A YOLOv7-based system for traffic object and road sign recognition

The rapid development of intelligent transportation and assistive technologies, the need for real-time environmental awareness has become increasingly important for vehicle drivers, individuals with visual impairments, and particularly motorcycle riders who are unable to install in-vehicle navigation systems, especially under low-light conditions. Addressing this need is crucial for improving road safety. This study proposes a real-time multi-class traffic object detection and road sign recognition system that integrates the YOLOv7 object detection model, OCR-based text recognition, and Google Text-to-Speech (gTTS) for auditory feedback. The system aims to provide drivers with timely and accurate visual and auditory support to enhance situational awareness and safety. YOLOv7 serves as the core detection module, identifying pedestrians, vehicles, traffic lights, and road signs. A localized dataset based on common traffic conditions in Taiwan improves recognition accuracy. For road sign reading, image preprocessing enhances text clarity under low-light or blurred scenarios, while OCR ensures precise parsing. Detected results are converted into natural speech by gTTS, with timestamp buffering and duplicate elimination mechanisms to prevent repetitive voice prompts. The modular design combines object detection, text recognition, speech feedback, and visualization, making the system scalable and suitable for embedded deployment. Potential applications include smart glasses, wearable aids, urban navigation systems, and intelligent intersections. This paper demonstrates the practical application of AI in assistive technology and its contribution to smart cities, inclusive environments, and technological equity. Finally, the recognition performance was assessed using Precision, Recall, and F1-score. The experimental results demonstrate that the proposed approach outperforms the comparison method, Fast R-CNN, in terms of recognition accuracy.