Human Pose Detection via Key Joint Segmentation and Statistical Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional model-based approaches for recognizing human body poses are inefficient in distinguishing complex poses, require high computational resources, and are not robust against segmentation errors, making them unsuitable for embedded environments and quick movements.
Innovation Solution
An apparatus and method that utilize a key joint detector to analyze images, a database to store object poses, and a pose retriever to identify the most likely pose based on detected key joint data, employing techniques like Inverse Kinematics and weighting factors to improve accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a conventional model-based approach is used to recognize human body poses, then pose recognition capability is provided, but computational complexity increases and processing speed decreases
Solution Approach 1:
The patent segments the human body into key joint points (e.g., head, shoulders, elbows, wrists, hips, knees, ankles) and detects their positions independently. This segmentation approach simplifies the overall pose recognition problem by breaking it down into multiple simple key point detection tasks, reducing computational complexity while maintaining pose recognition capability.
Solution Approach 2:
The patent extracts only the essential key joint points from the complete body model, rather than processing the entire body surface or volume. By taking out only the critical keypoints needed for pose representation, the system achieves efficient pose recognition with significantly reduced computational requirements.
2Adaptability or versatility
If a conventional model-based approach is used to distinguish complex poses with overlapping body parts, then pose differentiation capability is provided, but measurement precision decreases due to occlusions
Solution Approach 1:
The patent uses an intermediary statistical model that learns the natural correlations and constraints between different key joints from training data. This statistical model acts as a mediator that fills in missing or ambiguous key point information by reasoning about the probable positions based on learned body geometry and pose statistics, thereby maintaining precision even when body parts overlap or occlude each other.
3Adaptability or versatility
If a conventional model-based approach is used for pose recognition, then comprehensive pose analysis is provided, but processing speed decreases making it unsuitable for quick movements
Solution Approach 1:
The patent divides comprehensive pose analysis into independent key point detection tasks, where each key joint is detected separately using optimized algorithms. This segmentation enables parallel processing of multiple keypoints, significantly increasing processing speed while still providing comprehensive pose analysis through the collection of all detected key points.
Solution Approach 2:
The patent focuses on detecting only the essential key joints necessary for pose representation, rather than analyzing every detail of the body surface. This partial action approach provides sufficient pose analysis information at much higher speeds, making the system suitable for tracking quick movements.
4Adaptability or versatility
If a conventional model-based approach is used for pose recognition, then pose estimation capability is provided, but reliability decreases due to sensitivity to segmentation errors
Solution Approach 1:
The patent employs a statistical model that incorporates feedback mechanisms to correct detection errors. The model learns from training data the expected relationships between key joints and uses this knowledge to adjust and correct detected key point positions, making the pose estimation robust against segmentation errors and variations in image quality.
Solution Approach 2:
The patent pre-trains a statistical model on large datasets of human poses before deployment. This beforehand cushioning prepares the system to handle various pose variations and segmentation errors by pre-learning robust patterns, allowing the system to maintain reliable pose estimation even when individual key point detections are imperfect.
Data Source
AI summary
An apparatus and method detecting an object pose are provided. Key joint data of an object may be extracted, a candidate pose may be generated based on the extracted key joint data, and a most likely pose may be retrieved using a database, based on the generated candidate pose.