Research Of Feature Points Matching Algorithm Based On 3D Reconstruction From Portable Camera Images

With the development of image sensor technology,portable camera equipment has become more and more popular,such as mobile phones,digital cameras,which makes it easier for people to obtain picture information,and the accuracy of acquired pictures is also getting higher and higher.However,2D images cannot always restore the real 3D world.Professional 3D scanning devices are expensive and have high barriers to use.How to use portable camera images to quickly and effectively rebuild 3D models has become a problem that needs to be solved.This paper aims at the images based on portable cameras.The feature point matching algorithm in 3D reconstruction has been studied in depth.This paper focuses on the matching algorithm of feature points in 3D reconstruction.Feature point extraction of Scale Invariant Feature Transformation algorithm is the most widely used algorithm in portable camera images 3D reconstruction.The feature vector constructed by SIFT is invariant to image rotation,scaling and brightness change.It also maintain certain stability to the view change,affine transformation and noise of the picture.Because the number of feature point detected by SIFT is large and the feature vector descriptor based on each feature point is up to 128 dimensions,resulting in large amount of computation and severe time consuming in the matching process of features.In order to solve this problem,this paper proposes a nearest neighbor search algorithm based on euclidean distance and vector angle.First,computer the Euclidean distance from all vectors to the origin in high dimensional space and sort them,computer the angle between all vectors and stochastic selected reference vectors in high dimensional space and sort them.Then,the euclidean distance from the query vector to the origin is calculated,and a large number of non-nearest neighbors are eliminated by setting the query range parameter.As a result,the retrieval range has been narrowed.Finally,the angle between the query vector and the reference vector is calculated.In the narrowed range,take this angle as the center,retrieving the nearest neighbor of the query vector.The experimental results show that the proposed nearest neighbor search algorithm based on euclidean distance and vector angle can significantly improve the matching efficiency in the matching process of SIFT feature vectors.The final 3D Reconstruction experiment proves that this algorithm can effectively improve the matching efficiency and can also ensure good matching results.