Gyroscope-Assisted Motion Deblurring Network

• DOI: 10.48550/arxiv.2402.06854
• Publication Date: 2024-02-09
• Authors: Simin Luan; Cong Yang; Zeyd Boukhers; Xue Qin; Dongfeng Cheng; Wei Sui; Zhijun Li
• Keywords: Deblurring

Image research has shown substantial attention in deblurring networks recent years. Yet, their practical usage real-world deblurring, especially motion blur, remains limited due to the lack of pixel-aligned training triplets (background, blurred image, and blur heat map) restricted information inherent images. This paper presents a simple yet efficient framework synthetic restore images using Inertial Measurement Unit (IMU) data. Notably, includes strategy for triplet generation, Gyroscope-Aided Motion Deblurring (GAMD) network image restoration. The rationale is that through harnessing IMU data, we can determine transformation camera pose during exposure phase, facilitating deduction trajectory (aka. trajectory) each point inside three-dimensional space. Thus, our are inherently close natural strictly pixel-aligned, mass-producible. Through comprehensive experiments, demonstrate advantages proposed framework: only two-pixel errors between trajectories, marked improvement (around 33.17%) state-of-the-art method MIMO on Peak Signal-to-Noise Ratio (PSNR).