Action Recognition Method Based on Dense Trajectory Kernel Covariance Descriptor

Abstract

The invention discloses a behavior recognition method based on a dense trajectory kernel covariance descriptor, which mainly solves the problem that the prior art fails to consider the nonlinear correlation between different features, resulting in low accuracy of behavior recognition. The implementation steps are: 1) extract dense trajectory, extract features for each pixel in the trajectory cube to obtain the underlying feature matrix; 2) calculate the kernel covariance matrix of the underlying feature matrix, and map it to the Euclidean space to obtain vectorized features 3) Use all the feature representations in the trajectory cube to construct a dense trajectory-based kernel covariance matrix descriptor; 4) Encode the kernel covariance matrix descriptor with the BOW model to obtain a codeword histogram, and use the codewords of the training set Histogram training SVM, test the codeword histogram of the test set in the trained SVM, and obtain the behavior recognition result. The invention further improves the ability to describe behaviors, and can be used in complex environments such as video surveillance.

Description

technical field [0001] The invention belongs to the technical field of video processing, and in particular relates to a behavior recognition method, which can be used to describe video behavior in complex video monitoring environments. Background technique [0002] The wide application of video behavior recognition in the fields of human-computer interaction, virtual reality, video surveillance, and video retrieval and analysis has aroused the interest of more and more researchers, and has important academic research value and strong practical value. In the field of action recognition, the existence of factors such as viewing angle changes and complex backgrounds increases the difficulty of action recognition. In this case, due to the robustness of artificial local features to video noise, illumination changes and complex backgrounds, it has become a important research directions. At present, the most popular artificial local features are gradient direction histogram HOG ba...

AI Technical Summary

Problems solved by technology

Although this method takes into account the relationship between different features, it only considers the static features and does not consider the dynamic features of the behavior subject, resulting in low accuracy of behavior recognition

[0005] (2) Yi Y, Wang H. Motion keypoint trajectory and covariance descriptor for human action recognition [J]. The Visual Computer, 2017: 1-13, this method builds a trajectory-based covariance descriptor on the basis of the motion keypoint trajectory The variance descriptor can represent the linear relationship between different motion variables. However, this method ignores the nonlinear relationship between features and cannot capture the complex relationship between features in behavior recognition with complex environments.

Images