top of page
Task description
Find the relation between two cameras by analyzing the pictures taken with different angles
Random sample consensus(Ransac)
// retain most inliers
psuedocode:
for loop:
1. estimate model with part of raw data
2. calculate the error of all raw data with estimate model
3. keep the estimate model with most inliers(error smaller than threshold)
Draw epipolar lines between two images
We can have geometry information of correspondence points from epipolar lines.
Reconstruction the translation and rotation between two cameras
correspondence pairs in two camera frames, and their depth information
camera position2
camera position 1
There are four possible pairs of solution to rotation and translations. The most possible one is the one with most points in front of the camera, a.k.a depth is positive.
Projection of points to another camera frame
red: given correspondence points, blue: projection
To verify our solution, we can project the points on the one of camera onto the frame of the other camera. Then the points should be close to the correspondence points sifted by RANSAC in the previous step.
Robotics projects
Design works
bottom of page