基于注意力机制的道路交通标志检测方法

Road Traffic Sign Detection Method Based on Attention Mechanism

针对交通标志在道路场景中尺寸小、个数少、难以实现快速精准检测的情况,提出一种基于注意力机制的交通标志检测模型。模型以YOLOv4为基础,在主干网络中增加通道注意力机制来提高小尺寸样本学习能力,同时使用Ghost卷积核代替常规卷积核来降低特征提取过程的计算量;在颈部网络中,生成了更小感受野的特征图参与检测,进一步提高模型对小尺寸目标的检测能力。实验结果表明,本文模型能够对不同场景下的各类小尺寸交通标志实施精准的检测,其检测速度也能够达到实时检测的水平,能够在高级驾驶辅助系统、高精度地图数据自动化采集等多个领域发挥重要的应用价值。

Aiming at the small size and small number of traffic signs in road scenes,it is difficult to achieve fast and accurate detection.A traffic sign detection model based on attention mechanism is proposed.Based on YOLOv4,the model adds a channel attention mechanism in the backbone network to improve the learning ability of small-sized samples and uses the Ghost convolution kernel instead of the conventional convolution kernel to reduce the computational load of the feature extraction process;in the neck network,the generation of the feature map with a smaller receptive field is used to participate in the detection,which further improves the detection ability of the model for small-sized targets.The experimental results show that the model in this paper can accurately detect various small-sized traffic signs in different scenarios and its detection speed can also reach the level of real-time detection.It has important application values in advanced driving assistance systems,automatic collection of high-accurate map data and other application fields.