基于多尺度特征融合网络的交通标志检测

传统交通标志检测方法检测速度慢,且现有深度神经网络对小尺寸交通标志检测精度低。对此提出一个基于YOLOv3的新型端到端卷积神经网络。以YOLOv3为检测框架,对特征提取网络和特征融合网络加以改进,并应用K-means聚类算法生成更适合交通标志的锚点框。充分利用多尺度特征实现了对小尺寸交通标志检测性能的提升。在TT100K (Tsinghua-Tencent 100K)和GTSDB (German Traffic Sign Detection Benchmark)交通标志数据集上进行实验,获得的mAP分别为82.73%和92.66%,运行时间分别为0.037 s和0.033 s。实验结果验证了改进网络的有效性,表明了改进网络的整体性能优于其他检测方法。

TRAFFIC SIGN DETECTION BASED ON MULTI-SCALE FEATURE FUSION NETWORK

To solve the problems of slow detection speed of traditional traffic sign detection methods and poor detection effect of the deep neural network for small-scale traffic signs,a new end-to-end convolutional neural network based on YOLOv3 is proposed.It used YOLOv3 as the detection framework.The feature extraction network and feature fusion network were improved,and K-means clustering algorithm was applied to generate anchor boxes which were more suitable for traffic signs.The performance of small-scale traffic sign detection was improved by making full use of multi-scale features.The proposed network was evaluated on TT100K dataset and GTSDB dataset.The results show that the proposed detection network achieves state-of-the-art performance by obtaining mAP of 82.73%and 92.66%with average execution time of 0.037s and 0.033s in two datasets.The experimental results verify the effectiveness of the improved network,and the overall performance of the proposed network is better than other detection methods.