Cross-view gait recognition method based on subspace learning of joint hierarchical selection

Abstract

A cross-view gait recognition method based on subspace learning of joint hierarchical selection comprises the following steps: firstly, dividing gait samples of a target view and a registered view into a training set and a test set, simultaneously carrying out hierarchical block division on gait data of the two views, respectively vectorizing multi-level gait energy image blocks, then carrying out feature selection, and carrying out cascade connection; secondly, projecting registration view angle data and target view angle data to a public subspace at the same time, enhancing the relation between the registration view angle data and the target view angle data in a mode of constructing a cross-view-angle dual graph, performing effective feature selection and effective gait energy graph block selection in the projection process, removing redundancy, forming a registration sample set in the public subspace, and projecting test target visual angle data into the public subspace through the trained target visual angle projection matrix to form target sample sets in the public subspace, and performing gait recognition on the two sample sets by adopting a nearest neighbor mode of Euclidean distance. According to the method, the registered view gait data is introduced into the target view field, and the cross-view gait recognition effect is enhanced.

Description

technical field [0001] The invention relates to a cross-view gait recognition method based on subspace learning of joint hierarchical selection, and belongs to the technical field of pattern recognition. Background technique [0002] Gait is one of the long-distance perceivable biometrics, which can be captured from an unconscious and non-cooperative subject compared with other biometrics (such as face, fingerprint, palm, vein, etc.). Gait recognition has the advantages of non-contact, long-distance and not easy to camouflage. Research on gait recognition has been extensively carried out in recent decades. The state-of-the-art gait recognition techniques are mainly divided into two categories, namely model-based methods and motion-based methods. The model-based approach robustly extracts gait features and avoids noise interference issues. Motion-based methods characterize human motion patterns without fitting model parameters. However, due to changes in clothing, shoes, ...

AI Technical Summary

Problems solved by technology

However, this kind of method is usually more complicated to calculate. After the gait image is converted into a vector, the dimension is often as high as tens of thousands of dimensions, and the amount of calculation is very large, which greatly reduces the efficiency of gait recognition.

In addition, because the model-based method is limited by the data, the performance of this type of method drops greatly when the change of viewing angle is large.

In addition, some deep models have emerged, but the essence of deep learning is data-driven and requires a large amount of training data, while the existing database data volume is relatively small, which greatly limits the performance of deep models.

Images