KNN Non-Parametric Kernel Density Estimation Method for Motion Foreground Detection Based on Gaussian Filtering
Yang Xiao-qiang, Feng Tianju · 2019
Background subtraction is an important method in moving target detection. Its difficulty lies in the establishment of dynamic background model. All kinds of non-parametric density estimation methods are effective to solve this problem, but there are obvious shadow on the foreground image and can not clearly outline the contour of the target when the target moves forward and backward, left and right in a short distance. In order to solve this problem, K-Nearest Neighbor non-parametric kernel density estimation method based on Gaussian filter is proposed for target detection. K-Nearest Neighbor non-parametric kernel density estimation is used to analyze the probability density of pixels in the image and extract the dynamic target. Meanwhile, Gaussian filter is used to conduct high-pass filtering on the sampled video frame in the frequency domain. The resulting image is then fused with the foreground image in bit level to obtain the final target image. Experiments show that this method has certain generality, obvious effect of eliminating shadow, clear target contour, and can detect dynamic target in complex environment.