Deep learning-based quick dynamic human body action extraction and identification method

Abstract

The invention relates to a deep learning-based quick dynamic human body action extraction and identification method. At present, existing human body identification technology and application have the following deficiencies in multiple aspects: a human skeleton is a complicated structure body, and different human action habits correspond to different human action modes, so that human body identification is difficult generally. The method comprises the following steps of firstly describing whole information of size, color, edge, contour, shape and depth of a human body target; providing useful clues for action identification; extracting effective motion features from a video sequence; performing trace analysis by utilizing a motion trace of the target under the long-distance condition; and performing two-dimensional or three-dimensional modeling on four limbs and trunk of the target by utilizing information extracted from an image sequence under the short-distance condition. The method is used for extracting and identifying quick dynamic human body actions based on deep learning.

Description

Technical field: [0001] The invention relates to the field of action recognition, in particular to a method for extracting and identifying fast dynamic human action based on deep learning. Background technique: [0002] At present, the existing human body recognition technology and application have the following deficiencies. The human skeleton itself is a complex structure. Different people have different action habits, and their action methods are also different, which makes it difficult to identify the universality of the human body. Secondly, the position of the recognized target is limited and mandatory. The recognized target must adjust its position so that its front is aligned with the camera, and the side may not be recognized. Furthermore, the response speed and efficiency, for continuous human actions, frame and frame before There is redundancy in the data, which not only occupies a large storage space, but also increases the amount of calculation. [0003] Action...

AI Technical Summary

Problems solved by technology

[0004] In order to quickly acquire and identify human body behaviors and overcome the shortcomings of the above-mentioned prior art, the present invention provides a recognition method based on computer deep learning, which can extract and identify human bodies in a real-time, dynamic, fast, and large-scale Actions can be better applied to dangerous action alarms of enterprises, schools, and government agencies, and film and television production systems, reducing hardware requirements. Traditional methods have very strict hardware requirements in action recognition. In the recognition of ten thousand people, the general The computer can't meet the operation requirements at all. This method effectively solves the problem of excessive hardware occupation through single floating point and double floating point operations. In the random screening test of 100,000 people, the identification of people can be done in a very short time. real-time recognition

Images