Hybrid Pose Estimation Using Confidence-Based Recognition and Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current motion recognition technologies face challenges in accurately estimating user poses from depth images, particularly when confidence levels in recognition results are below predetermined thresholds, leading to inefficiencies in tracking and re-estimation processes.
Innovation Solution
A hybrid pose recognition and tracking method that utilizes a combination of recognition and tracking units, employing classification trees for initial pose recognition and user models for tracking, with confidence-based decision-making to switch between recognition and tracking processes, and selecting appropriate user models based on similarity to enhance accuracy and adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pose recognition is performed using classification trees, then pose estimation can be achieved, but accuracy deteriorates when confidence levels are below predetermined thresholds
Solution Approach 1:
A user model serves as an intermediary between the depth image and the pose estimation. When confidence from classification tree recognition is insufficient, the system uses the user model to fill in missing pose information, mediating between the unreliable recognition result and the required accurate pose estimation.
Solution Approach 2:
The system changes the approach based on confidence parameter thresholds. When confidence is above the threshold, classification tree recognition is used; when confidence is below the threshold, user model tracking is used. This parameter-based decision-making resolves the contradiction by selecting the appropriate method based on reliability conditions.
2Productivity
If tracking is performed using user models, then pose can be tracked, but the system complexity increases due to multiple models and confidence thresholds
Solution Approach 1:
The system dynamically switches between classification tree recognition and user model tracking based on confidence levels. This dynamic approach allows the system to adapt its complexity based on the situation, using simpler recognition when possible and more complex tracking only when necessary.
Solution Approach 2:
The user model is prepared in advance with predetermined body part configurations and relationships. This preliminary preparation allows the system to quickly switch to tracking mode when needed, reducing the computational overhead of model creation and enabling efficient operation.
3Reliability
If confidence-based switching between recognition and tracking is implemented, then pose estimation reliability improves, but processing time increases due to additional decision-making steps
Solution Approach 1:
The confidence threshold and switching logic are predetermined and prepared in advance. The system continuously monitors confidence levels and is ready to switch between recognition and tracking modes without delay, minimizing the time overhead of decision-making.
Solution Approach 2:
The system uses feedback from confidence level assessment to dynamically adjust the processing path. When confidence is sufficient, recognition is used; when confidence is insufficient, tracking is activated. This feedback mechanism ensures reliable pose estimation while optimizing processing time based on actual conditions.
Data Source
AI summary
A method and apparatus for estimating a pose that estimates a pose of a user using a depth image is provided, the method including, recognizing a pose of a user from a depth image, and tracking the pose of the user using a user model exclusively of one another to enhance precision of estimating the pose.