Wildlife Tracking with Classical Computer Vision

This repository contains implementations of classical computer vision tracking algorithms for wildlife monitoring. This project was developed as part of CMPUT 428 at the University of Alberta.

🌐 Live Demo

View Interactive Demo →

Explore the tracking algorithms, see performance comparisons, and understand the technical details through an interactive web interface.

Project Structure

The contents of this repository are organized as is decribed below:

• src/: Contains implementations of the Inverse Compositional Lucas Kanade Tracking Algorithm, The mean shift tracking algorithm, and the discrete Kalman Filter.
• data/: Contains video data with it's .txt annotations. These are derived from the AnimalTrack Dataset and the Zebra Dataset Cited.
• outputs/: Stores some sample outputs of the trackers and filters on different image sequences for a specific animal.
• test_*.py: These scripts in the root repository can be run to evaluate the performance of the tracker they are testing on a .mp4 file with it's ground truth .txt file.
• utils.py: Utility functions used across the testing algorithms. References were used for some of the functions inside. These are also cited.
• README.md: You are here!

Installation

For Running the Web Demo

# Clone the repository
git clone https://github.com/your-username/428-project-final.git
cd 428-project-final

# Install dependencies
pip install -r requirements.txt

# Run the interactive demo
streamlit run app.py

For Running Trackers Locally

1. Clone the repository
2. Install the required python packages using pip install -r requirements.txt

Use

To run a specific tracker or implementation of the Kalman filter, run the test scripts. For example: python test_mosse.py The path to the video and ground truth are located inside of the test file at the top of the script. In order to run the test on a different image sequence, update these static variables and run the script.

Findings

Preliminary runs of the trackers suggests that there needs to be a more robust implementation. The MOSSE tracker boasts the highest IoU and FPS so far with the Inverse Compositional Lucas Kanade tracker close behind. Initially, it was thought that the mean shift tracker would allow for animal patterns to be easily tracked but animals traveling in groups tend to confuse the tracker to the point of neglible tracking. The performance of the Discrete Kalman Filter is definitely dependent on the performance of the tracker that trains it. Tests with it have shown some minor increases in IoU.

Future Work

This project has a lot left to be done. Since the original classical trackers haven't proven to be adequate, the integration of a hybrid meta tracker and a re-identification strategy is necessary. Creator notes on it suggest:

• Implementing a trained YOLO model to detect animals and reidentify with the detected bounding box that is closest to the last traked region
• Implement a means to tell if the tracker is failing or not. Re-identification models would help with this. The Peak-to-Sidelobe ratio of the MOSSE tracker would also suffice. Future work should include a custom implementation of the tracker so as to be able to access the PSR value. The legacy opencv tracker implemented currently does not allow for that.
• More trackers should be implemented. The promise shown by the correlational filter based MOSSE tracker suggests that the KCF tracker could also be used.

References

1. David S Bolme, J Ross Beveridge, Bruce A Draper, and Yui Man Lui. Visual object tracking using adaptive correlation filters. In 2010 IEEE computer society conference on computer vision and pattern recognition, pages 2544–2550. IEEE, 2010.
2. Simon Baker and Iain Matthews. Lucas-kanade 20 years on: A unifying framework. International journal of computer vision, 56:221–255, 2004
3. Dorin Comaniciu, Visvanathan Ramesh, and Peter Meer. Real-time tracking of non- rigid objects using mean shift. In Proceedings IEEE Conference on Computer Vision and Pattern Recognition. CVPR 2000 (Cat. No. PR00662), volume 2, pages 142–149. IEEE, 2000.
4. Eric Price, Pranav C Khandelwal, Daniel I Rubenstein, and Aamir Ahmad. A framework for fast, large-scale, semi-automatic inference of animal behavior from monocular videos. bioRxiv, pages 2023–07, 2023.
5. Greg Welch, Gary Bishop, et al. An introduction to the kalman filter. 1995.
6. Libo Zhang, Junyuan Gao, Zhen Xiao, and Heng Fan. Animaltrack: A benchmark for multi-animal tracking in the wild. International Journal of Computer Vision, 131(2):496– 513, 2023.
7. https://www.youtube.com/watch?v=mwn8xhgNpFY&list=PLn8PRpmsu08pzi6EMiYnR-076Mh-q3tWr&ab_channel=MATLAB
8. https://github.com/zziz/kalman-filter
9. https://github.com/Blarc/mean-shift-tracking
10. Code from the UAlberta CMPUT 428 Assignments and LAbs was also used

Web Demo

This repository includes an interactive Streamlit web application that showcases the tracking algorithms.

Features:

• Algorithm demonstrations with side-by-side comparisons
• Performance metrics visualization (IoU, FPS)
• Technical explanations of each algorithm
• Interactive algorithm selection

Quick Start:

streamlit run app.py

Deployment Guide: See QUICKSTART.md for deployment instructions to Streamlit Cloud, Railway, or other platforms.

AI Tool Use Acknowledgement

In the creation of this project I did make use of the language model Claude to aid with detecting typos, and with formatting comments. The code in this repository was written by me though.