Traffic Sign Detection and Recognition

Detecting and recognizing traffic signs in images and video streams. It uses computer vision and machine learning techniques to identify and classify various traffic signs, such as stop signs, yield signs, and speed limit signs.

Video.Demo.mp4

NOTE: The left slider that show numbers is part of the original video our model draw the detection bounres and recognize the sign

Requirements

• Python 3.7 or later
• OpenCV
• TensorFlow 2.x
• Numpy

Usage

we used GTSRB & GTSDB datasets for this project

• The Recognition Part :
• This Part is responsible of classifying the signs. First, it cut the input image according to the surrounding box around the traffic sing. Get each sub image from the original image and classify each sign using Convolution layers, Batch Normalization layers and the RELU Notify each Box according to the class with a label (the number of the class which is the type of the traffic sign) The last output is image with all labeled Boxes. In case of videos, we combine all frames with labeled Boxes. use Classifier.ipynb in the colab folder to train and generate the weights file
• The Detection Part :
  • This model is responsible of detecting all the traffic sings in the input image or video through, using the YOLOv3 . The output of this model is bounding boxes surrounding the sings in the image or the frames of the video through using techniques of deep learning and probability.
  • You can change the configuration of the model through the YOLO Model Configuration folder
  • For preparing the dataset use the data preparation folder
  • use detector.ipynb in the colab folder to train and generate the weights file
• Merging
  • use Connection between Classifier and Detector.ipynb to connect the classifier with the detector and to test on images/videos
• The Application
  • For testing only images from the GTSRB dataset as my machine couldn't handle the detection part or the connection between the two models we depend on colab
• NOTES:
  • The notebooks are well commented and organized so no need to explain more details

Models

The repository includes YOLO for detecting traffic signs and a pure trained model for the recognizing.

conclusion

• the recognition model achieved 99.2% using Inception architecture With Data Augmentation and trained it for 200 epoch

• the detection model achieved 91%

  
• when we run the test video on colab it processing speed was 0.5 Frame/s we used a feature in openCV and the result was 1 Frame/s we didn't know where is the issue but if you want to contribute it will be a great help

Note

Since This Project is based on machine learning techniques this readMe file is written with the help of ChatGPT :)