| 摘 要: 在无人驾驶和驾驶辅助领域,交通标志识别是非常重要的。利用YOLOv4算法的实时性检测效果,本文 提出了一种基于YOLOv4的交通标志识别框架,主要识别LISA数据集中的四种交通标志:禁止标志、行人通过标志、 前进标志、限速标志,为了进一步提高YOLOv4的实验性能,采用K-means算法对实验数据进行聚类分析,确定适合 LISA数据集的先验框大小,实验结果表明改进后的框架对比原始的yolov4框架和YOLOv3框架,其mAP值分别提高了 0.37%和1.04%,说明改进后的YOLOv4框架在交通标志识别方面具有较高的实用价值。 |
| 关键词: 目标检测;交通标志识别;K-means算法;LISA数据集 |
|
中图分类号: TP311.5
文献标识码: A
|
|
| Traf c Sign Recognition based on Target Detection Network |
|
HE Kaihua
|
( Zhejiang Sci -Tech University, Computer Technology, Hangzhou 310018, China)
1137730657@qq.com
|
| Abstract: Traffic sign recognition plays a crucial role in the field of unmanned driving and driving assistance. In view of the real-time detection effect of YOLOv4 (You Only Look Once, YOLO) algorithm, this paper proposes a traf c sign recognition framework based on YOLOv4, which mainly identi es four types of traf c signs in LISA dataset: stop, crosswalk, go, and speed limit. In order to improve the experimental performance of YOLOv4, K-means algorithm is used to conduct cluster analysis on the experimental data and determine a suitable size for the LISA dataset. Experimental results show that compared with the original YOLOv4 and YOLOv3 framework, the improved framework's mAP (mean Average Precision) value is improved by 0.37% and 1.04%, indicating that the improved YOLOv4 is of great practical value in traf c sign recognition. |
| Keywords: target detection; traffic sign recognition; K-means algorithm; LISA dataset |
