A computer vision kit for algorithm verification. Far from stable.
Some universal functions and some definitions.
Algorithms for Optical flow
- Pyramidal Lucas-Kanada Algorithm
Function about Fundamental Matrix. There are two methods to find the fundamental matrix
-
8-Points Algorithm. Normalized 8-point algorithm
-
RANSAC. Self-adaptive sample by the inliers number of current best model. Solve the fundamental matrix by 8-Points algorithm.
Use Inverse Compositional and Efficient Second-order Minimization algorithm to align image patch in reference image to patch in current image. The model contains:
-
pixel 2D drift (IV)
$$I_c(\mathbf x + \mathbf u) = I_r(\mathbf x)$$ -
pixel 2D drift with bias(illumination or exposure differences) (IV, ESM)
$$I_c(\mathbf x + \mathbf u) = I_r(\mathbf x) + \beta$$
First of all, build the code.
mkdir build && cd build
cmake ..
make -j
Then run the demos
# Base
./test_base ../data/desk1.png ../data/desk2.png
# Optical Flow
./test_opticalflow ../data/floor1.png ../data/floor2.png
# Epipolar Geometry
./test_fundamental ../data/desk1.png ../data/desk2.png
# Image Alignment
./test_align2D ../data/floor1.png ../data/floor2.png
./test_align1D ../data/floor1.png ../data/floor2.png