针对路网兴趣点(point of interest,POI)数据传统获取方法人工依赖高、成本高、效率低等问题,基于遥感影像和车辆轨迹数据融合提出了一种典型路网POI自动识别方法.该方法通过卷积神经网络(convolutional neural network,CNN)提取遥感影像中的道路几何特征,并利用循环神经网络(recurrent neural network,RNN)有效捕捉车辆轨迹隐含的道路交通特征,最后通过神经网络,实现从多模态数据中识别路网中红绿灯路口、加油站、停车场等兴趣点.本研究以2019年重庆市中心城区道路兴趣区域的400张遥感影像数据和40000条出租车轨迹数据作为训练样本进行了实验验证.结果表明,对比单一使用车辆轨迹数据的算法,该算法识别精度提高了近11.83％;对比单一使用遥感影像数据的算法,该算法识别精度提高了近2.53％.
Automatic Recognition of Typical Road Network POIs Based on Multimodal Data Fusion
In order to overcome the inefficiencies,high costs and manual dependency associated with traditional data acquisition techniques,this study proposes an automatic rec-ognition method for typical road network POIs based on the fusion of remote sensing imagery and vehicle trajectory da-ta. The method utilizes convolutional neural networks (CNN) to extract road geometric features from remote sensing imagery and recurrent neural networks (RNN) to capture the implicit traffic characteristics in vehicle trajecto-ries. Finally,the method enables the accurate identification of POIs such as traffic lights,gas stations,and parking lots from multimodal data sources through neural network. Ex-perimental validation is conducted through 400 remote sens-ing images and 40000 taxi trajectory data recorded in 2019 from central Chongqing city. The results indicate that,compared to algorithms using only vehicle trajectory data,this method improves identification accuracy by approximate-ly 11.83％,and that compared to algorithms using only re-mote sensing imagery data,accuracy improves by approxi-mately 2.53％.
万方数据
维普
