基于多源信息融合的非接触式疲劳驾驶检测系统研究

发布时间：2018-08-29 10:10

【摘要】：随着社会的进步和人民生活水平的逐步提高,汽车保有量逐年增加,错综复杂的道路交通环境导致交通事故频繁发生,疲劳驾驶已成为交通事故的主要原因。目前的疲劳驾驶检测装置通常采用接触式生理信号检测驾驶员的疲劳状态,这种方式不仅导致驾驶员感觉不舒适,而且干扰驾驶员正常驾驶。非接触式图像传感器检测驾驶员面部信息的疲劳驾驶检测方法,虽克服了接触式检测的准确度低、可靠性差等缺点,但是容易受到外界环境的干扰。因此,研究多源信息融合的非接触式疲劳驾驶检测装置可以有效减少交通事故。本文设计了基于雷达探测生理信号和方向盘角度检测多源信息融合的检测系统,该系统主要包括多普勒雷达生理信号检测系统、方向盘角度检测系统、驾驶视频信号记录系统。首先根据多普勒雷达探测生理信号原理,设计硬件系统,通过对雷达信号进行预处理、放大、有源带通滤波、模数转换等一系列的信号处理,从中获取与生理信息相关的雷达数字信号。方向盘角度采集系统采用霍尔转角传感器和旋转编码器相结合的方法实现对方向盘角度信号的采集。其次针对雷达数字信号特点,采用FIR和IIR两种不同数字滤波方法对生理信号进行分离,对比分离算法的优缺点,采用零相位IIR滤波算法分离出呼吸信号和心跳信号。最后使用学生T检验法对疲劳驾驶视频信号和同步的呼吸、心跳信号特征值进行统计分析,得到与疲劳程度相关联的信号量,并设计极限学习机前馈神经网络算法识别疲劳程度。提取不同疲劳程度下生理信号和方向盘信号特征值,通过极限学习机算法训练识别,并对疲劳样本数据进行识别测试。学生T检验法统计驾驶员疲劳数据,结论表明驾驶员的呼吸深度与疲劳驾驶具有较强的依赖性。实验数据波形表明方向盘转角变化、驾驶员呼吸幅度和呼吸频率都随睡眠加深趋向减小。选取驾驶员疲劳样本进行识别测试,测试表明疲劳状态下的识别率达到81%。多源信息融合的非接触式疲劳驾驶检测系统具有预测性和较高的识别准确度。
[Abstract]:With the progress of society and the gradual improvement of people's living standard, the number of cars is increasing year by year. The complicated road traffic environment leads to frequent traffic accidents. Fatigue driving has become the main cause of traffic accidents. The current fatigue driving detection device usually uses the contact physiological signal to detect the driver's fatigue state, which not only causes the driver to feel uncomfortable, but also interferes with the driver's normal driving. The fatigue driving detection method based on contactless image sensor to detect driver's face information overcomes the shortcomings of low accuracy and poor reliability of contact detection, but it is easily disturbed by external environment. Therefore, the study of multi-source information fusion non-contact fatigue driving detection device can effectively reduce traffic accidents. In this paper, a multi-source information fusion system based on radar detection physiological signal and steering wheel angle detection is designed. The system mainly includes Doppler radar physiological signal detection system, steering wheel angle detection system and driving video signal recording system. Firstly, according to the principle of Doppler radar detecting physiological signal, the hardware system is designed, and a series of signal processing, such as preprocessing, amplifying, active bandpass filtering, A / D conversion and so on, are carried out. The radar digital signal related to physiological information is obtained from it. The steering wheel angle acquisition system combines Hall angle sensor and rotary encoder to collect angle signal of steering wheel. Secondly, aiming at the characteristics of radar digital signal, two different digital filtering methods, FIR and IIR, are used to separate the physiological signal. The advantages and disadvantages of the separation algorithm are compared, and the breathing signal and the heartbeat signal are separated by the zero-phase IIR filtering algorithm. Finally, the student T-test method is used to analyze the characteristic values of fatigue driving video signal and synchronized breathing and heartbeat signal, and the signal quantity associated with fatigue degree is obtained. And designed the extreme learning machine feedforward neural network algorithm to identify the fatigue degree. The characteristic values of physiological signal and steering wheel signal were extracted under different fatigue degree, and the identification was trained by extreme learning machine algorithm, and the fatigue sample data were identified and tested. The student T test method was used to calculate the fatigue data of the driver. The conclusion showed that the breathing depth of the driver was strongly dependent on the fatigue driving. The experimental data waveforms show that the steering wheel angle changes and the breathing amplitude and frequency of the driver decrease with the deepening of sleep. Driver fatigue samples are selected for identification test. The results show that the identification rate of fatigue state reaches 81%. The multi-source information fusion-based non-contact fatigue driving detection system has the advantages of predictability and high recognition accuracy.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP274

【参考文献】