computer-vision

The objective of this assignment was to detect and track meaningful features from a video.

The structure of the repository is the following:

• The keypoints folder contains the keypoints computed offline by the various detector used;
• The src folder contains the code;
• The video folder contains the video used for testing.

As already anticipated, the feature detection and tracking can be computed in two ways:

• Online detection
• Offline detection

The motion tracking algorithm instead is always computed online.

In this project, the feature detector developed are:

• SIFT
• Good features to track
• ORB
• FAST
• Brief

Two motion tracking algorithm has been developed:

• Lucas Kanade Optical flow
• Kalman filter with hungarian algorithm

Usage

To run the project, start the main.py script. By default, it runs the feature detection algorithms implemented with the Lucas Kanade Optical Flow tracking algorithm. To select only one method or to run the Kalman Filter tracker, open the manager.py script and change the line 20 according to your desire. The following is an example to select the ORB detector with the Kalman Filter tracker

if tracker.detector.name() == FeatureDetectorAlgorithm.ORB.value and tracker.tracking.name() == FeatureTrackingAlgorithm.KALMAN_FILTER.value:

To run the project using the keypoints computed offline, you need to follow these steps:

1. Run the create_keypoints.py script in the util package specifying your video
2. Run the main.py script by setting the online parameter to False.

To save the video, set the save_video flag to True in the main.py script and run it.