Dynamic Feature Database for Object Recognition Adaptability

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing object recognition systems for Augmented Reality on mobile platforms face challenges in real-time object detection and tracking, as their feature databases are static and do not adapt to changing viewpoints, leading to suboptimal detection performance.

Innovation Solution

A method for dynamically updating a feature database by extracting new features from images and selectively adding them if they meet certain criteria, such as uniqueness and relevance, to improve detection accuracy and adapt to changing object views without requiring deep learning or neural networks.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If the feature database is built pre-runtime with multiple sample images from various viewpoints, then the initial detection capability is established, but the detection performance remains static and cannot adapt to changing viewpoints

Engineering Contradiction:

Improvedetection adaptabilityVSAvoiddetection performance

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent transforms the static feature database into a dynamic one by implementing incremental learning. The system continuously updates the feature database with new features extracted from detected objects in real-time images, allowing the database to adapt to changing viewpoints and object variations while maintaining reliable detection performance through proven feature validation

Inventive Principle:

Principle #15Dynamics

2Reliability

If new features are continuously added to the feature database to improve detection accuracy, then detection performance improves, but the database size increases and processing overhead increases

Engineering Contradiction:

Improvedetection accuracyVSAvoiddatabase size

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies parameter changes by selectively adding only those new features that meet specific criteria (uniqueness, relevance, information content) to the feature database. This controlled approach improves detection accuracy by adding meaningful features while preventing database bloat through systematic filtering and consolidation processes

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent implements feature consolidation processes that identify and remove redundant or duplicate features from the database. By discarding unnecessary features and retaining only the most informative ones, the system maintains a manageable database size while preserving detection accuracy

Inventive Principle:

Principle #34Discarding and recovering

3Productivity

If the feature database is made static to reduce processing overhead, then processing efficiency is maintained, but the system cannot adapt to new object viewpoints or variations

Engineering Contradiction:

Improveprocessing efficiencyVSAvoidviewpoint adaptability

Core Design Contradiction:

ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent implements periodic updates to the feature database during runtime, rather than maintaining a completely static database. By periodically incorporating new features from detected objects and consolidating the database at intervals, the system adapts to new viewpoints while maintaining processing efficiency through controlled update frequencies

Inventive Principle:

Principle #19Periodic action

Data Source

PatentUS11263475B2Incremental learning for dynamic feature database management in an object recognition system

Publication Date: 2022.03.01 QUALCOMM INC

AI summary

A method of dynamically updating a feature database that contains features corresponding to a known target object includes providing an image, extracting a first set of features from within the captured image, and comparing the first set of features to the features stored in the feature database. If it is determined that the target object is present in the image then at least one of the extracted features of the first set that are not already included in the feature database are added to the feature database.