Traffic Sign Recognition and Auditory Alert System for Sri Lankan Drivers Using Deep-Learning

Traffic signs are essential for road safety, guiding and informing drivers about rules and conditions. Ignoring signs or traffic rules can lead to accidents, causing property damage, and even deaths. To ensure safety, drivers must remain aware of traffic signs, obey their instructions, and drive responsibly. Notably, Sri Lanka has unique traffic signs compared to other countries. This study aims to develop a deep learning-based traffic sign alert system with good accuracy and efficiency in traffic sign recognition and enhance driver awareness by integrating an auditory alert system. This alert system was implemented under two main sections: object detection system and auditory alert system. The traffic signs dataset images were labeled and pre-processed using the Roboflow software tool. The You Only Look Once version 5 (YOLOv5) model was used to train the system, which allowed it to understand complicated patterns and attributes associated with 15 different sign classes. To improve driver awareness, an auditory alert system was developed using the PyDub library to generate real-time alerts in the form of voice messages. These alerts are triggered whenever a traffic sign is detected, giving an additional layer of information to the driver. On the test dataset, the integrated and fine-tuned YOLOv5 model achieved the F1 score of 90.09% and mean average precision (mAP) of 87.55% resulting in good evaluation metrics values for detection. This YOLOv5 model-based traffic sign auditory alert system was highly effective and efficient in increasing driver awareness and also helped to reduce road accidents.