Pose Smoothing for Stable XR Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
XR devices suffer from noise and inaccuracies in object tracking due to camera motion and environmental factors, leading to degraded performance and user experience, particularly in extended reality applications.
Innovation Solution
A filtering algorithm that combines camera space-specific filter parameters with world space representation to adaptively filter pose data, reducing noise while maintaining accuracy and latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object tracking is performed in XR devices using camera-based methods, then the system can provide realistic and entertaining XR experiences by displaying virtual content based on tracked object positions, but noise and inaccuracies in measured position and orientation degrade tracking performance and user experience
Solution Approach 1:
The system performs preliminary actions by predicting the expected position of the tracked object based on its previous position and velocity. This prediction is done before the actual measurement is processed, allowing the system to have an expectation of what the measurement should be. The prediction serves as a reference to compare against the actual noisy measurement, enabling the system to filter out anomalies and noise by identifying measurements that deviate significantly from the predicted trajectory.
Solution Approach 2:
The system implements feedback by continuously comparing actual measurements with predicted positions and using this information to adjust tracking. The measurement prediction module uses feedback from previous tracking states to refine position estimates. When noise causes significant deviations, the feedback mechanism allows the system to recognize these as anomalies and correct them by relying more on the predicted position, thereby improving overall tracking accuracy and reducing the impact of noise.
2Stability of the object's composition
If filtering algorithms are applied to reduce noise in pose data, then tracking stability improves, but latency and processing time may increase
Solution Approach 1:
The measurement prediction module performs preliminary calculations of expected object position based on previous tracking data before actual measurements are fully processed. By having predictions ready in advance, the system can quickly compare and filter measurements without requiring complex real-time filtering algorithms, thus maintaining tracking stability while minimizing processing latency.
Solution Approach 2:
The system uses its own historical tracking data and physics-based motion models to generate predictions and filter measurements, rather than relying on external or computationally intensive filtering systems. This self-service approach allows the device to perform noise reduction using lightweight algorithms that maintain stability without introducing significant latency, as the filtering logic is embedded in the core tracking pipeline.
Data Source
AI summary
Examples in the present disclosure relate to systems and methods for reducing noise in object tracking data. Images of an object are obtained via one or more cameras. The images are processed to obtain first pose data indicative of a pose of the object over time. The first pose data is represented in a camera space. The first pose data is transformed to second pose data represented in a world space. The second pose data is filtered using a smoothing filter to generate filtered pose data. The filtering includes, for each pose data item in a time series of the second pose data, using a rotation transformation between the world space and camera space to apply one or more camera space-specific filter parameters to the pose data item that is represented in the world space. The pose of the object is dynamically tracked based on the filtered pose data.