Humanoid Pose Estimation from Depth Maps Using Patch-Based Descriptors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting high-level information from depth maps, particularly for tracking humanoid forms, face challenges in capturing and processing the dynamic poses and movements of human subjects in a robust and efficient manner.
Innovation Solution
A patch-based learning method is employed to estimate the pose of humanoid forms from depth-map data by extracting descriptors from spatial bins in the depth map, matching them to a database, and using a weighted voting process to determine joint locations, which allows for the reconstruction of a 3D skeleton and tracking of movements without requiring markers or beacons.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional methods are used to extract high-level information from depth maps, then the processing can be simpler, but the ability to robustly capture and track dynamic poses and movements deteriorates
Solution Approach 1:
The depth map is divided into multiple patches distributed across the humanoid form, with each patch processed independently to extract local descriptors. This segmentation allows the system to capture local pose information robustly while maintaining computational efficiency through parallel processing of individual patches.
Solution Approach 2:
A database of descriptors is pre-computed and stored before actual pose estimation occurs. During runtime, the system performs efficient matching by comparing extracted descriptors against this pre-built database, significantly reducing real-time computational complexity while maintaining high tracking reliability.
2Measurement precision
If detailed descriptors are extracted from multiple spatial bins to improve pose estimation accuracy, then the measurement precision improves, but the processing time and computational load increase
Solution Approach 1:
The system extracts descriptors from multiple spatial bins within each patch, using more information than the minimum required for basic pose estimation. This excessive action in descriptor extraction improves measurement precision by capturing detailed local features, while the pre-computed database enables efficient matching that compensates for the increased processing requirements.
3Adaptability or versatility
If a database of stored descriptors is used to match and estimate poses, then the system can handle unseen poses, but the initial setup and data storage requirements increase
Solution Approach 1:
Instead of storing complete 3D models or extensive training data, the system creates a compact database of descriptors that capture essential pose information. These descriptors serve as efficient copies or representations of pose configurations, enabling the system to generalize to unseen poses while minimizing storage requirements through compressed descriptor encoding.
Data Source
AI summary
A method for processing data includes receiving a depth map of a scene containing a humanoid form. Respective descriptors are extracted from the depth map based on the depth values in a plurality of patches distributed in respective positions over the humanoid form. The extracted descriptors are matched to previously-stored descriptors in a database. A pose of the humanoid form is estimated based on stored information associated with the matched descriptors.