A SECI Method Based on Improved YOLOv4 for Traffic Sign Detection and Recognition
| Autor: | Jing Zuo, Fengchun Han, Qingda Meng |
|---|---|
| Rok vydání: | 2022 |
| Předmět: | |
| Zdroj: | Journal of Physics: Conference Series. 2337:012001 |
| ISSN: | 1742-6596 1742-6588 |
| DOI: | 10.1088/1742-6596/2337/1/012001 |
| Popis: | Traffic sign detection and recognition (TSDR) has become extraordinarily significant in autonomous driving. To trade-off between high accuracy and fast detection speed, a four-stage method, Selecting-Extracting-Cleaning and Identifying (SECI) based on improved algorithm of You Only Look Once version 4 (YOLOv4) is proposed, including selecting the anchor boxes of traffic signs by using k-means++ clustering algorithm, extracting traffic sign features by using subsampled layer, cleaning the redundant information by amending structure of the Spatial Pyramid Pooling (SPP) and identifying traffic signs by adding auxiliary branches and multi-scale feature fusion. Compared to the origin YOLOv4, the results show that, the SECI method improves the mean average precision (mAP) by 6.71%, the detection speed is improved by 37%, and the F1 score is improved by 13%. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|