Action recognition method based on adaptive context region selection

Abstract

The invention relates to an action recognition method based on adaptive context region selection, which is used for recognizing character actions in an image and comprises the following steps: S1) extracting an overall feature map of a to-be-identified image and a character bounding box n of a to-be-identified action character in the to-be-identified image by utilizing the first four convolution blocks of a ResNet model; S2) adaptively selecting a context area bounding box of each person in the to-be-identified image according to the feature map and the related information of the person bounding box n; S3) carrying out feature extraction on the figure bounding box n and the context area bounding box, and calculating to obtain scores of the figure corresponding to each action category and scores of the context area corresponding to each action category; and S4) according to the scores of the action categories corresponding to the character and the context area, judging the action category of the character in the image, and completing the identification of the character action, compared with the prior art, the method has the advantages of high identification precision, high identification speed and the like.

Description

technical field [0001] The invention relates to the technical fields of computer vision and action recognition, in particular to an action recognition method based on adaptive context region selection. Background technique [0002] For decades, action recognition has been an important research branch in the field of computer vision. Its research scope covers many aspects such as image and video data. Related technologies are also widely used in human-computer interaction, information retrieval, security monitoring and other fields. [0003] Traditional action recognition mostly uses methods based on manual features. In recent years, thanks to the rapid development of deep learning, action recognition methods based on deep neural network learning and feature extraction have emerged in an endless stream. According to the features they extract and utilize, these methods can be divided into three categories: global feature-based methods, local feature-based methods, and context...

AI Technical Summary

Problems solved by technology

These areas may be mixed with motion-independent information. For example, a bicycle in the same picture will have an adverse effect on the recognition of the motion of the running person in the picture, which will affect the performance of the entire motion recognition method and reduce the performance of motion recognition. Accuracy

Images