The Improved Model Combines Feature Point Detection and Kernel Density Estimation for Moving Object Segmentation

摘要

Recent researches of dynamical object segmentation bring forward methods based on kernel density estimation model. This model needs a lot of recent samples of intensity values for the pixels. It is a method typically uses the intensity value in temporal domain, which ignores the spatial characteristics of the object itself. To address the limitation mentioned above, this paper proposes a new kernel density estimation model based on feature points detection. This result of feature point detection of the moving objects is used for this model to estimate the area where the moving object is located. In this way, we only need the samples for the area of moving objects to segment. The computing time can be shortened and the quality of segmentation can be improved. And, the experimental results presented prove this model is valid in normal conditions.