An Open View Action Recognition Method Based on Linear Discriminant Analysis

Abstract

The invention discloses an open-view action recognition method based on linear discriminant analysis, comprising the following steps: using the Kmeans algorithm to use the combined feature matrix for dictionary learning; converting the action sample feature matrix into an action sample feature vector; using linear discriminant analysis to utilize The feature vectors of the action samples under the training perspective and the auxiliary perspective are used to learn the projection matrix; combined with the projection matrix, a total of t action sample representation vectors under the training perspective and the test perspective are projected into the same space to obtain a new feature vector for each action sample; Use the new feature vector of the action sample to learn the action classification model by using the linear support vector machine, and finally use the action classification model to perform the action classification test under the test perspective. In the present invention, by learning the correlation of action samples under different viewing angles, the representation information of action samples under known and unknown viewing angles can be projected into the same vector space for action recognition.

Description

technical field [0001] The invention relates to the field of motion recognition, in particular to an open-view motion recognition method based on linear discriminant analysis. Background technique [0002] People began to pay attention to the research of human motion recognition a long time ago. In the 1970s, human motion recognition was studied for the first time, and psychologists conducted experiments on human motion perception. Movement in the dark was observed by marking bright spots on the joints of the person and then placing them in a dark environment. Experimental results show that the human visual system can distinguish the form of movement, such as walking, running, and even the gender of the athlete, for the sequence of light points in motion. This experiment shows that for the action recognition of the human body, it is not necessary to extract all the information of the movement, and only part of the information can be used to represent the movement. But this...

AI Technical Summary

Problems solved by technology

Because there are many perspectives in reality, and at the same time, there is no guarantee that there are samples of all action categories in each perspective.

Images