Visual-Inertial Fusion-Based Restoration of Image Degradation in High-Dynamic Scenes with Rolling Shutter Cameras

Area of Science:

• Computer Vision
• Robotics
• Image Processing

Background:

• Rolling shutter (RS) cameras are prevalent in mobile devices but suffer from motion blur and geometric distortions.
• Rapid motion and vibrations exacerbate these coupled degradations, impacting image quality and system performance.

Purpose of the Study:

• To develop a unified framework for estimating and correcting motion-related degradations in rolling shutter cameras.
• To improve both photometric (blur) and geometric (distortion) image quality in high-dynamic scenes.

Main Methods:

• A visual-inertial fusion framework integrating IMU and image data to estimate unified motion degradation parameters.
• An exposure-aware deblurring pipeline considering CMOS sensor photoelectric characteristics.
• A perspective-consistent rolling shutter compensation method addressing depth-motion coupling.

Main Results:

• Demonstrated significant improvements in image quality (photometric and geometric) on real-world mobile data.
• Achieved enhanced accuracy in downstream Simultaneous Localization and Mapping (SLAM) tasks.
• Outperformed existing representative baseline methods in correcting coupled degradations.

Conclusions:

• The proposed visual-inertial fusion framework effectively addresses coupled motion degradations in rolling shutter cameras.
• The method offers a robust solution for enhancing image quality and SLAM performance in dynamic environments.
• This work contributes to more reliable visual perception for mobile and embedded systems.