Research Of Geometric Features Analysis In Stereovision

The geometric feature analysis of stereovision has made great progress,especially in the three-dimensional reconstruction of binocular stereo vision based on image.Stereovision analyzes and studies the geometric characteristics of stereo pairs mainly by using feature detection operator to extract and match the image features.According to the polar geometry theory,we get the relative geometric relationship of the stereo pairs,then restore the structure of objects,and achieve the 3D reconstruction of objects.The traditional 3D reconstruction technology based on image is aimed at the whole image in the process of feature extraction and matching,which leads to the surrounding environment region of the object is also detected in the large scene image,which not only increases the computation time and the running time of the algorithm,but also has a certain influence on the modeling precision of the specific target.At the same time,for some mountainous areas,hills and other feature points difficult to extract,the matching density is not enough and the matching precision is low.In recent years,with the application of biological visual saliency in the field of computer vision,visual saliency detection technology has developed rapidly and has made great progress.In view of complex 3D scene images,visual saliency detection technology is introduced in the process of processing.At first,the region of interest is extracted.The corresponding analysis and processing for the region of interest ensures the accuracy of 3D reconstruction based on image,and effectively reduces the computational complexity of the algorithm and improves the efficiency.The main content of this paper is to analyze the region of interest.The experimental data of this paper are based on UAV images.In this paper,we first detect and process the UAV image feature,get the feature points,generate feature descriptors and match them.Then the epipolar geometry constraints are applied to image matching from coarse matching to accurate matching.The fundamental matrix is obtained by the iterative calculation of RANSAC algorithm,and the corresponding points are detected by the initial matching,and the mismatched points are eliminated,the exact coordinates of the corresponding points are obtained.Based on this,we use visual saliency detection method to detect salient regions and get saliency map.Then,we get saliency target segmentation map by using mean-shift segmentation method to segment saliency map.The feature area is detected and matched to the saliency area,and the feature points of the precise location of the object are obtained,and then the corresponding points are obtained through matching.In this paper,the method of matching based on the saliency region is put forward,which effectively solves the problem of low matching rate of point feature extraction in difficult area when the traditional stereo geometric feature analysis is analyzed and improves the matching accuracy and efficiency.In addition,this paper also analyzes the existing methods and theories of visual saliency detection,and compares the performance of the four algorithms of IT,AC,FT and HC through the corresponding experiments.At the same time,a lot of low frequency components are introduced into the content of FT algorithm in the process,and some high frequency information is suppressed.Thus,the difference of brightness between saliency and background areas is not obvious.What's more,a local inconsistency probability function NIF is introduced,and the FT algorithm is improved,so that the newly generated saliency map can retain the edge information of the saliency region,and also highlights the brightness difference between the saliency region and the background region.