AN AUTOMATIC CNN-BASED SYSTEM FOR DETECTING TRAFFIC SIGNS AND RECOGNIZING LICENSE PLATES

Abstract

Modern roadways are equipped with traffic signs to alert drivers to hazards including posted speed limits, upcoming road repairs, and
pedestrian crossings, among others. This study describes A picture segmentation, traffic sign detection, and input image classification
system for real-time Traffic Sign Recognition and classification. For this effect, we use the color boost technique to zero in on the reds in
the image. Traffic sign material is identified using Convolutional Neural Networks (CNN) for detection, classification, and recognition.
Signs that forewarn drivers of upcoming roadwork, sharp bends, and pedestrian crossings have greatly improved drivers' safety. The
three stages of this research are all about recognizing and categorizing traffic signs in real time: image classification, input image
segmentation, and traffic sign detection. The colour improvement method is used to isolate the red areas of the picture. Convolutional
Neural Networks (CNN), such as Faster R-CNN, Retina Net, YOLO V4, and YOLO V5, are used to detect, classify, and recognise the
traffic sign content.
The number of cars and trucks on the road has been growing at a staggering rate in recent decades. Typically, the transit regulation that
oversees parking garages requires that you check the identification of these cars before granting them permission to park there.
Physically inspecting such a massive fleet would be a herculean task. Accordingly, it is crucial to construct an accurate automated
licence plate identification model integrating character recognition in order to alleviate the aforementioned difficulties. We've built a
model using a wide variety of national licence plates. Yolov4, an implementation of CNN architectures, was used to train the dataset of
photos. After applying several methods of picture pre-processing and morphological changes, character recognition was performed
using the Tesseract OCR. The suggested system successfully detected 92% of licence plates.