Visual Positioning System Movable Object Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current indoor wayfinding technologies, such as visual positioning systems (VPS), face challenges in accurately navigating within buildings due to the presence of movable objects and the inability to seamlessly transition between indoor and outdoor coordinate systems, leading to inefficient use of computing resources and incorrect directions.
Innovation Solution
A machine learning model is trained to distinguish between movable and unmovable objects within images captured by user devices, allowing the VPS to disregard movable objects and utilize only unmovable objects for location determination, thereby enhancing navigation accuracy and conserving resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a visual positioning system uses all detected objects for location determination, then more reference points are available for positioning, but movable objects cause errors and reduce positioning accuracy
Solution Approach 1:
The system extracts and removes movable objects from the set of detected objects, keeping only unmovable objects as reference points for the visual positioning system. This extraction principle resolves the contradiction by selectively taking out the harmful movable objects while retaining the useful unmovable objects, thus maintaining positioning accuracy while still utilizing multiple reference points.
Solution Approach 2:
The system applies different treatment to different types of objects: movable objects are excluded from positioning calculations while unmovable objects are included. This local quality differentiation resolves the contradiction by assigning different roles to different objects based on their mobility characteristics, ensuring that only suitable reference points contribute to positioning accuracy.
2Device complexity
If the system processes all objects including movable ones, then processing is simpler without classification, but computing resources are wasted on incorrect location data
Solution Approach 1:
The system performs preliminary classification of objects into movable and unmovable categories before the visual positioning system processes them for location determination. This preliminary action resolves the contradiction by filtering out inappropriate reference points in advance, preventing waste of computing resources on movable objects while maintaining relatively simple processing for the remaining unmovable objects.
3Adaptability or versatility
If the VPS uses all detected objects for navigation, then more features are available for location determination, but incorrect directions are provided due to movable objects
Solution Approach 1:
The system extracts movable objects from the navigation reference set, using only unmovable objects for determining navigation directions. This extraction ensures that the navigation system relies on stable, permanent features of the environment, thereby maintaining direction accuracy while still utilizing multiple reference points for flexible navigation.
Solution Approach 2:
The system incorporates feedback about object mobility characteristics to determine which objects should be used for navigation. By continuously identifying and excluding movable objects based on their detected properties, the system ensures that navigation directions are based on reliable, unmovable reference points, thus maintaining both accuracy and flexibility.
Data Source
AI summary
A device may receive images identifying interiors of buildings and movable objects and unmovable objects located in the interiors, and may train a machine learning model with the images to generate a trained machine learning model. The device may receive an image identifying an interior portion of a building and objects located in the interior portion, and may process the image, with the trained machine learning model, to identify a movable object and an unmovable object. The device may disregard the movable object to generate an image in which data identifying the movable object has been disregarded, and may process the image in which the data identifying the movable object has been disregarded, with a visual positioning system, to determine a location of the user device in the interior portion of the building. The device may perform one or more actions based on the location of the user device.