【作者】 陈学军；

【导师】 杨永明；

【作者基本信息】 重庆大学 ， 电气工程， 2011， 博士

【摘要】 研究和实现眼震分析的方法随着科学技术的发展而呈现出多样化,相继出现了裸眼检查法、Frenzel氏眼镜检查法、眼震电图描记法、视频眼震图法。前两种方法虽然简便,但无法对眼震进行定量分析,而后两种方法可以进行定量分析。目前,这些检测方法常用的使用工具分别有生物电极线圈、电磁线圈、视频眼震图仪。然而,生物电极线圈和电磁线圈都是采集生理电信号,易引起测量误差。视频眼震图仪虽克服了上述的大部分缺点,但本身存在线缆连接、设备大型、价格昂贵,不利于临床应用推广。为此,首次研制了一种基于WIFI的无线视频眼震分析系统。研制包括了基于WIFI眼震视频采集头盔系统的硬件和软件以及基于PC端的眼震软件分析系统。采用8层电路板制作技术,实现头盔系统电路小型化;研究并移植了嵌入式Linux内核,完成了整个嵌入式系统从硬件到软件系统平台的构建。研制的系统解决了眼电图仪和有线视频眼震仪不便于临床中一些动态试验项目的实施,避免受试者被线缆干扰而影响试验结果。研究基于WIFI眼震视频实时传输,利用新一代编码标准H.264对眼震视频进行压缩,通过流媒体协议实现无线眼震视频实时通信。提出了基于C/S模式构建头盔系统与眼震视频分析系统进行通信。针对本系统的眼震视频通信,定义了WIFI通信数据格式,并采用认证方式保证眼震视频安全传输。此外,研究和比较了H.264算法与其它视频压缩算法,并实验验证了其优越性;并基于块匹配的帧间运动估计,主要研究分析了7种运动估计算法,实验比较了它们的PSNR特性和计算搜索数,提出了基于块匹配的自适应十字模式搜索ARPS作为块运动估计快速匹配算法。针对眼震时瞳孔运动轨迹提取,提出了基于形态学和Canny算法的瞳孔中心定位算法和基于形态学和左上邻域法的瞳孔中心定位算法。实验分析了这两种瞳孔定位算法,结果表明两者均能够很好地提取瞳孔运动轨迹,并且运算速度比Hough变换快。同时,针对眼震临床试验过程中,常常采集到的眼震视频图像会有眨眼数据帧,首次提出了基于灰色预测模型对眼震眨眼时的瞳孔中心进行预测,并与卡尔曼滤波处理进行了实验比较。实验结果表明,灰色预测值的相对剩余误差低于3%,预测效果与卡尔曼滤波相当,有的甚至更优,能满足该系统眼震分析需求,有效地消除了眨眼数据,避免了临床重复实验。首次提出了一种基于HHT变换的眼震信号特征分析处理算法。研究不同的眼震信号经过EMD分解和Hilbert变换后,各IMF分量在时域、频域、时频域以及能量谱出现的不同差异,并与小波分解结果进行比较。实验结果表明,该算法可以很好获得不同种类眼震信号特征量,比小波分解能够自适性地实现眼震信号多分辨率分析,为各种类型的眼震信号特征提取提供了保障。为了使系统能够辅助临床医生实现自动识别眼震类型和特征,首次提出了基于RBF神经网络的眼震信号分类识别模型,并进行了初步研究探讨。研究分析并设计了临床实验方法。同时,对所研制的无线视频眼震分析系统进行了神经内科和眼科临床实验,验证了该系统实验样机的性能和有效性。更多还原

【Abstract】 With the development of science and technology, the methods of nystagmus analysis are diversification. Nowadays, there are many methods used in hospital and research institute, including naked eye examination, Frenzel glasses examination, electronystag-mography, videonystagmography. Although the first two methods are simple, they are not quantitative analysis methods of nystagmus. The latter two methods are quantitative analysis of nystagmus. The tools which are usually used for these detection methods include biological electrode, electromagnetic coils and video-oculography. However, biological electrode and electromagnetic coils are based on physiological signal, and easily lead to measurement error. Although the video-oculography overcomes this shortcoming, but it is not conducive to widely used in clinic with calbe, large, expensive and other shortcomings. Therefore, a WIFI-based video-oculography system was proposed and developed.The deveploped system includes hardware and software of WIFI-based helmet system for video capturing, and PC-based software for nystagmus analysis. The 8-layer circuit board technology was used to implement miniaturization for circuit of helmet system. Linux kernel was reduced and transplanted, and then the platforms for hardware and software of the embedded system were built. The developed system solved the shortcomings of cable video-oculography which are not convenience to implement some dynamic clinical tests, and then avoided the interference of cable.In order to transmit video in realtime, the H.264 was used to compress nystagmus video, and then compressed video were tramsmited through streaming media protocol. The client/sever mode for communication between helmet system and PC-based systagmus analysis system was built. The data format for nystagmus video communication was defined. The authentication was used to ensure transmission secure of video nystagmus. The H.264 was studyed and compared with other algorithms, and the experiments were implemented. The results showed that H.264 could provide significantly better compression performance than others. Based on block-matching, seven kinds of motion estimation algorithms were studyed. The experiments were done to compare their PSNR and number of searches. An adaptive rood pttern search algorithm for fast block-matching motion estimation was proposed.To extract the nystagmus trajectory, two pupil center location algorithms were propsed. The first method was based on morphology and left-up the neighborhood, and the sedcond was based on morphological and Canny algorithm. The experiments were done to test the two algorithms. The results showed that both algorithms can well extract the nystagmus trajectory, and the processing speeds were faster than the Hough transform. During the clinical tricals, because the tested object usually blinks, it would be difficult to acquire the pupil center of the blinking video frames. A grey forecasting model was constructed to patch blinking frames. The experiments were implemented and compared with the processed results by the Kalman filter. The results showed that the average relative residual error of predicted value by the Grey forecasting model was lower than 3%, and was similar to the results by the Kalman filter or even better. The method can meet the needs of nystagmus analysis, effectively patch the blink data, and avoid re-testing.A feature analysis algorithm for nystagmus signal based on Hilbert-Huang transform was proposed. After the different nystagmus signals were processed through the EMD decomposition and Hilbert transform, the intrinsic mode functions showed the differences in the domain, frequency domain, time-frequency domain and energy spectrum. The method was compared the results by the wavelet decompostion. The experimental results showed that the method could get good features for different types of nystagmus signal, and achieve better self-adaptive multiresolution than wavelet decomposition. In order to automatically recognise nystagmus, a nystagmus signal classification and recognition algorithm based on RBF neural network was proposed, and a preliminary study and experiment of the algorithm were implemented.The clinical trials methods were analyzed and designed. To test the validity and performance of the developed WIFI-based video-oculography system, the clinical experiments were repectively operated in neurology department and phthalmology department.更多还原