tempate / detecttrack

An object detector & tracker in Python.

DetectTrack

This project allows users to select an object and track it in subsequent frames by combining an object detector (opencv's MeanShift) with an object tracker (opencv's KalmanFilter).

Running the code

1. Install the requirements:

pip install -r requirements.txt

2. Run main.py with a video file:

python main.py video.mp4

Interacting with the GUI

Users can:

• Skip to the next frame of the video by pressing the SPACEBAR.

• Select a region to be tracked by clicking on two points.

  The detector works best when all four corners of the selected region are inside the object that is being tracked.

• Quit the program by pressing q.

Trackers

The program displays three different trackers so that the user can compare them frame by frame.

• In green, a reference tracker moving upwards at a constant speed.
• In blue, the region detected by Mean Shift that resembles the original section.
• In red, the tracked region resulting from applying the Kalman Filter to Mean Shift.

Structure of the code

The code consists of four files:

1. Box.py: defines the box objects that delimit the regions both selected by the user and detected and tracked by the program.
2. Detector.py: is a wrapper around opencv's Mean Shift function.
3. Predictor.py: is a wrapper around opencv's KalmanFilter function.
4. main.py: combines the other files to display the three trackers.

Diary

The file diary.txt contains my thought process while working on the project. I have tried to edit it as little as possible so it remains faithful to my first incursion into Computer Vision.