Upper-Body Pose Extraction from Depth Maps Using Medial Axes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for processing depth maps struggle to accurately extract and track the pose of humanoid forms, especially when only the upper body is captured or when the form is seated against a background, due to difficulties in separating limbs and torso from the background and identifying their positions and orientations.
Innovation Solution
A method using a digital processor to identify the head and arms in a depth map by locating ridges and depth edges, and estimating the upper-body pose by finding 3D medial axes that satisfy a predefined body model while minimizing error, without referencing the lower body, to extract 3D coordinates of shoulder joints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing methods process depth maps to identify humanoid forms, then object identification is achieved, but accuracy deteriorates when only upper body is captured or when seated against background
Solution Approach 1:
The patent segments the depth map processing into distinct stages: initial identification of head and arms, extraction of 3D medial axes, and estimation of upper-body pose parameters. This segmentation allows the system to focus computational resources on upper-body features specifically, improving accuracy in scenarios where only the upper body is visible.
Solution Approach 2:
The patent extracts and processes only the relevant upper-body components (head and arms) from the depth map, discarding irrelevant lower-body information. This extraction approach enables the system to maintain high accuracy in upper-body pose estimation while being adaptable to scenarios where the lower body is not visible or relevant.
2Measurement precision
If depth map processing includes locating ridges and depth edges to identify head and arms, then pose extraction accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary identification of head and arm regions by locating ridges and depth edges before proceeding to 3D medial axis extraction and pose estimation. This preliminary action simplifies subsequent processing by pre-segmenting the region of interest, reducing the overall computational complexity despite the added initial step.
Solution Approach 2:
The patent applies different processing techniques to different parts of the depth map: ridge and depth edge detection is applied specifically to regions containing the head and arms, while other regions receive minimal or no processing. This localized approach improves identification accuracy for critical features while minimizing unnecessary computational complexity across the entire depth map.
3Adaptability or versatility
If the system extracts upper-body pose without reference to lower body, then adaptability to seated users is improved, but information completeness for full-body pose estimation deteriorates
Solution Approach 1:
The patent performs partial pose estimation by focusing exclusively on upper-body components (head and arms) rather than attempting to estimate the complete full-body pose. This partial action approach enables the system to maintain high adaptability and accuracy for seated users while accepting that lower body information is not recovered, representing a deliberate trade-off rather than information loss.
Data Source
AI summary
A method for processing data includes receiving a depth map of a scene containing at least an upper body of a humanoid form. The depth map is processed so as to identify a head and at least one arm of the humanoid form in the depth map. Based on the identified head and at least one arm, and without reference to a lower body of the humanoid form, an upper-body pose, including at least three-dimensional (3D) coordinates of shoulder joints of the humanoid form, is extracted from the depth map.