Cross-Camera Image Alignment Using Adversarial AI Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Supervised learning-based deep learning models require large amounts of ground truth image data and struggle to provide consistent outputs when using cameras with different properties, such as varying color, scale, and distortion, which is particularly challenging for autonomous mobility devices.
Innovation Solution
An image transformation AI model using adversarial training generates aligning parameters to transform image data from cameras with different properties, adjusting for intrinsic and distortion coefficients, brightness, and resizing to align images, utilizing an encoder, transformer, and discriminator.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If image data from cameras with different properties is used directly in a supervised learning model, then the model can process diverse camera inputs, but the output consistency deteriorates due to variations in color, scale, and distortion
Solution Approach 1:
The patent applies preliminary action by transforming image data from cameras with different properties into a unified reference coordinate system before feeding it to the supervised learning model. The coordinate transformation unit pre-processes the images by calculating transformation parameters (rotation, translation, scaling) based on camera intrinsic parameters and extrinsic parameters, ensuring that all images are aligned to a common reference frame before model processing, thus guaranteeing output consistency
Solution Approach 2:
The patent introduces an intermediary coordinate transformation system that acts as a mediator between diverse camera inputs and the supervised learning model. This transformation unit serves as an intermediate processing layer that standardizes different camera coordinate systems into a unified reference system, allowing the model to receive consistent input regardless of the original camera properties
2Reliability
If large amounts of ground truth image data are collected for supervised learning, then model performance improves, but the economic cost increases
Solution Approach 1:
The patent applies parameter changes by transforming the coordinate system parameters of images from different cameras to match a reference coordinate system. By changing the spatial parameters (position, orientation, scale) of the input images through mathematical transformation, the system enables the supervised learning model to process diverse camera inputs as if they were from a single reference camera, thereby improving model performance without requiring separate ground truth data for each camera
Solution Approach 2:
The patent achieves universality by creating a single supervised learning model that can process images from multiple cameras with different properties through coordinate transformation. The transformation system makes the model universal by enabling it to handle various camera types, positions, and orientations through a unified coordinate reference framework, eliminating the need for camera-specific models or extensive additional ground truth data
Data Source
AI summary
A method for generating aligned image data through an aligning parameter generated by an image transformation artificial intelligence (AI) model includes, through an encoder of the image transformation AI model, generating at least one or more aligning parameters from a first camera property and a second camera property related to a first camera and a second camera respectively. The method also includes, through an image transformer of the image transformation AI model, transforming, based on the at least one aligning parameter and a brightness parameter, first image data photographed by the first camera to be aligned with second image data photographed by the second camera. The method also includes training the encoder and a discriminator of the image transformation AI model by adversarial training. The image transformation AI model discriminates between the transformed first image data and the second image data.