Object identification method based on p-SIFT (Scale Invariant Feature Transform) characteristic

Abstract

The invention provides an object identification method based on a p-SIFT (Scale Invariant Feature Transform) characteristic. The object identification method comprises a template library training phase and a test picture matching phase. The template library training phase comprises the following steps of: respectively calculating an SIFT characteristic point of each training picture in a template library to obtain a characteristic matrix; and calculating a covariance matrix of the characteristic matrix to obtain a p-SIFT characteristic descriptor. The test picture matching phase comprises the following steps of: calculating an SIFT characteristic matrix of a test picture; and calculating the similarity of the test picture and the characteristic matrix of a template library training picture. According to the object identification method disclosed by the invention, the characteristic points described by the p-SIFT characteristic are the region relativity and the direction relativity, so that the characteristic points are changed into a relative direction and a relative direction from an absolute position and an absolute direction, and the identification accuracy is improved.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a p-SIFT feature-based object recognition method suitable for real-time 2D or 3D object recognition application scenarios that require a balance between recognition accuracy and computational complexity. Background technique [0002] At present, rigid object recognition in 2D or 3D pictures is more and more widely used in various computer vision algorithms and many related application systems have been developed, such as defective product detection in industrial production, license plate recognition at traffic intersections and Internet image retrieval etc. When measuring the pros and cons of object recognition algorithms, recognition accuracy and computational complexity are two important indicators. Usually, the improvement of recognition accuracy is achieved by introducing more features, which will also increase computational complexity. , that is, recognition accurac...

AI Technical Summary

Problems solved by technology

Although this algorithm can effectively reduce the amount of calculation, there are two problems. One is that the incompleteness of its feature space makes the discrimination of feature points decrease compared with SIFT, which affects the accuracy of recognition. The other is its training process. Not easy to expand

Images