mrobert3456/traffic_sign_recognition

Traffic sign detection and recognition using Yolov3 and Convolutional Neural Networks

Traffic Sign Recognition

⚙ How it works

1. Traffic sign detection using YoloV3, trained with GTSDB dataset using Darknet framework
2. Traffic sign detection trained with GTSRB dataset using Convolutional Neural Network

For detailed description of how it works, please check out my publication on IEEE: https://ieeexplore.ieee.org/abstract/document/10158539

📦 Installation

This Repository

Download this repository by running:

git clone https://github.com/mrobert3456/Traffic_sign_recognition.git
cd Traffic_sign_recognition

⚡ Software Dependencies

This project utilizes the following packages:

• Python 3
• OpenCV 2
• Matplotlib
• Numpy
• Pandas
• h5py
• Tensorflow-gpu
• Filterpy

To setup the environment,you need to install conda, then run the following commands:

conda env create -f environment.yaml
conda activate GPU_ENV

🚀 Usage

Traffic sign recognition

1. Before you run the preprocessing methods and training, create a folder called GTSRB in the root folder

2. Download the GTSRB dataset, which contains the following zip files, that you need to unzip in the GTSRB folder:

   • GTSRB_Final_Test_Images.zip
   • GTSRB_Final_Training_Images.zip

3. Unzip the GTSRB_Final_Test_GT.zip file into the GTSRB/Final_Test/Images folder

4. In the root folder, create a folder called 'ts'

   • Inside the 'ts' folder create two subfolders 'aug' and 'orig'
     • This is where the preprocessing and the training results will be saved

To run the preprocessing for the GTSRB dataset, just run the following command:

python preprocess_tsr_dataset.py

To create and train the recognition model, run the following jupyter notebook file:

Build_Train_TSR.ipynb

Traffic sign detection

1. Create a folder called GTSDB under the root folder
2. Download the GTSDB dataset, unzip the FullIJCNN2013.zip file under the GTSDB folder
3. To run the preprocess method for the GTSDB dataset, just run the following command:

python prepare_tsd_dataset.py

After the process is finished, the following files will be generated:

• test.txt
• train.txt
• classes.names
• ts_data.data

The classes.names, ts_data.data, yolov3_ts_test.cfg and the yolov3_ts_train.cfg files needs to be placed under the darknet-master/build/darknet/x64/cfg folder, where you installed the Darknet framework

• yolov3_ts_test.cfg and the yolov3_ts_train.cfg files can be found under 'ts' folder

4. Create a 'weights' folder under darknet-master/build/darknet/x64
   • Copy the darknet53.conv.74 file under it

To start the training process run the following command:

darknet.exe detector train cfg\ts_data.data 
cfg\yolov3_ts_train.cfg weights\darknet53.conv.74

The necessary model and weight files can be found here: https://drive.google.com/file/d/1nLjbzzL77rTNSduE5hl7cByP0lgvai2i/view?usp=sharing

These files need to be placed under ts folder

python detect_recogn.py <input_file.mp4> <output_file.mp4>

The output file will be saved as output.mp4