3D Skeleton Mapping for Crowded Pose Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional pose and gesture detection systems perform poorly in crowded environments where multiple bodies are in close proximity, as they rely on visual features that are perspective-dependent and expensive to compute, making it difficult to accurately distinguish between individuals.
Innovation Solution
The solution employs 3D skeleton mapping using geometric constraints, formulated as a series of matrix multiplications, which can be processed on GPU hardware, and does not require the detection of personally identifiable information, allowing for automated systems to track and distinguish between individuals without relying on expensive neural processing or visual features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional visual feature-based pose detection is used, then the system can detect pose in simple environments, but it performs poorly when multiple bodies are in close proximity
Solution Approach 1:
The patent transitions from 2D visual feature detection to 3D skeleton mapping by introducing depth information through multiple camera views. By reconstructing skeletons in 3D space and applying geometric constraints across multiple perspectives, the system achieves accurate pose detection in crowded environments where 2D methods fail due to occlusion and proximity.
Solution Approach 2:
The patent introduces geometric constraints as an intermediary mechanism to bridge multiple camera perspectives. These constraints (epipolar geometry, triangulation) serve as mathematical mediators that correlate skeletal keypoints across different views, enabling reliable 3D reconstruction and individual distinction even when bodies are in close proximity or occluding each other.
2Measurement precision
If expensive neural processing is used for visual feature detection, then detection may be more accurate, but computational costs increase significantly
Solution Approach 1:
The patent replaces complex neural network-based visual feature processing with geometric constraint-based skeleton correlation. Instead of using expensive deep learning models to extract and match visual features, the system uses mathematical geometry (epipolar constraints, triangulation) to directly compute 3D skeleton positions from multiple 2D projections, dramatically reducing computational requirements while maintaining accuracy.
Solution Approach 2:
The patent changes the fundamental parameters of detection from visual features (color, texture, shape) to geometric parameters (skeletal keypoint coordinates, spatial relationships). This parameter transformation enables the use of efficient geometric algorithms instead of computationally intensive neural networks, achieving low-cost real-time processing.
3Loss of information
If visual features are used for tracking, then individual identification may be achieved, but perspective dependence makes matching between distant perspectives difficult
Solution Approach 1:
The patent resolves perspective dependence by transforming 2D perspective-dependent visual features into 3D perspective-independent skeleton coordinates. By reconstructing skeletons in 3D space using geometric constraints from multiple views, the system creates a representation that is invariant to camera perspective, enabling reliable individual tracking across distant and varied perspectives.
4Device complexity
If conventional 2D pose detection is used, then processing is simpler, but reliable distinction between individuals in 3D space cannot be achieved
Solution Approach 1:
The patent enhances 2D pose detection by adding the depth dimension through multi-view geometry. By projecting 3D skeleton models onto multiple 2D camera planes and enforcing geometric consistency across views, the system achieves accurate 3D individual distinction while building upon relatively simple 2D keypoint detection foundations.
Data Source
AI summary
A system includes processing hardware and a memory storing software code. When executed, the software code receives first skeleton data including a first location of each of multiple skeletal key-points from the perspective of a first camera, receives second skeleton data including a second location of each of the skeletal key-points from the perspective of a second camera, correlates first and second locations of some or all of the multiple skeletal key-points to produce correlated skeletal key-point location data for each of at least some skeletal key-points. The software code further merges the correlated skeletal key-point location data for each of those at least some skeletal key-points to provide merged location data, and generates, using the merged location data and the locations of the first, second, and third cameras, a mapping of the 3D pose of a skeleton.