Action intelligent comparison method

Abstract

The invention discloses an action intelligent comparison method which is used for intelligently judging the similarity between technical actions and standard actions in table tennis. The method comprises the following steps: data preprocessing: determining a reference coordinate and calculating an angle of a joint point; human body coordinate transformation, wherein the human body coordinate transformation comprises coordinate translation and a rectangular coordinate system rotating sphere coordinate system; action splitting: during action data capture, splitting one action into a plurality ofaction templates to respectively identify and record the action templates as joint point 3D position data at different action stages; and, when an action curve is drawn, finding a segmentation pointthrough an algorithm, and splitting the segmentation point into a plurality of segments of data; calculating motion deviation, wherein the calculation of the motion deviation comprises calculation ofmotion track local deviation and calculation of motion track overall deviation. According to the method, the similarity of the postures is measured based on the 3D position information of the joints,the human body posture error is prevented from being measured through the joint angles, and the calculated amount is small.

Description

technical field [0001] The invention relates to the technical field of action comparison, in particular to an action intelligent comparison method, which is used for intelligent determination of the similarity between technical actions and standard actions in table tennis. Background technique [0002] Existing human action comparison methods include: naked eye observation method, video frame-based method and time series similarity measurement; among them: [0003] 1. The naked eye observation method can only be pointed out by experienced trainers to point out the differences between movements. [0004] 2. The method based on video frames belongs to the field of pattern recognition. It uses the difference between frame images to perform action comparison. This method is limited to plane comparison. Human actions are three-dimensional and involve the relative positions of joints. , Relative rotation and other issues. [0005] 3. The similarity measure of time series has sev...

AI Technical Summary

Problems solved by technology

[0005] 3. The similarity measure of time series has several characteristics: large amount of data, high dimensionality, and complex structure

[0013] 1. The human eye observation method requires experienced coaches, and basically cannot be applied in large quantities and cannot accurately measure movement deviation

[0014] 2. Human body movements are three-dimensional, and the video frame method is limited to plane comparison, which cannot meet the standard of accurate measurement and cannot determine the deformation of movements at different angles

[0015] 3. Euclidean distance can only represent the distance between points in multidimensional space, and cannot reflect the similarity between gestures, nor can it be used as a judgment of sequence matching in the longest common subsequence. Specific shortcomings: 1. Noise and time series very sensitive to fluctuations

3. Does not support time series deformation, such as time axis bending, stretching, etc.

[0016] 4. Compared with the Euclidean distance, the dynamic time warping distance can better describe the similarity of the time series, and can solve the problems that the Euclidean distance cannot solve, but there are also certain defects: 1. The solution is more complicated, and the calculation Large amount; 2. Cannot satisfy the triangle inequality

[0017] 5. The longest common subsequence can avoid similarity problems caused by short-term mutations or discontinuities in time series, but it cannot handle deformations such as amplitude translation, time axis stretching and bending

Images