Unsupervised cross-domain action recognition method based on channel fusion and classifier confrontation

Abstract

The invention discloses an unsupervised cross-domain action recognition method (CAFCCN) based on channel fusion and classifier confrontation. Efficient action recognition of a target domain test set based on a source domain labeled data set and a target domain unlabeled training set is achieved. The method comprises the following specific steps of: (1) selecting an action recognition model; (2) optimizing a double-flow deep network structure; (3) constructing an objective function based on the double-flow network; (4) building an unsupervised cross-domain action recognition model based on thedouble-flow network; and (5) constructing a data set. The method has the advantages that unlabeled data sets of other training sets can be subjected to efficient action recognition on the basis of theknown data set, and the problem of unlabeled data of the training set of the target data set can be effectively solved. By applying the confrontation method, confusion of categories and domains can be achieved at the same time, domain-level and class-level invariant features are obtained, the convergence speed of the method is high, and efficient recognition of actions can be achieved.

Description

technical field [0001] The invention belongs to the technical field of computer vision and pattern recognition, and relates to an unsupervised cross-domain action recognition method (CAFCCN) based on channel fusion and classifier confrontation. There is no question of labels. Aided by source domain data, the effectiveness of the model is verified when the target domain data training set is unlabeled. Background technique [0002] In recent years, with the rapid development of deep learning technology, many scholars have proposed many action recognition methods based on deep learning technology, which can extract robust video representations. Classical action recognition methods include 3D convolution-based methods and two-stream-based methods. In the 3D convolution-based method, C3D has achieved great success. In the C3D method, the input is a continuous 16-frame image, and the spatial features and temporal features are simultaneously obtained through 3D convolution, and g...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to solve the problem of unlabeled action recognition in the training set of the target data set, which is different from the existing data set. The data sets used in the previous action recognition methods all satisfy the independent and identical distribution of the training set and the test set, generalization weak ability

Images