Cross-modal evaluation method for shadowboxing training actions

Abstract

The invention discloses a cross-modal evaluation method for shadowboxing training actions, and provides a method for aligning a video sequence and an action capture sequence for a cross-modal sequence registration problem of original shadowboxing training actions with a nonlinear relationship. Denoising and reconstruction recovery are carried out on Tai Chi motion sequences of different lengths and dimensions, then staged space-time registration is carried out, the motion difference between different subjects is solved, space-time alignment of cross-modal Tai Chi long sequences is achieved, and the Tai Chi motion recruitment completion degree evaluation accuracy is improved.

Description

technical field [0001] The invention belongs to the technical field of sports training, and in particular relates to a cross-modal evaluation method for Tai Chi training actions. Background technique [0002] Taijiquan is a form of Chinese martial arts, the most representative of which is the simplified 24 forms, and long-term training can strengthen the body. In the interactive training scene, the training effect needs to be evaluated, and a notable feature of Taijiquan is the slow and seamless transition between postures, and different people and learning stages may bring about differences in time and movements. It brings difficulties to segmentation-based move recognition and registration methods. The representation of each object varies over time, resulting in non-linear temporal differences in captured data. Due to differences in skill levels, height, weight, and gait, both objects and between-object pairings require non-linear transformations for temporal alignment. ...

AI Technical Summary

Problems solved by technology

[0004] DTW requires data to have the same dimensions and cannot be used for multimodal data

In addition, DTW uses dynamic programming to solve the registration matrix, and the time complexity is the quadratic of the sequence length

Both GCTW and CTW use the shallow expression learning method to select feature expressions based on the linear projection of the original image, but their nonlinear expression capabilities are limited, especially when the original data has nonlinear relationships and local geometric structures. Insufficient consideration of modal data characteristics

The commonly used camera pose estimation method PnP has good processing ability for the video recording of the same object at the same time, but the effect of camera pose estimation for different objects and different times is poor.

Images