Research And Implementation Of Target Recognition And Three-dimensional Registration Algorithms In Augmented Reality

With the advancement of science and technology,an emerging technology has developed.that is,augmented reality technology.Augmented reality technology is the use of computer vision technology to superimpose virtual objects into real-life scenes,allowing users to perceive a new environment where virtual scenes are fused in real scenes.Let us see a world of science fiction.At present,augmented reality is in an evolving stage.Which from the beginning of two-dimensional code augmented reality,to the current augmented reality in the three-dimensional world.Compared to 2D images,the 3D world is more complex and the technical requirements involved are even higher.The accuracy of multi-angle recognition of objects and the stability and real-time performance of augmented reality all need further improvement.In view of the above problems,this paper studies the accuracy of image feature points matching in augmented reality and the stability and real-time performance of three-dimensional registration of objects in a real scene.Firstly,this paper studies the accuracy of object recognition matching in augmented reality.There is a large number of wrong matching pairs in the matching pairs obtained by the traditional feature point matching method.After the general method of filtering out the wrong matching pair,the obtained matching pair still has some wrong matching pair,which affects the recognition accuracy.To solve this problem,this paper improves the method of eliminating false matching pairs.After studying various feature point extraction algorithms and comparing them,the ORB algorithm was selected as the image feature point detection method in this paper because of its fastest computing speed.The initial matching pair is obtained by the traditional feature point matching method,and one matching point matching a plurality of feature points is eliminated.Three pairs of matching pairs are selected,and using the principle of "similar triangles" to select the correct matching pairs.The homography matrix of the image is further calculated to filter out the correct matching pairs.The method can effectively improve the accuracy of the matching pair and further improve the recognition accuracy of the image.Next,this article studies the three-dimensional registration method in augmented reality.The current development of augmented reality is gradually changing from the two-dimensional code phase to the three-dimensional world.Research has been done on the three-dimensional registration of objects in real-world scenes to achievereal-time stable augmented reality effects.A three-dimensional registration algorithm based on natural feature points is introduced.In this paper,three-dimensional reconstruction of the target object is performed using the feature point data of the keyframe images at various angles of the object and camera parameters to form a feature point matching library for the target object in three-dimensional space.When the camera captures a real scene,the image in the real scene is matched with the target object,and the key image frame that most closely matches the target object with the real scene is found.When the camera captures a real scene,the image in the real scene is matched with the target object,and the key image frame that most closely matches the target object with the real scene is found.According to the camera pose information of the keyframe image,a three-dimensional registration matrix is calculated to realize the augmented reality in a real scene.And the use of optical flow tracking method to eliminate the camera's jitter problem,enhance the real-time augmented reality.After experimental comparison,it is found that the method described in this paper removes almost all of the erroneous feature point matching pairs,improves the accuracy of feature point matching,and ensures that the augmented reality display effect is good.The three-dimensional registration method based on natural feature points in the real scene which introduced in this article can achieve a good tracking registration of three-dimensional objects.Real-time stable augmented reality effects can also be achieved.At the same time,it also meets the development needs of augmented reality.