Skeleton Data Correction for Action Recognition Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
It is challenging to generate data sets for identifying action categories without requiring individuals to perform each action category, and operators must classify actions, which is inefficient.
Innovation Solution
An information processing apparatus acquires training moving images and generates skeleton data indicating joint positions, correcting the data to match the joint distances of a target person, allowing for the creation of an inference model that identifies action categories without the need for individuals to perform each action.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a person performs actions in each category to generate training data, then the data set can be created, but it is difficult and time-consuming for the person to perform all actions
Solution Approach 1:
The patent uses skeleton data from a first person performing actions to generate corrected skeleton data representing a second person. The system copies and transforms existing data rather than requiring the second person to physically perform all actions, thereby reducing time requirements while maintaining training data quantity and quality.
2Measurement precision
If operator classifies data showing actions into each category, then action categories can be identified, but the classification process is inefficient
Solution Approach 1:
The system automatically generates and corrects skeleton data without requiring operator intervention for classification. The processor autonomously performs the correction process by calculating joint distances and transforming skeleton data, thereby maintaining high measurement precision while significantly improving productivity by eliminating manual classification work.
3Quantity of substance
If training skeleton data is generated from a first person, then action data can be collected, but the distance between joints does not match a second person's anatomy
Solution Approach 1:
The patent modifies the skeleton data parameters by calculating the ratio of joint distances between the first and second persons. The processor corrects the skeleton data by applying this ratio transformation, thereby changing the parameters to match the second person's anatomy while preserving the action data structure and training availability.
Data Source
AI summary
An information processing apparatus includes an interface configured to acquire a training moving image obtained by photographing a first person performing an action, and a processor configured to generate training skeleton data indicating positions of joints in the first person in time series from the training moving image and to correct the training skeleton data such that a distance between joints in the training skeleton data matches a distance between joints in a second person different from the first person.