The rapid development of the automobile resulted in the traffic infrastructure becomes more and more complicated. Both type and the number of traffic signs on the streets are increasing. Therefore, there is a need for applications to support drivers to recognize the traffic signs on the streets to help them avoid missing traffic signs. This paper proposes an approach to detecting and classifying Vietnamese traffic signs based on the YOLO algorithm, a unified deep learning architecture for real-time recognition applications. An anchor boxes size calculation component based on the k-means clustering algorithm is added to identify the anchor boxes size for the YOLO algorithm. A Vietnamese traffic sign dataset including 5000 images containing 5704 traffic signs of types was collected and used for evaluation. The F1 score of the tiny model (the fastest but the most inaccurate model) achieved is about more than 92% and the detection time is approximately 0.17s (in our testing environment: laptop CPU Intel 3520M, 8GB RAM, no GPU). In comparison with similar research in Vietnamese traffic sign recognition, the proposed approach in this paper shows a potential result that provides a good trade-off between the recognition accuracy and recognition time as well as the feasibility for real applications.