深度学习的应用简化了数字条纹投影三维测量的过程,在传统数字条纹投影三维测量技术条纹投影、相位计算、相位展开、相位深度映射的流程中,研究者们已经成功证明了前三个环节以及整个流程结合深度神经网络的可行性。基于深度学习,PDNet(...深度学习的应用简化了数字条纹投影三维测量的过程,在传统数字条纹投影三维测量技术条纹投影、相位计算、相位展开、相位深度映射的流程中,研究者们已经成功证明了前三个环节以及整个流程结合深度神经网络的可行性。基于深度学习,PDNet(Phase to Depth Network)神经网络模型被提出,用于绝对相位到深度的映射。结合多阶段深度学习单帧条纹投影三维测量方法,通过分阶段学习方式依次获得物体的绝对相位与深度信息。实验结果表明,PDNet能较准确地测量出物体的深度信息,深度学习应用于相位深度映射步骤具有可行性。并且,相较于直接从条纹图像到三维形貌的单阶段深度学习单帧条纹投影三维测量方法,多阶段深度学习单帧条纹投影三维测量方法可以明显提升测量精度,仅需单帧条纹图像输入即可获得毫米级测量精度,且能适应具有复杂形貌物体的三维测量。展开更多
Any tidal defense engineering involves the collection and analysis of massive information about engineering structures and their surrounding environment. Traditional method, which is carried out mainly by means of two...Any tidal defense engineering involves the collection and analysis of massive information about engineering structures and their surrounding environment. Traditional method, which is carried out mainly by means of twodimensional drawings and textures, is not efficient and intuitive enough to analyze the whole project and reflect its spatial relationship. Three-dimensional visual simulation provides an advanced technical means of solving this problem. In this paper, triangular irregular network (TIN) model simplified by non-uniform rational B-splines (NURBS) technique was used to establish the digital terrain model (DTM) of a super large region. Simulation of dynamic water surface was realized by combining noise function with sine wave superposition method. Models of different objects were established with different modeling techniques according to their characteristics. Application of texture mapping technology remarkably improved the authenticity of the models. Taking the tidal defense engineering in the new coastal region of Tianjin as a case study, three-dimensional visual simulation and dynamic roaming of the study area were realized, providing visual analysis and visible demonstration method for the management and emergency decision-making associated with construction.展开更多
文摘深度学习的应用简化了数字条纹投影三维测量的过程,在传统数字条纹投影三维测量技术条纹投影、相位计算、相位展开、相位深度映射的流程中,研究者们已经成功证明了前三个环节以及整个流程结合深度神经网络的可行性。基于深度学习,PDNet(Phase to Depth Network)神经网络模型被提出,用于绝对相位到深度的映射。结合多阶段深度学习单帧条纹投影三维测量方法,通过分阶段学习方式依次获得物体的绝对相位与深度信息。实验结果表明,PDNet能较准确地测量出物体的深度信息,深度学习应用于相位深度映射步骤具有可行性。并且,相较于直接从条纹图像到三维形貌的单阶段深度学习单帧条纹投影三维测量方法,多阶段深度学习单帧条纹投影三维测量方法可以明显提升测量精度,仅需单帧条纹图像输入即可获得毫米级测量精度,且能适应具有复杂形貌物体的三维测量。
基金Supported by Tianjin Research Program of Application Foundation and Advanced Technology (No.12JCZDJC29200)Foundation for Innovative Research Groups of National Natural Science Foundation of China (No.51021004)National Key Technology R&D Program in the 12th Five-Year Plan of China(No.2011BAB10B06)
文摘Any tidal defense engineering involves the collection and analysis of massive information about engineering structures and their surrounding environment. Traditional method, which is carried out mainly by means of twodimensional drawings and textures, is not efficient and intuitive enough to analyze the whole project and reflect its spatial relationship. Three-dimensional visual simulation provides an advanced technical means of solving this problem. In this paper, triangular irregular network (TIN) model simplified by non-uniform rational B-splines (NURBS) technique was used to establish the digital terrain model (DTM) of a super large region. Simulation of dynamic water surface was realized by combining noise function with sine wave superposition method. Models of different objects were established with different modeling techniques according to their characteristics. Application of texture mapping technology remarkably improved the authenticity of the models. Taking the tidal defense engineering in the new coastal region of Tianjin as a case study, three-dimensional visual simulation and dynamic roaming of the study area were realized, providing visual analysis and visible demonstration method for the management and emergency decision-making associated with construction.
