A Human Action Recognition Method Based on Global Features and Sparse Representation Classification

Abstract

The invention relates to a human behavior recognition method based on global features and sparse representation classification. Gaussian kernel convolution filter preprocessing is performed on the video frame, and the difference method is used to extract the moving foreground pixels; according to the parameters, the pixel value is sampled in the space-time dimension to determine the moving area, the size of the video frame is adjusted for preliminary dimensionality reduction, and each frame of video is spliced ​​into a vector by column The feature vectors are obtained by combination; after splicing the feature vectors into feature matrices by column, the dimension is reduced for the second time, and the feature matrix is ​​obtained to form the initial feature dictionary. The dictionary obtains the sparse code of the input signal, and the code is sent to the classifier to output the behavior category; the dictionary learns parameters and realizes real-time behavior recognition. The invention obtains a dictionary and a linear classifier with both reconstruction performance and classification performance, which can be used to improve the efficiency of human behavior recognition, and is suitable for scientific fields such as security monitoring, content-based video retrieval, and virtual reality.

Description

technical field [0001] The invention relates to the technical field of video surveillance images and video processing, in particular to a human behavior recognition method based on global features and sparse representation classification. Background technique [0002] In recent years, with the rapid development of intelligent video surveillance systems, moving object detection, as an important link in intelligent video surveillance systems, has become a hot research issue in current computer vision. Moving object detection is to extract the foreground moving area from the background image in the video sequence frame. In an intelligent video surveillance system, the effect of moving target detection plays a decisive role in later target tracking, behavior understanding, and target classification. At present, video moving target detection methods mainly include optical flow method, inter-frame difference method, background subtraction method, etc. Among them, background subtr...

AI Technical Summary

Problems solved by technology

[0005] At present, there are some problems in the field of human behavior recognition, such as the complexity of the scene in the video, the intra-class difference of behavior, the inter-class similarity of behavior and the lack of comprehensive and accurate motion feature representation, which leads to the fact that in the actual complex environment, the feature representation is vulnerable. Influenced by changes in the external environment, the classification model has poor anti-interference ability, and the accuracy of human behavior recognition is low

Images