Image Tilt Correction Using Machine Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image correction methods, such as using computer vision algorithms to detect lines and seams, are limited in accuracy and flexibility, especially when capturing images with camera pitch and roll errors exceeding ±10 to ±15 degrees, particularly in dynamic scenes.
Innovation Solution
A computer-implemented method utilizing a machine learned model, like a convolutional neural network, to estimate and correct image tilt by inputting image data, providing a more accurate and flexible solution that can handle various fields of view and reduce distortion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If computer vision algorithms are used to detect lines and seams for image correction, then the method works well for specific scenes with axis aligned buildings or walls, but the accuracy deteriorates when camera pitch and roll errors exceed ±10 to ±15 degrees or in dynamic scenes
Solution Approach 1:
The patent replaces traditional computer vision algorithms (mechanical/algorithmic approach) with a machine learning model (neural network) that learns to estimate tilt angles from image data. This substitution enables the system to handle dynamic scenes and larger tilt errors that exceed the capabilities of rule-based line detection methods.
Solution Approach 2:
The patent changes the approach from detecting geometric features (lines and seams) to directly estimating tilt parameters (pitch and roll angles) using a neural network. This parameter change allows the system to operate effectively beyond the ±10 to ±15 degree limitation of traditional methods.
2Ease of manufacture
If traditional computer vision algorithms are used for tilt correction, then the implementation is simpler, but the adaptability to various scenes and tilt conditions is reduced
Solution Approach 1:
The neural network model is trained to handle multiple scene types and tilt conditions universally, replacing the need for scene-specific algorithms. The model can process images from various environments (indoor, outdoor, dynamic, static) and correct tilts beyond the limited range of traditional methods, achieving multi-functionality in a single system.
3Ease of operation
If accelerometer data is used to capture pitch and roll, then the correction can be applied to images, but the captured values often have errors of +/−10 to +/−15 degrees especially when the camera is moving
Solution Approach 1:
The patent introduces a neural network as an intermediary between the raw accelerometer data and the final tilt correction. The neural network processes the accelerometer readings along with image features to produce a corrected tilt estimate, filtering out the noise and errors present in raw sensor data especially during camera movement.
Data Source
AI summary
Systems and methods for adjusting an image are provided. In one example embodiment, a computer-implemented method includes obtaining data representing an image. The method includes inputting the image into a machine learned model. The method includes obtaining, in response to inputting the image into the machine learned model, an output of the machine learned model that includes an estimated tilt of the image. The method includes correcting a tilt of the image based at least in part on the estimated tilt obtained from the machine learned model.