Image Positioning via Neural Network Upsampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image positioning technologies require extensive computing resources and time due to the need for high-resolution wide region images, resulting in low positioning efficiency and high hardware costs.
Innovation Solution
An image positioning system and method utilizing upsampling based on machine learning, where a neural network data model processes a low-resolution region image to generate a super-resolution image, allowing for precise target positioning with reduced computational requirements and hardware costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution wide region image is used for object recognition and positioning, then positioning precision is improved, but computing resources and time are excessively consumed
Solution Approach 1:
The patent divides the wide region image into multiple small region images, each covering a potential target area. This segmentation allows the system to process only relevant regions rather than the entire high-resolution image, significantly reducing computational load while maintaining positioning precision through subsequent super-resolution processing of selected regions.
Solution Approach 2:
The patent performs preliminary low-resolution target detection on small region images to identify potential target locations before generating super-resolution images. This preliminary action filters out irrelevant regions, allowing the computationally intensive super-resolution process to be applied only to regions containing targets, thereby improving overall positioning efficiency.
2Measurement precision
If high-resolution wide region image is used for object recognition and positioning, then positioning precision is improved, but hardware cost increases
Solution Approach 1:
The patent uses software-based super-resolution processing to generate high-resolution images from low-resolution inputs. Instead of relying on expensive high-resolution camera hardware, the system creates high-resolution copies through neural network-based image processing, thereby achieving the same positioning precision with simpler, lower-cost hardware.
Solution Approach 2:
The patent employs computationally generated super-resolution images as a temporary, software-based solution to replace expensive physical high-resolution imaging hardware. These generated images serve the same purpose as captured high-resolution images but are produced on-demand through processing, eliminating the need for costly camera equipment.
3Device complexity
If low-resolution region image is used for target detection, then hardware cost is reduced, but positioning precision deteriorates
Solution Approach 1:
The patent changes the resolution parameter of the image dynamically. Low-resolution images are used for initial target detection to reduce hardware requirements, then the resolution is enhanced through super-resolution processing for precise positioning. This parameter transformation allows the system to achieve high positioning precision without requiring expensive high-resolution imaging hardware throughout the entire process.
Data Source
AI summary
An image positioning system based on upsampling and a method thereof are provided. The image positioning method based on upsampling is to fetch a region image covering a target from a wide region image, determine a rough position of the target, execute an upsampling process on the region image based on neural network data model for obtaining a super-resolution region image, map the rough position onto the super-resolution region image, and analyze the super-resolution region image for determining a precise position of the target. The present disclosed example can significantly improve the efficiency of positioning and effectively reduce the required cost of hardware.