Head Pose Guided Body Pose Estimation for Large Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Body-pose estimation techniques are computationally intensive, especially when operating on large images, leading to unnecessary processing of non-relevant areas and individuals, which wastes computational resources.
Innovation Solution
Estimate body poses by determining a search region based on head pose information, using a trained pose-estimation model that operates only on this region, reducing computational load and focusing on individuals of interest.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If body-pose estimation is performed on large images using a trained machine-learning model, then pose estimation accuracy is improved, but computational resource consumption increases
Solution Approach 1:
The patent divides the large image into multiple smaller regions based on head pose information. The machine-learning model processes only these relevant regions instead of the entire large image, maintaining pose estimation accuracy while significantly reducing computational resource consumption and energy usage.
Solution Approach 2:
The patent extracts and processes only the relevant portions of the image (regions containing bodies of interest) based on head pose data. By taking out and processing only these essential regions rather than the complete large image, the system achieves accurate pose estimation with reduced computational load.
2Reliability
If body-pose estimation is performed on large images, then comprehensive pose detection is improved, but processing time increases
Solution Approach 1:
The patent segments the large image into multiple smaller regions based on head pose information and processes only these relevant regions. This segmentation approach maintains comprehensive pose detection capability while significantly reducing processing time compared to analyzing the entire large image.
Solution Approach 2:
The patent performs preliminary processing by identifying head poses and determining relevant regions before executing the full body-pose estimation. This preliminary action of locating regions of interest first allows the system to maintain comprehensive detection while reducing overall processing time.
3Reliability
If the machine-learning model processes the entire large image, then detection completeness is improved, but computational efficiency deteriorates
Solution Approach 1:
The patent divides the large image into multiple smaller regions based on head pose information and processes only these segments. This segmentation maintains detection completeness for bodies of interest while dramatically improving computational efficiency by avoiding processing of irrelevant areas.
Solution Approach 2:
The patent extracts and processes only the relevant regions containing bodies of interest from the large image, based on head pose data. This extraction approach ensures detection completeness for target objects while improving computational efficiency by eliminating processing of non-relevant image areas.
Data Source
AI summary
Systems and techniques are described herein for pose estimation of a person. For instance, a method for pose estimation of a person is provided. The method may include obtaining an image of the person; obtaining a head pose of the person; determining a search region of the image based on the head pose; and determining a pose of the person based on the search region.