optical flow estimation

**Optical Flow Estimation** is the **task of calculating the apparent motion of image brightness patterns** — determining a displacement vector $(u, v)$ for every pixel between two consecutive video frames, representing how pixels "move" over time. **What Is Optical Flow?** - **Definition**: Dense 2D motion field. - **Assumption**: Brightness Constancy (the pixel's color doesn't change, it just moves). - **Output**: A color-coded map where color indicates direction and intensity indicates speed. **Why It Matters** - **Video Compression**: "This block just moved 5 pixels left", saving massive bandwidth (MPEG). - **Stabilization**: Smoothing out shaky camera footage. - **Action Recognition**: Two-stream networks use flow to "see" motion explicitly. **Key Models** - **Classical**: Lucas-Kanade, Horn-Schunck. - **Deep Learning**: FlowNet, PWC-Net, RAFT (Recurrent All-Pairs Field Transforms). **Optical Flow Estimation** is **pixel-level motion tracking** — the foundational signal processing step that underpins most modern video analysis algorithms.