Detecting driving stress in physiological signals based on multimodal feature analysis and kernel classifiers

Monitoring driving status has great potential in helping us decline the occurrence probability of traffic accidents and the aim of this research is to develop a novel system for driving stress detection based on multimodal feature analysis and kernel-based classifiers. Physiological signals such as electrocardiogram, galvanic skin response and respiration were record from fourteen drives executed in a prescribed route at real drive environments. Features were widely extracted from time, spectral and wavelet multi-domains. In order to search for the optimal feature sets, Sparse Bayesian Learning (SBL) and Principal Component Analysis (PCA) were combined and adopted. Kernel-based classifiers were employed to improve the accuracy of stress detection task. Analysis I used features from 10 s intervals of data which were recorded during well-defined rest, highway and city driving conditions to discriminate three levels of diving stress achieving an averaging accuracy over 99% at per-drive level and 89% in cross-drive validation. Analysis II made continuous stress evaluation throughout a complete driving test attaining a high coincidence with the true road situation especially at the switching interval of traffic conditions. Experimental results reveal that different levels of driving stress can be characterized by specific set of physiological measures. These physiological measures could be applied to in-vehicle intelligent systems in various approaches to help the drivers better manage their negative driving status. Our design scheme for driving stress detection could also facilitate the development of similar in-vehicle expert systems, such as driver's emotion management, driver's sleeping onset monitoring, and human-computer interaction (HCI).     

Insurance Classifications and Drivers' Galvanic Skin Response

Drivers' galvanic skin responses were recorded in town traffic and on country roads. Two groups of drivers were compared, those who would pay higher insurance premiums, by reason of their ago or occupation, and the rest. The high insurance group did not differ from the rest in the magnitude of the GSR. but when the ratio of the GSR in town traffic to that on a country road was considered there was a significant difference such that the high insurance group had a relatively greater GSR on the country road.

It is suggested that in town traffic most of the GSR arc generated by the actions of other drivers, whereas on the open road the GSRs are self generated by the subjects' driving. The GSR on the open road will therefore be more affected by the individual method of driving and by the risks the driver takes. The ratio of town GSR to country GSR can be used to eliminate the large unexplained differences between subjects and to expose the differences in driving behaviour     

A method for fatigue detection based on Driver's steering wheel grip

We propose a simple method to measure a driver's fatigue state by detecting the driver's grip force on the steering wheel while driving. We tested the grip force of 36 drivers on the steering wheel in conscious states (Alert) and fatigue states under actual road driving conditions. Using the Stanford sleepiness scale (SSS), we divided drivers into Alert Group A, fatigue Group A, and fatigue Group B. During 20-min real-road driving trials, we measured the steering wheel grip force, electroencephalogram index (R = (α + θ)/β), and blink frequency of each driver synchronously. We found that ΔF, the difference between the maximum/minimum grip force and the standard deviation of the grip force, σF, for each driver, strongly correlated with the driver's fatigue state. In the fatigue state, both ΔF and σF increased significantly. We examined these force indices using analysis of variance (ANOVA) and validated them against the R-value, blink frequency, and the driver's self-reported fatigue state. Using the grip force in fatigue detection, our method can achieve an overall recognition rate of 86.6% and an individual recognition rate of 88.3%. These results indicate that this method can effectively detect a driver's fatigue state during actual road driving. This new method has several advantages, such as a high signal-to-noise ratio, simple data collection, and no influence on daily driving. Thus, our proposed method may provide a theoretical foundation for the development of fatigue-detecting steering wheels     

Approaching and stopping behaviors to the intersections of aged drivers compared with young drivers

According to the many reports of the traffic accidents, the number of the accidents near the intersections was increased in the cases of the aged drivers. The purpose of the research was to measure the sensitivity of elder drivers through behavioral responses in approaching and stopping to the intersections and to obtain the difference of the responses for braking and stopping comparing with young drivers. In the field tests of real running on experimental proving ground, elder and young drivers were observed their driving behaviors in various conditions of approaching and stopping at intersections. Comparing the results of the elder with the young, the unstable driving behaviors were examined in elderly and they were apt to run fast in approaching the intersection and also stopping rapidly. The rates of deceleration change (Jerk) with braking operation were unstable and the driving behaviors were affected by the environments or conditions of the intersection. These driving performances should be interfered with the traffic flow and exposed to the risk of accidents. If the features of these aged driver's behaviors were clarified better, the more effective driving assistant systems for elderly could be developed based on characteristics of elderly driver's driving performances and their physical and psychological features of driving.     

DRIVERS' GALVANIC SKIN RESPONSE AND THE RISK OF ACCIDENT

Galvanic skin responses of 20 drivers were measured in two studies covering a wide range of roads and road conditions. Accident histories wore obtained for the roads in one of the studies. It is shown that the level of GSR activity does not depend primarily on the nature of the road or conditions. Consistent sources of variation in the GSR are observed, one of them apparently being the subject's experience of driving. The distribution of GSR per unit distance travelled was found to be similar to the distribution of accidents per unit total distance of vehicle travel (the accident rate). The results support a view that driving is a self-paced task governed' by the level of emotional tension or anxiety which the driver wishes to tolerate. The possible effects of this on the distribution of accidents is discussed.     

基于眼部自商图—梯度图共生矩阵的疲劳驾驶检测

目的 疲劳驾驶是引发车辆交通事故的主要原因之一,针对现有方法在驾驶员面部遮挡情况下对眼睛状态识别效果不佳的问题,提出了一种基于自商图—梯度图共生矩阵的驾驶员眼部疲劳检测方法。方法 利用以残差网络（residual network,ResNet）为前置网络的SSD（single shot multibox detector）人脸检测器来获取视频中的有效人脸区域,并通过人脸关键点检测算法分割出眼睛局部区域图像;建立驾驶员眼部的自商图与梯度图共生矩阵模型,分析共生矩阵的数字统计特征,选取效果较好的特征用以判定人眼的开闭状态;结合眼睛闭合时间百分比（percentage of eyelid closure,PERCLOS）与最长闭眼持续时间（maximum closing duration,MCD）两个疲劳指标来判别驾驶员的疲劳状态。结果 在六自由度汽车性能虚拟仿真实验平台上模拟汽车驾驶,采集并分析驾驶员面部视频,本文方法能够有效识别驾驶员面部遮挡时眼睛的开闭状态,准确率高达99.12%,面部未遮挡时的识别精度为98.73%,算法处理视频的速度约为32帧/s。对比方法1采用方向梯度直方图特征与支持向量机分类器相结合的人脸检测算法,并以眼睛纵横比判定开闭眼状态,在面部遮挡时识别较弱;以卷积神经网络（convolutional neural network,CNN）判别眼睛状态的对比方法2虽然在面部遮挡情况下的准确率高达98.02%,但眨眼检测准确率效果不佳。结论 基于自商图—梯度图共生矩阵的疲劳检测方法能够有效识别面部遮挡时眼睛的开闭情况和驾驶员的疲劳状态,具有较快的检测速度与较高的准确率。     

Estimation method for the consciousness level while driving vehicles

Recently, the driver's attention while driving a vehicle has to be taken seriously in a modernized society. Although some studies of attention while driving are being conducted now, the character of human activity is complicated for estimating attention while driving a vehicle. In the present study, the driver's attention was studied by driving performance and meandering of the vehicle. Two sets of drivers were used to compare with higher and lower states of consciousness. For driving performance, the degree of steering and the degree of acceleration were measured. For meandering, the shoulder line on the road was detected by a CCD camera to calculate the coordinates of the vehicles. These three values showed the dynamical degree of the driver's attention. The results show that the meandering values and the degree of steering values correlated with the degree of attention of the driver, and these results can be applied to make an alert system for drivers during decreased consciousness or concentration in order to realize a safe society for our modern roadways.     

Effect of driving experience on drivers' decoding process of roadway interpersonal communication

Sixty-three participants (32 novice, 31 experienced drivers) evaluated meanings of road users' signals in 24 traffic situations such as blinkers, headlights, hazard lamps and hand gestures. The traffic scenes were projected with a slide projector in a laboratory. Confidence in answers was also evaluated by using a five-point scale. The signals were classified into three categories: Formal Device-based Signal (Formal Signal), Informal Device-based Signal (Informal Signal), and Informal Gesture-based Signal (Everyday Signal). The total comprehension scores demonstrated that experienced drivers could understand the signals better than novice drivers. There was a large difference in the comprehension scores for Informal Signal between experienced and novice drivers. Novice drivers could understand Formal Signal and Everyday Signal better than Informal Signal. Similar results were also obtained in the confidence scores. Experienced drivers were more confident in their answers than novice drivers. An effect of gender was found in the scores of confidence. The discussion focuses on how driver's skill in interpersonal communication on roads develops in real traffic situations.     

Haptic perceptions in the vehicle seat

The perceptual overloads of visually and auditorily based information and their interference phenomena within vehicles led to research for the applicability of haptically based information and the haptic interfaces to intelligent vehicles. Because seats are the interface that touches the largest area of the driver's body, the driver's seat in vehicles has been the focus of a promising haptic interface that can improve the safety of drivers and the effectiveness and efficiency of the information transfer between vehicles and drivers. This study aims to provide practical guidelines as a building block for designing the haptic (or vibrotactile) interface in a vehicle's driver's seat by investigating, through four experiments, 1) proper intensity of vibration, 2) minimum distance of spatially distinguishable vibrations, 3) proper position and direction of vibration, and 4) proper rhythm of vibration. Twenty participants took part in the experiments, which were conducted in driving simulation environments. These environments consisted of a real car seat, commercial vibration actuators (i.e., the eccentric motors), and a monitor that showed scenes of the road while driving. This study recommended the proper intensity (approximately 26 to 34 Hz and 2.0 to 3.4 G), position (seat pan or back support), direction (horizontal or indirect), intervibration distance (8 to 9 cm), and rhythm of vibration (3‐s duration with 0.5‐s interval), and showed how the characteristics of drivers, such as gender and age, had effects on setting the design variables of the haptic interface in the vehicle seat. © 2010 Wiley Periodicals, Inc.     

Drivers' communicative interactions: on-road observations and modelling for integration in future automation systems

Social interactions with other road users are an essential component of the driving activity and may prove critical in view of future automation systems; still up to now they have received only limited attention in the scientific literature. In this paper, it is argued that drivers base their anticipations about the traffic scene to a large extent on observations of social behaviour of other ‘animate human-vehicles’. It is further argued that in cases of uncertainty, drivers seek to establish a mutual situational awareness through deliberate communicative interactions. A linguistic model is proposed for modelling these communicative interactions. Empirical evidence from on-road observations and analysis of concurrent running commentary by 25 experienced drivers support the proposed model. It is suggested that the integration of a social interactions layer based on illocutionary acts in future driving support and automation systems will improve their performance towards matching human driver's expectations.

Practitioner Summary: Interactions between drivers on the road may play a significant role in traffic coordination. On-road observations and running commentaries are presented as empirical evidence to support a model of such interactions; incorporation of drivers' interactions in future driving support and automation systems may improve their performance towards matching driver's expectations.     

智能驾驶中的行为辅助压力感知方法

人的压力与其行为紧密相关,特别是在智能驾驶时,驾驶员压力感知对实现辅助驾驶具有巨大的应用潜力.现有压力感知方法多用于静态环境,检测过程也缺乏便捷性,难以适应高度动态的智能驾驶应用需求.为了实现智能驾驶中自然、准确和可靠的压力检测,提出一种基于可穿戴系统的行为辅助压力感知方法.该方法基于行为伴随实现压力检测,并基于多指标执行压力状态判别,能够有效提高压力检测准确度.其基本原理在于每个人在不同压力状态下的生理特征和行为模式不同,会对压力相关的PPG数据和行为相关的IMU数据产生独特影响.首先使用嵌入多传感器的可穿戴手套测量驾驶员的生理和运动信息,通过多信号融合技术获得可靠的生理行为指标,最终使用泛化性能较好的SVM模型分类驾驶员的压力状态.基于所提出的方法在模拟驾驶环境下部署了验证实验,实验结果显示,压力分类精确度可达到95%.     

Mental load and risk in traffic behaviour

Abstract

Car driving means accomplishing a variety of continuously varying driving subtasks which constitute workload on the driver. Total workload can be analysed by type; for instance, the amount of information to be processed while driving, or the effort of car control. One of the driver's main tasks is to cope with the hazards with which he can be confronted on each particular route. This task places demands on his mental capacities. Therefore, coping with hazards is part of the total workload of car drivers. Based on this premise, the following topics are discussed in this paper.

(1)The mental-load approach in modelling traffic behaviour is described in some detail.

(2)Procedures and results of investigations into load factors in car driving are presented, emphasizing workload by reference to hazards. These include a job-analytic study of driving behaviour, a simulation study of hazard perception, and a field study of drivers' exposure to different road conditions.

(3)Hazards and risks must be perceived before making adequate decisions. Some aspects of hazard perception are therefore mentioned.

(4)To understand ‘risky decision making’ in car driving it is necessary to consider the opportunities drivers have to engage in risky situations. Those opportunities are discussed.

(5)To complement these discussions of traffic behaviour, certain issues of risk-taking behaviour in non-traffic situations are considered in order to assess whether knowledge from these areas can be applied to driving.     

基于深度学习的驾驶员换道行为预测

车道变换在交通安全中起着至关重要的作用,准确预测驾驶员的车道变换行为可以显著提高驾驶安全性.本文提出了一种基于全连接神经网络和循环神经网络的混合神经网络,用于精准预测车道变换行为.并且提出动态时间窗口,提取包括驾驶员生理数据和车辆运动学数据的车道变换特征.最后,通过真实交通场景下的数据验证了所提出模型的有效性.此外,将所提出的模型与五种其他预测模型进行了比较,结果表明,与其他模型相比,本文所提出的预测模型具有更高的精确率和前瞻时间.     

基于道路摩擦系数估计的智能小车紧急避障策略研究

雾霾环境下驾驶员的视野受到限制,无法准确估计周围的环境信息,对行车安全具有重大影响.自主紧急制动(AEB)系统是一种重要的车辆主动安全功能,用来避免碰撞或减轻碰撞程度.通常,AEB系统利用一个碰撞时间TTC衡量与障碍物发生碰撞的危险程度.通常设计用于制动的TTC门槛值时假设道路摩擦系数为常数,然而,道路情况复杂多变,道路摩擦系数也是变化的,驾驶员在雾霾环境下更难准确估计道路摩擦系数.因此,开发了一个考虑不同摩擦系数对TTC门槛值影响的AEB控制策略.首先用一个复合滑移率轮胎模型来估计峰值道路摩擦系数,再用该系数计算TTC的门槛值,进而利用该Trc门槛值衡量与障碍物发生碰撞的危险程度.因为可以实时识别道路摩擦系数,提出的AEB策略可以自适应雾霾环境下不同的道路表面.仿真结果表明了该方法的有效性.     

Real-time driving danger-level prediction

This paper introduces a driving danger-level prediction system that uses multiple sensor inputs and statistical modeling to predict the driving risk. Three types of features were collected for the research, specifically the vehicle dynamic parameter, the driver's physiological data and the driver's behavior feature. To model the temporal patterns that lead to safe/dangerous driving state, several sequential supervised learning algorithms were evaluated in the paper, including hidden Markov model, conditional random field and reinforcement learning. Experimental results showed that using reinforcement learning based method with the vehicle dynamic parameters feature outperforms the rest algorithms, and adding the other two features could further improve the prediction accuracy. Based on the result, a live driving danger-level prediction prototype system was developed. Compared to many previous researches that focused on monitoring the driver's vigilance level to infer the possibility of potential driving risk, our live system is non-intrusive to the driver, and hence it is very desirable for driving danger prevention applications. Subjective on-line user study of our prototype system gave promising results.     

A review of: “Indoor air pollutants: exposure and health effects; Euro Reports and Studies 78” (1983)

This study investigated whether gradual or abrupt music change towards more calming music is most effective in calming drivers during high-demand driving situations. Twenty-eight participants were subjected to two types of music change (gradual, abrupt) in a within-subject design. First, a relatively happy mood was induced with personally selected music during an eight-minute simulated high-demand drive. The drive then continued and the mood was changed either gradually or abruptly. Subjective results showed successful music mood induction irrespective of gradual or abrupt changes. The results further showed lower skin conductance (less arousal) and more facial corrugator muscle tension (more sadness) during the abrupt music change. Fewer accidents occurred during the abrupt music mood change. To conclude, the results support the abrupt way of changing music type to down-regulate one's mood: during high-demand driving, abrupt changes in music led to more physiological calmness and improved driving performance, and were thus safer and more effective.

Practitioner Summary: The current study shows that during high-demand drives, drivers are calmed more effectively using abrupt music changes compared to gradual music changes. This is illustrated by reductions in physiological arousal and improved driving behaviour. Hence, in-car music presentation can be used as a tool to improve driver's mood and behaviour.     

Development of statistical geometric models for prediction of a driver's hip and eye locations

The Society of Automotive Engineers (SAE) J1517 and J941 models of a driver-selected seat position and a driver's eye location mainly rely on their statistical linear relationships with seat configuration and package variables. Although the SAE models are useful for vehicle interior design, their prediction performance was not provided. The present study was intended to develop accurate prediction models of a driver's hip location (HL) and eye location (EL) based on their statistical geometric relationships with anthropometric dimensions and driving postures. A driving simulation experiment was conducted with 40 Korean drivers (20 males and 20 females) in a seating buck reconfigurable to various package conditions. The anthropometric measurements, HLs, ELs, and joint angles of the participants were collected using an anthropometer, a motion capture system, and a digital human model simulation program. Two types (full model and simplified model) of statistical geometric models (SGMs) for HL and EL prediction were developed by multiple regression analysis of the anthropometric measurements and driving postures on the HLs and ELs. The average adjusted R² and RMSE of the SGMs were .82 (± .06) and 25.7 (±3.3) mm, respectively. The SGMs showed accurate and stable prediction performance because the SGMs additionally incorporated the geometric relationships of HL and EL with anthropometric dimensions and joint angles. The SGMs would be useful to predict the HLs and ELs of drivers with various body sizes and joint angles in occupant packaging.     

Partially automated driving has higher workload than manual driving: An on-road comparison of three contemporary vehicles with SAE Level 2 features

Vehicles with SAE Level 2 automated features are already in active use on the road, and vehicles with Level 3 or 4 will be with us soon. Although the vehicles provide support for longitudinal and lateral control, partially automated driving experience is sometimes more demanding than manual driving. However, the effects of automated driving on workload in naturalistic conditions have not been extensively investigated, as most studies have been undertaken in driving simulators. This study aims to extend the current understanding about workload in partially automated driving on public roads. Drivers' perceived workload was assessed after conducting manual and automated driving activities using a small sample (N = 8). They performed driving tasks in three contemporary vehicles with SAE Level 2 features, in highway and urban environments. The comparative findings revealed that drivers' perceived workload was higher in partially automated driving than manual driving. Furthermore, perceived workload was higher in urban environments than highway environments and in less experienced drivers than more experienced drivers. Although the findings may need to be interpreted with caution due to the small sample size, they provide a future research agenda that can be built upon.     

A hierarchical fuzzy system for modeling driver’s behavior

The study of human behavior during driving is of primary importance for improving the driver??s security. In this study, we propose a hierarchical driver_vehicle_environment fuzzy system to analyze driver??s behavior under stress conditions on a road. We include climate, road and car conditions in fuzzy modeling. For obtaining fuzzy rules, experts?? opinions are benefited by means of questionnaires on effects of parameters such as climate, road and car conditions on driving capabilities. The number of fuzzy rules is optimized by Particle Swarm Optimization (PSO) algorithm. Also the frequency of pressing on brake and gas pedals and the number of car??s direction changes are used to determine the driver??s behavior under different conditions. Three different positions are considered for driving and decision making; one position in driving lane and two positions in opposite lane. A fuzzy model called Model 1 is presented for modeling the change of steering angle and speed control by considering time distances with existing cars in these three positions, the information about the speed and direction of car, and the steering angle of car. The behaviors of different drivers under two stress conditions are investigated. Also we obtained two other models based on fuzzy rules called Model 2 and Model 3 by using Sugeno fuzzy inference. Model 2 has two linguistic terms and Model 3 has four linguistic terms for estimating the time distances with other cars. The results of three models are compared. The comparative studies have shown that simulation results are in good agreement with the real world situations.     

双车道公路弯道驾驶员轨迹决策行为研究

车辆行驶轨迹是道路路线安全性和舒适性评价的重要指标之一。采用计算机仿真技术,模拟驾驶员操纵车辆行驶时的轨迹决策行为,研究在自由流交通状态下,道路几何线形对车辆行驶轨迹的影响规律。为此,在郭孔辉院士提出的预瞄最优曲率模型的基础上,引入了SK道路几何模型,建立了实际道路条件输入的驾驶员方向控制模型。得到以下主要研究结论：当驾驶员轨迹决策行为只受道路本身条件影响时,且车辆在进入曲线段时,行驶轨迹存在着朝曲线内侧偏移的运动趋势;在出曲线段时,车辆行驶轨迹趋向朝曲线外侧偏移。平曲线半径大小对驾驶员轨迹决策行为有着显著的影响,圆曲线半径越大,行车轨迹侧向偏移量越小,反之越大。