Vehicle Counting and Tracking using YOLOv8 and ByteTrack

1. Introduction

This project aims to create a vehicle tracking and counting system using the YOLOv8 and ByteTrack models. It offers a reliable and efficient system for analyzing traffic flow, monitoring congestion, and enhancing overall road safety. The main components of the project are:

• YOLOv8: The YOLOv8 model from Ultralytics is utilized for accurate and real-time vehicle detection.
• ByteTrack: ByteTrack algorithm is employed for multi-object tracking, ensuring smooth and reliable tracking of vehicles across frames.
• Line Counter: We use the supervision library to count the number of vehicles entering or leaving a region.
• TensorRT Inference: Export the model to TensorRT format to speed up inference.

2. Installation

To use this repository, we need to set up our environment with its required libraries. The steps are:

1. Clone the repository:

   git clone https://github.com/arief25ramadhan/vehicle-tracking-counting.git
   

2. Go to the repository, and install dependencies:

   cd vehicle-tracking-counting
   pip install -r requirements.txt
   

3. Inside this current repo, clone the ByteTrack Libraries:

   git clone https://github.com/ifzhang/ByteTrack.git
   

4. Install ByteTrack dependencies:

   cd ByteTrack
   
   # workaround related to https://github.com/roboflow/notebooks/issues/80
   sed -i 's/onnx==1.8.1/onnx==1.9.0/g' requirements.txt
   
   pip install -r requirements.txt
   
   python3 setup.py develop
   pip install cython_bbox onemetric loguru lap thop
   

3. Usage

To perform inference using the vehicle tracker and counter pipeline:

1. Go to main.py. In the last few lines, change the input_video, output_video and use_tensorrt variables accordingly. The input_video refers to the video we want to perform tracking and counting on, while the output_video is the desired path of the prediction. The use_tensorrt is a boolean variable indicating whether to use TensorRT format for a quicker inference time.

2. Save changes.

3. Run detection and tracking by executing this command in your terminal:

   python main.py
   

4. The predicted video should be available in the output_video path. Figure below displays the example frame of the predicted video.

References

This project is heavily based on tutorial by Roboflow in this colab notebook. It works by combining the YOLOv8 model from Ultralytics and ByteTrack model developed by Yifu Zhange, et al. The links to the YOLOv8 and and ByteTrack repository are:

• YOLOv8: Link to YOLOv8 repository
• ByteTrack: Link to ByteTrack repository