结合计算集成成像与相关器对局部遮挡三维目标识别方法研究

Object recognition and tracking is the hot topic in the research of military defense and civil security. In some occasions, the interested target may be partially occluded by some barriers, which will result it hard to be recognized based on 2D image processing due to the noise of interference. This project will study the theory and method research of partially occluded 3D object recognition based on computational integral imaging and joint transform correlator. We will set up and optimize the integral imaging system, enlarging the viewing angle and depth of focus of the imaging system, to obtain the high quality of elemental images. We will study the effective method of removing the barrier information in the elemental images according to the statistical characteristics of the barriers and the occluded targets and the disparity between the elemental images. A high efficient reconstruction method will be researched to obtain the reconstructed images of the target with high resolution. Finally we will study the 3D target image recognition based on joint transform optical correlator. Occasions when the target object has rotation, scaling and translation relative to the reference object will be studied. The filter and discrimination function will be optimized. We will study how to improve the recognition accuracy, speed and robustness. This project will provide some theoretical basis to the 3D recognition of the partially occluded targets, and have significance to the defense and civil applications.

目标识别与跟踪是当前军事国防与民事安防监控研究的热点。在某些场合下,感兴趣的目标物会被障碍物局部遮挡,使得基于二维图像的识别在受到干扰噪声影响下变得困难。本项目将研究基于计算集成成像技术和联合变换光学相关器的局部被遮挡三维目标识别理论与方法。建立及优化集成成像系统,扩大成像系统的视场,增加成像系统的景深,得到高质量的元素图像阵列；根据障碍物与目标物的数理统计特征及元素图像间的差异性，研究有效去除障碍物的算法，研究高效快速重建方法,得到高分辨率的目标重构图像；研究基于联合变换光学相关器的三维目标图像识别技术，研究当目标物与参考物之间存在旋转、缩放及平移变换时的识别,优化滤波器及判别函数,增强目标识别的精度、速度及鲁棒性。本项目的顺利研究将为部分被遮挡三维物体的识别奠定一定的理论基础，对于国防和民事应用均具有重要的意义。

实现物体跟踪无目标识别是当前热门的研究方向，但在物体被部分遮挡的情况下实现二维追踪会非常困难。本项目实现了基于集成成像的高质量图像获取，建立自由曲面微透镜阵列集成成像系统，获取参考物与目标物的元素图像阵列，优化光学成像系统，提高成像系统的视场及景深，提高元素图像的成像质量；对元素图像阵列进行特征分析、特征增强，分析元素图像阵列间的特性与约束，计算视差场信息，去除遮挡物体的影响；优化计算成像重构算法，得到无障碍物遮挡的三维目标物图像。. 为了进一步提高重建三维图像质量和效率，本项目研究了基于计算机生成集成成像（CGII）的光子计数集成成像方法。提出一种基于稀疏表示的光场图像超分辨率重建算法，利用光场多视角图像之间的冗余信息对光场图像进行超分辨率重建。研究了三维集成成像技术的景深增强及分辨率增强技术。研究了一种可变空间滤波及中间视图重建技术，用来同时提高集成成像技术中的景深及图像分辨率。与传统的重建方法相比，在能量梯度（EOG）和Tenengrad值方面，所提算法能够提高重建图像质量评价7.63倍和4.81倍。所提算法能够很好地增强集成成像图像质量与景深。