Virtual Point Projection for VSLAM Self-Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing self-localization methods, such as those disclosed in Patent Literature 1, are unable to perform accurate self-localization when latitude and longitude information is unknown, leading to inaccurate position estimation of robots using visual simultaneous localization and mapping (VSLAM).
Innovation Solution
An information processing apparatus and method that projects virtual points onto images to extract a second image for improved feature point matching, allowing for accurate estimation of the position and pose of an imaging device using feature points in both the target image and the second image, thereby enhancing self-localization accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If feature point matching is performed using only the immediately preceding image in VSLAM, then real-time processing is enabled, but position estimation accuracy deteriorates when three-dimensional position errors occur
Solution Approach 1:
The system pre-generates an environment map containing multiple images and their three-dimensional position information before self-localization is needed. This preliminary preparation allows the robot to perform accurate self-localization by comparing current images against the pre-built map, rather than relying solely on incremental matching with the immediately preceding image, thus resolving the contradiction between real-time processing and accuracy.
2Measurement precision
If geotag images are used for self-localization, then position accuracy can be improved, but the method becomes inapplicable when latitude and longitude information is unknown
Solution Approach 1:
The system creates a virtual environment map that copies and stores multiple images along with their three-dimensional position and pose information. This virtual map serves as a reference for self-localization, replacing the need for geotag images that require latitude and longitude. The copying approach allows accurate self-localization without dependency on geographic coordinate systems, thus resolving the contradiction between accuracy improvement and adaptability.
3Measurement precision
If multiple images are used for feature point matching to improve accuracy, then computational complexity increases
Solution Approach 1:
The environment map is constructed in advance, organizing multiple images and their corresponding three-dimensional position information into a structured format. During self-localization, the system efficiently queries this pre-organized map rather than performing complex real-time processing of multiple images, thus reducing computational complexity while maintaining high accuracy.
Solution Approach 2:
The system extracts and utilizes only the necessary features from the environment map (images and their three-dimensional positions) that are relevant to current self-localization needs. This selective extraction approach reduces the computational burden by focusing processing on essential data rather than analyzing all available information, resolving the contradiction between accuracy and computational complexity.
Data Source
AI summary
An information processing apparatus according to the present disclosure includes: a projection unit configured to arrange a predetermined number of virtual points in a front direction of an imaging device that captures a first image, by using an estimated position of the imaging device as a starting point, and project the virtual points onto each of a plurality of images used for generating an environment map being three-dimensional information; an image extraction unit configured to extract a second image from the plurality of images, based on a result of the projection of the virtual points; and an estimation unit configured to estimate, by using a feature point included in a target image being a target for estimating a position and a pose of an imaging device and a feature point included in the second image, aa position and a pose of the imaging device that captures the target image.