Static image behavior identification method based on main and auxiliary clues

Abstract

The invention discloses a static image behavior identification method based on main and auxiliary clues. The technical scheme is characterized in that a behavior recognition network containing a mainclue branch and an auxiliary clue branch is established; the feature frames corresponding to all the candidate auxiliary areas and the feature frames corresponding to the human body areas are found inthe feature graph; the main clue branch calculates a judgment score that the image I belongs to the behavior category ak according to the human body region information; the auxiliary clue branch calculates a judgment score that I belongs to the behavior category ak according to the auxiliary area information; and the sum of the scores of the two branches is used as a behavior score of which I belongs to ak, so that a behavior score of each type of behavior in the behavior type set of I is obtained, and all the scores are normalized to obtain the probability that I belongs to each type of behavior, and the maximum probability is the behavior type described by I. By adopting the method, the human body behavior in the image can be effectively identified, and the speed and the accuracy of thestatic image behavior identification are improved.

Description

technical field [0001] The invention relates to a method for static image behavior recognition in the technical field of multimedia information processing, and its essence is a method for extracting and classifying behavior-related features in an image, and is a recognition method that can adapt to many different behaviors. Background technique [0002] The advent of the image society or the visual era has become a dominant and comprehensive cultural landscape and an extremely important academic theoretical hotspot in the era of globalization. Visual data such as videos and images have gradually replaced text as the mainstream form of information and are more easily accepted by people. With the continuous maturity of web2.0 technology and the rapid development of artificial intelligence, the computer's understanding of pictures gradually tends to the semantic level, and the research on the recognition and classification of image content is no longer limited to the traditiona...

AI Technical Summary

Problems solved by technology

In the existing static image behavior recognition algorithms, the behavior recognition research of static images is usually carried out around these clues, specifying a certain clue, or combining a number of specified clues for behavior recognition, have achieved certain recognition effects, but The applicability of this identification method of specifying cues is not strong, and it is difficult to deal with the situation that the key cues are different in different behaviors

In different behavior categories, the clues that can provide key information describing the behavior are different, and the specified clues are not applicable to all behavior categories, which will affect the overall recognition rate

[0007] The existing static image behavior recognition research cannot effectively meet the needs of automatic image labeling, human-computer interaction, and video intelligent monitoring. It is particularly important to study a fast and effective static image behavior recognition method

Images