Visual Odometry

Visual Odometry¶

Visual Odometry
- Visual Tracking
  - Optical Flow
- Pose Estimation

Visual Tracking¶

Optical Flow¶

LK

Pose Estimation¶

2D-2D: Epipolar Geometry¶

Fundamental Matrix
Essential Matrix
Homography Matrix

3D-3D: ICP¶

SVD
非线性优化

3D-2D: PnP¶

The Perspective-n-Point (PnP) pose problem is the problem of estimating the relative pose – 3D position and orientation – between a calibrated perspective camera and a 3D object (or between the camera and the entire 3D scene) from a set of n visible 3D points with known (X,Y,Z) object (or scene) coordinates and their 2D projections with known (u,v) pixel coordinates.

The PnP problem is relevant in 3D object tracking and camera localization / tracking and is often used for example in Structure-from-Motion (SfM), Visual Odometry (VO), Simultaneous Localization and Mapping (SLAM) and image-based localization pipelines.

Perspective-n-Point (Wikipedia)
Code
midjji/pnp: A RANSAC and BA based pnp wrapper for the Lambdatwist p3p solver
ydsf16/PnP_Solver: Personal implementations of solvers for PnP problem, including DLT and EPnP

Solver¶

DLT
P3P
EPnP
UPnP
Motion Only BA（非线性优化）
RANSAC