融合面部特征的机动车驾驶人疲劳检测

为降低因驾驶人疲劳驾驶导致的交通事故,需要开展驾驶人疲劳检测研究.为满足在线实时检测的要求,本文提出了融合面部特征的机动车驾驶人疲劳检测方法,首先通过背景差分缩小检测区域、减少图像金字塔层数等方法对MTCNN人脸检测网络进行优化加速,加速后的速度与之前相比提升了258％.其次通过多级级联的残差回归树对人脸进行特征点检测,得到了人脸的特征点,最后通过融合面部嘴、眼开合度特征的方式建立驾驶人疲劳检测模型并进行训练.实验表明,该检测方法的准确率可达95.4％,每帧检测平均速度64 ms,检测速度快,能满足实时性的要求.     

融合面部特征的机动车驾驶人疲劳检测

为降低因驾驶人疲劳驾驶导致的交通事故,需要开展驾驶人疲劳检测研究.为满足在线实时检测的要求,本文提出了融合面部特征的机动车驾驶人疲劳检测方法,首先通过背景差分缩小检测区域、减少图像金字塔层数等方法对MTCNN人脸检测网络进行优化加速,加速后的速度与之前相比提升了258％.其次通过多级级联的残差回归树对人脸进行特征点检测,得到了人脸的特征点,最后通过融合面部嘴、眼开合度特征的方式建立驾驶人疲劳检测模型并进行训练.实验表明,该检测方法的准确率可达95.4％,每帧检测平均速度64 ms,检测速度快,能满足实时性的要求.     

一种基于移动机器人视觉的服务目标人确认算法

本文提出一种基于移动机器人视觉的服务目标人的确认算法.首先提出采用基于标准混和高斯模型与人眼特征相结合的方法检测人脸,提高了光线不均情况下人脸的检测率;在检测到人脸的基础上,再采用基于Gabor特征变换法的HMM人脸识别方法确认指定的服务目标,采用几何标准化方法将检测到的目标人脸进行标准化处理,然后对检测到的每个人脸分块并提取其Gabor、特征,建立HMM模型.通过大量的实验验证了目标人寻找方法不仅对本实验室采集的人脸库有较好的识别效果,而且在线实时验证也取得较好的效果.     

基于Harris角点检测算法的红外人脸器官定位

本文提出了运用Harris角点检测的算法来进行红外人脸的特征点定位。实验结果表明,这种算法能很好地检测出人脸器官的特征点,有效地进行人脸特征的定位,进而达到很好的检测效果。     

基于支持向量机的彩色图像人脸检测

本文提出了一种彩色图像中的人脸检测方法,该方法先通过综合定位皮肤区域的一维Haar小波变换和幅度投影来进行特征判别分析(DFA),然后再利用支持向量机(SVM)分类器和贝叶斯统计分析法来区分人脸和非人脸.实验结果表明,该算法鲁棒性强,正确检测率高.     

改进的高斯肤色模型及其在人脸检测中的应用

针对传统高斯肤色模型在肤色和光照变化较大情况下不能有效提取肤色区域的问题,提出一种改进的高斯肤色模型,并将其应用于人脸检测中。模型参数采用一种自适应更新的参数选择方法,通过对相似度人脸和灰度人脸在对应像素点加权相乘的方式,得到将肤色相似度信息和灰度分布信息有效结合的人脸肤色模型,并结合Adaboost算法设计了人脸检测方法。在FERET(facial recognition technology database)、LFW(labeled faces in the wild)、GTFD(Georgia Tech face database)和多人脸图库上的实验结果表明,该模型的肤色提取正确率比传统高斯肤色模型提高了27.1%,提出的人脸检测方法的检测率比Adaboost算法提高了5.5%。     

用DSP实现实时人脸检测

介绍了人脸识别技术中人脸检测的基本原理,分析了现有基于PC机实现方法的不足,提出了基于TMS320C62xxDSP实现人脸检测算法,阐述了用CCS中对C程序进行性能优化的方法.这一实现方法降低了人脸识别系统对CPU和内存的要求,从而解决了人脸检测实时性要求这一技术难题.     

基于Haar分类器和AAM算法的人脸基准点定位

人脸基准点定位可应用于人脸识别、疲劳检测等领域。针对人脸基准点定位中常用的主动表观模型(AAM)的局限性,提出了Haar分类器和AAM算法相结合的人脸基准点定位方法。先是计算图像积分图,然后采用基于Haar特征的AdaBoost级联检测器快速定位出人脸区域,最后将检测到的位置和图像信息传递给AAM进行人脸基准点定位。该方法在抽取的AFLW(annotated facial landmarks in the wild)人脸测试集上表现出良好的性能。实验结果表明,采用该方法能准确、快速定位出人脸基准点。     

基于注意力模型的人脸关键点检测算法

人脸关键点定位因受到表情、光照、姿态等的影响,常常会出现大的误差。为了准确地定位到人脸的关键点,提出了一种基于注意力模型的人脸关键点检测算法。先是利用可变型模型(DPM)算法检测出图片中的人脸区域,然后结合残差网络(ResNet)和收缩激励网络(SeNet)对该区域进行人脸关键点定位。实验结果表明,该算法在人脸数据集上获得了较高的准确率,证明了该算法的有效性。     

融合Adaboost和光流算法的视频人脸实时检测

提出一种基于AdaBoost和光流的融合算法实现视频人脸的实时检测,利用AdaBoost算法对人脸进行检测,结合光流算法对视频图像帧间的连续性进行估计,获得人脸在图像中的位置.实验证明,本文介绍的人脸检测和追踪的方法利用了Ada-Boost和光流算法各自的优点,一定程度上解决了人脸偏转、部分遮挡、表情变化以及光照等因素对人脸检测的影响.     

基于遗传神经网络的智能复合材料损伤检测传感器位置优化的研究

具有损伤自检测功能的智能复合材料是一个多传感器体系结构,对其传感器进行数目及位置优化,具有重要的实用价值,值得深入研究。采用神经网络建立了复合材料冲击损伤检测方法,运用遗传算法并结合神经网络对复合材料损伤检测的3个传感器布置进行了优化,结果得到了穷举法的验证。该遗传神经网络方法具有一般性,可有效地推广到类似的更多传感器位置优化问题。     

基于最小二乘的极低频微弱信号实时检测方法

极低频信号由于对金属介质具有良好的穿透性,现被应用于管道机器人的示踪定位系统中.实现极低频信号的可靠检测是保证示踪定位系统正常工作的关键,现有检测系统中使用的检测方法易受噪声干扰,经常出现误报与漏报,可靠性低.根据实际工程中极低频信号动态范围大、观测时间有限的特点,提出了一种基于最小二乘的极低频信号实时检测方法,构造了检验统计量并通过MATLAB仿真分析了窄带高斯噪声情况下的检测性能.以DSP为核心设计实现了基于最小二乘的极低频磁场信号检测系统,通过仿真与工程实验验证了所提出方法的优越性.在信噪比为3 dB,恒虚警概率为1%的条件下,检测概率可达90%,完全满足管道机器人示踪定位系统中极低频信号实时检测的要求.     

Detection of abrupt change and applications in sensor decalibration monitoring

The primary objective of fault detection is to detect abrupt undesirable changes in a process at an early stage. This early detection has a potential of preventing loss of production and equipment damage due to these undesirable changes, thus reducing process downtime. This paper details the implementation of some parametric fault detection techniques for sensor decalibration monitoring. A parametric fault detection approach that is handled in depth in this paper is the local approach. This approach developed by Benveniste, Basseville, and Moustakides [Benveniste, A., Basseville, M., and Moustakides, G., The asymptotic local approach to change detection and model validation. IEEE Trans. Autom. Control AC-32 (7), 583-592 (1987)] offers a computationally inexpensive way to attain the objective of monitoring changes in model parameters. However, the algorithm in its original formulation is not applicable to certain processes such as sensors. Therefore, the local approach is coupled with other estimation algorithms such as the input independent Kalman filter to derive a robust sensor decalibration monitoring algorithm. The proposed fault detection algorithm is applied to a pilot scale process for evaluation of its performance.     

基于离群算法组合曲面特征点提取的研究

为了克服目前组合曲面提取特征点算法中阈值选取困难导致边界特征点误判的缺点，在对组合曲面特性进行分析的基础上，提出了一种基于离群算法的组合曲面特征点提取算法。该算法根据曲面特性定义了曲面域和曲面域深度，在空间统计学基础上引入正态分布的标准单位数和置信系数，采用空间数据挖掘中的离群算法提取组合曲面特征点。通过在某型摩托车零件中的应用，表明了该方法可以有效地避免阈值选取问题，且证明了该算法的有效性和实用性。     

An experimental study on damage detection of structures using a timber beam

Using vibration methods for the damage detection and structural health monitoring in bridge structures is rapidly developing. However, very little work has so far been reported on timber bridges. This paper intends to address such shortcomings by experimental investigation on a timber beam using a vibration based method to detect damage. A promising damage detection algorithm based on modal strain energy was adopted and modified to locate/evaluate damage. A laboratory investigation was conducted on a timber beam inflicted with various damage scenarios using modal tests. The modal parameters obtained from the undamaged and damaged state of the test beam were used in the computation of damage index, were then applied using a damage detection algorithm utilising modal strain energy and a statistical approach to detect location of damage. A mode shape reconstruction technique was used to enhance the capability of the damage detection algorithm with limited number of sensors. The test results and analysis show that location of damage can be accurately identified with limited sensors. The modified method is less dependent on the number of modes selected and can detect damage with a higher degree of confidence.     

Quantitative damage detection for planetary gear sets based on physical models

Planetary gear set is the critical component in helicopter transmission train,and an important problem in condition monitoring and health management of planetary gear set is quantitative damage detection.In order to resolve this problem,an approach based on physical models is presented to detect damage quantitatively in planetary gear set.A particular emphasis is put on a feature generation and selection method,which is used for sun gear tooth breakage damage detection quantitatively in planetary gear box of helicopter transmission system.In this feature generation procedure,the pure torsional dynamical models of 2K-H planetary gear set is established for healthy case and sun gear tooth-breakage case.Then,a feature based on the spectrum of simulation signals of the dynamical models is generated.Aiming at selecting the best feature suitable for quantitative damage detection,a two-sample Z-test procedure is used to analyze the performance of features on damage evolution tracing.A feature named SR,which had better performance in tracking damage,is proposed to detect damage in planetary gear set.Meanwhile,the sun gear tooth-chipped seeded experiments with different severity are designed to validate the method above,and then the test vibration signal is picked up and used for damage detection.With the results of several experiments for quantitative damage detection,the feasibility and the effect of this approach are verified.The proposed method can supply an effective tool for degradation state identification in condition monitoring and health management of helicopter transmission system.     

生乳中次氯酸钠的检测研究

建立了生乳中次氯酸钠的检测方法。试样中次氯酸根在弱酸性条件下与碘化钾生成游离碘,以淀粉溶液为指示剂,硫代硫酸钠标准滴定溶液滴定,计算氯的含量。方法处理简单,易操作,回收率高,试验成本低,分析周期短,因此该方法可作为生乳中次氯酸钠的快速检测方法。     

Underwater cable detection in the images using edge classification based on texture information

In this paper, a new approach is proposed for detection of an underwater cable, which makes an Autonomous Underwater Vehicle (AUV) capable for automatic tracking. In this approach instead of traditional image segmentation, first, edges of the images are extracted. Then they are classified using Multilayer Perceptron (MLP) neural network and Support Vector Machine (SVM) using texture information. Then the edge points belonged to the background information are removed and the remaining ones are used for the next processes. Finally, the filtered edges are repaired by morphological operators and are fed into the Hough transform for cable detection. Some texture information methods are used for feature extraction but the results confirm that the 2D Fourier transform in combination with MLP network is the best method for edge classification in this environment. Hough transform, is used in two strategies, which in the first one, the whole information of the edges in the image, are used for line detection, and in the second approach because of curve like shape of the cable, a center part of the image, is used for line detection. In the experiments, many different scenes was used for testing the cable detection algorithm, which first method, resulted to good accuracy but the second one, provided better recognition rate for the cable detection task.     

结构化道路中车道线的单目视觉检测方法

车道线实时检测是智能车辆视觉导航系统中的重要研究内容.提出了一种基于单目视觉的车道线实时检测方法.首先用Canny边缘检测方法对给定图像进行边缘检测.然后提出一种基于方向优先级的车道线搜索方法,分别对左右车道独立完成搜索,在增强车道线特征的同时削弱其他边缘特征.接下来用霍夫变换计算每条线段的直线度,过滤具有复杂纹理的边缘.最后利用图像的亮度信息及其变换识别出车道线.对提出方法进行了实验验证,结果显示提出方法定位准确,在PIII 933 MHz的CPU上的处理速度平均达每秒13帧,能够满足车辆驾驶的安全性和实时性要求.     

大型复杂结构损伤识别两步法研究

基于模型修正技术及模态柔度曲率差方法提出了一种解决大型复杂结构损伤识别问题的两步法。首先运用基于模型修正的损伤识别方法进行模糊识别,通过建立带约束边界非线性最小二乘目标函数,极小化结构实测模态与解析模态之间的误差,将损伤识别问题转化为优化问题,并采用信赖域方法求解该优化问题,识别出损伤所属单元组。然后运用模态柔度曲率差方法,对损伤进行精确定位。对某导弹发射台骨架的数值仿真及试验研究结果表明,该损伤识别两步法识别效果较为理想,为解决大型复杂结构的损伤识别问题提供了新思路。