基于光子计数的非视域激光成像方法的优化研究

发布时间：2018-03-05 00:21

本文选题：非视域成像　切入点：单光子探测　出处：《哈尔滨工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：非视域激光成像技术是利用散射光携带的飞行时间信息对视域外的目标进行三维像重建的技术。该技术是近几年兴起的一种新型的激光成像技术,在机器视觉、反恐侦查、医学成像、历史考古以及飞行器导航等众多领域有着很强的应用前景。本文从基于椭球逆投射方法的非视域成像出发,理论分析了该方法自身原因导致最终恢复像存在不确定区域的原因。通过仿真分析出该方法存在以下问题:成像体制单一、场景信息利用率低,横纵向分辨率不一致以及滤波算法单一、对噪声敏感、恢复像畸变的问题。之后本文即针对上述问题分别从成像体制和恢复算法上进行了优化。针对逆向椭球投射方法恢复像在横纵向分辨率不一致的问题,根据逆向椭球投射方法叠加时使用的是偏心率较大的椭球面情况,本文提出球投射恢复方法,从三维图像频域空间信息获取范围角度证明球投射方法成像质量比椭球投射高的原因。对于单点目标,同样系统设定情况下逆向椭球投射恢复像横向半峰宽占26个体素,而逆向球投射的恢复像占12个体素。典型的双点目标实验结果也和理论和仿真结果一致。针对非视域激光成像在非视域场景中的信息利用率低的问题,本文提出了使用两个互相垂直的墙面进行双中介面的非视域激光成像的体制。首先分析两个中介面存在时可能存在的4种成像方式,使用双中介面各自恢复像归一化叠加的方式进一步减少恢复像的不确定区域。仿真和实验的结果显示使用双中介面进行非视域成像对提高最终恢复像的位置准确度、强度均匀性有着重要的意义。最后针对椭球投射算法恢复像存在弯曲畸变以及在噪声存在情况下边缘模糊的问题,本文首先引入了聚类算法中的mean shift滤波方法,仿真结果表明使用mean shift滤波方法的逆向椭球投射算法在多目标或者带噪声的情况下具有优秀地聚类、抑制噪声、边缘保持以及保留强度低目标不被阈值掉的特点。之后从非视域激光成像的另一种恢复算法-代数迭代法出发,通过最优化方法中的梯度投影算法恢复图像,通过建立上述算法的仿真得到的恢复像质量优于逆向椭球投射算法,代数迭代法在参数选取合适的情况下恢复像弯曲畸变的情况得到改善。
[Abstract]:Non-visual laser imaging technology is a 3D image reconstruction technology based on time-of-flight information carried by scattered light. It is a new laser imaging technology developed in recent years, which is used in machine vision, anti-terrorism detection. Many fields such as medical imaging, historical archaeology, aircraft navigation and so on have strong application prospects. This paper starts from the non-horizon imaging based on ellipsoidal inverse projection method. This paper theoretically analyzes the reason that the method itself leads to the uncertain region of the final restoration image. The simulation results show that the method has the following problems: the imaging system is single and the scene information utilization ratio is low. The horizontal and vertical resolution is inconsistent and the filtering algorithm is single and sensitive to noise. The problem of image restoration distortion is discussed in this paper. Then, we optimize the imaging system and restoration algorithm for the above problems, and aim at the problem that the resolution of image restoration in horizontal and vertical direction is not consistent with the reverse ellipsoidal projection method. According to the case of ellipsoidal surface with high eccentricity when the reverse ellipsoidal projection method is superimposed, a spherical projection restoration method is proposed in this paper. The reason that the imaging quality of the spherical projection method is higher than that of the ellipsoidal projection is proved from the angle of the frequency domain spatial information acquisition range of the 3D image. For a single point target, the transverse half-peak width of the reverse ellipsoidal projection restoration image accounts for 26 individual elements under the same system setting. The restoration image of reverse sphere projection accounts for 12 individual elements. The typical experimental results of two-point target are in agreement with the theoretical and simulation results. In view of the problem of low information utilization rate of non-visual laser imaging in non-visual scene, In this paper, we propose a method of using two vertical walls to carry out non-visual laser imaging with two mediating surfaces. Firstly, four imaging methods that may exist in the presence of two interfacial surfaces are analyzed. The method of normalized superposition of restoration images with double mediums is used to further reduce the uncertain regions of restoration images. The simulation and experimental results show that the use of double mediums for non-visual imaging can improve the accuracy of the final restoration images. The intensity uniformity is of great significance. Finally, aiming at the problems of bending distortion and edge blur in the restoration image of ellipsoidal projection algorithm, this paper first introduces the mean shift filtering method in clustering algorithm. The simulation results show that the inverse ellipsoidal projection algorithm using mean shift filter has excellent clustering and noise suppression in the case of multiple targets or with noise. Edge preservation and retention of low intensity targets are not lost by threshold. Then, from another restoration algorithm of non-visual laser imaging, algebraic iterative method, gradient projection algorithm in optimization method is used to restore images. The restoration image quality obtained by the simulation of the above algorithm is better than that of the reverse ellipsoid projection algorithm, and the algebraic iterative method can improve the image bending distortion when the parameters are selected properly.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN249

【参考文献】