New higher dimensional distributions are introduced in the framework of Clifford analysis.They complete the picture already established in previous work, offering unity and structuralclarity. Amongst them are the buil...New higher dimensional distributions are introduced in the framework of Clifford analysis.They complete the picture already established in previous work, offering unity and structuralclarity. Amongst them are the building blocks of the principal value distribution, involvingspherical harmonics, considered by Horvath and Stein.展开更多
在三维视觉任务中,三维目标的未知旋转会给任务带来挑战,现有的部分神经网络框架对经过未知旋转后的三维目标进行识别或分割较为困难.针对上述问题,提出一种基于自监督学习方式的矢量型球面卷积网络,用于学习三维目标的旋转信息,以此来...在三维视觉任务中,三维目标的未知旋转会给任务带来挑战,现有的部分神经网络框架对经过未知旋转后的三维目标进行识别或分割较为困难.针对上述问题,提出一种基于自监督学习方式的矢量型球面卷积网络,用于学习三维目标的旋转信息,以此来提升分类和分割任务的表现.首先,对三维点云信号进行球面采样,映射到单位球上;然后,使用矢量球面卷积网络提取旋转特征,同时将随机旋转后的三维点云信号输入相同结构的矢量球面卷积网络提取旋转特征,利用自监督网络训练学习旋转信息;最后,对随机旋转的三维目标进行目标分类实验和部分分割实验.实验表明,所设计的网络在测试数据随机旋转的情况下,在ModelNet40数据集上分类准确率提升75.75%,在ShapeNet数据集上分割效果显著,交并比(Intersection over union,IoU)提升51.48%.展开更多
The study of marine data visualization is of great value. Marine data, due to its large scale, random variation and multiresolution in nature, are hard to be visualized and analyzed. Nowadays, constructing an ocean mo...The study of marine data visualization is of great value. Marine data, due to its large scale, random variation and multiresolution in nature, are hard to be visualized and analyzed. Nowadays, constructing an ocean model and visualizing model results have become some of the most important research topics of ‘Digital Ocean'. In this paper, a spherical ray casting method is developed to improve the traditional ray-casting algorithm and to make efficient use of GPUs. Aiming at the ocean current data, a 3D view-dependent line integral convolution method is used, in which the spatial frequency is adapted according to the distance from a camera. The study is based on a 3D virtual reality and visualization engine, namely the VV-Ocean. Some interactive operations are also provided to highlight the interesting structures and the characteristics of volumetric data. Finally, the marine data gathered in the East China Sea are displayed and analyzed. The results show that the method meets the requirements of real-time and interactive rendering.展开更多
目的激光雷达采集的室外场景点云数据规模庞大且包含丰富的空间结构细节信息,但是目前多数点云分割方法并不能很好地平衡结构细节信息的提取和计算量之间的关系。一些方法将点云变换到多视图或体素化网格等稠密表示形式进行处理,虽然极...目的激光雷达采集的室外场景点云数据规模庞大且包含丰富的空间结构细节信息,但是目前多数点云分割方法并不能很好地平衡结构细节信息的提取和计算量之间的关系。一些方法将点云变换到多视图或体素化网格等稠密表示形式进行处理,虽然极大地减少了计算量,但却忽略了由激光雷达成像特点以及点云变换引起的信息丢失和遮挡问题,导致分割性能降低,尤其是在小样本数据以及行人和骑行者等小物体场景中。针对投影过程中的空间细节信息丢失问题,根据人类观察机制提出了一种场景视点偏移方法,以改善三维(3D)激光雷达点云分割结果。方法利用球面投影将3D点云转换为2维(2D)球面正视图(spherical front view,SFV)。水平移动SFV的原始视点以生成多视点序列,解决点云变换引起的信息丢失和遮挡的问题。考虑到多视图序列中的冗余,利用卷积神经网络(convolutional neural networks,CNN)构建场景视点偏移预测模块来预测最佳场景视点偏移。结果添加场景视点偏移模块后,在小样本数据集中,行人和骑行者分割结果改善相对明显,行人和骑行者(不同偏移距离下)的交叉比相较于原方法最高提升6.5%和15.5%。添加场景视点偏移模块和偏移预测模块后,各类别的交叉比提高1.6%Institute)上与其他算法相比,行人和骑行者的分割结果取得了较大提升,其中行人交叉比最高提升9.1%。结论本文提出的结合人类观察机制和激光雷达点云成像特点的场景视点偏移与偏移预测方法易于适配不同的点云分割方法,使得点云分割结果更加准确。展开更多
In this paper, the wavelet inverse formula of Radon transform is obtained with onedimensional wavelet. The convolution back-projection method of Radon transform is derived from this inverse formula. An asymptotic rel...In this paper, the wavelet inverse formula of Radon transform is obtained with onedimensional wavelet. The convolution back-projection method of Radon transform is derived from this inverse formula. An asymptotic relation between wavelet inverse formula of Radon transform and convolution-back projection algorithm of Radon transform in 2 dimensions is established.展开更多
文摘New higher dimensional distributions are introduced in the framework of Clifford analysis.They complete the picture already established in previous work, offering unity and structuralclarity. Amongst them are the building blocks of the principal value distribution, involvingspherical harmonics, considered by Horvath and Stein.
文摘在三维视觉任务中,三维目标的未知旋转会给任务带来挑战,现有的部分神经网络框架对经过未知旋转后的三维目标进行识别或分割较为困难.针对上述问题,提出一种基于自监督学习方式的矢量型球面卷积网络,用于学习三维目标的旋转信息,以此来提升分类和分割任务的表现.首先,对三维点云信号进行球面采样,映射到单位球上;然后,使用矢量球面卷积网络提取旋转特征,同时将随机旋转后的三维点云信号输入相同结构的矢量球面卷积网络提取旋转特征,利用自监督网络训练学习旋转信息;最后,对随机旋转的三维目标进行目标分类实验和部分分割实验.实验表明,所设计的网络在测试数据随机旋转的情况下,在ModelNet40数据集上分类准确率提升75.75%,在ShapeNet数据集上分割效果显著,交并比(Intersection over union,IoU)提升51.48%.
基金supported by the Natural Science Foundation of China under Project 41076115the Global Change Research Program of China under project 2012CB955603the Public Science and Technology Research Funds of the Ocean under project 201005019
文摘The study of marine data visualization is of great value. Marine data, due to its large scale, random variation and multiresolution in nature, are hard to be visualized and analyzed. Nowadays, constructing an ocean model and visualizing model results have become some of the most important research topics of ‘Digital Ocean'. In this paper, a spherical ray casting method is developed to improve the traditional ray-casting algorithm and to make efficient use of GPUs. Aiming at the ocean current data, a 3D view-dependent line integral convolution method is used, in which the spatial frequency is adapted according to the distance from a camera. The study is based on a 3D virtual reality and visualization engine, namely the VV-Ocean. Some interactive operations are also provided to highlight the interesting structures and the characteristics of volumetric data. Finally, the marine data gathered in the East China Sea are displayed and analyzed. The results show that the method meets the requirements of real-time and interactive rendering.
文摘目的激光雷达采集的室外场景点云数据规模庞大且包含丰富的空间结构细节信息,但是目前多数点云分割方法并不能很好地平衡结构细节信息的提取和计算量之间的关系。一些方法将点云变换到多视图或体素化网格等稠密表示形式进行处理,虽然极大地减少了计算量,但却忽略了由激光雷达成像特点以及点云变换引起的信息丢失和遮挡问题,导致分割性能降低,尤其是在小样本数据以及行人和骑行者等小物体场景中。针对投影过程中的空间细节信息丢失问题,根据人类观察机制提出了一种场景视点偏移方法,以改善三维(3D)激光雷达点云分割结果。方法利用球面投影将3D点云转换为2维(2D)球面正视图(spherical front view,SFV)。水平移动SFV的原始视点以生成多视点序列,解决点云变换引起的信息丢失和遮挡的问题。考虑到多视图序列中的冗余,利用卷积神经网络(convolutional neural networks,CNN)构建场景视点偏移预测模块来预测最佳场景视点偏移。结果添加场景视点偏移模块后,在小样本数据集中,行人和骑行者分割结果改善相对明显,行人和骑行者(不同偏移距离下)的交叉比相较于原方法最高提升6.5%和15.5%。添加场景视点偏移模块和偏移预测模块后,各类别的交叉比提高1.6%Institute)上与其他算法相比,行人和骑行者的分割结果取得了较大提升,其中行人交叉比最高提升9.1%。结论本文提出的结合人类观察机制和激光雷达点云成像特点的场景视点偏移与偏移预测方法易于适配不同的点云分割方法,使得点云分割结果更加准确。
文摘In this paper, the wavelet inverse formula of Radon transform is obtained with onedimensional wavelet. The convolution back-projection method of Radon transform is derived from this inverse formula. An asymptotic relation between wavelet inverse formula of Radon transform and convolution-back projection algorithm of Radon transform in 2 dimensions is established.
