AI Camera Stabilization Using Multi-Sensor Tremor Compensation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image stabilization methods using simple feedback algorithms are limited in performance and consume excessive power, and struggle to effectively compensate for various tremor patterns in different usage environments.
Innovation Solution
An artificial intelligence-based image stabilization method utilizing an artificial neural network (ANN) model trained on tremor detection data from multiple sensors, including gyro and Hall sensors, to generate compensation signals for image shaking, while also considering temperature and defocus amounts, thereby improving stabilization speed and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If simple feedback algorithms are used for image stabilization, then device complexity is reduced, but processing speed and stabilization accuracy deteriorate
Solution Approach 1:
The patent applies preliminary action by pre-training the artificial neural network model offline to learn optimal stabilization patterns from training data. During actual image stabilization operations, the pre-trained model rapidly processes sensor data without requiring complex real-time computations, thus achieving high processing speed while maintaining algorithm sophistication.
Solution Approach 2:
The patent replaces traditional mechanical feedback control systems with an AI-based neural network model. The neural network learns complex stabilization patterns through training and substitutes the need for complex real-time algorithmic processing, achieving both high processing speed and accurate stabilization across various tremor patterns.
2Use of energy by moving object
If simple feedback algorithms are used for image stabilization, then power consumption is reduced, but stabilization accuracy deteriorates
Solution Approach 1:
The neural network model is pre-trained offline to capture complex stabilization patterns. During actual operation, the pre-trained model requires minimal computational resources to process sensor data and generate stabilization signals, achieving high stabilization accuracy without excessive power consumption during runtime.
Solution Approach 2:
The neural network model serves itself by learning optimal stabilization strategies during offline training. During actual image stabilization, the model autonomously processes sensor data and generates appropriate compensation signals without requiring complex real-time control algorithms, thereby achieving high accuracy with low power consumption.
3Productivity
If traditional feedback algorithms are used, then processing speed is reduced, but power consumption increases
Solution Approach 1:
The neural network model is pre-trained offline to learn optimal stabilization patterns from training data. During actual image stabilization operations, the pre-trained model rapidly processes sensor data with minimal computational overhead, achieving both high processing speed and low power consumption simultaneously.
4Device complexity
If simple feedback algorithms are used, then device complexity is reduced, but adaptability to various tremor patterns deteriorates
Solution Approach 1:
The neural network model is pre-trained offline using diverse training data representing various tremor patterns and usage environments. This preliminary training enables the model to adapt to different tremor characteristics without requiring complex real-time adjustments, achieving high versatility while maintaining algorithm efficiency.
Solution Approach 2:
The neural network model learns to adjust its internal parameters and weights during offline training to adapt to various tremor patterns. During actual operation, the model automatically applies the appropriate learned patterns based on input sensor data, achieving high adaptability without requiring complex real-time parameter adjustments or multiple specialized algorithms.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach maximizes processing speed and reduces power consumption while providing improved image stabilization accuracy across various tremor patterns and environments, enabling efficient focus adjustment and rapid image stabilization.
Implementation Method 1
outputting stabilization data for compensating for an image shaking, the stabilization data outputted using an artificial neural network (ANN) model trained to output the stabilization data based on the tremor detection data
Implementation Method 2
acquiring tremor detection data with respect to the image, the tremor detection data acquired from two or more sensors... a gyro sensor
Implementation Method 3
generates a compensation signal using a Hall sensor that detects a change in separation between a coil and a magnet
Data Source
AI summary
A method for stabilizing an image based on artificial intelligence includes acquiring tremor detection data with respect to the image, the tremor detection data acquired from two or more sensors; outputting stabilization data for compensating for an image shaking, the stabilization data outputted using an artificial neural network (ANN) model trained to output the stabilization data based on the tremor detection data; and compensating for the image shaking using the stabilization data. A camera module includes a lens; an image sensor to output an image captured through the lens; two or more sensors to output tremor detection data with respect to the image; a controller to output stabilization data based on the tremor detection data using an ANN model; and a stabilization unit to compensate for an image shaking using the stabilization data. The ANN model is trained to output the stabilization data based on the tremor detection data.