嵌入式驾驶疲劳检测系统设计

随着机动车交通事故数量的快速增长,交通安全已成为研究的热点.疲劳驾驶是引起交通事故的一个关键原因.为了减少疲劳驾驶带来的损失,论文利用TI公司的DSP芯片TMS320DM642,设计了嵌入式驾驶疲劳检测系统.该系统集图像采集,图像处理,图像识别以及疲劳预警于一体.实验结果表明此设计能进行实时疲劳检测,可以有效地减少交通意外,并具有尺寸小,成本低,实时好的优势.     

疲劳驾驶状态检测系统的设计

司机疲劳驾驶容易引起严重的交通事故,疲劳状态检测系统的研究成为计算机应用研究的重要领域.为了满足疲劳状态检测系统实时性要求,选择实时性较好的Adaboost算法来识别人眼,采用单位时间内眼睛闭合时间所占的百分率来判断疲劳状态,应用该算法后进行疲劳驾驶训练和识别研究.选择DSP移植方案并将疲劳状态检测算法移植到DSP中.经过实时检测实现的系统基本能满足实时检测疲劳驾驶的要求.     

一种改进型KCF的疲劳驾驶检测方法

随着人民生活水平的提高,汽车人均持有量持续上升,从而也就导致了交通事故的增加,根据统计可知,大多数交通事故都是由疲劳驾驶导致的,因此为了减少司机因为疲劳驾驶导致的交通事故,对司机的疲劳程度检测是十分有必要的;目前现有的疲劳检测方法虽然种类较多,但是大多数存在着部署困难、实时性差、检测精度不高等缺点,因此难以在实际生活中应用;为了克服以上缺点,在这里提出了一种基于视觉的实时疲劳检测方法;该方法通过优化KCF算法实现实时人脸跟踪,再利用现有的人脸关键点检测手段,实现对眼睛和嘴巴的状态识别;最后结合SVM实现对驾驶员疲劳状态的检测;实验结果表明,提出的方法具有较好的实时性,并且精度较高,能够满足日常需求。     

驾驶员疲劳检测系统设计

以DM642为主控芯片设计了一套驾驶员疲劳检测的硬件系统,包括主控器模块、视频采集模块、视频输出模块和报警模块等相关电路;综合国内外的研究现状,确定了了疲劳状态判断的理论基础;交叉运用图像处理技术、人脸检测技术和PERCLOS疲劳检测方法,根据眼睛的疲劳特征,实时判断驾驶员的疲劳状态,进行报警,有效防止交通事故的发生;经过对系统的软硬件测试,结果表明,该方案可以有效地识别出驾驶员的疲劳状态,运行速度快、实时性好,具有较高的鲁棒性。     

疲劳驾驶识别算法与技术

针对驾驶员长途驾驶汽车时,会因难以克服的生理疲劳时常发生交通事故的残酷现状,提出一种疲劳驾驶的识别算法,能够有效阻止驾驶员在疲劳状态下,持续行驶的不良行为,以避免事故的发生。该算法通过对驾驶员脸部图像的采集与特征空间建立,采用与疲劳训练样本的类比计算,实时确认当前驾驶状态的类别属性。依据驾驶状态的类别属性即可确定是否需要对驾驶员实施干预。系统识别运算周期小于50ms,识别准确率高于96%。     

基于5G和车联网技术下的疲劳驾驶预警系统

通过逐年报告分析,疲劳驾驶一直是交通事故中的一大重要隐患。本文结合5G和车联网的技术背景,对实时监测疲劳驾驶预警系统进行了一系列设计和构思,从项目概述、实施方案到项目特色和对未来展望进行了深入的探索。     

基于FPGA的多特征融合司机疲劳状态检测系统设计

驾驶员的疲劳驾驶是造成交通事故的重要因素,为了实时有效地检测驾驶员的驾驶状态,设计了一种融合多种疲劳特征进行疲劳状态判定的检测算法,并构建了车载的基于现场可编程门阵列(FPGA)的嵌入式检测平台。该多检测算法融合了眼睛和嘴巴的疲劳特征,当某一特征的检测受到影响时可以使用另外的特征进行疲劳状态的判定,较传统的单一特征疲劳检测算法拥有更高的检测效率。实验结果表明:系统的算法简单、可靠、实时性强。     

驾驶员疲劳检测系统研究综述

实时检测驾驶员的状态,可以避免大量交通事故的发生.通过机器视觉技术可以识别出驾驶员眼睛张开闭合状态,判断驾驶员是否疲劳.基于视频图像分析的驾驶员疲劳检测方法成为当前研究的一个重要方向.本文对国内外驾驶员疲劳检测方法及发展趋势进行了综述.     

校车监控终端系统的设计与实现

针对校车经常发生交通事故的现状,提出了校车监控终端系统的设计方案。该系统以处理器S3C6410为硬件核心,以WinCE6．0操作系统为软件平台,通过USB摄像头实现视频采集并对校车进行无线网络实时监控;该系统也利用MMA7361L加速度传感器和STC12C5410AD单片机采集加速度信号,监控服务器端可通过设定加速度报警阈值来控制加速度的输出,实现了校车的加速度测量与监控。     

基于视频的人眼跟踪方法研究

疲劳驾驶是交通事故发生的主要原因之一.在车载的驾驶员疲劳检测系统中,对于人眼的高效、实时、准确检测和跟踪是关键技术之一.本文研究基于视频信息的驾驶员人脸图像中眼睛的跟踪方法.采用Viola-Jones算法和AdaBoost算法设计了Haar级联人脸分类器,根据人脸的几何结构和眼睛的特性,通过两次定位的方法实现了对眼睛的定位和跟踪.实验表明,本文方法具有检测精度高、鲁棒性强的特点.     

疲劳驾驶多源性智能监测预警方法研究

对疲劳驾驶监测预警方法进行研究,可以避免驾驶员因疲劳驾驶产生的交通事故,减少因疲劳驾驶造成的人员伤亡和经济损失。当前的疲劳驾驶监测预警方法存在监测灵敏度低、可靠性差等问题,不能及时对疲劳驾驶的驾驶员进行报警,来避免交通事故的发生。为此,提出了疲劳驾驶多源性智能监测预警方法,首先将摄像头采集的驾驶员图像进行预处理,通过计算驾驶员图像信息的灰度值,得到驾驶员图像中像素的分布密度,为后续的监测和预警工作提供信息。其次,采用卡尔曼滤波算法对驾驶员的图像信息进行跟踪,得到驾驶员各个时间内的状态估计值,最后,通过计算驾驶员状态估计值判断驾驶员是否存在疲劳状态。实验结果表明,该方法的丢包率低、多源性高、抗干扰能力强、计算效率高。     

基于面部动作时空特征的疲劳预警算法

目前疲劳预警算法多采用实时监测报警的方式,这在高速行驶中具有很大的安全隐患。鉴于人类疲劳状态的时序相关性,提出一种基于面部动作时空特征提取的预警算法。首先,构建加入空间变换结构的卷积神经网络,识别人脸区域,对脸部特征点进行检测标记;其次,建立时空特征提取网络,利用采集的人脸图像序列,对未来图像序列进行预测并输出;最后,在输出的图像序列中根据眼部、嘴部综合状态判断是否发出警告。实验结果表明,以15 fps的速率采集图像,预测未来2 s 30帧图像的方式下,该算法能以90%以上的准确率提前26帧（约1.5 s）预警,且提前15帧（1 s）预警的准确率达到97%。在我国高速公路平均100 km/h的车速下,相当于提前40 m预警,能进一步减少交通事故的发生。     

恶劣光照下驾驶人疲劳及注意力视觉监测算法

驾驶人驾驶行为监测及预警系统对于提高行车安全性及降低交通事故等问题具有重要作用,而利用机器视觉对驾驶人疲劳状态及注意力状态进行监测和分析是安全辅助驾驶领域内的研究热点之一。机器视觉受光照条件影响非常严重,甚至直接影响到图像处理算法的可靠性。在弱光和强光两种特殊照明条件下,介绍了驾驶人疲劳状态的检测思路,提出利用普尔钦斑点及投影曲线极点位置分割面部器官独立区域,最终获得眼睛的轮廓状态,利用PERCLOS判断驾驶人的疲劳状态。在驾驶人面部及面部器官定位的基础上,对驾驶人的头部旋转运动进行分析,提出了计算驾驶人头部旋转角度的计算方法,以驾驶人头部旋转角度为依据判断驾驶人的注意力是否分散。实践证明,在弱光和强光两种特殊照明条件之下,算法实时性好,准确率较高,效果非常理想,能够为驾驶人疲劳状态及注意力状态分析提供较为准确的依据。     

Drivers’ Performance Examination using a Personalized Adaptive Curve Speed Warning: Driving Simulator Study

ABSTRACT

A great number of traffic accidents occur during curve negotiation maneuvers. Most of these accidents could be avoided if drivers are provided with information that better guide them through the maneuver. Advanced Driver Assistance Systems (ADAS) such as Curve Speed Warning (CSW) have shown to be effective to improve safety on curve maneuvers by warning drivers of the speed required to make a safe curve maneuver. The effectiveness of such warning systems can potentially improve if the warnings are not only adapted to road and traffic condition but also adapted to individual drivers’ behavior. In this study, an Adaptive Curve Speed Warning (ACSW) system is developed that presents drivers a two-level visual and audio warning considering the variation in individual drivers’ perception-reaction time (PRT). The warning timing is adjusted according to a reward/punishment function to reinforce safer actions while providing an individualized in-time warning. Next, within a driving simulator environment, drivers’ performance using ACSW is compared to a CSW that does not consider PRT variation among drivers. Further, variation in drivers’ performance on curve maneuvers is discussed with respect to drivers’ approaching speed, variation in PRT, and braking behavior. Results show that drivers’ interaction with warning systems varies significantly based on their age and gender. In addition, results show how drivers approaching speed to a curve varies significantly based on road characteristics such as curve advisory speed and curve direction. Results from this study contribute to the development of more intelligent ADAS that could improve drivers’ comfort and safety.     

Monitoring drivers’ sleepy status at night based on machine vision

Driver fatigue is a chief cause of traffic accidents. For this reason, it is essential to develop a monitoring system for drivers’ level of fatigue. In recent years, driver fatigue monitoring technology based on machine vision has become a research hotspot, but most research focuses on driver fatigue detection during the day. This paper presents a night monitoring system for real-time fatigue driving detection, which makes up for the deficiencies of fatigue driving detection technology at night. First, we use infrared imaging to capture a driver’s image at night, and then we design an algorithm to detect the driver’s face. Second, we propose a new eye-detection algorithm that combines a Gabor filter with template matching to locate the position of the corners of the eye, and add an eye-validation process to increase the accuracy of the detection rate. Third, we use a spline function to fit the eyelid curve. After extracting eye fatigue features, we use eye blinking parameters to evaluate fatigue. Our system has been tested on the IMM Face Database, which contains more than 200 faces, as well as in a real-time test. The experimental results show that the system has good accuracy and robustness.     

基于Spark的电网工控系统流量异常检测平台

针对传统的电力网络流量检测安全预警系统在面对海量高维度数据时，其在精度、实时性、扩展性以及效率上都无法满足需求的问题，建立出一种基于Spark的电网工控系统流量异常检测平台.该平台以Spark为计算框架，主要由数据采集与网络流量深度包检测协议解析模块，实时计算数据分析处理模块，安全预警预测模块和数据存储模块组成，为流量异常检测提出了一套完整的流程.实验结果表明，该平台能够有效地检测出异常流量，做出安全预警，方便工作人员及时做出决策，这充分说明该平台非常适用于电力控制系统，能够应对海量高维复杂数据做出实时分析以及安全预警，极大地提高了电网工控系统的安全性能.     

Si472x射频收发芯片的交通状况提示装置

在汽车拥有量剧增的今天,前方的道路交通状况是人们关注的焦点。现有交通状况提示装置功能较单一,不能反映复杂多变的道路交通状况。为此我们设计了一种采用单芯片射频收发器Si472x系列芯片、高性能C8051F310单片机与录放语音芯片ISD2560相结合的新型道路交通状况提示装置方案,并描述了该新型装置的功能、硬件实现和软件设计流程。     

空管理系统中低高预警算法改进

预警和告警模块是空管系统的重要组成部分,对保证空中交通安全意义重大.及时准确地进行各类预警、告警是该模块的基本要求.本文对基于马赛克检测的低高告警算法做了分析,并针对该算法的不足给出了改进算法.     

Ensemble hill climbing optimization in adaptive cruise control for safe automated vehicle transportation

The Journal of Supercomputing - Road traffic crashes are currently one of the leading causes of death. Many accidents occur when the attention of a driver is diverted owing to fatigue or external...     

Countermeasures to prevent detection failure of a vehicle approaching on collision course

Objective

This study investigated causal factor of perceptual failure and possible countermeasure of crossing path crashes at clear-sighted unsignalised intersections.

Background

Crossing path crashes involving two vehicles at intersections are a common and serious problem, and perceptual failure has been identified as a predominant causal factor. Previous studies have showed that late detection of a crossing vehicle frequently occurs even when there are no visual obstructions, at such as rural intersections.

Method

With using a fixed-based driving simulator, three experiments were performed to investigate a driver’s ability to detect a periphery presented cross traffic while approaching an intersection. In Experiment 1, drivers’ ability to detect crossing vehicles in their peripheral field of view was studied, both in conditions of vehicles following a collision and a non-collision trajectory. In Experiment 2, we examined whether abrupt appearance of a vehicle on collision course would improve detection performance. In Experiment 3, we tested potential of collision warning, if it affects voluntary visual scanning, improving the detection performance regarding hazards cross traffic.

Results

The results of Experiment 1 showed that vehicles on collision course vehicles were detected late. This suggested that the late detection could be related to the lack of motion visible in the peripheral view. In Experiment 2, it was found that abrupt appearance effect (“pop-up” from road side occluding furniture) improves detection performance of a crossing vehicle. The results of Experiment 3 demonstrated that cross traffic collision warnings were beneficial for preventing late detection responses by means of encouraging voluntary visual scanning.

Conclusion

Less attention attractive visual properties of hazardous cross traffic attributed to mutual approaching course can cause recognition failure. Drivers’ gaze shift to potential conflicting direction, either reflective or voluntary manner, is crucially important for preventing crossing path crashes at such as rural intersections.