一种基于事件相机的图像去模糊方法及系统
By constructing a scale-aware neural network and utilizing a self-supervised framework, an image deblurring method based on event cameras is proposed. This solves the problem of poor deblurring performance caused by inconsistent data distribution in existing technologies and achieves stable deblurring results in real-world scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- WUHAN UNIV
- Filing Date
- 2023-03-14
- Publication Date
- 2026-07-17
AI Technical Summary
Existing motion blur removal algorithms based on frame images perform poorly in real-world scenarios, limited by the inconsistent data distribution between simulation and real data, making effective generalization difficult.
An image deblurring method based on event cameras is adopted. By collecting blurred image datasets and event stream datasets, a scale-aware neural network model is constructed. A self-supervised framework is used for image deblurring generalization. A self-supervised training data augmentation dataset is constructed by combining multilayer perceptron and deformable convolution. The self-supervised training framework is then used for training the data augmentation dataset.
It achieves stable deblurring performance at different spatial and temporal scales in real-world scenarios, improves the generalization ability of image deblurring, and adapts to motion blur removal effects under different conditions.
Smart Images
Figure CN116385283B_ABST