基于灰度投影的实时电子稳像算法研究

【作者】 汪小勇；

【导师】 冯华君； 徐之海；

【作者基本信息】 浙江大学 ， 光学工程， 2006， 硕士

【摘要】 电子稳像是新一代的稳像方法,它利用计算机图像处理技术实现图像序列的稳定。因为具有稳像精度高、体积小、重量轻、功耗低以及成本低等优点,它在军事和民用中都有广泛的应用前景。实时电子稳像是指用电子稳像的方法及时地稳定图像序列,使得摄像系统和跟踪系统的操作者不会受到摄像机本身运动的干扰,因此实时电子稳像算法具有稳像速度快精度高的特点。 本课题在嘉兴市科技项目资助下,对实时电子稳像技术进行了深入研究,内容涵盖原理探讨、常用算法分析、实时电子稳像算法构建、稳像系统实现、试验论证和稳像评价等方面,取得了一系列成果,部分成果对于实时电子稳像技术的发展有影响作用。 论文分析了图像序列间产生差异的原因,明确了电子稳像的必要性。阐述了电子稳像基本原理、基本方法和处理过程,并讨论其实现的关键技术。提出实时电子稳像算法的要求,并分析了目前国外电子稳像常见算法的特点,选取灰度投影算法作为本文实时电子稳像算法的基础。 建立了电子稳像的软件仿真平台。生成了几种不同类型运动矢量的模拟图像序列,用来分析和验证灰度投影算法的稳像特性,为以后构建实时电子稳像算法奠定了重要的基础。 针对实时稳像中灰度投影算法不能处理的几种情况,如对比度差、内部运动物体和旋转图像序列等。在灰度投影基础上分别构建了新的算法,并用试验证明其可行性。论文首次提出了运用两次匹配的算法稳定内部运动物体的图像序列和运用先补偿平移后补偿旋转的算法快速稳定旋转图像序列。并且也提出了一种判断摄像系统是否存在扫描运动的策略。 建立了电子稳像系统。阐述了电子稳像算法的评价方法。用CCD摄像机采集不同类型运动矢量的视频图像序列,对本文构建的实时电子稳像算法进行评价和验证。试验结果表明,本文的非旋转实时电子稳像算法帧处理率达到50帧/秒同时,稳像精度和块匹配算法的精度接近;旋转实时电子稳像算法也在PSNR大大提高的基础上,帧处理率达到30帧/秒。更多还原

【Abstract】 Electronic image stabilization is a new method to stabilize image sequences with computer image processing technique. It has a lot of advantages, such as high image stabilizing precision, compact size, lightweight, low power consumption and reasonable price. Therefore the electronic image stabilization has been applied to martial and civil equipments all over the world. Real-time electronic image stabilization is a method to timely stabilize image sequences in electronic image stabilization algorithm, which remove the unwanted movements of camcorders. So real-time electronic image stabilization algorithm is characterized by highly precision and quickly matching performance.This thesis, supported by the Innovation Fund of Jiaxing city in Zhejiang province, mainly discussed the real-time image stabilization, concerning the theoretical research, common algorithm analysis, real-time algorithm design, image stabilization system realization, experimental testifying and algorithm evaluation. A series of result are obtained and parts of them will influence the development of real-time electronic image stabilization.This thesis analyzes the blurring cause of image sequences and proposes the necessity of electronic image stabilization. It expatiates principle and method of electronic image stabilization in detail and discusses key technique of realization. Based on study for several typical algorithms of motion vector estimation adopted by overseas electronic image stabilization system and real-time electronic image stabilization algorithm’s requirements, pray projection algorithm is chosen as the base part of real-time electronic image stabilization algorithm.A software simulated platform of electronic image stabilization is set up in the computer. Several simulated image sequences with different types motion vector are created to analyze and testify the performance characterization of pray projection algorithm. These analysis provide an important theoretical basis for designing real-time electronic image stabilization algorithm.Gray projection algorithm can not deal with certain image sequences, like as low contrast ratio, athletic objects inside and rotary images. New algorithms based on pray projection are designing for these situations and are testified by specifically experiments. The twice matchingalgorithm to stabilize image sequences including athletic objects inside and the algorithm compensating translation vector firstly and compensating rotary vector secondly to stabilize rotary image sequences are firstly presented in this thesis. A method to judge whether the camcorder has scanning motion is presented also.Electronic image stabilization system is set up and evaluation methods of image stabilization are presented. Image sequences with different types motion vector are obtained by CCD camcorder to evaluate and testify the performance characterization of real-time image stabilization algorithm designed in this thesis. Experimental results show that the non-rotary real-time algorithm can process 50 frames per second with similar image stabilizing precision and that the rotary real-time algorithm can process 30 frames per second with improving PSNR.This thesis discusses infrared electronic image stabilization. Characteristics are known by analyzing the video obtained by infrared camera. Experimental results of processing infrared image sequences show that gray projection algorithm can restrain the noise of infrared images and is feasible.更多还原