【作者】 梁静坤；

【导师】 徐桂芝；

【作者基本信息】 河北工业大学 ， 电气工程， 2012， 博士

【摘要】 脑机接口技术在康复领域的应用是为那些有肢体残疾但大脑思维正常的人群提供一种辅助控制方式。本文对基于想象驾驶的脑电特征及控制接口系统进行了研究，目的是为有肢体障碍的人群提供一种辅助驾驶方式。为了模拟真实的驾驶环境，实验设计将外界动态的交通场景融入到研究中，同时从驾驶者自身感知包括视觉、听觉及视听多模态感知出发，设计了基于多模态感知的驾驶场景并对采集的相关脑电信号进行特征分析。实验中将受试者本身作为黑盒系统，受试者在交通信息的提示下利用左右手想象运动，实现对车辆或轮椅的启动和制动控制。利用Neuroscan软件，本文对采集的三类感知下的脑电数据分别进行了预处理，并利用公共空间模式（common spatial patterns, CSP）算法进行了特征提取分析和比较。同时，将线性回归模型引入了特征信号的分类中，取得了较好的分类效果。鉴于听觉感知的良好效果，利用SPEC061a单片机及外围系统，设计了基于语音识别技术的脑电驾驶模型，利用语音系统对车辆模型进行命令驱动，并将其作为基于受试者本身的反馈信息，以便及时对错误信息进行调整。全文主要研究工作如下：1、虚拟交通环境的设计采用三维动画软件Maya技术，设计了虚拟的动态交通环境。结合现实交通场景，从视觉、听觉以及视听融合的交通信息出发，分别设计了基于不同感知的三种环境。其中视觉设计采用了红绿灯系统，听觉设计采用刹车鸣笛提示音频系统，而视听环境则采用二者环境的融合。虚拟交通环境的设计，即能够从实际驾驶环境出发，充分考虑人类感知器官对脑电的影响，又能够提高了脑电采集中受试者对实验的沉浸感。2、脑电的预处理脑电的采集和预处理采用美国Neuroscan公司的脑电采集系统，利用该软件进行了去眼电、肌电、工频干扰等预处理，并利用脑电地形图软件进行了定位分析，得出不同感知下的脑电在反应强度和持续时间上存在着不同，听觉感知下得到的响应脑电持续时间较长，超过1s，为本文中脑电驾驶的进一步研究奠定了基础。3、脑电的特征提取选用CSP算法，对预处理的数据分别从时域、频域进行了特征提取。结果表明：听觉感知下得到的脑电信息时域特征明显，在与左右手想象运动相关的电极C3、C4上，均表现出右手任务得到的脑电幅值高于左手任务，且最大幅值差达120μV，且脑电响应时间较长，这一点与脑电地形图分析结果类似。4、分类处理文中将线性回归算法引入到脑电的分类处理中，视觉和视听感知下的最高平均识别率达83.3%，而对于特征比较明显的听觉感知下的脑电数据来说，最高平均识别率达96.67%。从处理结果看，线性回归算法在对听觉感知下脑电数据分类中，取得了较好的效果。5、基于语音识别的脑电驾驶模型的设计根据三类脑电处理结果，针对听觉感知实验获得的良好效果，本文结合语音识别技术，利用SPCE061a单片机系统，设计了以语音作为反馈效果的脑机接口车辆控制模型，实现了利用想象脑电控制车辆前进和停止的目的，为残疾人车辆驾驶提供了实验数据。更多还原

【Abstract】 BCI can provide an auxiliary control mode for the people with physical disabilities, butnormal brain thinking. In this paper, EEG features and control interface system based onmotor imagery under driving behavior are researched. In order to simulate a real driving scene,the experiment designed into the characteristics of the traffic scene and driver’s feelings Thisresearch involves the state of perception of visual, auditory and audio-visual multi-mode, inwhich the related EEG is collected and featured, while the subjects are looked as black boxsystems. In the tips of the traffic information, the subjects imagine the right or left handmovement to the driving behavior to start or brake the vehicle, then the EEG is collected andanalysed.With Neuroscan software, three types of EEG are preprocessed, then the feature isextracted and classified using the CSP algorithm and the linear regression algorithm,respectively. The result is better.Considering the good results of the auditory perception, a BCI system model for drivingis designed with SPEC061a and peripheral systems based on speech recognition technology.The speech system can produce voice command to drive the vehicle, and also can adjust thedriving behavior as a feedback-based on subjects in a timely manner.The main work is as follows:1. Design of virtual traffic environmentA virtual traffic environment is designed by Maya software. Combining of the realistictraffic scenarios, three environments are designed for simultaneous auditory, visual oraudio-visual. The visual design uses a traffic light system, while the auditory mode uses thebrake whistle and audio-visual environment with the integration of both. The design of virtualtraffic environment can give full consideration to the human sensory organs, and also can improve the immersion of subjects which proves a favorable effect on the experiment.2. EEG pretreatmentUsing Neuroscan, the driving EEG is collected and preprocessed, after which the EOG,EMG, and frequency interference are removed. The EEG graph shows the differences inreaction intensity and duration in the three perception mode, while the EEG of auditory modehas a longer duration more than1s, which lays the foundation for further study.3. Feature extractionUsing CSP, the feature of pretreatment data is extraced from the time and frequencydomain. Results: EEG under auditory perception has an obvious feature from the time domain.The signals sampled from C3and C4all shows that the voltage amplitude caused by the rightimagery is higher than the others, the difference between the maximum amplitude is120μV;and the longer duration is similar to the result of EEG graph.4. Classification of EEGUsing linear regression algorithm，the highest average recognition rate is83.3%underthe visual and audio-visual perception, but the rate under auditory perception is96.67%. Thelinear regression algorithm obtains a better result in classifying the data under the auditoryperceptive.5. Design of speech recognition driving modelAccording to the good results obtained from the auditory perceptive, the controllingmodel with moive control and feedback of BCI vehicle is designed using SPCE061a systems.The system is able to achieve the control of vehicle to go forward or stop by motor imagery,which can provide the experimental data for the BCI vehicle.更多还原