Feature Tracking

This a Feature Tracking Computer Vision project using openCV. It's intended to be applied to the Mapping a new road with senseFly Corridor data set available at senseFly.

Dependencies

In order to build the project first install OpenCV 3.3 with support for the non-free modules from the OpenCV contrib repository as specified here.

Input images

You can download the Mapping a new road with senseFly Corridor data set from here. The data set comes in a zip file. Therefore, to use the images first unzip the archive.

Build the code

To build the code you'll simply have to run;

  cd <project-root-dir>
  mkdir build
  cd build
  cmake ..
  make

Running tests

This project uses Google Test for as testing framework. After compiling you can run our tests by running;

  cd <project-root-dir>
  cd build
  ./tests_exec

Feature Tracking Demo

After compiling you can run the tracking demo by running;

  cd <project-root-dir>
  cd build
  ./tracking_demo -indir=<path-to-input> [-outdir=<path-to-output> -show -extract -match -v -finder=<org|surf> -help]

Options

• help - displays help
• v - enable verbose mode.
• indir - path to the input data set
• outdir - path to where the output will be written
• show - enables the GUI and displays images in a pair basis
• extract - enable feature extraction. If not specified the features will be read from <project-root-dir>/build/features.yml. If the features.yml file doesn't exist then the extraction is automatically enabled.
• match - enable feature matching. If not specified the matches will be read from <project-root-dir>/build/features.yml. If the features.yml file doesn't exist then the matching is automatically enabled.
• finder - specified the feature finder to be used. Only SURF was tested.

Output GUI

The GUI is enabled by adding the show parameter. The GUI will open 4 different windows described below;

1. Image 1: First image with the extracted feature points drawn as rich points.
2. Image 2: Second image with the extracted feature points drawn as rich points.
3. Matches: Both images with the extracted feature points drawn as rich points and the matches joining the points from one image to the other.
4. Warped: First image with the second image with the homography applied superposed.

Output Directory

If the output directory is specified using outdir parameter then a separate directory per image pair will be created named as the first image, but without the extension. For example the image pair EP-00-00012_0366_0001.JPG and EP-00-00012_0366_0002.JPG will create a directory named EP-00-00012_0366_0001 for storing their outputs. The directory will contain;

• homography.yml - with the obtained homography stored using OpenCV's FileStorage.
• decomposedHomography.yml - with the obtained homography, camera matrix, translational and rotational matrix. Also stored using OpenCV's FileStorage.
• <first-image-file-name> - with the first image file scaled as used in the algorithm.
• <second-image-file-name> - with the second image file scaled as used in the algorithm.
• features_<first-image-file-name> - with the first image and the found SIFT features drawn as rich points.
• features_<second-image-file-name> - with the second image and the found SIFT features drawn as rich points.
• matches_<first-image-file-name> - with the both image side-by-side, the found SIFT features drawn as rich points, and the drawn matches as lines.
• warped_<first-image-file-name> - with the superposition of the first image and the second with the homography applied.