Abstract: Traffic sign detection and identification not only support driver assistance technologies but also play a vital role in enhancing traffic management. These capabilities are essential for ensuring safe transportation and effective operation of self driving vehicles, aiding in real-time decision making for both human and autonomous drivers. Recognizing traffic signs in Bangladesh is especially difficult due to the country’s unique driving environment. This includes non-standard signage, diverse road conditions, and the frequent presence of pedestrians and livestock on roadways. By utilizing technologies like machine learning and computer vision, these systems can be tailored to local contexts, ultimately enhancing roadway safety in Bangladesh. This research proposes a smart assistant system that utilizes a dataset of 6,000 diverse traffic sign images from Bangladeshi road environments to improve road safety, especially in regions with potential driver compliance issues. The dataset encompasses 41 traffic sign categories and presents various real-world challenges, including faded color, weather conditions, blurring and vibration, occlusion, variable size of traffic sign and low light conditions. To enhance the dataset’s robustness, data augmentation techniques such as random rotations, shearing and zooming were applied. We trained the YOLOv10 deep learning model, renowned for its real-time object detection capabilities, on this dataset. The model achieved significant results, with a mean Average Precision (mAP) of 0.80, a recall of 0.87, a precision of 0.92, and an F1 score of 0.89, demonstrating its effectiveness in real-world traffic sign detection and classification.
Keywords: TSR, Smart Traffic, Feature Extraction,Traffic Sign,YOLOv10,Computer Vision
|
DOI:
10.17148/IJARCCE.2025.14101
