Moving Objects Detection Method Based On Gaussian Mixture Model

Moving object detection is the basis for tracking and behavior analysis tasks.In moving target detection,eliminating the interference of background and noise,separating moving targets out from the image has been the focus of the study.Gaussian mixture model method is widely used in object detection.It has a better anti-interference ability for existence of small amplitude motion of background,can extract more complete moving target,but at the same time has disadvantage of noise presenting,the large amount of calculation and poorer shadow suppression effect.This paper based on Gaussian mixture model mainly gives three improvements.The first improvement,it is background initialization.To avoid false detections that initial frame containing moving objects produced,extract the video sequence of the front N frame images and choose this N frame images corresponding point average as each pixel of the initial background.The second improvement,aiming at the problem of large amount of calculation of Gaussian mixture model,a method of interval pixels modeling is proposed in this paper.The third improvement,four-frame differencing and improved gaussian mixture model are combined.Then an improved algorithm for moving target detection based on Gaussian mixture model is proposed,which make up deficiencies of the poorer noise and shadow suppression effect of gaussian mixture model method.The experimental results indicate that the proposed method can eliminate noise and shadow effectively in a simple environment,reduce amount of calculation,increase the accuracy of original algorithm in some extent,improve the reliability of the test results and truly reflect the situation of moving targets.