• keywords: ORB, RANSAC, normalized direct linear transformation (DLT) algorithm

• keywords: normalized direct linear transformation (DLT) algorithm, inverse warping (bilinear interpolation), GUI mouse click

• keywords: epipolar geometry, stereo processing

• keywords: image stitching, image warping

• keywords: rasterization, mesh, pygame

• keywords: Perspective-n-Point (PnP) pose computation, DLT, EPnP, RANSAC, trajectory, augmented reality (AR)

• keywords: chessboard, camera intrinsic matrix, distortion coefficients, RMS reprojection error

• keywords: camera relative pose, epipolar geometry, triangulation, absolute orientation problem

• keywords: NICE-SLAM, SA-ConvONet, Mask R-CNN
• external link: github, report, video

• Some notes and references are listed at the bottom of this page.

python3 HomographyEstimation.py --img1 images/1-b0.jpg --img2 images/1-b1.jpg --n 4 --descriptor o

python3 DocumentRectification.py --img1 images/book1.jpg 

python3 StereoRectification.py --img1 images/bike1.png --img2 images/bike2.png

python3 image_stitching.py --imgDir successive_images/scottsdale --outputDir results 

• Objective: To display the polygons
• Language: Python
• Library: pygame
• Input: .obj files

The parameters can be modified in main_displayer.py. Then run

python3 main_displayer.py

• Movement: W, A, S, D
• Rotation: ← (left arrow) and → (right arrow)

First, merge separate data inputs.

python3 merge_data.py

After runnung merge_data.py, please make sure that 4 pkl files, namely images.pkl, point_desc.pkl, points3D.pkl, and train.pkl, are in the data directory.

python3 trajectoryPlot.py --pnp epnp_gauss
python3 trajectoryPlot.py --pnp epnp_gauss  --onlyshow 1 

• Options for the command-line argument --pnp for PnP pose computations are provided:
  • p3p_Grunert_ransac
  • normalized_DLT
  • epnp
  • epnp_gauss
  • opencv_PnPRansac

python3 cubeDrawing.py --pnp epnp_gauss --videopath results/cubeVideo.mp4
python3 cubeDrawing.py --pnp epnp_gauss --onlyshow 1 --videopath results/cubeVideo.mp4 
python3 cubeDrawing.py --onlyshow 2 --videopath results/cubeVideo.mp4 

python3 camera_calibration.py

This program estimates the camera intrinsic matrix and the distortion coefficients.

python3 main_vo.py --imgDir input_vo_frames --camParams results_calibration/camera_params.npy --features 1000 --trackerConfig orb --poseMethod opencv --show True

where the --trackerConfig has the following options:

orb
brisk
sift
sift_root
akaze
LK_SHI_TOMASI
LK_FAST

• The following two videos are performed with sift_root.
  • Demonstration 1: https://youtu.be/wlfBCzuvmBQ
  • Demonstration 2: https://youtu.be/HYALMOclFIc

python3 poseError.py

python3 transform_cube.py

• All output files are saved in the results folder.
• Some of projects are modified from the class of 3D Computer Vision with Deep Learning Applications at National Taiwan University.

[1] Lecture Notes from Chu-Song Chen's class of 3D Computer Vision with Deep Learning Applications (Fall 2022)
[2] Homography Estimation from hughesj919's Github
[3] Inverse Warping from makkrnic's Github
[4] (Project 3) images
[5] Stitching Pipeline from OpenCV
[6] Image Stitching from apoorva-dave's Github
[7] (Project 5) 3D Rasterizer from lcox74's github
[7] (Project 6) Dataset
[8] EPnP and Gauss-Newton Optimization from WeiyanCai's Github
[9] Direct Linear Transform (DLT) from acvictor's Github
[10] Visual Odometry from luigifreda's github
[11] Pose Recovery from laavanyebahl's github
[12] Absolute Orientation Problem in Cyrill Stachniss's lecture notes