AR Object Detection via Homography and Feature Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing augmented reality systems lack the ability to create configurable, context-sensitive functions that minimize misdetection of objects, are not computationally expensive, and avoid confusing multiple objects.
Innovation Solution
A system and method that utilize a computer to process images of monitored devices, determine feature points, and match them with reference feature points to compute a homography matrix, thereby identifying object zones and determining object states based on associated properties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object state detection is performed using deep learning networks, then detection accuracy is improved, but computational cost increases
Solution Approach 1:
The patent divides the image processing into distinct stages: first extracting feature points using a lightweight method, then performing homography transformation, and finally applying deep learning only to the transformed image regions of interest. This segmentation reduces computational cost by applying the expensive deep learning method only where necessary rather than to the entire image.
Solution Approach 2:
The patent performs preliminary feature point extraction and homography transformation before applying deep learning for object state detection. By preparing the image data in advance through these preprocessing steps, the system optimizes the subsequent deep learning inference, improving accuracy while managing computational resources more efficiently.
2Productivity
If multiple objects are detected in an image, then comprehensive monitoring is improved, but misdetection and confusion between objects increases
Solution Approach 1:
The patent segments the monitoring process into distinct phases: feature point extraction for each detected object, individual homography transformation for each object's region, and separate object state determination. This segmentation allows the system to maintain comprehensive monitoring of multiple objects while reducing confusion by processing each object's state independently through its own transformed image region.
3Adaptability or versatility
If context sensitive functions are added to AR experiences, then functionality is improved, but system complexity increases
Solution Approach 1:
The patent creates a universal image processing pipeline that handles multiple object types and detection scenarios through a single consistent approach. The homography transformation and feature point matching mechanism works across different objects and contexts, providing adaptable AR functions without requiring separate complex systems for each object type, thus managing overall system complexity.
Data Source
AI summary
A system for performing functions in AR experiences is provided having a computer receiving an image of a monitored device having at least one object; the computer processes said image to determine a plurality of feature points; a database is in data communication with said computer and stores a plurality of reference images with reference feature points and reference object zones, each having a property; the computer matches at least some of the plurality of feature points with at least some of the plurality of reference feature points to compute a homography matrix; the computer transforms the image into a transformed image using the homography matrix; the computer identifies an object zone in the transformed image based on the reference object zone; the computer determines an object state by processing a sub-image in the object zone of the transformed image based on the property associated with the reference object zone.