【作者】 张伟；

【导师】 罗大庸；

【作者基本信息】 中南大学 ， 交通信息工程及控制， 2013， 博士

【摘要】 摘要：随着人类社会的老龄化及文化发展的多元化,交通事故的主观诱因日趋复杂,这些诱因不仅包含传统意义下的疲劳及分神行为,同时包括饮酒、吸毒、疾病、药物等新型诱因。这些诱因致使驾驶主体进入多诱因非合适驾驶状态(Various Inducement Unfitted Driving States,VIUDS),降低了对驾驶环境的认知能力,并最终导致交通事故。目前,传统的方法只能实现单一诱因的检测,而无法在统一的技术构架下实现多种诱因的同时检测。针对这一问题,本文以生物医学及认知科学的最新研究成果为基础,在分析了多种主观诱因对驾驶主体中枢神经系统认知能力负面共性影响的基础上,选取了面部表情及脑前额叶血氧饱和度作为基本特征量,通过贝叶斯网络实现了认知能力的融合评估,并最终在同一技术构架下实现了VIUDS多诱因的统一检测。本文的主要创新如下：1、在分析交通事故多种主观成因的基础上,得出了在多诱因作用下VIUDS本质上是中枢神经系统认知能力下降的状态；在分析基于单特征量及现有检测方法局限性的基础上,选取了面部表情及脑血氧饱和度作为反映认知能力的基础特征量,为认知能力的融合评估奠定了基础；2、提出了一种认知能力的评估方法,该方法首先采用去趋势波动分析的方法计算脑额叶血氧饱和度时间序列的标度指数(Cerebral Oxygenation Saturation Scaling Exponents,COSSE),并通过该指数实现了主体中枢神经系统认知能力的分阶评估；3、提出了一种基于石英音叉及椭球面声聚焦光声传感结构的脑额叶血氧饱和度检测方法。在检测过程中,额叶组织及石英音叉分别位于椭球面的两个焦点,在光激励脉冲的刺激下,焦点处额叶组织产生的弱光声信号经椭球面聚焦后被另一焦点处的石英音叉以共振的方式拾取,最后根据激励光强、光声信号强度及血氧饱和度之间的定量关系,可实现前额叶脑血氧饱和度的远场无创测量；4、提出了一种正交流形保持投影的表情方法。针对流形保局投影方法的局限性,即不能在低维投影空间保持流形全局结构的不变性及投影矩阵非正交的问题,重新定义了目标约束函数,并采用格拉姆-施密特正交化过程重新求解投影矩阵,不仅在投影空间有效地保持了流形的全局及局部结构,同时有效地降低了特征维数,提高了表情识别的准确率；5、提出了一种基于贝叶斯网络的认知能力多特征融合评估方法,该方法采用脑额叶血氧饱和度时间序列的标度指数、疲劳表情、眨眼的频率等作为融合的基本特征,通过贝叶斯网络计算了这些独立随机量对应的后验概率,并采用后验概率作为认知能力的评估依据,从而有效地克服了采用单特征进行检测的局限性,提高了评估的精度。论文在最后一章对全文工作进行了总结,并指出了后继的工作及研究方向。本文包括图33幅,表10个,参考文献159篇。更多还原

【Abstract】 Abstract:With the aging of human society and the diversity of cultural development, the subjective reasons of traffic accident become more and more complex, which include not only the traditional fatigue and distraction behavior, but also alcohol, drugs, disease and medicine etc. All these incentives take the drivers into Various Inducement Unfitted Driving States(VIUDS), which reduces the driver’s cognitive abilities of the driving environment, and eventually leads to the traffic accidents. At present, the traditional methods can only detect a single inducement but not all the inducements under the unified technical architecture at the same time. In this dissertation, in order to solve the problem, the commonality negative effects of cognitive ability of the central nervous system was analyzed based on the latest research of biomedical and cognition. By selecting the facial expressions and the prefrontal cerebral oxygen saturation as the basic characteristic to implement the fusion assessment of cognitive ability with Bayesian networks, the uniform detection of the inducements of VIUDS was ultimately achieved. The main innovations are summarized as follows.1. The nature of VIUDS under various subjective inducements is found to be a status of cognitive decline of central nervous system based on the analysis of the traffic accidents. The facial expressions and cerebral oxygen saturation are selected as the basic characteristic of the reflection of the cognitive ability after analyzing the limitations of the detection with a single characteristic and the existing methods, which establishes the foundation for the fusion evaluation of the cognitive ability.2. The assessment method of the cognitive ability was proposed. In this method, the prefrontal cerebral oxygenation saturation scaling exponents (COSSE) of the prefrontal blood oxygen saturation time series is calculated by the detrended fluctuation analysis method. As a result, the evaluation of cognitive abilities of driver’s central nervous system is ultimately realized.3. A detection method of the prefrontal cerebral oxygenation saturation was proposed based on the photoacoustic sensing architecture with quartz tuning fork and ellipsoid acoustic focusing. In the process of detection, the prefrontal organization and the quartz tuning fork are located at the ellipsoid’s two different focuses. Under the stimulus of the excitation laser pulse, the prefrontal organization at one of the focuses generates the weak photoacoustic signal, which is focused by the ellipsoid and collected by the quartz tuning fork located at the other focus. At last, the far-field non-invasive measurement of the prefrontal cerebral oxygenation saturation is fulfilled according to the quantitative relationship between the excitation light intensity, ultrasonic intensity and the oxygenation saturation.4. An orthogonal manifold preserving projections method was proposed. In order to solve the limitations of the locality preserving projections method that can’t preserve the invariance of the global manifold structure in the low dimensional space, and the projection matrix is non-orthogonal, an orthogonal manifold preserving projections method was proposed by redefining the objective constraint function and recalculating the projection matrix by Gramm-Schmidt orthogonalization process. This improvement does not only effectively maintain the global and local manifold structure in the projection space, but also reduces the characteristic dimension and improves the accuracy of the expression recognition.5. A multi-feature fusion evaluation method of the cognitive ability was proposed based on the Bayesian network. In this method, COSSE, fatigue expressions and blink frequency are used as the basic characteristics for the fusion detection. The posterior probability corresponding to these random variables is calculated by the Bayesian network. Using the posterior probability as the basis for the cognitive ability assessment, the detection limitations with the single feature is overcome and the precision of the evaluation is improved.In the last chapter, the contents of this dissertation were summarized and the future research direction and work were also discussed. This dissertation contains33figures,10tables and159references.更多还原