Research On Spatial Location Method Of Video Image

With the rapid development of the machine vision industry,the camera as the second pair of human "eyes" is widely used in various geographic scenes,providing a large amount of video image data for humans.The spatialization of video images is one of the basic propositions of video spatiotemporal information extraction.At present,the camera's geographic location and attitude information are mainly obtained through camera calibration or equipped with external positioning equipment,so as to realize the mapping of video images to two-dimensional geographic space.However,not all video images have camera calibration conditions or can obtain initial positioning information of the images.Images at the same or similar scales can use the geometric features of the images to match to achieve spatial positioning.For video images that do not have initial positioning and do not include landmarks,it is difficult to retrieve a correct match from a large-scale and large-scale mass video image database,which cannot complete the task of spatial positioning of video images.Based on the above research background,this paper takes the unmanned aerial vehicle(UAV)video image as the main research object,and uses open source images as the positioning benchmark.Through the topological relationship and spatial distribution relationship between various types of feature objects in the video image,the corresponding geospatial map model is constructed.Realize the spatial positioning of video images by means of graph matching,and expand the application scenarios of the spatial positioning method of video images.The main research contents and results are as follows:(1)Construction rules for graph models of multi-scenario video images.This article divides video images into side-view images and top-view images according to their shooting angles.The top view video can not only represent the features of the image content,but also provide the spatial relationship between the features,which is suitable for expressing in the form of a graph.According to the categories and spatial distribution characteristics of features in different urban scenes,the video images are divided into three types: urban central area,suburbs,and rural areas.The corresponding graph model construction rules are formulated respectively.(2)A Graph Model Construction Method Based on Remote Sensing Image Segmentation and Classification.In this paper,an object-oriented multi-scale segmentation and classification method of remote sensing image is used to re-combine adjacent segmented regions of the same class to construct an adjacency graph model oriented to the image class.According to the geometric properties and spatial structure of the segmented area,a method for constructing a spatial distribution map model of image objects is designed.(3)Video image spatial positioning method based on graph model matching.The adjacency matrix is used to express the topological relationship between the nodes in the adjacency relationship graph,and the similarity measure of the adjacency graph model of the video image is realized by calculating the difference between the feature values.Aiming at objects with different levels of contribution in video content expression,construct a multi-level spatial distribution graph model.Design a graph matching method that combines the relationship between the image structure and the spatial position to complete the coarse positioning of the video image space.Finally,the correctly matched graph model nodes are used as image control points to realize the spatialization of UAV video images.This paper provides a new method for spatial positioning of video images,enriches and expands the theories and methods of video image feature matching and positioning.From an application point of view,the spatialized daily video can help track missing persons and provide auxiliary location information for heterogeneous video images for city management,thus providing effective support for further analysis and decisionmaking.