Visual Localization Using Vertical Feature Lines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional visual localization methods face challenges in achieving high precision due to changes in illumination conditions, rain, or snow, as they rely on feature points extracted based on luminance changes, which leads to ineffective matching and low localization accuracy.
Innovation Solution
A visual localization method that generates a descriptor based on the horizontal viewing angle between vertical feature lines of a building, using a ring array or circle representation, to match with a preset descriptor database and determine the photographing location, even under varying conditions, and constructs descriptor databases on a large scale using satellite images without on-site image capture.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If feature points are extracted based on luminance change of image pixels, then the conventional visual localization method can be implemented, but the localization precision deteriorates when illumination conditions change greatly or weather conditions (rain/snow) occur
Solution Approach 1:
The patent changes the extraction parameters from luminance-based features to geometric-based features (vertical feature lines). By using geometric parameters (line positions, angles, intersections) instead of luminance parameters, the system achieves invariance to illumination and weather changes while maintaining localization precision
Solution Approach 2:
The patent substitutes the optical/luminance-based feature extraction mechanism with a geometric/structural mechanism. Instead of relying on pixel intensity variations (luminance), the system uses geometric properties of vertical lines in the image, which are derived from building structures and remain stable under varying illumination and weather conditions
2Measurement precision
If a large quantity of images are pre-captured and feature points are extracted to construct a three-dimensional spatial map, then visual localization can be performed, but the computational complexity and storage requirements increase
Solution Approach 1:
The patent extracts only the essential geometric features (vertical feature lines and their intersections) from images, discarding redundant information. This extraction approach reduces the data dimensionality and complexity while retaining the key structural information needed for accurate localization
Solution Approach 2:
The patent creates simplified geometric representations (copies) of building structures using vertical feature lines and their intersection points. These geometric copies serve as efficient descriptors that can be stored and matched without requiring the original complex images, reducing storage and computational requirements
Data Source
AI summary
The disclosure provides example visual localization methods and devices. One method includes that a terminal device obtains an image of a building. The terminal device generates a descriptor based on the image. The descriptor includes information about a horizontal viewing angle between a first vertical feature line and a second vertical feature line in the image. The first vertical feature line indicates a first facade intersection line of the building, and the second vertical feature line indicates a second facade intersection line of the building. The terminal device performs matching in a preset descriptor database based on the descriptor, to obtain localization information of a photographing place of the image.