Introduction

The project consists of 03 versions of Human Tracking system: DeepSORT, DeepOCSORT and a version that relies only on Human Re-identification model.

The project is part of the Internship Program of COTAI from 19/06/2023 to 12/07/2023.

Process Description

• Flowchart of the DeepOCSORT and DeepSORT system:

• Flowchart of the Human Re-identification system:

Installing packages

DeepOCSORT

The requirements are listed in the requirement_boxmot.txt file. Note:

• The boxmot module is modified from the original module to fit the project. It is maintained locally so installing boxmot is not required.
• The ultralytics module is installed without its dependencies. Hence, it must be installed seperated from the other modules.

1. First, install ultralytics without its dependencies.

pip install ultralytics==8.0.132 --no-deps

2. Next, install the other modules in the requirement_boxmot.txt file

pip install -r requirement_deepocsort.txt

DeepSORT

The requirements are listed in the requirement_deepsort.txt file.

To install the modules in the requirement_deepsort.txt file, run the command:

pip install -r requirement_deepsort.txt

Human Re-identification

The requirements are listed in the requirement_reid.txt file.

To install the modules in the requirement_reid.txt file, run the command:

pip install -r requirement_reid.txt

Run the project

After installing the required packages, run the command

python deepocsort.py

to run the DeepOCSORT system.

OR

python deepsort.py

to run the DeepSORT system.

OR

python reid.py

to run the Human Re-identification system.

Reference List

• Human re-identification definition
• Pre-trained models ONNX
• Pre-trained Reidentification models OpenVino
• Pre-trained Object Detection models OpenVINO
• Multi Camera Multi Target Python Demo
• Simple Online and Realtime Tracking (SORT) paper
• SORT GitHub
• DeepSORT paper - arxiv.org
• DeepSORT GitHub
• DeepSORT explanation - OpenCV