Human action recognition method based on semi-supervised dictionary learning based on similarity weight

Abstract

The invention discloses a human body behavior recognition method based on semi-supervised dictionary learning of similarity weights, which mainly solves the problem of low recognition rate of human body behavior by supervised methods in the prior art. The identification process is as follows: (1) Divide the input data set into test samples and training samples; (2) Perform local feature detection on all samples, and randomly select the local features of labeled samples to obtain the initialization dictionary; (3) According to the initialization dictionary , using a semi-supervised method for dictionary learning; (4) use the learned dictionary to perform group sparse coding on all samples to obtain the coding matrix of each sample; (5) vectorize the coding matrix of each sample to obtain the final representation; (6) Use the final representation of each sample and the sparse representation classification method to classify the test samples, and complete the recognition of human behavior in the test samples. The invention enhances the discrimination of dictionary learning, improves the recognition rate of human behavior, and can be used for target detection in video.

Description

technical field [0001] The invention belongs to the technical field of pattern recognition, in particular to a method for recognizing the behavior of a target person in a video, which can be used for target detection in the video. Background technique [0002] Human behavior recognition refers to identifying the behavior information of the target in the video sequence and preparing for the subsequent processing work, which includes detecting relevant target visual information from the video sequence, expressing it in an appropriate way, and finally interpreting the information To achieve learning and recognition of human behavior. [0003] In recent years, unsupervised and supervised dictionary learning have been successfully applied in the fields of image classification and action recognition. In the field of human action recognition, whether they are distinguished using labeled video sequences, unsupervised dictionary learning does not use video label information, and sup...

AI Technical Summary

Problems solved by technology

[0009] Although the above method can obtain a more discriminative dictionary and improve the recognition accuracy, the shortcomings of this method are also obvious: it only considers the marked samples, does not consider the information of unmarked samples, and does not make full use of the sample information; and the actual On the Internet, it is often very difficult to obtain labeled samples, but unlabeled samples can be easily obtained and exist in large numbers. How to fully extract and utilize the information of a large number of unlabeled samples has become the key to this field.

Images