IMU-Aided Adaptive Mesh-Grid Based Video Motion Deblurring

This repository contains the implementation of the algorithm presented in the paper "IMU-Aided Adaptive Mesh-Grid Based Video Motion Deblurring". The code is designed to handle non-uniform motion blur in videos by utilizing an adaptive mesh-grid approach, with a focus on balancing deblurring quality and computational efficiency.

Overview

Traditional deblurring methods often assume uniform blur, which is computationally efficient but ineffective for non-uniform blur scenarios. Our proposed method addresses this limitation by:

• Dividing the video frame into a mesh-grid structure.
• Estimating the blur Point Spread Function (PSF) using data from an Inertial Measurement Unit (IMU).
• Adapting the size of the mesh-grid cells based on the spatial variance of blur magnitude in each frame.
• Offering two versions of the algorithm: one optimized for the best deblurring quality and another for a balance between quality and computational cost.
• Providing a trade-off parameter to adjust the mesh-grid size according to specific application needs.

Features

• Adaptive Mesh-Grid: The grid size dynamically adjusts to handle areas with varying levels of blur, improving PSF estimation accuracy where needed.
• IMU Integration: Inertial sensor data is used to aid in accurate blur estimation, particularly for non-uniform motion.
• Performance Options: Choose between high-quality deblurring or a balanced approach with reduced computational requirements.
• Trade-Off Parameter: Customize the algorithm to fit your application's performance and quality needs.

Dataset

The image dataset used in this study is generated from the videos that are available in the following links:

• https://www.youtube.com/watch?v=wz3BGn--UIA
• https://pixabay.com/videos/buddha-buddhism-asia-zen-203408/

The raw blurred images of castle image sequence is made available in the repository