Image-Based Positioning Using Environmental Maps for VSLAM Drift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing VSLAM systems suffer from accumulation errors in feature point matching and reduced calculation accuracy of relative attitude due to low similarity between images, leading to inaccurate robot positioning.
Innovation Solution
A positioning device and method that utilize image processing techniques to extract similar images with matching feature vectors, estimating the position and attitude of an imaging device using environmental maps and feature point groups to improve matching accuracy and reduce errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature point matching is performed between target image and frame image in VSLAM, then robot position can be estimated, but accumulation error increases over time leading to reduced positioning accuracy
Solution Approach 1:
The patent applies preliminary action by pre-generating an environmental map with pre-extracted feature points and descriptors before the robot operates. This pre-processing creates a reference database that can be quickly queried during positioning, avoiding the need for real-time feature matching between arbitrary frame images and reducing accumulation error in position estimation.
2Measurement precision
If matching is performed on feature points in target image and frame image, then relative attitude can be calculated, but error in matching occurs when degree of similarity between images is low
Solution Approach 1:
The patent uses the pre-generated environmental map as an intermediary between the target image and frame images. Instead of directly matching features between similar frame images, the system queries the environmental map which contains pre-extracted feature points and descriptors, serving as a stable reference that improves matching reliability even when image similarity is low.
Solution Approach 2:
The patent creates a copy of the environmental structure by pre-extracting feature points and descriptors from training images to build the environmental map. This copied representation serves as a reference model that can be queried during operation, improving matching accuracy by comparing against a stable, pre-processed reference rather than directly matching variable frame images.
3Productivity
If immediate processing is performed in VSLAM to estimate camera position, then real-time positioning is achieved, but detailed calculation cannot be performed leading to accumulation error
Solution Approach 1:
The patent performs detailed feature extraction, descriptor generation, and environmental map construction as preliminary actions before real-time operation. This pre-processing creates a ready-to-use reference database, allowing the system to perform only query and match operations during real-time positioning, thus achieving both fast processing speed and high accuracy.
Data Source
AI summary
A positioning device includes an acquisition unit that acquires an environmental map of a target area being generated by using an image processing technique for estimating a three-dimensional shape of a target included in a two-dimensional image, and a plurality of captured images being used for generating the environmental map, an image retrieval unit that extracts, from a plurality of the captured images, a similar image having a feature vector similar to a feature vector determined based on a feature amount included in an input image, and an estimation unit that estimates a position and an attitude of an imaging device that has captured the input image by using a first feature point group of the input image and a third feature point group in the environmental map associated to a second feature point group of the similar image being matched with the first feature point group.