Traffic sign detection system is vital to improving road safety, recognizing and following important signs, and preventing accidents. Balancing precision and recall has been a persistent challenge for conventional traffic sign detection algorithms, especially when applied to real-time scenarios. High precision can result in missed detections (low recall), and high recall can raise false positives (low precision), reducing detection model performance. This paper presents a new Indian Traffic Sign Detection Dataset (ITSDD) that addresses the lack of consistent benchmark data for Indian traffic sign scenarios. Furthermore, the model proposes an iterative training process that incorporates YOLOv10, with the F1-score as the leading parameter. During each iteration, data augmentation is applied to the failed detection test samples to enhance the model and improve its generalization performance. To check the proposed model performance, the benchmark CCTSDB 2021 data has been tested along with ITSDD samples. Experimental results shows that precision–recall balance of the YOLOv10 model significantly improves the F1-score to 98%, which is suitable for the deployment of proposed model in complex real-world scenarios.