Image as an IMU: Estimating Camera Motion from a Single Motion-Blurred Image

2025-03-21

AI Summary

Conventional pose estimation fails under high-speed motion due to severe motion blur degrading image quality. Method: This paper proposes a novel approach to directly estimate six-degree-of-freedom instantaneous linear and angular velocities from a single motion-blurred image. Instead of treating blur as noise, it models motion blur as geometric motion cues. Within an end-to-end differentiable framework, the method jointly predicts dense optical flow and monocular depth, then enforces geometric consistency via the small-motion assumption to formulate linear constraints—solved via linear least squares to recover IMU-like motion states. Contribution/Results: To our knowledge, this is the first method enabling direct mapping from a single blurred frame to continuous 6-DoF motion quantities. Pretrained on synthetic ScanNet++v2 data and fine-tuned jointly on real-world scenes, it achieves state-of-the-art accuracy in both angular and translational velocity estimation on real-world benchmarks, significantly outperforming MASt3R, COLMAP, and other SOTA methods.

Technology Category

• Computer Vision: Motion & Tracking
• Intelligent Robots: State Estimation
• Machine Learning: Multimodal Learning

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Security and Privacy: Large-scale security measurements

Abstract

In many robotics and VR/AR applications, fast camera motions cause a high level of motion blur, causing existing camera pose estimation methods to fail. In this work, we propose a novel framework that leverages motion blur as a rich cue for motion estimation rather than treating it as an unwanted artifact. Our approach works by predicting a dense motion flow field and a monocular depth map directly from a single motion-blurred image. We then recover the instantaneous camera velocity by solving a linear least squares problem under the small motion assumption. In essence, our method produces an IMU-like measurement that robustly captures fast and aggressive camera movements. To train our model, we construct a large-scale dataset with realistic synthetic motion blur derived from ScanNet++v2 and further refine our model by training end-to-end on real data using our fully differentiable pipeline. Extensive evaluations on real-world benchmarks demonstrate that our method achieves state-of-the-art angular and translational velocity estimates, outperforming current methods like MASt3R and COLMAP.

Problem

• Estimating camera motion from motion-blurred images
• Predicting motion flow and depth from blurred images
• Recovering camera velocity using linear least squares

Innovation

• Estimates motion from motion-blurred images
• Predicts dense flow and depth maps
• Solves linear least squares for velocity