基于Mask R-CNN的道路交通标志识别  

Road Traffic Sign Recognition Based on Mask R-CNN

作　　者：王之博 赵双明[1] WANG Zhibo;ZHAO Shuangming(School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,China)

机构地区：[1]武汉大学遥感信息工程学院,湖北武汉430079

出　　处：《测绘地理信息》2022年第3期119-122,共4页Journal of Geomatics

基　　金：北京市科技计划(Z191100004319001)。

摘　　要：在遥感无人机影像中,道路交通标志经过透视投影呈现出形变大、尺度变化大和干扰多等特点,传统的检测方法只关注标志的形状和颜色,应用于无人机影像时易出现漏检、误检等问题。针对上述问题,先利用透视变换对数据集进行增广,再基于Mask R-CNN框架对交通标志进行实例分割。在原框架中增加掩膜得分的策略,分割精度约提高了2%。实验结果表明,基于Mask R-CNN的方法具有较高的准确度,在解决无人机影像交通标志检测问题上具有较好的性能。Through perspective projection,the road traffic signs in remote sensing unmanned aerial vehicle(UAV)images show the characteristics of large deformation, large scale change and many interferences. However,the traditional detection method only pays attention to the shape and color of signs. When using UAV images,problems such as missed detection and false detection are easy to occur. To solve these problems,the data set is first augmented by perspective transformation,and then the traffic signs are segmented based on Mask R-CNN framework. The strategy of adding mask scores in the original framework improves the segmentation accuracy by about 2%. The experimental results show that the method based on Mask R-CNN has higher accuracy and it has better performance in solving the problem of UAV image traffic sign detection.

关 键 词：无人机影像 交通标志 Mask R-CNN 实例分割 

分 类 号：P237[天文地球—摄影测量与遥感]