Rotation Invariant Fast Feature Descriptor for Mobile Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Handheld devices face limitations in processing power for image processing tasks, such as tracking and recognition in computer vision applications, due to their limited computing capabilities.
Innovation Solution
The development of rotation-invariant feature descriptors, specifically the Rotation Invariant Fast Feature (RIFF) descriptor, which uses radial gradient transforms and approximations to generate descriptors that are computationally efficient and robust for tracking and recognition tasks, allowing for real-time or near real-time processing without the need for orientation assignment or pixel interpolation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional feature descriptors are used for tracking and recognition, then recognition accuracy is maintained, but computational complexity increases and processing speed decreases
Solution Approach 1:
The feature descriptor is segmented into multiple bins (e.g., 8 bins) that represent different gradient orientations. This segmentation allows the system to process and compare only relevant orientation components, reducing the overall computational complexity while maintaining recognition accuracy through selective feature comparison.
Solution Approach 2:
The patent transforms the feature representation by computing gradient histograms in a rotated coordinate system that aligns with the dominant orientation. This parameter transformation enables rotation invariance while reducing the dimensionality of comparison operations, thereby improving processing speed without sacrificing recognition performance.
2Reliability
If rotation invariant feature descriptors are implemented, then robustness to orientation changes is improved, but computational cost increases
Solution Approach 1:
The system performs preliminary orientation estimation by computing gradient histograms in all directions and identifying the dominant orientation before final descriptor computation. This preliminary action allows subsequent descriptors to be computed in a rotated coordinate system, achieving rotation invariance more efficiently by avoiding redundant computations across all orientations.
Solution Approach 2:
The patent introduces a rotational dimension by transforming the gradient computation into a rotated coordinate system. This dimensional transformation allows the descriptor to inherently capture rotation invariance by expressing gradients relative to the dominant orientation rather than fixed image coordinates, reducing computational cost while maintaining robustness.
3Measurement precision
If detailed feature extraction is performed for accurate recognition, then recognition precision is improved, but processing time increases
Solution Approach 1:
The patent extracts only the most discriminative features by computing gradient histograms in specific orientation bins and selecting dominant orientations. This selective extraction of essential features maintains recognition precision by focusing on the most informative characteristics while reducing processing time by eliminating redundant feature computations.
Data Source
AI summary
Various methods for tracking and recognition with rotation invariant feature descriptors are provided. One example method includes generating an image pyramid of an image frame, detecting a plurality of interest points within the image pyramid, and extracting feature descriptors for each respective interest point. According to some example embodiments, the feature descriptors are rotation invariant. Further, the example method may also include tracking movement by matching the feature descriptors to feature descriptors of a previous frame and performing recognition of an object within the image frame based on the feature descriptors. Related example methods and example apparatuses are also provided.