Visual Positioning via Semantic Graphs and Random Walk
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image retrieval technologies for visual positioning, particularly in VR, AR, and unmanned driving, face challenges with poor performance in scenes with significant changes in perspectives and illuminations, especially those with many indoor dynamic objects and repetitive textures, leading to inaccurate coarse positioning and reduced robustness.
Innovation Solution
A visual positioning method that involves collecting and analyzing images through semantic segmentation to generate graphs representing entity types and distances, using random walk paths to determine description information, and matching this information with pre-stored data to identify candidate entities and accurately position image collections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If image retrieval technologies are used for visual positioning, then positioning functionality is provided, but positioning accuracy deteriorates in scenes with significant changes in perspectives and illuminations
Solution Approach 1:
The patent segments the image into multiple superpixels and identifies entities within each superpixel, then constructs a semantic graph representing spatial relationships between entities. This segmentation approach enables the system to handle perspective changes by focusing on local entity characteristics rather than global image features, thereby maintaining positioning accuracy under varying perspectives and illuminations.
2Reliability
If general image retrieval technologies are adopted, then positioning is achieved, but robustness deteriorates in environments with many indoor dynamic objects and repetitive textures
Solution Approach 1:
The patent transitions from traditional 2D image retrieval to a semantic graph representation that incorporates spatial relationships and entity types as additional dimensions. By constructing a semantic graph that captures the topological relationships between entities, the system can distinguish between different scenes even when they contain similar repetitive textures or dynamic objects, thereby improving robustness against these harmful factors.
3Productivity
If coarse positioning is performed first, then positioning process is completed, but positioning precision deteriorates
Solution Approach 1:
The patent performs preliminary semantic segmentation and entity identification to construct a semantic graph before conducting the matching process. This preliminary action extracts meaningful semantic information and spatial relationships that are then used to guide the precise matching between query and database images, enabling the system to achieve high positioning precision without sacrificing efficiency through a two-stage approach.
Data Source
AI summary
The disclosure provides a visual positioning method and apparatus, an electronic device and a computer-readable storage medium. The method includes: generating a semantic graph by semantically identifying collected images; determining description information of each entity through a random walk algorithm in the established semantic graph; determining candidate entities matching each entity in a preset entity map based on the description information; and positioning a collection area of the current image based on an area where the candidate entities are located in the preset entity map. The description information of each entity node constructed with the random walk algorithm not only contains semantic information of the corresponding node, but also local constraint information between semantics.