Advanced Driver Assistance Systems (ADAS) Based on Machine Learning Techniques for the Detection and Transcription of...

Area of Science:

• Computer Science
• Artificial Intelligence
• Road Safety Engineering

Background:

• Driver distraction from variable message signs (VMSs) is a significant cause of traffic accidents.
• Advanced driver assistance systems (ADAS) aim to improve driver comfort and safety by perceiving the environment.
• Current ADAS solutions do not extensively utilize machine learning for VMS interpretation.

Purpose of the Study:

• To develop a prototype VMS reading system using machine learning.
• To create an assistant that recognizes VMS and converts their text content into speech.
• To mitigate driver distraction caused by VMS.

Main Methods:

• A dataset of VMS images was created using manual annotation, scraping, and data augmentation.
• A RetinaNet model, based on ResNet50 and pre-trained on COCO, was trained for VMS recognition.
• Image preprocessing and binarization were applied for optimal quality before text extraction.
• Tesseract OCR (version 4.0) was used for text extraction, and IBM Watson Text to Speech for audio conversion.

Main Results:

• A functional prototype system for VMS recognition and text-to-speech conversion was developed.
• The machine learning approach demonstrated potential for accurately interpreting VMS content.
• The system successfully processed images, extracted text, and generated speech output.

Conclusions:

• Machine learning techniques can be effectively applied to VMS reading systems.
• This technology has the potential to reduce driver distraction and improve road safety.
• Further development of such systems could enhance the capabilities of advanced driver assistance systems.