Mobile Apparatus Scale Prediction for Position Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional localization and object recognition systems face challenges in accurately determining the position of a robot after external forces are applied, especially when changes in pose and lighting occur, and struggle with recognizing moving objects or images with varying sizes, leading to poor recognition performance with limited query images.
Innovation Solution
A mobile apparatus and control method that extract feature points by predicting scale changes using correlation coefficients between original and blurred images, generating multi-scale image patches, and matching feature points across different scales to enhance object and position recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional localization is used to determine robot position, then the system is simple to implement, but it fails to recognize position after external forces are applied
Solution Approach 1:
The patent segments the image processing into multiple scales by generating image patches at different resolutions (e.g., 1/2 scale, 1/4 scale, 1/8 scale). This allows the system to analyze the same image data at multiple levels of detail, improving position recognition accuracy after external forces are applied without requiring additional sensors or complex hardware.
Solution Approach 2:
The patent adds a scale dimension to the image analysis by creating multi-scale image patches from the same original image. This transforms the problem from analyzing a single image at one resolution to analyzing multiple representations of the same image at different resolutions, enabling better detection of position changes caused by external forces.
2Reliability
If object recognition is performed with limited query images, then the processing speed is fast, but the recognition performance is significantly lowered
Solution Approach 1:
The patent performs preliminary processing by generating multi-scale image patches from the original image before feature point extraction and matching. This pre-processing step organizes the image data into multiple scales in advance, allowing the recognition algorithm to efficiently search through different scales without increasing the overall processing time significantly, thereby improving recognition accuracy with limited query images.
3Measurement precision
If feature point matching is performed without scale prediction, then the processing is simple, but the matching accuracy is poor for moving objects or images with varying sizes
Solution Approach 1:
The patent changes the scale parameter by generating image patches at different resolutions (e.g., 1/2, 1/4, 1/8 of original size). This allows the system to adapt to varying object sizes and movements by comparing feature points across multiple scales, significantly improving matching accuracy for moving objects without requiring complex real-time scale adjustment mechanisms.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of controlling a mobile apparatus includes acquiring a first original image and a second original image, extracting a first feature point of the first original image and a second feature point of the second original image, generating a first blurring image and a second blurring image by blurring the first original image and the second original image, respectively, calculating a similarity between at least two images of the first original image, the second original image, the first blurring image, and the second blurring image, determining a change in scale of the second original image based on the calculated similarity, and controlling at least one of an object recognition and a position recognition by matching the second feature point of the second original image to the first feature point of the first original image based on the change in scale.