基于深度学习的LGPR数据处理及定位方法

LGPR data processing and localization method based on deep learning

为解决自动驾驶方案中的GPS导航、视觉和激光传感器在恶劣环境下可能会面临的失效问题,提出一种基于深度学习的定位探地雷达(localizing ground-penetrating radar,LGPR)数据定位方法,补充现有车辆定位的方法。在离线建库阶段,对提前采集到的LGPR数据中特征与非特征数据训练生成粗分类网络模型,对特征数据训练生成细分类网络模型并结合GPS信息形成地下特征指纹库。在线定位阶段,将再次采集的LGPR数据先输入到粗分类网络模型中过滤掉非特征数据,通过置信度阈值判断后输入到地下特征指纹库中获取具体位置实现定位。实验结果表明,所提方法能够较好实现定位。

To solve the problem that GPS navigation,vision and laser sensors in autonomous driving schemes may face failure in harsh environments,a localizing ground-penetrating radar(LGPR)data positioning method based on deep learning was proposed to supplement the existing vehicle positioning methods.In the offline library building phase,the feature and non-feature data from the LGPR data collected in advance were trained to generate coarse classification network models,and the feature data were trained to generate fine classification network models,and combined with GPS information,a subsurface feature fingerprint library was formed.In the online localization stage,the LGPR data collected again was input into the coarse classification network model to filter out the non-feature data,and then input into the underground feature fingerprint database to obtain the specific location by judging the confidence threshold.Experimental results show that the proposed method can better achieve localization requirements.