【作者】 陈杰春；

【导师】 丁振良；

【作者基本信息】 哈尔滨工业大学 ， 仪器科学与技术， 2008， 博士

【摘要】 在任何风洞试验中,精确测量出模型在加载后的实际攻角是提高风洞试验数据准确度的一个重要途径。此外,精确测量出风洞模型的其他运动参数也很重要,比如对于磁悬挂系统来说,需要将悬浮模型的位置和姿态信息作为反馈信号,控制模型稳定悬浮于风洞试验段中。目前,国内的风洞试验多数还在使用加速度攻角传感器和干涉式激光测角仪测量模型攻角,而国外已经将三维视觉测量技术应用于风洞试验中。与传统的测量方法相比,基于三维视觉的风洞模型位姿参数测量方法能够同时测量出模型的位置和姿态。本文试图探索基于三维视觉的风洞模型位姿参数测量方法,并就其中所涉及的几个关键问题进行深入的研究。给出了两种基于三维视觉的风洞模型位姿参数测量方法:基于主动视觉的模型位姿参数测量方法和基于被动视觉的模型位姿参数测量方法,第一种方法可用于测量大型风洞模型的位姿参数,而第二种方法则适合用于测量小型和微型风洞模型的位姿参数。由于这两种方法都以向量的坐标变换为基础,因此首先从欧拉角的定义和坐标变换原理出发,不但给出了模型位姿参数测量系统的结构,还给出了确定风洞模型位置和姿态的解析表达式。对这两种方法分别作了仿真分析,分析结果为测量系统结构参数的优化提供了依据,此外也通过实验对这两种方法进行了验证。对空间点目标的三维重构方法作了较深入的研究,其间提出了一种大视场条件下的摄像机精确标定方法,此外对图像视觉特征提取、立体匹配和三维重构算法也进行了较深入的研究。相比较而言,传统摄像机标定方法可以获得较高的精度,但如何精确测量控制点的3D坐标是应用传统摄像机标定方法的关键,文中给出了一种大视场条件下生成控制点并精确测量其3D坐标的方法。通过评定图像视觉特征检测和空间点目标三维重构的不确定度,可为三维视觉测量系统的优化提供依据。首先提出了一种估计图像视觉特征检测不确定度的新方法,建立了一个像素强度不确定度数学模型,一旦由统计实验确定了该数学模型的未知参数和图像内不同像素之间的强度相关系数,由像素强度不确定度的估计值、不同像素之间的强度相关系数和图像视觉特征提取算子的表达式就可以估计出图像视觉特征检测的不确定度。接着给出了一种估计三维重构不确定度的新方法,该方法同时考虑了摄像机标定过程和重构3D点过程中的不确定度传递,不仅给出了不确定度传递表达式,还用蒙特卡罗方法对该方法作了验证。最后,提出了一种基于遗传算法的摄像机优化布置方法,该方法以空间点目标的三维重构不确定度最小为目标,实现三维视觉测量系统中的摄像机优化布置。更多还原

【Abstract】 In any wind tunnel test, measuring angle of attack accurately is crucial to improve the accuracy of experiment data. In addition, measuring other motion parameters of a wind tunnel model accurately is also important. For instance, a wind tunnel model in magnetic suspension and balance system is suspended stably by using the feedback signal of position and attitude (P&A) signal of wind tunnel model. Accelerometer and interferometric laser goniometer are usually used to measure angle of attack of a wind tunnel model in most domestic wind tunnel test nowadays, whereas stereo vision technology has been applied in wind tunnel test abroad. Comparing with traditional method for measuring P&A of a wind tunnel model, stereo vision based method can meaure P&A of a wind tunnel model simultaneously. This dissertation tries to explore the stereo vision based mehods for P&A measurement of a wind tunnel model, and study on some key problems associated with this method.Two stereo vision based methods for P&A measurement of wind tunnel models are proposed: active vision based method and passive vision based method. The first method is suitable for measuring P&A of huge wind tunnel models, while the second one is suitable for measuring those of miniature or micro wind tunnel models. The configurations of both P&A measurement systems are presented, and the corresponding P&A determination algorithms are also derived from the definition of Eular angles and the principle of coordinate transformation. Experiments with synthetic data have been made, and the experiment results can provide useful guidance for designing of a P&A measurement system. In addition, experiments with real data have also been made to validate the accuracy of both methods.The methods associated with stereo reconstruction of a point target are studied deeply. A mehod for calibrating camera accurately under broad field of view condition is proposed firstly. Then the methods for image feature detection, stereo matching and stereo reconstruction are also studied deeply. Comparing with other methods for camera calibration, the traditional one can meet the requirment of high accuracy. Whether can meaure the 3D coordinates of control points accurately is crucial to use the traditional method for camera calibration. A new approach to setting control points in three-dimensional space and measuring their 3D coordinates in the world coordinate system is proposed in this dissertation.Uncertainty of image feature detection and stereo reconstruction can be regarded as criterions for optimization of a stereo vision based measurement system. A mathematical model to evaluate the uncertainty associated with pixel intensity is poposed firstly. Once the parameters of the model to evaluate the uncertainty of pixel intensity and the correlation coefficients between the intensity of different pixels used to detect an image feature are determined by a statistical experiment, the uncertainty associated with feature detection can be determined according to the analytical expressions of a feature detection operator. Then an approach to evaluating the uncertainty associated with stereo reconstruction is proposed. The uncertainty propagation while calibrating cameras and reconstructing a 3D point from its image pairs are both taken into account, and the analytical expressions to evaluate the uncertainty of stereo reconstruction are presented. At last, genetic algorithm is used to solve the camera optimal placement problem in order to obtain minimal uncertainty of stereo reconstruction, and experiments with synthetic data have also been made.更多还原