SORT + PF (Particle Filter)

This code is a Python implementation of the SORT + PF (Particle Filter) algorithm for object tracking in challenging environments.

Installation

To install the required packages, run the following command:

Note: Python 3.7 is required to run this code due to libraries conflicts.

pip install -r requirements.txt

Dataset

The code was tested on the UAVPorpoise dataset.

Usage

To run the code, use the following command:

python3 main.py --task viz --tracker sort-pf --source ./dataset/test_videos/000.avi --engine cpu --cache-yolo

Evaluation

To evaluate the tracker, use the following command:

bash ./scripts/track_all.bash <tracker> <motion_model>

• available trackers: base (SORT), pf (SORT + PF)
• available motion models: none, flow (Sparse Optical Flow, aka SOF)

Then run the following command to evaluate the tracker:

python3 eval_track.py

Results

Method	HOTA ⬆️	ID-SW ⬇️	ID-F1 ⬆️
Strong-SORT	0.514	10.600	0.535
OC-SORT	0.526	3.400	0.597
BoT-SORT	0.580	8.633	0.628
BoT-SORT (ECC)	0.451	32.833	0.466
BoT-SORT (ORB)	0.454	4.533	0.495
BoT-SORT (SOF)	0.604	4.400	0.650
SORT	0.568	4.533	0.627
SORT (SOF)	0.568	4.567	0.628
SORT-PF	0.657	3.567	🔥 0.699 🔥
SORT-PF (SOF)	🔥 0.660 🔥	🔥 3.300 🔥	0.687