使用局部二进制相似模式改进背景减除

需积分: 10 176 浏览量更新于2024-09-10 收藏 462KB PDF 举报

"这篇论文提出了一种改进的背景减除方法，通过利用局部二进制相似模式（LBSP）来增强VIBE算法的性能，提高了前景目标提取的运算速度。" 正文：背景减除是视频分析中的关键步骤，主要用于识别视频序列中的运动对象，即前景目标。传统的像素级背景减除方法主要依赖于单个像素的强度变化来判断是否为前景，这种方法简单但可能受光照、阴影等影响，效果并不理想。VIBE（ViBe）是一种低开销且高效的背景建模算法，它通过构建像素的混合模型来处理这些变化，但其速度和准确性仍有提升空间。论文“Improving Background Subtraction using Local Binary Similarity Patterns”提出了一种基于LBSP（Local Binary Similarity Patterns）的适应性背景减除方法，该方法在VIBE的基础上进行改进。LBSP是一种描述符，它考虑了像素邻域内的相对亮度关系，而非仅仅依赖像素强度。这种空间-时间二进制相似度描述符能够更好地捕捉像素间的上下文信息，从而提高对复杂环境变化的鲁棒性。具体来说，LBSP通过对像素邻域内像素值进行比较，生成一种二进制模式，然后统计这些模式的出现频率，以此来表示像素的局部特征。在背景建模中，这种局部相似性的分析可以更准确地识别正常背景的变化，减少误报前景的可能性。同时，由于LBSP的计算相对简单，它也提升了整体算法的处理速度。实验部分，作者将提出的LBSP-VIBE方法与原版VIBE进行了对比，结果显示LBSP-VIBE在运算速度上优于VIBE+，同时保持了良好的前景提取效果。这表明，引入LBSP不仅增强了背景减除的准确性，还降低了计算复杂度，使得该方法更具实用性。论文提出的LBSP-VIBE方法是对传统背景减除技术的一次重要改进，它利用LBSP的特性，有效地结合了空间和时间信息，提高了在复杂动态环境下的背景建模和前景检测能力。这一工作对于视频监控、智能交通等领域有着广泛的应用价值，特别是在要求实时性和准确性兼顾的场景下。

Improving Background Subtraction using Local Binary Similarity Patterns

Pierre-Luc St-Charles, Guillaume-Alexandre Bilodeau

LITIV lab., Dept. of Computer & Software Eng.

Ecole Polytechnique de Montr

eal

Montr

eal, QC, Canada

{pierre-luc.st-charles,gabilodeau}@polymtl.ca

Abstract

Most of the recently published background subtraction

methods can still be classiﬁed as pixel-based, as most of

their analysis is still only done using pixel-by-pixel com-

parisons. Few others might be regarded as spatial-based

(or even spatiotemporal-based) methods, as they take into

account the neighborhood of each analyzed pixel. Although

the latter types can be viewed as improvements in many

cases, most of the methods that have been proposed so far

suffer in complexity, processing speed, and/or versatility

when compared to their simpler pixel-based counterparts.

In this paper, we present an adaptive background subtrac-

tion method, derived from the low-cost and highly efﬁcient

ViBe method, which uses a spatiotemporal binary similarity

descriptor instead of simply relying on pixel intensities as

its core component. We then test this method on multiple

video sequences and show that by only replacing the core

component of a pixel-based method it is possible to dramat-

ically improve its overall performance while keeping mem-

ory usage, complexity and speed at acceptable levels for

online applications.

1. Introduction

Background subtraction based on change detection is the

ﬁrst step used in many video analysis applications to de-

tect and segment foreground objects. Hence, the end results

quality often depends largely on the method used. A naive

approach based on directly comparing pixel intensities over

multiple video frames would not be practical in most cases,

since real video sequences often contain outliers in the form

of noise and dynamic elements (varying illumination, undu-

lating objects, etc.) that should not be considered as fore-

ground. Consequently, more precise and complex reference

models and methods have been developed over the years to

ﬁll the need for robust applications.

What we propose in this paper is an “all-in-one”, single

model, single update scheme, training-less, spatiotemporal-

based background subtraction method which is in part de-

rived from the non-parametric ViBe method[1]. The key as-

pect to spatial and temporal neighborhood analysis here is

that instead of solely using pixel-level intensities as the core

component to create our reference model (in a ViBe-like

paradigm), we complement it with the output of a modiﬁed

Local Binary Similarity Pattern (LBSP) descriptor[3]. In

other words, we show that even without modifying a generic

method’s initialization, maintenance and foreground detec-

tion rules, it is possible to achieve proper spatiotemporal

analysis by only incorporating a minimalistic feature-based

component in its reference model. We also present in the

paper a few general improvements that were implemented

in our method, coined LOBSTER (LOcal Binary Similar-

ity segmenTER), over a more “straightforward” integration

strategy.

Altogether, using the proposed approach, standardized

evaluations using the recently-published CDNet dataset[7]

demonstrate that we extract enough spatial and temporal

neighborhood information in video sequences to dramati-

cally increase the base algorithm’s performance across mul-

tiple complex scenarios, without compromising baseline re-

sults. Also, we manage to keep extra memory, processing

time and post-processing operations to a minimum while

outperforming other much more complex state-of-the-art

methods also tested in [7], including ViBe+[5].

2. Related Work

The most simple and earliest examples of change de-

tection algorithms are the ones based on the idea that the

background can be modeled by independently observing ev-

ery pixel’s intensity subsequently (i.e. the so-called pixel-

based algorithms) in order to then determine if a given in-

tensity value is a true outlier (to be considered foreground).

Some of these methods, such as the ones based on den-

sity estimation (via parametric Mixture of Gaussians[23]

or non-parametric Kernel Density Estimation[6]) are now

下载后可阅读完整内容，剩余6页未读，立即下载

sunny_develop

粉丝: 124
资源: 15

使用局部二进制相似模式改进背景减除

sss.rar_CAS peal_CAS-PEAL_Face It_YCbCr_feature lbp

921164-MB51-Improving the runtime using database

Improving working memory using EEG biofeedback

Improving your application performance using telemetry tools

Improving Distributional Similarity with Lessons Learned from Word Embeddings

Improving-Healthcare-Using-NLP---COMP-5360-Final

Improving Contextual Advertising Matching by Using Wikipedia Thesaurus Knowledge

Improving Distantly-Supervised Neural Relation Extraction using Side Information

Improving

CMT-NC: Improving the Concurrent Multipath Transfer Performance using Network Coding in Wireless Networks

最新资源