Chapter 9: Motion Estimation
Translational alignment, parametric motion, optical flow, and layered motion techniques.
Chapter Overview
Video adds a new dimension: time. Motion estimation extracts how pixels move between frames, enabling tracking, stabilization, and understanding of scene dynamics.
What is this chapter about? We learn to estimate motion at different granularities—from global camera motion to dense per-pixel optical flow. These techniques analyze how the visual world changes over time.
Why does this matter? Motion understanding enables:
- Video stabilization: Removing camera shake
- Object tracking: Following targets through video
- Action recognition: Understanding what people are doing
- Video compression: Exploiting temporal redundancy
How the topics connect: We start with optical flow—estimating motion for every pixel. Then video stabilization shows how to smooth out camera shake. Finally, we explore tracking algorithms that follow specific objects through time.
Chapter Roadmap
Click any topic to jump in
Optical Flow
Per-pixel motion estimation — brightness constancy, Lucas-Kanade, Horn-Schunck, and coarse-to-fine pyramids.
Two applications of dense motion
Video Stabilization
Smoothing camera trajectories — motion filtering, rolling shutter correction, and crop trade-offs.
Frame Interpolation
Synthesizing intermediate frames — motion-compensated warping, bidirectional flow, and neural methods.
Object Tracking
Following targets through video — Kalman filtering, template correlation, and deep Siamese trackers.
Sign up to unlock this chapter
This chapter is part of PixelBank Premium. Create a free account, then upgrade to read the full lesson — concepts, walkthroughs, and exercises.