An Efficient r-KDE Model for the Segmentation of Dynamic Scenes
朱青松, 宋展, 谢耀钦 · 2012
This study presents a recursive Kernel Density Estimation model (r-KDE) based method for the segmentation of dynamic scenes. In the algorithm, local maximum in the density functions is approximated recursively via mean shift method firstly. Via the proposed thresholding scheme, components and parameters in the mixture Gaussian distributions can be determined adaptively. The coarse foreground is obtained by background subtraction, and the Bayes classifier is then adopted to eliminate the misclassified points to refine the segmentation result. Experiments with two typical video clips are used for the comparison and to demonstrate the improvements.