This repository is a companion page for the following paper:
Sanne Eeckhout, Pratik Phadte , Gokul Ramalingan, Hussein Hesham Hassan Abdelaziz Gad. 2024. Man Overboard Object Tracking. Image Processing and Computer Vision. University Twente.
It contains all the material required for replicating the study, including: stabilization, camera calibration, object tracking using a customized CSRT tracker and distance calculation to the object using horizon detection.
The result of the object tracking and distance calculation can be seen below (compressed version), or here.
buoy_tracking.mp4
The packages used by this repository can be installed using pip install opencv-python numpy
The code can be run using the buoyTracker.py file in the src folder.
The main result of the tracking model is the generated video, which can be seen below.
This is the root directory of the repository. The directory is structured as follows:
.
|
|--- src/ Source code used in the paper
|
|--- buoyTracker.py Main file containing the tracker model.
|--- depthCalculation.py Main script to run the inference and track power metrics
|--- stabilization.py Stabilizes the image using Lucas-Kanade Optical Flow
|--- motionModel.py Creates a motion model for the tracker model,
|--- documentation/ Contains the figures presented in the paper.
|
|--- data/ Contains coordinates of the detected object, manual ground truth and detected distances.