Mobile Robot Position Estimation Using Visual Feature Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the position of a moving vehicle, such as mobile robots, are often unreliable due to errors in GPS and the failure of LIDAR to track position in dynamic environments, particularly outdoors.
Innovation Solution
A method and apparatus using a single camera to estimate the position of a moving vehicle by generating vectors between feature points in sequential image frames, clustering these vectors based on spatial correlations, and removing noise to determine the vehicle's position without relying on GPS or odometry, employing techniques like optical flow, Delaunay triangulation, and RANSAC.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If GPS is used for position estimation, then position information can be obtained, but the position estimation is erroneous and only works properly in outdoor environment
Solution Approach 1:
The patent introduces feature points extracted from image frames as an intermediary to establish spatial relationships. Instead of relying directly on GPS signals, the system uses visual features from the camera to create vectors and determine fundamental matrices, which serve as mediators to infer position information in a reliable manner across different environments
Solution Approach 2:
The patent replaces the GPS satellite-based electromagnetic signal system with a camera-based optical measurement system. By substituting the mechanical/electromagnetic GPS infrastructure with a visual processing system using feature point matching and fundamental matrix calculation, the system achieves environment-independent position estimation
2Measurement precision
If LIDAR is used for position estimation, then accurate position information can be obtained, but it easily fails to track position due to non-reflection of light when used outdoors
Solution Approach 1:
The patent uses readily available camera images as disposable, easily obtainable data sources instead of relying on expensive LIDAR systems. By extracting feature points from standard camera frames and using them to compute fundamental matrices, the system achieves reliable position tracking without the complexity and cost of LIDAR hardware
Solution Approach 2:
The patent makes the camera serve multiple functions: it not only captures images for visual processing but also provides the fundamental data source for position estimation through feature point matching. This multi-functional use of the camera eliminates the need for separate LIDAR hardware while maintaining tracking reliability across indoor and outdoor environments
3Measurement precision
If multiple sensors are used for position estimation, then measurement accuracy can be improved, but device complexity and cost increase
Solution Approach 1:
The patent makes the camera serve multiple functions: it not only captures images for visual processing but also provides the fundamental data source for position estimation through feature point matching. This multi-functional use of the camera eliminates the need for separate LIDAR or GPS hardware while maintaining tracking reliability
Solution Approach 2:
The patent merges the position estimation function with the existing camera system by integrating feature point extraction and fundamental matrix calculation into the image processing pipeline. This combining of functions reduces overall system complexity while achieving accurate position estimation without requiring additional specialized sensors
Data Source
AI summary
An apparatus of estimating a position of a moving vehicle such as a robot includes a feature point matching unit which generates vectors connecting feature points of a previous image frame and feature points of a current image frame, corresponding to the feature points of the previous image frame, and determines spatial correlations between the feature points of the current image frame, a clustering unit which configures at least one motion cluster by grouping at least one vector among the vectors based on the spatial correlations in a feature space, and a noise removal unit removing noise from each motion cluster, wherein the position of the moving vehicle is estimated based on the at least one motion cluster.