Client-Side Visual Localization via Compressed Inverted Files
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current mobile visual location recognition systems face challenges in rapidly and accurately searching for images in large georeferenced databases due to complex retrieval tasks and communication delays, which are exacerbated by the need for low query times and efficient data processing and storage.
Innovation Solution
The proposed solution involves shifting the retrieval task from the server to the client by transferring the quantizer and compressed inverted file system of the most suitable subregion, and periodically updating the client with relevant visual words and their associated inverted files to enable local pose estimation, thereby reducing network delays and improving query times.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If the retrieval task is performed on the server with a large georeferenced database, then the comprehensiveness of the database is improved, but the communication delay and query time increase
Solution Approach 1:
The large georeferenced database is divided into multiple subregion databases, each covering a specific geographic area. The client device stores quantized visual word vocabularies and inverted file indexes for these subregions locally, enabling distributed storage and parallel query processing across multiple subregion databases without requiring centralized server access for every query operation.
Solution Approach 2:
Visual word vocabularies and inverted file indexes are pre-computed and stored on the client device before actual location recognition queries are executed. This preliminary preparation of quantized descriptors and indexing structures enables rapid local querying without requiring real-time server processing, significantly reducing query time while maintaining comprehensive database coverage.
2Loss of time
If the quantizer and inverted file system are transferred to the client, then the query time is reduced, but the data storage requirements on the client increase
Solution Approach 1:
Different components of the visual localization system are distributed between server and client based on their specific requirements. The server maintains the complete original image database and performs periodic updates, while the client stores compressed quantized vocabularies and inverted file indexes optimized for rapid local querying. This asymmetric distribution allows each system to operate in its optimal performance zone.
Solution Approach 2:
Instead of transferring entire high-resolution images and complete database structures to the client, the system creates compressed copies in the form of quantized visual word vocabularies and inverted file indexes. These compressed representations retain the essential information needed for location recognition while occupying minimal storage space on the client device.
3Measurement precision
If more visual words and inverted files are stored on the client, then the location recognition precision is improved, but the processing complexity increases
Solution Approach 1:
The system transforms the continuous visual feature space into a discrete quantized vocabulary, changing the parameter representation from continuous image descriptors to discrete visual word indices. This parameter transformation simplifies the matching process by converting complex image similarity comparisons into efficient index-based lookups, reducing processing complexity while maintaining location recognition precision.
Data Source
AI summary
There is provided a visual localization method comprising: (a) transmitting data representative of one or more detected visual features from a mobile device to a server; (b) estimating the location of the mobile device at the server based on the visual features received from the mobile device; (c) transmitting reference data associated with the estimated location from the server to the mobile device; and (d) the mobile device determining its location based on the reference data received from the server.