基于改进Faster R-CNN的交通标志检测算法

Traffic sign detection algorithm based on improved Faster R-CNN

针对大视野交通场景下背景复杂和交通标志目标较小的问题,提出一种改进Faster R-CNN检测网络的算法。首先采用深度残差网络ResNet50作为骨干网络,提取交通标志的特征;然后设计了在两个不同层级特征图上使用合理尺度滑动窗口的策略来生成目标建议区域,增强多尺度交通标志的检测能力;最后在残差块中引入注意力机制模块,强化图像的关键信息,抑制图像的背景信息。在中国交通标志数据集上验证了算法的有效性,取得了98.52%的平均检测精度和每幅图像0.042 s的检测速率。本文算法检测效果明显优于原Faster R-CNN检测方法,更适用于复杂场景下的交通标志检测,鲁棒性较强。

Aiming at the problems of complex background and small traffic sign target in large view traffic scene,an improved Faster R-CNN detection network algorithm is proposed.Firstly,the deep residual network ResNet50 is used as the backbone network to extract the features of traffic signs.Secondly,the strategy of using reasonable scale sliding window on two different level feature maps is designed to generate the target proposal region to enhance the detection ability of multi-scale traffic signs.Finally,the attention mechanism module is introduced into the residual block to strengthen the key information of the image and suppress the image background information.The validity of the algorithm is verified on the Chinese traffic sign dataset,with an average detection accuracy of 98.52%and a detection rate of 0.042 s per image.The detection effect of the improved algorithm is obviously better than the original Faster R-CNN detection method,and is more suitable for traffic sign detection in complex scenes,with strong robustness.