Motion identification method based on human bone joint point distance

Abstract

The invention relates to a motion identification method based on a human bone joint point distance. The method comprises the following steps of (1) projecting each frame of a bone sequence on three planes of a Cartesian orthogonal system and generating a human bone distribution map; (2) extracting bone joint points in the human bone distribution map; (3) calculating a euclidean distance among the bone joints of each frame of the bone sequence and making the distances of all the joint points form frame distance vectors; (4) making all the frame distance vectors of the bone sequence form a sequence distance matrix according to a time sequence; (5) using a pseudo-color coding method to carry out color coding on a distance two-dimensional matrix and acquiring a color texture map; and (6) using a depth learning method to carry out picture classification so as to complete a human motion detection identification task. The motion identification method based on the bone joint point distance is not influenced by an environment and interaction motion and is suitable for interactive motion identification under a complex scene so that the method possesses a wide application value.

Description

technical field [0001] The invention belongs to the field of multimedia information processing, relates to computer intelligence, pattern recognition and machine learning, and is an action recognition method based on the distance of human skeleton nodes. Background technique [0002] With the continuous development of computer intelligence technology, human action recognition has broad application prospects in future life. For example: intelligent monitoring, human-computer interaction somatosensory games, video retrieval, etc. Therefore, the study of action recognition methods has far-reaching research value. In recent years, with the maturity of computer vision technology, easy-to-use, low-cost depth sensors, such as Kinect cameras, have been widely used. Human action detection and recognition based on deep skeletal video sequences has attracted increasing attention due to the advantages of depth cameras, which are insensitive to illumination changes and capable of relia...

AI Technical Summary

Problems solved by technology

However, the current bone-based action technology recognition is not yet mature

[0003] The two patents CN106203503A and CN106228109A previously applied by the inventor of this application are the most relevant public patents to this application. Their disadvantage is that they can only recognize actions in a single subject scene and cannot meet the needs of interactive action recognition in complex scenes.

Images