首页> 外国专利> A learning method and a learning device for updating an HD map by reconstructing a 3D space by using depth prediction information for each object and class information for each object acquired by the V2X information fusion technology, and the same are used. Testing method and testing apparatus

A learning method and a learning device for updating an HD map by reconstructing a 3D space by using depth prediction information for each object and class information for each object acquired by the V2X information fusion technology, and the same are used. Testing method and testing apparatus

机译:使用了用于通过使用针对每个对象的深度预测信息和通过V2X信息融合技术获取的针对每个对象的类别信息来重建3D空间来更新HD地图的学习方法和学习设备。测试方法及测试装置

摘要

PROBLEM TO BE SOLVED: To provide a method for reconstructing a 3D space and updating an HD map. A learning device has a coordinate neural network to apply a coordinate neural network operation to a coordinate matrix to generate a local feature map and a global feature vector; the learning device has a decision neural network 140. , Applying a decision neural network operation to the integrated feature map to generate a first predictive fitness score to an Nth predictive fitness score map; and a learning device with a loss layer A loss is generated by referring to the predicted fitness score and the first GT fitness score to the Nth GT fitness score, and the back propagation using the loss is performed to determine the neural network and the coordinate neural network. Learning the parameters of the network. [Selection diagram] Fig. 3
机译:要解决的问题:提供一种重建3D空间和更新HD地图的方法。一种学习设备,其具有坐标神经网络,以将坐标神经网络操作应用于坐标矩阵,以生成局部特征图和全局特征向量;学习设备具有决策神经网络140。将决策神经网络操作应用于集成特征图,以生成第一预测适应度得分至第N预测适应度得分图;通过将预测的适应度得分和第一GT适应度得分参考第N GT适应度得分来产生损失,并使用该损失进行向后传播以确定神经网络和坐标神经。网络。学习网络参数。 [选择图]图3

著录项

相似文献

  • 专利
  • 外文文献
获取专利

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号