Feature Descriptor Clustering for Accurate Low-Memory Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems, such as those used in autonomous vehicles and extended reality applications, face inaccuracies due to changing lighting conditions and object movement, leading to increased memory and processing requirements, which affect system cost and performance.
Innovation Solution
A method involving a first and second clustering process to determine descriptor clusters and cluster centers, generating descriptor cluster data characterizing similarity, and storing this data in a repository, with iterative updates based on performance criteria.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the database of features is increased to improve recognition accuracy, then the accuracy of feature matching is improved, but the memory required and processing power consumed increase
Solution Approach 1:
The patent segments the feature database into multiple clusters using clustering algorithms. Instead of storing and processing all individual features, the system organizes them into representative clusters characterized by cluster centers and membership probabilities. This segmentation reduces the effective database size while maintaining recognition accuracy through the clustered representation.
Solution Approach 2:
The patent extracts the essential characteristics of feature clusters by computing cluster centers and their corresponding probability distributions. These extracted cluster representations capture the most important information from the original feature database, allowing the system to work with a reduced set of data that preserves the essential recognition capabilities.
2Measurement precision
If the database of features is increased to improve recognition accuracy, then the accuracy of feature matching is improved, but the processing power required increases
Solution Approach 1:
By segmenting features into clusters and using cluster centers as representatives, the patent significantly reduces the number of features that need to be processed during recognition tasks. This segmentation maintains accuracy while reducing computational load compared to processing the full feature database.
Solution Approach 2:
The patent changes the representation parameters from individual features to cluster centers with associated probability distributions. This parameter transformation allows the system to maintain rich representational information while working with a more compact data structure that requires less processing power.
3Productivity
If traditional feature comparison methods are used to handle changing lighting conditions and object movement, then the system can process features, but the accuracy of comparisons deteriorates
Solution Approach 1:
The patent incorporates feedback mechanisms where the system continuously learns from matching results and updates cluster representations. This feedback loop allows the system to adapt to changing lighting conditions and object movement by refining cluster centers and membership probabilities based on actual performance data.
Solution Approach 2:
The patent uses probability distributions as parameters to represent cluster memberships, allowing for flexible matching that accounts for variations in lighting and object position. This probabilistic parameter representation enables the system to maintain high accuracy under changing conditions by comparing probability distributions rather than relying on fixed feature matches.
Data Source
AI summary
Methods, systems, and apparatuses are provided to cluster and match image feature descriptors for use in various systems. For example, a computing device receives a location from a remote device. The computing device applies a first clustering process to a plurality of descriptors associated with the location to determine a number of descriptor clusters. The computing device also applies a second clustering process to the number of descriptor clusters to determine a descriptor cluster center for each of the number of descriptor clusters. Further, the computing device generates descriptor cluster data characterizing a similarity between the plurality of descriptors and the descriptor cluster centers. The computing device then transmits the descriptor cluster data to the remote device. The remote device may match descriptors to the descriptor cluster centers based on the descriptor cluster data.