期刊名称：International Journal of Computer Science and Information Technologies
电子版ISSN：0975-9646
出版年度：2012
卷号：3
期号：2
页码：3787-3792
出版社：TechScience Publications
摘要：Nowadays automated surveillance system has become a trend in field of security. Video processing algorithms are utilized to implement these systems. For any video processing system, first task is to detect moving object or subtract a background. In Computer Vision, many techniques are available for detection of the moving object, but Mixture of Gaussian (MoG) models [1] is best suited for system having static and complex background with clutters. MoG technique is more accurate but has a larger time complexity which is unrealistic for real time processing. In this paper, we present a fast technique to extract moving object from background using MoG model and Haar wavelet. In this technique, before applying the MoG we down sample each video frame to acceptable resolution using Haar wavelet decomposition. Selection of wavelet decomposition level depends on original video resolution. The technique has been implemented and tested on videos of PETS [2] and CAVIAR [11] databases. For PETS sample, Original video sample frames having resolution 768 X 576 are down sampled to resolution 192 X 144 using level three Haar wavelet decomposition. Then MoG model is applied to subtract the background. Our result shows this technique is able to detect all moving objects from video in presence of complex background and clutters. We observed that this technique works almost three times faster than using only MoG model without sacrificing the quality of results.
关键词：Video surveillance; Background subtraction;Mixture of Gaussian model; Wavelet decompositionon
