Sleeping Posture Detection Using Cross-Frame Error Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing posture detection techniques using machine learning, such as OpenPose, often erroneously detect positional information of human body parts due to inconsistencies between frame images, leading to false detection of body motions that do not actually occur.
Innovation Solution
A posture detection apparatus and method that utilizes a receiver, position estimator, and determination processor to identify and correct erroneous detections by comparing positional information across frame images, using a trained model to analyze the degree of difference and matching between frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If posture detection is performed using machine learning on each frame image independently, then real-time detection capability is achieved, but detection accuracy deteriorates due to inconsistent results between frames
Solution Approach 1:
The system performs preliminary detection on each frame image independently to maintain real-time capability, then applies post-processing correction by comparing consecutive frame results to identify and correct detection errors, achieving both speed and accuracy
Solution Approach 2:
The system uses feedback from consecutive frame comparisons to detect and correct erroneous detections. By comparing positional information across frames and identifying inconsistent changes, the system corrects detection errors while maintaining real-time performance
2Productivity
If positional information is detected for each frame image independently without considering frame relationships, then processing speed is maintained, but detection reliability deteriorates due to false motion detection
Solution Approach 1:
The system performs independent detection on each frame to maintain processing speed, then applies preliminary correction by comparing with previous frame results to eliminate false detections before final output
Solution Approach 2:
The system introduces an intermediary correction step that compares positional information between consecutive frames. This intermediary process identifies false motions by detecting inconsistent positional changes that do not correspond to actual body movements
Data Source
AI summary
A position estimator outputs, by inputting information obtained from the frame images into a trained model, positional information of a plurality of parts of a human body from the trained model. A determination processor outputs information on erroneous detection by the position estimator. The position estimator outputs a first position group in response to an input of an obtained first frame image, the first frame image being obtained by the receiver and outputs a second position group in response to an input of an obtained second frame image different from the first frame image, the second frame image being obtained by the receiver. The determination processor outputs the information on erroneous detection by the position estimator based on a degree of difference between the first position group and the second position group and a degree of matching between the first frame image and the second frame image.