Object Detection Algorithm Based on Gaussian Mixture Model and Sift Keypoint Match

摘要

A fusion background modeling method based on GMM and scale invariant feature transform (SIFT) key point matching technology is proposed. The scale invariant feature conversion (Scale-invariant feature transform or SIFT) is a visual algorithm used to detect and describe the local featuresof the image. It searches for the extreme points in the spatial scale, and extracts the invariant of position, scale and rotation. SIFT is widely used in object recognition, robot map perception and navigation, image mosaic, 3D modeling, gesture recognition and so on. Our basic idea is that,on the basis of the GMM background model, we first conduct the morphological processing and then apply the SIFT key point matching algorithm to detect the moving object pixels. Morphological processing includes closed operation and disconnection operation. They are composed of corrosion andexpansion and the only difference between them is order. We see that corrosion helps to eliminate the boundary points and shrink the boundaries inward, which can be used to eliminate small and meaningless pixels. The expansion is the process of merging all the background points. It is usedto fill the holes in the object. It combines open operation and closed operation, and can form a morphological noise filter, which can achieve the effect of filtering all kinds of noise in the bright and dark areas. Finally, by judging whether the segmentation image is matched with the backgroundtemplate, the SIFT key point matching algorithm is used to eliminate the misjudgment of the segmented image. Experiments show that the above method can improve the accuracy of moving object detection compared with the original GMM.