Adaptive Gaussian mixture learning for moving object detection

摘要

Adaptive Gaussian mixture learning has been used for moving object detection in video surveillance applications for years. However, the method suffers from low convergence speed in the learning process, especially in complex environments. This paper proposed a novel method which improves adaptive Gaussian mixture leaning from four aspects including calculating the learning rate of means and variances respectively, employing a default minimal value for variances, selecting the optimal match for new pixel and improving renewal equation of weights. Experimental results show that our algorithm is promising, compared with conventional methods.