Human Skeleton Image Normalization for Viewpoint-Robust State Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques for monitoring systems lack robustness in state recognition processing due to limited viewpoint retrieval and vulnerability to variations in orientation and angle of view, particularly when dealing with unknown postures or partial body concealment.
Innovation Solution
An image processing apparatus that detects a two-dimensional skeleton structure, estimates the upright height of a person, and normalizes the structure based on this height to enhance robustness, using methods like OpenPose for skeleton estimation and normalization techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If skeleton structures are detected without normalization, then the detection process is simple and fast, but the robustness of state recognition processing is low due to variations in orientation and view angle
Solution Approach 1:
The patent applies parameter changes by normalizing skeleton structure parameters (coordinates, lengths, angles) based on the detected upright height. This transforms the raw skeleton data into a standardized form that is invariant to viewpoint and orientation changes, thereby improving robustness without requiring complex additional hardware or processing systems
Solution Approach 2:
The patent performs preliminary normalization of the skeleton structure before state recognition processing. By pre-processing the skeleton data to remove viewpoint and orientation dependencies, the system enables more reliable and accurate state recognition without needing to handle these variations during the main recognition process
2Measurement precision
If feature amounts indicating posture features are used for retrieval, then retrieval from a specific viewpoint is accurate, but robustness against retrieval from various viewpoints is low
Solution Approach 1:
The patent transforms posture feature parameters by normalizing them with respect to the upright height. This parameter transformation creates viewpoint-invariant features that maintain accuracy across different retrieval scenarios, enabling the system to accurately retrieve postures regardless of the viewpoint or orientation in the captured image
3Productivity
If skeleton detection is performed without height estimation, then the processing speed is faster, but the normalization accuracy is reduced
Solution Approach 1:
The patent performs preliminary height estimation from the detected skeleton structure before normalization. This preliminary action of estimating upright height provides the necessary reference for accurate normalization, ensuring that the subsequent state recognition processing achieves high precision without requiring additional slow processing steps
Data Source
AI summary
An image processing apparatus (10) according to the present disclosure includes: a skeleton detection unit (11) configured to detect a two-dimensional skeleton structure of a person based on an acquired two-dimensional image; an estimation unit (12) configured to estimate the height of the person when the person stands upright in a two-dimensional image space based on the two-dimensional skeleton structure detected by the skeleton detection unit (11); and a normalizing unit (13) configured to normalize the two-dimensional skeleton structure detected based on the height of the person when the person stands upright estimated by the estimation unit (12).