Vehicle-Classification

Vehicle detection and classification project (2024)

Introduction

In this project, the aim is to train machine learning models for detection and classification of vehicles and run them in a web application for real-time inference. The project focuses on utilizing computer vision techniques to automatically identify vehicles in a road and classify vehicles based on their make and model.

Project Components

1. Data Collection: Download the Roboflow-100 Vehicles Collection for detection tasks and Stanford Car Dataset for classification tasks.
2. Data Preprocessing: Clean and preprocess the dataset to prepare it for training the models.
3. Model Training: Train computer vision models using state-of-the-art techniques and frameworks such as PyTorch.
4. Evaluation: Evaluate the trained models on test datasets to measure their accuracy, precision, and other measures.
5. Integration: Integrate the trained models into an application for real-time vehicle detection and classification.

Installation in Windows/Linux

Prerequisites (only in Windows)

1. Prior to installing WSL in Windows

• Search in Windows: Control Panel > Programs > Activate or deactive Windows features.
• Enable "Virtual Machine Platform", "Windows Hypervisor Platform" and "Windows Subsytem for Linux".
• Restart your PC after making those changes.

2. Steps to install WSL in Windows

• Go to Microsoft Store and install "Windows Subsystem for Linux".

• Open Windows Powershell as Administrator and check if WSL is installed:

  wsl --status

• Go back to Microsoft Store and install "Debian" or "Ubuntu".

• Open the chosen OS terminal and create a UNIX user and password. Then install git:

  sudo apt update && sudo apt install git

• Open VS Code and select the blue button in the bottom left corner to create a Remote Connection with WSL option.

Steps to Install the Environment

1. Clone the repository:

• Open a new terminal in VS Code and execute:

  git clone https://github.com/carvarsou/Vehicle-Classification

2. Install Miniconda:

• Execute these commands to install Miniconda:

  wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh 
  bash Miniconda3-latest-Linux-x86_64.sh

• Check if Miniconda is added to the PATH:

  nano ~/.bashrc
  export PATH="$HOME/miniconda3/bin:$PATH"

• Reload the Shell Configuration:

  source ~/.bashrc

• Verify installation:

  conda --version

3. Create a Conda Environment:

• Install a new environment:

  conda create --name env python=3.10

  If you already have an environment to use, downgrade the version as so:

  conda activate <your_env>
  conda install python=3.10

4. Activate the Environment:

   conda init
   conda activate env

   To deactivate:

   conda deactivate env

5. Install the requirements:

• Make sure Conda Environment is now activated. Then install Torch and other packages:

  pip install -r requirements.txt

• Install npm and node.js:

  # installs nvm (Node Version Manager)
  curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.0/install.sh | bash
  # download and install Node.js (you may need to restart the terminal)
  cd ~/.nvm && nvm install 22
  npm install -g pnpm

Usage

Training the models (optional)

• Run the train file of each model in the core folder.

  python -m core.detect_vehicles.yolov5.train
  python -m core.detect_vehicles.yolov8.train
  python -m core.classify_vehicles.efficientnet_b1.train
  python -m core.classify_vehicles.yolov8.train
  ...

Running the webapp API locally

• The app uses Swagger as the API framework.

  cd <path_to>/Vehicle-Classification
  # Generates .env file for local settings
  echo -e "DJANGO_KEY='$(python -c 'from django.core.management.utils import get_random_secret_key; print(get_random_secret_key())')'\nDEBUG=True" > .env
  # Runs SQLite server
  ./webapp/vc_backend/manage.py makemigrations 
  ./webapp/vc_backend/manage.py migrate
  ./webapp/vc_backend/manage.py runserver

Running the application UI locally

• The app uses Vue.js 3 + Vite as the frontend framework.

  Open another terminal:

  cd <path_to>/Vehicle_Classification/webapp/vc_frontend/perseidai
  pnpm up
  pnpm dev

License

This project is released under the Apache 2.0 License. For more information, check: https://www.apache.org/licenses/LICENSE-2.0.

Contributors

Carlos Varela Soult (carvarsou)

Acknowledgements

Roboflow 100 Project

Stanford Car Dataset by J. Utrera