Lens Apparatus Adaptive Control Using Transfer Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional lens apparatuses face challenges in achieving optimal performance for autofocus, diaphragm control, and zoom operations, particularly in balancing positioning precision, velocity, power consumption, and quietness, which vary depending on user requirements and lens apparatus states, and require extensive machine learning-driven adjustments.
Innovation Solution
A lens apparatus comprising an optical member, a driving device, and a controller that utilizes a learned model obtained from a different apparatus, allowing for adaptive control based on a neural network algorithm to manage drive commands and optimize performance according to user requests and device constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning is performed using the lens apparatus itself, then the learned model can be optimized for the specific apparatus, but a large number of drives are required which increases wear and learning time
Solution Approach 1:
The patent applies preliminary action by performing machine learning using a different apparatus (camera body) before using the learned model in the lens apparatus. This allows the learning process to be completed in advance, reducing the number of drives required and minimizing wear on the driving device while still achieving optimized control performance.
2Object-affected harmful factors
If the velocity and acceleration of the optical member are limited to achieve quietness, then noise is reduced, but positioning precision and velocity performance deteriorate
Solution Approach 1:
The patent applies dynamics by using a learned model that can dynamically adjust driving parameters based on operating conditions. The controller selects from multiple learned models or adjusts parameters in real-time, allowing the system to achieve quiet operation when needed while maintaining positioning precision and velocity performance when required, rather than being constrained by fixed limits.
3Speed
If high-speed operations are performed for autofocus and zoom to meet snapshot requirements, then responsiveness is improved, but operation sound increases and is recorded as noise
Solution Approach 1:
The patent applies dynamics by using learned models that can adapt driving speed and acceleration profiles based on the shooting mode. For still image photography, the learned model optimizes for high-speed response to meet snapshot requirements. For moving image photography, the learned model adjusts parameters to reduce operation sound, preventing noise from being recorded while still achieving acceptable response performance.
4Reliability
If extensive machine learning drives are performed to obtain an optimized learned model, then control performance is improved, but wear on the driving device increases
Solution Approach 1:
The patent applies preliminary action by completing the machine learning process using a different apparatus (camera body) before deploying the learned model to the lens apparatus. This preliminary learning phase allows the system to achieve optimized control performance without requiring extensive drives from the lens apparatus's driving device, thereby minimizing wear and extending device lifespan.
Data Source
AI summary
A lens apparatus is provided that includes: an optical member; a driving device configured to drive the optical member; and a controller configured to control the driving device based on a first learned model, wherein the first learned model is a learned model obtained by learning with respect to the lens apparatus using a second learned model as an initial learned model, the second learned model being obtained by learning with respect to an apparatus different from the lens apparatus.