Contrastive Landmark Localization for Adaptive Indoor Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing localization methods for mobile devices, such as AGVs, in environments like factories or warehouses, face challenges with error accumulation, high cost, impracticality in indoor spaces, and complex object detection, especially when using topological maps, which require manual updates and rely on resource-intensive deep neural networks or geometric representations.
Innovation Solution
A method utilizing contrastive learning to automatically detect landmarks, create topological maps, and update them by retraining the model when new landmarks are detected, using a novel Siamese network architecture that performs multiple similarity analyses efficiently, suitable for resource-constrained devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep neural networks are used for landmark detection in topological maps, then detection accuracy is improved, but processing cost and device resource requirements increase
Solution Approach 1:
The patent replaces expensive deep neural networks with a simpler, more efficient neural network architecture that consumes fewer computational resources. This simpler model achieves adequate detection accuracy without the high processing costs of complex DNNs, effectively using a 'cheaper' computational approach for the same task.
Solution Approach 2:
The patent extracts and uses only the essential features from images for landmark detection, rather than processing complete high-dimensional data through complex networks. By focusing on key discriminative features, the system achieves accurate detection with reduced computational overhead.
2Reliability
If manual intervention is used to update topological maps, then map accuracy is maintained, but operation time and complexity increase
Solution Approach 1:
The system automatically detects when environmental changes occur and triggers map updates without manual intervention. The neural network continuously monitors landmarks and autonomously determines when the topological map needs regeneration, making the system self-sufficient and eliminating the need for manual map maintenance.
Solution Approach 2:
The patent implements a feedback mechanism where the system continuously compares current landmark detections with the stored topological map. When discrepancies indicate environmental changes, the system automatically initiates map updates, creating a closed-loop control system that maintains accuracy without human intervention.
3Measurement precision
If complete retraining of the contrastive learning model is performed when new landmarks are detected, then localization accuracy is improved, but processing time increases
Solution Approach 1:
Instead of performing complete model retraining, the patent uses incremental learning to update only the necessary portions of the contrastive learning model when new landmarks are detected. This partial update approach maintains localization accuracy while significantly reducing the processing time and computational resources required compared to full retraining.
Data Source
AI summary
A method in a contrastive learning node includes obtaining a set of landmark feature images representing a set of landmark features in the environment; obtaining a first feature image derived from a first image captured by the mobile device; determining, using a contrastive learning model, whether the first feature image is similar to any of the set of landmark feature images, wherein the contrastive learning model is trained based on a first set of feature images; responsive to determining that a landmark feature image is similar to the first feature image, initiating determination of a position of the mobile device; and responsive to determining that none of the set of landmark features images are similar to the first feature image, retraining the contrastive learning model based on an updated set of feature images comprising the first feature image.