CHMM-based background subtraction algorithm
Guangtai Ding · Jisuanji gongcheng yu sheji · 2012
To address the sensitivity of the background subtraction algorithm in the procedure of the foreground extraction to the change of light illumination and the salt and pepper noise which is prone to be generated in the foreground extraction,a CHMM-based background subtraction algorithm is proposed.Firstly,the Markov nature of pixels is analyzed.Then,the CHMM(coupled hidden Markov model) is modeled according to the pixel,and the transition probability of the pixel's hidden state is determined based on the statistical method.Proper standard deviations of the foreground and background are determined by the experimental method.Finally,the optimal hidden state problem of CHMM is solved by the Viterbi algorithm.By using this algorithm to analyze a traffic surveillance video,it is proved that the algorithm is very effective to suppress the changes of light illumination and the generated foreground salt and pepper noise to a certain extent.