摘要
针对自动驾驶场景下,提高交通标志检测速度和准确率的问题,提出一种基于卷积神经网络(Convolutional Neural Network,CNN)的交通标志检测算法,与传统的图像检测算法相比拥有明显的优势。首先解析影响交通标志检测准确性的因素,并对算法提出了两项改进:使用101层的残差网络作为特征提取的基础网络以获得高精度的特征提取和物体检测,同时优化网络的区域候选框特征提取方式以提高交通标志图像的检测效果。在GTSDB德国交通标志检测基准数据集上的实验结果表明,该算法实现在复杂背景下交通标志的精准检测。
A traffic sign detection algorithm based on convolutional neural network（CNN） is proposed in this paper.Compared with the traditional image detection algorithm,our algorithm has obvious advantages.Firstly,this paper analyzes the factors affecting the accuracy of traffic signposting and proposes two improvements to the algorithm：it uses the 101-layer residual network as the basic network for feature extraction to obtain high-precision feature extraction and object detection while optimizing the network region proposals extraction method of feature to improve the traffic sign image detection effect.
出处
《工业控制计算机》
2018年第5期99-101,共3页
Industrial Control Computer
关键词
卷积神经网络
交通标志
图像检测
convolutional neural network
traffic sign
object detection
