Computer Vision Assignment2: Structure From Motion (SFM)

Programming assignment 2 for computer vision course

• Structure from motion using SIFT, RANSAC and 5-point algorithm
• Triangulation method

Dependencies

• MATALB version: at least 2022a
• VLFeat: at least v.0.9.21

Usage

# add paths inside matlab console
addpath('Data')
addpath('Calibration_Data')
addpath('match_plot')
addpath('Step2')
addpath('Step5')
addpath('Step7')
addpath('vlfeat-0.9.21');
addpath('external_source_code');

# run codes using matlab
# you have to run with this order
cd Codes
step1
step2
step3
step4

Directories

• Codes: Source code for this project
• external_source_code: External source code for visualizing
• Data: Image data that used for the report
• Calibration_Data: Calibration image and my own data to reproduce the 3D coordinates
• Results: Results from my code, please see "the_real_result" file
• camera_calibration_results: My camera calibration results

Components

• Codes

  • step_1.m: Imaging feature extraction and basic matching
  • step_2.m: Running RANSAC to achieve essential matrix E and its reasonable matchings
  • step_3.m: Receive 4 camera matricies candidate
  • step_4.m: Triangulation and the 3D coodinate estimation
  • matrix_normalize: Feature vector normalization based on camera intrinsic K
  • RANSAC.m: RANSAC function
  • BEST_SCORES_RANSAC: RANSAC based on the best scores (not that meaningful)
  • Triangulation.m: Triangulation function
  • additional_tri.m: Triangulation in backwards (switching camera)
  • color_reshape.m: Function that reconstructs the color of 3D mesh
  • Plotting.m: Plotting a graph for my report

• Images

  • the_real_result.ply: The submission file for .ply
  • my_image_result.ply: Result from my own image