TrafficSignClassification (Björn Bulkens, Miriam Lorenz, Daniel Fauland)

Aim of the project

The aim of the project is the implementation of a picture classification model for traffic signs. It's simultaneously necessary for the model to be accurate, fast and as light as possible since the Use Case requires the environment to work with a low level of computing power.

Requirements

• Necessary packages can be found in 'requirements.txt'
• If you do not have a NVIDIA GPU install TF2 via this command: 'pip install tensorflow'
• If you have a NVIDIA GPU make sure to install the following:
  • Latest NVIDIA GeForce drivers
  • NVIDIA CUDA 10.1
  • NVIDIA cuDNN v7.6.5 for CUDA 10.1: Available here
  • Note: You have to create a free NVIDIA developer account in order to download the software mentioned above
• Tested with Python 3.6.4 / Windows 10 version 1909 / RTX 2080 Ti

Get the training data

• To reduce the size of the GitHub repo the training data has to be downloaded separately
• Download the data from Google Drive
• Extract the folder 'trainingData' to the main project directory

Train the model yourself

• Note: Make sure you have at least 16GB of Ram installed or the program will crash during training!
• If you want to train the model yourself you should run the 'createAugmentedImages' file in the 'run' folder first (This will take several minutes)
• Then you can run the 'trainModel' file in the same directory (This process can take very long without a GPU!)
• After the training is complete you can see the accuracy and validation accuracy in a graph as well as the optimal amount of epochs for the validation accuracy

Predict the validation data

• To predict the validation data you have to run the file 'predictValidationData' in the run folder
• This will show the a series of pictures with the prediction as title and the actual label as caption
• You can change the number of images with the variable 'num'

Predict real world data

• To predict real data you have to run the file 'predictRealData'
• Any picture that is inside the folder 'realData' will be predicted
• You can add more pictures to the folder without having to change anything in the code
• All 'png' or 'jpeg' files are accepted no matter the size or resolution

Useful git commands

- 'git add -A'  # adds all files directories and subdirectories to the queue (dependent on your current directory)
- 'git commit -am "Mesaage"'  # commits changes to the local repo
- 'git push'  # Pushes all changes to the online repo
- 'git pull' or 'git pull <link>'  # pulls newest version from github (Necessary before push command if changes werde made to the github repo)
- 'git branch'  # Shows all available branches
- 'git checkout <branch-name>'  # Switches to different branch