A modified Gaussian mixture background model via spatiotemporal distribution with shadow detection. Issue 2 (February 2016)

• Record Type: Journal Article
• Authors: Xia, Haiying; Song, Shuxiang; He, Liping
• Is Part Of: Signal, image and video processing. Volume 10:Issue 2(2016)
• Page Start: 343
• Page End: 350
• Publication Date: 2016-02
• Subjects: GMM -- Intelligent video surveillance -- Spatial background model -- Dynamic background -- Shadow detection; Signal processing -- Digital techniques -- Periodicals; Image processing -- Digital techniques -- Periodicals; Digital video -- Periodicals; 621.3822
• DOI: 10.1007/s11760-014-0747-z
• Languages: English
• ISSNs: 1863-1703

Abstract

Abstract This paper proposes a modified Gaussian mixture model designed to improve sensitivity in highly dynamic environments, overcome the low background recovery rate of the traditional Gaussian mixture model (GMM). This model uses spatial information to compensate for time information, and the neighborhood of each pixel is sampled using a random number generation method to complete the spatial background modeling. The time distribution of each pixel is used to model the Gaussian mixture background. For foreground detection, a spatial background model and time background model are both utilized by a fusion decision-making method. We conduct experiments on a dataset consisting of 31 real-world videos. Through a series of comparisons between our improved GMM algorithm, frame difference algorithm, Stauffer and Grimson's, T2F-MOG and Zivkovic's, we measure that the average running time of our algorithm is 0.0428 s/frame, faster than T2F-MOG, and the Recall is significantly improved with our method. We conclude that the experimental results show that the proposed algorithm is real time and accurate.