Deblurring 3D Reconstruction with Event Cameras: Enhancing Noise Robustness and Modeling Pixel-Wise Nonlinear Response
3D reconstruction with event cameras is a rapidly growing research area, but traditional methods often struggle with high-noise events and fail to accurately model the nonlinear response of real event cameras. To address these challenges, we propose a novel framework for deblurring 3D reconstruction that enhances noise robustness and models pixel-wise nonlinear camera response. Our method integrates event double integral (EDI) and event-image cross-modal attention (EICA) mechanisms, alongside a Kolmogorov-Arnold Network (KAN) with Radial Basis Function (RBF) as the basis function to capture complex response characteristics. Our approach effectively handles low signal-to-noise ratio and complex response in real events, providing high-fidelity environmental perception in high-speed motion scenarios.
☐ The code and data will be made public once the paper is accepted. Stay tuned!