【作者】 崔彦平；

【导师】 林玉池；

【作者基本信息】 天津大学 ， 测试计量技术及仪器， 2007， 博士

【摘要】 本课题是与二炮某部合作研究的预研项目,研究目的是测量靶场运动目标落地前的姿态与速度。论文对国内外靶场测量的现状和发展进行了阐述,深入分析了现代靶场测量的各种方法,结合二炮某部提出的任务和要求,提出了采用凝视等待式双目视觉摄像机,高速摄像测量回转体类运动目标的姿态与速度,本测量方法可充分利用视觉检测与图像处理等先进测量原理与处理技术,具有系统组成简单、响应速度快、测量精度高等优点。论文首先建立了双目交汇测量系统的数学模型,对双目传感器的布站方式和测量系统相关参数进行了详细的分析。采用RAC两步法摄像机标定技术,对考虑径向畸变的摄像机模型进行了标定。通过仿真实验研究了两摄像机对称放置时,有效视场和成像像素个数与两摄像机的内部参数和结构参数之间的关系,并且完成了双目摄像机标定实验。提出了基于改进BP神经网络的复杂背景下目标识别方法。采用中值滤波滤除图像噪声,用改进的最大方差阈值法进行图像分割。建立了基于BP神经网络的目标识别系统,提取目标七个不变矩特征作为神经网络输入的特征向量,进行目标识别,实验结果说明论文提出的图像预处理方法可有效去除复杂背景图像噪声、准确的分割图像;基于BP神经网络的目标识别方法也获得了较高的识别准确率。研究了基于空间矩的像面直线亚像素提取与空间直线三维重建的新方法。提出了首先利用LOG算子进行快速像素级定位,再在包含边缘点的领域内利用空间矩进行精确定位的两步亚像素定位法。在对边缘点进行去畸处理的基础上,采用Hough变换与最小二乘相结合的方法获得像面直线,由双目视觉实现空间直线重建。实验结果说明,该像面直线提取与空间直线重建方法具有较高的精度。提出了回转体类运动目标落地前姿态与速度的测量方法。通过最小二乘法拟合目标在各个采样时刻的质心,建立了目标在三维空间运动轨迹的最小二乘多项式函数,对多项式函数求导得到目标运动速度。根据回转体类运动目标特点,建立了回转体类目标姿态测量中轴线法的数学模型,并根据目标在摄像机像面成像的大小,提出了不同的姿态测量方案。系统实验结果说明本文提出的姿态与速度测量方法具有较高的精度和稳定性,能够满足运动目标姿态与速度测量的需要。论文对运动目标姿态与速度测量方法从理论和实验两个方面进行了深入研究和探讨,提出了一些创造性的理论和方法,并完成了相应的实验验证工作。更多还原

【Abstract】 The research work of the dissertation is a pre-research project cooperated with PLA Second Artillery Forces to measure the attitude and velocity of moving target before landing. The actuality and development of shooting range measuring in the world are introduced, and the modern measurement methods in shooting range are analyzed in detail. According to the task and request, the binocular intersection measuring system has been set up to detect the attitude and velocity of moving target by high-speed image cameras synchronously. The scheme which takes full advantage of the advanced measuring and processing technique of machine vison and image processing is with simple structure, rapid responding speed, high measuring precision, and so on.The mathematics model of binocular intersection measuring system is set up firstly, and then the distribution modes and correlative parameters of binocular cameras are analyzed detailedly. The camera model taking account of radial distortion and the binocular cameras are calibrated by the RAC technique. The relations between effective viewing and image pixels, as well as inner parameters and structure parameters of binocular cameras are discussed as the two cameras are placed symmetrically, and the binocular cameras are calibrated too.Target recognition in complex background based on improved BP neural network is proposed in the dissertation. Median filter is adopted to get ride of the noise and an improved method of maximum classes square error is used to compute the threshold of the image segmentation. The target recognition system based on improved BP neural network is established as seven invariant moments serve as the input feature vector. The experiment results show that the image noise could be gotten rid of effectively and the image could be segmented exactly by the image preprocessing method put forward in the dissertation, and the recognition system acquires an excellent recognition result.Subpixel detection of beeline on image plane based on spatial moment and 3D reconstruction of spatial beeline is researched. First the LOG operator is used to locate the edge at the level of pixel quickly, and then the spatial moment is used to detect subpixel location of the discrete points of edge. The least square fitting is adopted to fit the subpixel edge points which have been detected by Hough transform, after having gotten rid of the camera lens distortion. Mathematics model of reconstruction of spatial beeline from image lines on two image planes is set up. The experiments show that the detection of beeline on image plane and 3D reconstruction of spatial beeline have acquired high precision.The measuring methods of attitude and velocity of rotary moving target are discussed. The binocular intersection measuring system has been set up near the landing area to detect the landing parameters of moving target by high-speed photograph. The polynomial function of target track 3D curve is fitted with the method of least square according the 3D coordinate of centroid at the different moments, so the moving speed could be obtained by calculating the derivatives of the polynomial function. Based on the characteristic of rotary targets, mathmatics model of axes-rotary method to measure the attitude of rotary targets is established. Three measuring methods were proposed to measure the pose according to the size of images in the camera planes. The systemic experiments show that the method proposed in the dissertation could meet the needs of measuring attitude and velocity of moving target with high precision and stability.Both theoretical and experimental researches are taken, then some creative theories and means are proposed and relevant experiments are completed.更多还原