Indoor Positioning via BIM Virtual Camera Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing indoor positioning technologies face challenges in construction sites due to the lack of telecommunications and network equipment, making it difficult to implement broadband and Wi-Fi-based methods, and GPS is inaccurate indoors due to satellite signal unavailability.
Innovation Solution
An indoor positioning system utilizing a computing device with a BIM model, virtual cameras, and deep learning networks to generate and match virtual and captured images, combined with visual-inertial odometry for accurate positioning, allowing for semi-automatic feature extraction and manual correction of errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If GPS positioning method is used, then outdoor positioning accuracy is improved, but indoor positioning accuracy deteriorates due to satellite signal unavailability
Solution Approach 1:
The patent introduces an intermediary system consisting of virtual cameras, BIM models, and deep learning networks that mediate between the unavailable satellite signals and the positioning requirement. The system captures images through virtual cameras, processes them through deep learning networks to extract features, and matches these features with BIM models to determine position, thereby replacing the direct GPS signal path with an indirect but functional alternative.
Solution Approach 2:
The patent creates virtual copies of the physical environment through BIM (Building Information Modeling) and virtual camera images. These digital twins replicate the spatial and visual characteristics of the actual construction site, allowing the system to perform positioning by matching captured images against the virtual model copies, thus achieving accurate indoor positioning without satellite signals.
2Adaptability or versatility
If broadband or Wi-Fi based positioning methods are used, then indoor positioning capability is improved, but ease of implementation deteriorates due to lack of network equipment in construction sites
Solution Approach 1:
The system performs self-service by using the mobile device's own camera and processing capabilities to capture images and extract features locally. The deep learning network runs on-device or edge-computing, eliminating the need for external network infrastructure. The system serves itself by generating and processing its own visual data without requiring broadband or Wi-Fi networks.
Solution Approach 2:
The patent employs a universal approach by using standard image capture and processing technologies that can function across diverse environments without requiring specialized infrastructure. The deep learning-based feature extraction and BIM matching system can operate in any indoor environment with visual features, making it universally applicable whether or not network equipment is present.
3Productivity
If automated feature extraction is used, then productivity is improved, but measurement precision deteriorates due to repetitive and symmetrical structures in construction sites
Solution Approach 1:
The system implements feedback mechanisms where the deep learning network's automated feature extraction results are continuously refined through comparison with BIM model data. The matching process provides feedback that helps correct errors introduced by repetitive structures, as the system can distinguish between identical-looking features by their spatial relationships and contextual information from the BIM model.
Solution Approach 2:
The patent combines multiple processing approaches into a composite system: automated deep learning feature extraction is combined with manual or semi-manual verification steps, and both are integrated with BIM model matching. This composite approach leverages the speed of automated extraction while compensating for its inaccuracies through additional processing layers, achieving both high productivity and precision.
Data Source
AI summary
An indoor positioning system and method are provided. The indoor positioning method includes: establishing an image database through a BIM model of a target area, and using a trained deep learning model to extract features of a virtual image; after obtaining a captured image in the target area, using the trained deep learning model to extract features thereof, and performing similarity matching with the image database to calculate a spatial position of a most similar image; calculating the most similar image and its essential matrix through multiple sets of feature points, and obtaining capturing positions and capturing pose parameters as positioning results; projecting the BIM model to a tracking captured image, and updating the positioning results and the capturing pose parameters with a visual inertial odometer; and continuously correcting the positioning results and the capturing pose parameters by detecting horizontal and vertical planes from the tracking captured image.