Object Identification Using Homography Matrices for Accurate Counting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In object identification using local feature values, existing methods struggle to accurately determine the number of objects in a query image when multiple database images are registered for one object, leading to inaccurate identification results.
Innovation Solution
An object identification device and method that compare captured images against multiple identification images, using geometric relation parameters to determine if multiple objects in the image are the same by analyzing feature information and homography matrices between database images and the captured image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple database images are registered for one object to enhance recognition precision, then the amount of information in the database increases and recognition precision improves, but the accuracy in determining the number of objects decreases because multiple database images may be output for a single object in the query image
Solution Approach 1:
The patent introduces homography matrices as an intermediary mechanism to bridge the relationship between database images and query images. By calculating and comparing homography matrices, the system can determine whether multiple detection results correspond to the same physical object, thus resolving the contradiction between using multiple database images for precision and accurately counting objects.
Solution Approach 2:
The patent changes the parameter used for object differentiation from simple image matching scores to geometric relationship parameters (homography matrices). This parameter transformation allows the system to distinguish between multiple detections of the same object versus multiple different objects, maintaining high recognition precision while improving object counting accuracy.
2Quantity of substance
If multiple database images are used for identification, then more comprehensive object features are captured, but the complexity of determining whether detected objects are the same increases
Solution Approach 1:
The patent replaces complex logical reasoning about object identity with mathematical computation of homography matrices. Instead of using mechanical or manual methods to determine if objects are the same, the system uses computational geometry to automatically resolve identity relationships, reducing operational complexity while handling comprehensive database information.
3Productivity
If traditional object identification methods are used without geometric relation analysis, then the processing speed is faster, but the accuracy in identifying the number of objects is lower
Solution Approach 1:
The patent performs preliminary calculation of homography matrices for database images before actual object identification. This preliminary action prepares the geometric relationship data in advance, allowing the system to quickly compare and determine object identity during query processing, thus maintaining high processing speed while improving identification accuracy.
Data Source
AI summary
An object identification device includes: at least one memory configured to store instructions; and at least one processor configured to execute the instructions to: compare a captured image against a plurality of identification images for identifying objects; and determine, after the comparison result indicates that a plurality of objects are included in the captured image, whether or not the plurality of objects are same objects, based on a first parameter indicating a geometric relation between the identification images and a second parameter indicating a geometric relation between identification image related to each identified object and the captured image.