Method for comparing similarity of motion postures in video

Abstract

The invention belongs to the field of computer vision, and relates to a method for comparing similarity of motion postures in a video, which comprises the following steps of: carrying out extracting data, and generating a skeleton key node graph containing a plurality of human body nodes; preprocessing the data to generate a node time sequence state diagram; removing a non-change part in the nodetime sequence state diagram, only retaining change states of the nodes at different time, and generating a node time sequence change characteristic diagram; calculating a difference value through a difference algorithm according to the node time sequence change characteristic graph, and synthesizing the difference values of each node in different time periods to form a difference matrix of all nodes; and calculating the similarity of the two actions in three dimensions of action posture, action amplitude and action speed through a similarity algorithm according to the difference matrix. The similarity comparison technology provided by the invention can be used for efficiently, quickly and accurately comparing the similarity of the motion postures including the three dimensions of the motion posture, the motion speed and the motion amplitude, and can be applied to the fields of physical exercise, fitness error correction and the like.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a method for comparing the similarity of motion postures in videos. Background technique [0002] The motion posture analysis in the video is a basic topic in the field of computer vision research and application. With the development of artificial intelligence technology, this topic is also attracting more and more people's interest. On the one hand, it is because the motion posture in the video has a wide application prospect in sports analysis, motion error correction and other fields. On the other hand, the comparison and analysis of motion poses in different videos contains a lot of challenges, such as: changes in the angle of motion video extraction, the influence of light, and so on. [0003] The invention patent application CN108734104A published on November 2, 2018 discloses an image recognition motion error correction system based on deep learning. The huma...

AI Technical Summary

Problems solved by technology

This invention is limited by the wearing of sensors, and cannot be applied to the field of action recognition on a large scale. At the same time, the difference in motion speed and amplitude is not considered in the comparison process, resulting in mediocre performance in the comparison of motion posture similarity.

Images