Vehicle Tracking using YOLOV8 + Bytetrack

This project is built by combining YOLOv8 and Bytetrack for multiple vehicle tracking in the streets.

Documentation

News

• [✅] Add tracking by detection.
• Coming soon: Add tracking by segmentation using SAM( Segment Anything Model).

Installation

Step 1: Create a conda enviroment

conda create -n tracking python=3.8

Step 2: Clone the repository

git clone https://github.com/tanphan07/Vehicle-Tracking.git

Step 3: Install package dependencies

cd Vehicle-Tracking
pip3 install -r requirements.txt

Dataset Preparation

In this project, I have used COCO dataset to train the model. You can download the dataset from here then filter the vehicle image in the dataset or you can use this file to automating download COCO

python3 download_dataset.py

After downloading the dataset, you need to convert the dataset to YOLO format. You can use prepare_dataset.py file to convert the dataset to YOLO format (change the path to your dataset path).

Training

In this project, I have used YOLOv8 to train the model. You can follow the instruction in this to train the model. After training, you will get the weight file in runs/train/exp/weights/best.pt. For the better performance, I used TensorRT to optimize the model. You can convert the weight file to TensorRT engine following the instruction in YOLOv8 repository. I have train a model and convert it to TensorRT engine. You can download the engine here.

Tracking

After getting the TensorRT engine, you can use tracking.py to track the vehicle in the video. You can change the path to your video in the file.

python3 tracking.py

You can change your detection model by changing detection.py file. This project allow you use another detection model for testing.

Result

Here is a demo video of the project. The original video here. And the result video here.