【作者】 周力；

【导师】 谭毅华；

【作者基本信息】 华中科技大学 ， 控制工程， 2012， 硕士

【摘要】 背景建模是运动目标检测方法中最重要也是应用最多的一种方法。国内外很多学者在理论上和实践应用上都对此进行了广泛而深入的研究和探索，并取得了大量研究成果。这些算法大多是针对特定场景下的特殊问题，如果换成其它环境将缺乏适应性；少量的算法虽然考虑到一般场景，但它们的应用范围很难界定，只能定性描述，缺乏精确的定量指标。目前，很少有学者对背景建模算法的性能进行评估，给出算法使用的场合、各自的优缺点等。对大量已形成或正在形成的目标检测算法进行有效的评估并通过性能评估来改进算法和发展新算法已成为日益迫切的需要。目标检测技术性能评价对研究和发展目标检测识别算法与系统、界定算法与系统的边界与适应性、提高系统的性能十分关键。本文结合具有代表性的五种序列图像背景建模算法研究一般情况下目标检测算法性能评价的理论、方法与支撑环境，为算法与系统的研制提供理论依据和试验平台。本文从三个方面展开研究：一是背景建模目标检测算法性能评估的基本框架；二是序列图像背景建模方法的一般框架；三是结合试验结果分析现有算法，用于支撑发展新算法。本论文通过对目标检测算法性能评估现状的分析，提出了将算法研究与性能评估统一到一个框架的观点。论述了性能评估的基本原理和方法，并探讨了算法性能评估的组织结构和软件支撑平台。更多还原

【Abstract】 Background modeling is the most important and most widely used method of movingtarget detection. Many researchers have carried out extensive and in-depth study in thetheory and practical application of this method, and get a lot of achievements. Most ofthese algorithms are developed to solve the special problem in the specific scene, so theycannot adapt to the change of scene. A few algorithms take into account the general scene,but it’s difficult to define their scope. For the lack of precise quantitative targets, they canonly do qualitative description. At present, few scholars could evaluate performance of thebackground modeling algorithms, clarify the right occasions of the algorithms and theiradvantages and disadvantages. It’s more and more urgently necessary to effectivelyevaluate a large number of target detection algorithms which have been proposed or arebeing proposed, and evaluate their performance to improve the algorithms and proposenew algorithms. Performance evaluation of target detection algorithms is very critical tothe research and development of target detection and recognition algorithms and systems,and also critical to the boundary definition and adaptability of the algorithms and systems,and so to the improvement of the system performance.In this paper, we evaluate five representative background modeling algorithm appliedto image sequences and study the performance evaluation theories, methods andenvironmental support of the general target detection algorithm, in order to provide atheoretical basis and testing platform for development of algorithm and system. We studyfrom three aspects: first, the framework of background modeling algorit hm andperformance evaluation; second, the general framework of background modelingalgorithm applied to image sequences; finally, analysing the existing algorithms throughthe test results to support the development of new algorithms.In this paper, through analysing status quo of the performance evaluation of thebackground modeling algorithm, we present the view on putting the study and performance evaluation of algorithm into a unified framework. The basic principles andmethods of performance evaluation are discussed, and we also explore the organizationalstructure and software support platform of the performance evaluation of the algorithm.更多还原