Title :
Eigenbackground Bootstrapping
Author :
Hughes, Kit ; Grzeda, Victor ; Greenspan, Marshall
Author_Institution :
Electr. & Comput. Eng., Queen´s Univ., Kingston, ON, Canada
Abstract :
-A new method for initializing Eigenbackground is proposed. The approach does not require supervised or lengthy training, but instead is bootstrapped as a single unobstructed background frame is used to exploit spatial information in place of gathering a temporal history to generate pixel statistics. Experimental results indicate that the bootstrapped Eigenbackground performed comparable to and sometimes better than the supervised Eigenbackground on a standard background subtraction data set.
Keywords :
eigenvalues and eigenfunctions; image segmentation; statistical analysis; video signal processing; Eigenbackground bootstrapping; pixel statistics; spatial information; standard background subtraction data set; supervised Eigenbackground; unobstructed background frame; Adaptation models; Heuristic algorithms; Principal component analysis; Road transportation; Training; Training data; Vectors; Background Subtraction; Principal Component Analysis; Subspace Learning;
Conference_Titel :
Computer and Robot Vision (CRV), 2013 International Conference on
Conference_Location :
Regina, SK
Print_ISBN :
978-1-4673-6409-6
DOI :
10.1109/CRV.2013.47
