Motion Matching Based On Architectural Feature Of Corner

The goal of image matching algorithm is to find homogeneous regions between the images,and then establish their spatial correspondence accordingly.It is not only a hot topic in the field of computer vision research,but also the foundation of many vision applications,such as image matching,robot vision and 3D reconstruction etc..However,in the moving scene,local feature matching algorithms(such as distributed algorithms SIFT,SURF and KAZE,etc.)failed at the edge of the moving target as the motion background changes sharply.Therefore,this paper focused on the inner information of the angular structure,established a model of architectural feature,which integrates the attribute information of the corner points,such as the angle's size,orientation and color,proposed a motion matching algorithm which is suitable for the non-background change,the matching of rigid moving targets at normal speed is solved.The specific work is as follows:(1)Firstly,this paper analyzed the relationship between background and feature points extraction and motion matching.Then we studied the mechanism of architectural feature extraction and motion matching.(2)Secondly,based on the observation of the corner's interior information,this paper proposed a point pair structure which can express the independent corner background,and the corner detection is realized by the local clustering of the point to the structure.(3)This paper established a model of architectural feature,through the point pairs structure in the corner range,the property information of the location,angle,orientation and color of the corner point is calculated.(4)Through the description of the architectural feature,this paper proposed the process of the local feature's matching in the motion scene,and the local features are not affected by the background changes.In the motion scene,considering the continuity of time and the variability of the background,a motion matching algorithm based on the corner point structure is proposed in this paper.Experiments show that the algorithm proposed in this paper is feasible in motion scene matching,and it is more comprehensive and continuous in moving targets.