High-Dynamic Visual Systems Utilizing Mamba-Integrated Event-Based Solutions for Motion Deblurring

Motion blur in dynamic scenes poses a significant challenge for computer vision applications. Traditional image sensors capture complete frames at fixed intervals, often failing to precisely record the positions of objects during rapid motion, resulting in blurred images. Event-Based Vision Sensors (EVS) offer a novel solution by asynchronously capturing pixel-level changes in luminosity, generating event data highly sensitive to temporal dynamics. This paper presents an enhanced motion deblurring framework that integrates the Mamba module within the architecture. The proposed model leverages the high temporal resolution and asynchronous data acquisition capabilities of EVS to combine event data with frame-based imagery, enriching motion representation and improving image clarity. Key components, such as the Event-Image Fusion-Mamba Module (EIFMB) and the Event-Guided Masking Module (EGMM), are introduced to adaptively integrate and process dynamic and static visual information. Experimental results on benchmark datasets demonstrate significant improvements in deblurring performance, with the model achieving a Peak Signal-to-Noise Ratio (PSNR) of 36.62 and a Structural Similarity Index (SSIM) of 0.977, validating the effectiveness of this approach.