摘要
针对运动目标检测中背景模型的提取和更新,本文首先提出了基于改进K-均值聚类算法的背景提取方法。该方法给出了动态三元组(DTDG)的概念,并且对每个像素用3个动态三元组进行建模,实现了原始背景的提取。其次,提出了一种新颖的自适应背景建模方法。对每个像素维护一个新的动态三元组,根据像素的动态变化信息决定更新策略,实现了背景的自动更新,可以适应光照的突变、缓变和场景本身的变化。实验验证了该方法的有效性。
To solve the generation and update of background model in moving target detection, firstly, an approach to background generation based on modified K-means clustering algorithm was proposed in this paper. The concept of Dynamic Three-dimension Group (DTDG) was given and each pixel was modeled using three DTDGs. The approach got background extraction. Secondly, a novel approach of adaptive background model was presented. Any pixel was modeled by one new DTDG. Updating decisions were made according to the pixel dynamic change information. The approach can update the background to adapt slowness, sudden changes in illumination and moving background targets automatically. The experimental results demonstrate the effectiveness of the proposed approach.
出处
《光电工程》
EI
CAS
CSCD
北大核心
2008年第1期26-30,共5页
Opto-Electronic Engineering
关键词
背景提取
背景建模
K
均值聚类
计算机视觉
动态三元组
background generation
background model
K-means clustering
computer vision
dynamic three-dimension group