Action Recognition via Node Coordinates and Time-Series Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional action recognition technologies struggle to accurately recognize user actions from images that do not cover the entire body of the user, often due to obstacles or unfavorable camera angles.
Innovation Solution
An action recognition device that acquires images, estimates node coordinates of the user, calculates time-series feature vectors connecting the trunk and head of the user, and compares these vectors with reference vectors to decide the user's action and any changes in that action.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional action recognition technology is used, then the system can process images, but it fails to recognize user actions with high accuracy when the image does not cover the whole body
Solution Approach 1:
The patent divides the user's body into multiple node points (joints, landmarks) and processes them individually. Instead of requiring the entire body to be visible, the system segments the visible portions into detectable nodes and uses these segmented data points to infer the complete action, resolving the contradiction between accuracy and adaptability to partial images
Solution Approach 2:
The patent introduces an intermediary processing stage that converts raw image data into node coordinate information, then transforms these coordinates into time-series feature vectors. This intermediary representation allows the system to work effectively with partial body visibility while maintaining high recognition accuracy through comparative analysis against reference vectors
2Ease of operation
If the camera is placed in restricted positions, then the installation flexibility is improved, but the image coverage of the whole body is reduced
Solution Approach 1:
By segmenting the body into multiple detectable nodes rather than requiring full body visibility, the system enables accurate action recognition even when cameras are placed in restricted positions that only capture partial body views
Solution Approach 2:
The patent applies partial action by demonstrating that complete action recognition can be achieved from partial body observations. The system processes only the visible portions of the body (partial input) and uses this incomplete information to accurately determine the complete action through feature vector comparison
3Adaptability or versatility
If obstacles block parts of the user, then the practical deployment scenarios are expanded, but the visibility of the whole body is reduced
Solution Approach 1:
The patent converts the harmful effect of obstacles blocking body parts into a beneficial feature by designing a system that specifically processes partial visibility scenarios. The node-based approach and feature vector comparison methodology transform the limitation of blocked views into an acceptable input condition that maintains recognition accuracy
Solution Approach 2:
Segmentation of the body into nodes allows the system to process only the unobstructed portions visible in the image, ignoring blocked areas while still accurately recognizing the complete action through temporal analysis of the visible nodes
Data Source
AI summary
An action recognition device performs: estimating node coordinates of a user from an image; calculating, on the basis of the node coordinates, time-series feature vectors each indicating a feature vector connecting a trunk of the user and a head of the user to each other in time series; deciding an action of the user and a change in the action of the user by comparing input time-series feature vectors being the calculated time-series feature vectors with reference time-series feature vectors; and outputting information indicating the action and the change in the action that are decided.