Local Invariant Feature Extraction And Motion Matching Based On Background Invariant

The image feature is the basis of machine vision.And good image feature not only can accurately reflect the nature of image,but also is the key to solve problems such as image analysis,image understanding.According to the related physiological research,human visual system is sensitive to local information in image processing.Therefore,local invariant feature of the image is consistent with the human visual mechanism.Moreover,local feature of the image has following advantages,such as low redundancy,not need to pre segment the image and enhancing the real-time performance.Now,the local invariant feature of image has been used in fields of image matching,robot vision and 3D reconstruction.In local invariant feature,how to extract it,how to describe it and how to achieve the matching of it are key problems.To solve these factors,this thesis proposed background invariant pair points based on analyzing the relationship between local feature and background.Then we proposed background invariant feature detection method and its descriptor based on pair points.Finally,we achieved the matching of this feature in motion scenes.The main work we have done is as follows.(1)Firstly,we analyzed the relationship between background and feature points extraction,motion matching.Then we studied the mechanism of background invariant feature extraction and motion matching.(2)This thesis extracted background invariant pair points based on the consistency of inside gray,and got the pair points energy graph by the aggregation of pair points.(3)With the pair points energy graph,this thesis extracted the local background invariant feature points by seeking local extremum.(4)First,we use the properties of aggregated points as descriptors of feature points.Then with the principle of minimum distance,we use these feature points to achieve image matching under motion scenes.From the inside of the corner,we proposed the background invariant local image feature based on pair points,and achieved the matching of this feature in motion scenes.The experimental results show that,the background invariant feature points based on pair points is feasible for image matching in motion scene.And this feature can get more matching points from moving objects.