An Improved Scale-invariant Feature Conversion Algorithm And Its Applications

The scale invariant feature transform(SIFT)algorithm has good invariance in many situations such as image rotation transformation and scale scaling,and can provides robust invariant features for image registration.However,because the algorithm analyzes the image through the scale space,the registration time is longer and the accuracy is relatively unsatisfactory.In order to improve the efficiency of the SIFT algorithm and the accuracy of feature extraction,we propose a new improved algorithm in this paper and take binocular stereo vision image matching as an example since the SIFT algorithm is widely used in image registration.The improved algorithm is utilized to match images to verify its effectiveness.Our main research works are as follows:1.The mathematical model of visual imaging is first briefly introduced.It mainly analyzes the mathematical and physical models such as visual imaging principle and vision system coordinate system conversion.On the basis discussing the calibration principle of the binocular camera,Zhang Zhengyou's calibration method is used to compare the results obtained by the two software platforms MATLAB and Open CV,and then the more accurate internal and external parameters can obtained for the parallel optical axis camera system.2.An improved SIFT algorithm is proposed.Based on the research of the SIFT algorithm,we can improve the SIFT algorithm in two ways.One is to simplify the feature descriptors and reduce the key point descriptors from the original 128 to 56 dimensions,which improves the matching efficiency of the key points.Another is to add constraints in the matching process and uses the RANSAC algorithm to filter and remove matching points so that the efficiency of correct matching of key points increases.Through the experimental results,it can be seen that the matching efficiency and accuracy of the improved algorithm are significantly increased when compared with the original algorithm.3.The improved SIFT algorithm is applied to the binocular visual stereo matching as an example.Firstly,the binocular visual stereo matching is briefly introduced.Then the validity of the obtained camera parameters and the improved SIFT algorithm are verified through the GRB-400 SCARA robot vision system.Experimental results show that the improved SIFT algorithm can effectively meet the real-time requirements of target matching and positioning,and at the same time the improved SIFT algorithm can achieve the purpose of improving the accuracy and accuracy of stereo matching.These experiment effects are satisfactory.