面向静态相机的背景减除法分析比较

发布时间：2018-02-14 14:17

本文关键词： 视频监控背景减除法运动检测静态相机后处理　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：背景减除法在许多计算机视觉系统中是一个非常关键的步骤,它可以检测视频流中有意义的运动物体或区域,背景减除法主要应用于视频监控、遥感和人机交互等。本文主要研究在静态摄像机条件下针对静态背景的背景减除法。首先讨论了一般背景减除法所面临的主要挑战包括逐渐光照条件变化、光照条件突变、动态背景、相机抖动、前景伪装、前景阴影、Ghost区域等;其次介绍了目前主流背景减除法的基本原理和特点,分析不同类型背景减除法的优缺点。在简单帧间差法中通常采用线性平滑平均更新方法来更新背景模型,本文提出了一种基于非线性更新方法来更新背景模型。在一像素点被判定为背景的情况下,根据该像素点当前输入和该像素点前一时刻背景之间的差异来自适应调整更新率,从而使得背景模型随时间推移更加稳定,不会轻易将一些前景内容包含进来。我们同时还把该非线性更新策略推广到ViBe算法中,使得ViBe中原先离散随机背景更新策略拓展成连续随机背景更新策略,而原先背景更新策略是新的背景更新策略的一种特殊情况。为了客观分析比较不同背景减除法在不同场景下的检测结果,本文采用Change Detection 2012和2014的数据集对算法的检测结果进行量化分析和比较,发现采用非线性背景更新策略的两种具有代表性算法在一些视频类别上检测精度比原算法有大幅度的提高。但是由于背景减除法所面临的挑战,设计一种背景减除算法能适用于所有的场景和光照条件仍是一个非常困难的问题。
[Abstract]:Background subtraction is a key step in many computer vision systems. It can detect significant moving objects or regions in video streams. Background subtraction is mainly used in video surveillance. Remote sensing and human-computer interaction, etc. In this paper, the background subtraction method for static background under the condition of static camera is studied. Firstly, the main challenges of the general background subtraction method are discussed, including the gradual change of illumination condition, the abrupt change of illumination condition, and so on. Dynamic background, camera shake, foreground camouflage, foreground shadow Ghost region, etc. Secondly, the basic principle and characteristics of current mainstream background subtraction and division are introduced. The advantages and disadvantages of subtraction and division of different types of background are analyzed. The linear smooth average updating method is usually used to update the background model in the simple inter-frame difference method. In this paper, a nonlinear updating method is proposed to update the background model. In the case of a pixel being judged as background, the difference between the current input of the pixel and the background at the previous moment of the pixel is derived from the adaptive updating rate. So that the background model is more stable over time, and some foreground content will not be included easily. We also extend the nonlinear updating strategy to ViBe algorithm. The original discrete random background updating strategy in ViBe is extended to continuous random background updating strategy. The original background updating strategy is a special case of the new background updating strategy. In order to objectively analyze and compare the detection results of different background subtraction methods in different scenarios, In this paper, the data sets of Change Detection 2012 and 2014 are used to quantitatively analyze and compare the detection results of the algorithm. It is found that the two representative algorithms with nonlinear background updating strategy have much higher detection accuracy than the original algorithm in some video categories. However, because of the challenges faced by the background subtraction method, It is still a very difficult problem to design a background subtraction algorithm that can be applied to all scenes and lighting conditions.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】