建立了视觉感知模型。在模型中,将视觉感知分为感觉和知觉两个过程,并引入了注意力、记忆力、驾驶疲劳和驾驶经验等因素,对每个因素进行分析。通过对虚拟驾驶员视觉感知模型的仿真实验,验证了该模型的可靠性和有效性,为驾驶行为研究提供了基础。     

基于车辆行驶数据的驾驶人行为谱分析方法

针对我国驾驶人行为谱的研究尚不完善，专业领域内没有相应的行为谱分析工具的问题，提出了一套针对营运客车的完整的驾驶人驾驶行为谱体系并设计了一套分析工具。首先，设计并定义了驾驶人行为谱的特征指标和评价指标；其次，给出了驾驶人行为谱的特征指标分析、计算方法，采用基于马尔可夫链蒙特卡洛采样和离群点剔除的K-means算法对驾驶人的驾驶风格进行分析，采用回归学习对驾驶人的驾驶技能进行分析；然后，设计了基于车联网、大数据的驾驶人行为谱的基础数据采集和预处理方法；最后，采用Java语言、Spring MVC架构开发出驾驶人行为谱分析工具。将机器学习中的数据挖掘、数据分析算法与交通安全领域相结合，对完善我国驾驶人行为谱框架体系具有理论意义，为我国驾驶人行为谱的研究提供了一个科学、定量化分析的工具，对交管部门规范驾驶人驾驶行为、提高道路安全指数、制定合理的交通安全管理策略具有指导意义。     

基于人-车-路-环境综合计算的驾驶员期望车速

微观交通流仿真是智能运输系统研究与开发的重要手段。期望车速是微观交通流仿真研究的一个重要参数,受到驾驶员特性、车辆特性、道路条件、交通干扰、天气和承运任务急缓等多种因素的影响,准确确定期望车速是驾驶员行为研究的难点。从研究驾驶员心理-物理特性的角度出发,利用层次分析法,对驾驶员决策思维的递阶层次进行量化,建立基于人-车-路-环境综合计算的驾驶员期望车速模型。经过实测数据验证,该方法用于驾驶员期望车速模型的研究是可行的。     

基于驾驶员面部时序数据的疲劳驾驶检测算法

针对传统疲劳驾驶检测方法识别准确率低、泛化能力差的问题,提出了一种基于CNNs和LSTM的端到端可训练网络,检测驾驶员的疲劳状态。根据驾驶员面部特征点提取ROI,将在其他计算机视觉任务上表现较好的深度网络迁移到疲劳检测任务中,并结合LSTM处理时序数据的能力,提出一种新的疲劳检测网络,该网络能够读入视频流中的时序数据并检测出驾驶员的疲劳状态。实验证明所提方法和模型在公开数据集中具有较高的识别准确率,并且在不同的数据集间具有很好的泛化能力,对于减少路面车祸、保障人身安全具有很重要的意义。     

出租车载客点个性化推荐算法

不同的出租车司机在寻找乘客选取载客点时会有不同倾向,利用三种推荐算法对上海出租车司机载客点选取行为进行分析,根据司机对载客点的喜好程度进行个性化推荐.首先,利用基于用户和基于项目的协同过滤的算法来对出租车司机的载客点进行推荐,利用正确率指标来验证算法,实验证实了这两种算法的可行性;之后,考虑到出租车的载客行为受到时间的影响,在上述两种算法基础上增加了时间因子;最后,利用隐含因子模型（LFM）,将出租车与载客点的共现矩阵进行分解,根据分解所得矩阵进行兴趣度的分析.实验结果证明,三种方法可有效形成推荐,且LFM算法推荐准确率较高.     

基于VR的驾驶仿真及安全教育系统开发

针对全国道路交通事故高发现状及传统驾驶安全教育方式单一、培训效果差的缺点,基于虚拟现实技术（VR）,在引发交通事故人为因素理论基础上,开发驾驶仿真及安全教育系统。系统基于Unity3D引擎,构建了基于道路实景数据的虚拟场景,并联合SUMO实现了道路交通流仿真,通过VR技术仿真驾驶环境及驾驶行为;基于碰撞检测原理,建立了关卡违规触发机制,编码自定义屏幕空间渲染方式模拟驾驶员视觉效果,并构建了基于图像的交通事故现场三维全景,从认知、感知层面培训驾驶员安全驾驶。实用性测试结果表明,系统实现了不同道路场景、气象条件与交通状况下的驾驶模拟及安全培训,增强了使用者的学习兴趣,提高了使用者驾驶安全素养,具有较强的实用性。