技术领域Technical Field
本发明涉及点云数据处理技术领域，尤其涉及一种基于语义级拓扑结构的点云特征提取方法、系统及设备。The present invention relates to the technical field of point cloud data processing, and in particular to a point cloud feature extraction method, system and device based on a semantic level topological structure.
背景技术Background technique
随着数字化、网络化、智能化的新型城市建设的大力发展，城市全要素数字化和虚拟化逐渐提上日程。而实现智慧城市的前提是利用数字信息表达现实的世界，使得各类空间结构更为数字化和透明化。在此过程中常用激光雷达等设备采集相关数据，比如点云数据。点云数据是指在一个三维坐标系统中的一组向量的集合。数据以点的形式记录，每一个点包含有三维坐标，并且可以携带有关该点属性的其他信息，例如颜色、反射率、强度等。点云数据的主要特点是具有高精度、高分辨率和高维度的几何信息，可以直观地表示空间中的物体形状、表面和纹理等信息。点云数据的处理和分析通常需要使用计算机视觉。With the vigorous development of new digital, networked and intelligent urban construction, the digitization and virtualization of all elements of the city are gradually on the agenda. The premise of realizing a smart city is to use digital information to express the real world, making various spatial structures more digital and transparent. In this process, laser radar and other devices are often used to collect relevant data, such as point cloud data. Point cloud data refers to a set of vectors in a three-dimensional coordinate system. The data is recorded in the form of points, each of which contains three-dimensional coordinates and can carry other information about the attributes of the point, such as color, reflectivity, intensity, etc. The main characteristics of point cloud data are high-precision, high-resolution and high-dimensional geometric information, which can intuitively represent the shape, surface and texture of objects in space. The processing and analysis of point cloud data usually requires the use of computer vision.
然而，现有采集到的点云数据同时往往具有离散无序性、非结构性、稀疏性等特点。在结构复杂、特征丰富、形态多样的各式各样的场景下，这类数据很难直接使用端到端的模型进行处理。点云数据的处理第一步是点云特征提取，点云数据的特征提取对目标检测、识别、分割等下游任务的开展起到至关重要的作用。另外，在处理点云数据时，涉及到的信息量太少，没有充分表示现实世界中临近物体间丰富的邻接关系。However, the existing point cloud data collected often has the characteristics of discrete disorder, unstructured, and sparsity. In a variety of scenarios with complex structures, rich features, and diverse forms, it is difficult to directly use end-to-end models to process such data. The first step in processing point cloud data is point cloud feature extraction. The feature extraction of point cloud data plays a vital role in the implementation of downstream tasks such as target detection, recognition, and segmentation. In addition, when processing point cloud data, the amount of information involved is too small, and the rich adjacency relationship between adjacent objects in the real world is not fully represented.
因此，现有技术中在进行点云数据处理的过程中，存在信息损失大、无法表征高频率邻接关系的问题。Therefore, in the process of processing point cloud data in the prior art, there are problems such as large information loss and inability to represent high-frequency adjacency relationships.
发明内容Summary of the invention
有鉴于此，有必要提供一种基于语义级拓扑结构的点云特征提取方法、系统及设备，用以解决现有技术中在进行点云数据处理的过程中，存在的信息损失大、无法表征高频率邻接关系的问题。In view of this, it is necessary to provide a point cloud feature extraction method, system and equipment based on semantic-level topological structure to solve the problems of large information loss and inability to characterize high-frequency adjacency relationships in the process of point cloud data processing in the prior art.
为了解决上述问题，本发明提供一种基于语义级拓扑结构的点云特征提取方法，包括：In order to solve the above problems, the present invention provides a point cloud feature extraction method based on a semantic level topological structure, comprising:
获取点云数据，并对点云数据进行局部聚类，得到多个点集；Acquire point cloud data, and perform local clustering on the point cloud data to obtain multiple point sets;
提取各点集的特征向量和全局特征，并对特征向量和全局特征进行融合，得到点集的嵌入向量；Extract the feature vector and global features of each point set, and fuse the feature vector and global features to obtain the embedding vector of the point set;
确定各点集和与之相邻的邻接点集的注意力系数，并根据注意力系数和邻接点集的嵌入向量，确定点集的拓扑结构特征向量；Determine the attention coefficient of each point set and its adjacent point set, and determine the topological structure feature vector of the point set based on the attention coefficient and the embedding vector of the adjacent point set;
对嵌入向量和拓扑结构特征向量进行融合，确定点集的目标点云特征。The embedding vector and the topological structure feature vector are fused to determine the target point cloud features of the point set.
进一步地，提取各点集的特征向量和全局特征，并对特征向量和全局特征进行融合，得到点集的嵌入向量，包括：Furthermore, the feature vector and global features of each point set are extracted, and the feature vector and the global features are fused to obtain the embedding vector of the point set, including:
将点集输入至点云网络模型，得到点集的特征向量；Input the point set into the point cloud network model to obtain the feature vector of the point set;
将点集输入至自注意力网络模型，得到点集的全局特征；Input the point set into the self-attention network model to obtain the global features of the point set;
将特征向量和全局特征进行特征拼接并输入至多层感知器，得到点集的嵌入向量。The feature vector and the global feature are concatenated and input into the multi-layer perceptron to obtain the embedding vector of the point set.
进一步地，确定各点集和与之相邻的邻接点集的注意力系数，并根据注意力系数和邻接点集的嵌入向量，确定点集的拓扑结构特征向量，包括：Furthermore, the attention coefficient of each point set and its adjacent point set is determined, and the topological structure feature vector of the point set is determined according to the attention coefficient and the embedding vector of the adjacent point set, including:
计算点集和与之相邻的邻接点集的注意力系数；Calculate the attention coefficient of the point set and its adjacent point set;
将注意力系数进行归一化处理，并和邻接点集的嵌入向量相乘，聚合更新点集的节点特征；The attention coefficient is normalized and multiplied with the embedding vector of the neighboring point set to aggregate and update the node features of the point set;
由激活函数激活节点特征，得到语义级别的点集的拓扑结构特征向量。The node features are activated by the activation function to obtain the topological structure feature vector of the point set at the semantic level.
进一步地，计算点集和与之相邻的邻接点集的注意力系数，包括：Furthermore, the attention coefficients of the point set and its adjacent point sets are calculated, including:
根据注意力系数计算公式，计算点集和与之相邻的邻接点集的注意力系数。According to the attention coefficient calculation formula, the attention coefficient of the point set and its adjacent point set is calculated.
进一步地，注意力系数计算公式为：Furthermore, the attention coefficient calculation formula is:
其中，为注意力系数，/>为点集，/>为邻接点集，/>为共享参数，[/>||/>]表示对两个向量进行拼接，/>为点集与邻接点集的邻接特征，/>表示多层感知器，/>( )表示将高维特征映射到一个实数。in, is the attention coefficient, /> is a point set, /> is the set of adjacent points, /> For shared parameters, [/> ||/> ] means concatenating two vectors, /> is the adjacency feature between the point set and the adjacent point set, /> represents a multilayer perceptron, /> ( ) represents mapping a high-dimensional feature to a real number.
进一步地，在计算点集和与之相邻的邻接点集的注意力系数之前，还包括：Furthermore, before calculating the attention coefficient of the point set and its adjacent point set, it also includes:
构建Delaunay三角网，得到点集的二值邻接矩阵；Construct a Delaunay triangulation network and obtain the binary adjacency matrix of the point set;
根据二值邻接矩阵确定点集的邻接点集。Determine the adjacent point set of a point set based on the binary adjacency matrix.
进一步地，对嵌入向量和拓扑结构特征向量进行融合，确定点集的目标点云特征，包括：Furthermore, the embedding vector and the topological structure feature vector are fused to determine the target point cloud features of the point set, including:
将嵌入向量和拓扑结构特征向量拼接后输入至多层感知器，确定点集的目标点云特征。The embedding vector and the topological structure feature vector are concatenated and input into a multi-layer perceptron to determine the target point cloud features of the point set.
为了解决上述问题，本发明还提供一种基于语义级拓扑结构的点云特征提取系统，包括：In order to solve the above problems, the present invention also provides a point cloud feature extraction system based on a semantic level topological structure, comprising:
点集获取模块，用于获取点云数据，并对点云数据进行局部聚类，得到多个点集；The point set acquisition module is used to acquire point cloud data and perform local clustering on the point cloud data to obtain multiple point sets;
嵌入向量获取模块，用于提取各点集的特征向量和全局特征，并对特征向量和全局特征进行融合，得到点集的嵌入向量；The embedding vector acquisition module is used to extract the feature vector and global features of each point set, and fuse the feature vector and the global features to obtain the embedding vector of the point set;
拓扑结构特征向量获取模块，用于确定各点集和与之相邻的邻接点集的注意力系数，并根据注意力系数和邻接点集的嵌入向量，确定点集的拓扑结构特征向量；A topological structure feature vector acquisition module is used to determine the attention coefficient of each point set and its adjacent point set, and determine the topological structure feature vector of the point set according to the attention coefficient and the embedding vector of the adjacent point set;
目标点云特征获取模块，用于对嵌入向量和拓扑结构特征向量进行融合，确定点集的目标点云特征。The target point cloud feature acquisition module is used to fuse the embedding vector and the topological structure feature vector to determine the target point cloud features of the point set.
为了解决上述问题，本发明还提供一种基于语义级拓扑结构的点云特征提取设备，包括存储器和处理器，其中，In order to solve the above problems, the present invention also provides a point cloud feature extraction device based on a semantic level topological structure, comprising a memory and a processor, wherein:
存储器，用于存储程序；Memory, used to store programs;
处理器，与存储器耦合，用于执行存储器中存储的程序，以实现如上文所述的基于语义级拓扑结构的点云特征提取方法、系统及设备中的步骤。The processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps in the point cloud feature extraction method, system and device based on semantic-level topological structure as described above.
为了解决上述问题，本发明还提供一种计算机可读存储介质，用于存储计算机可读取的程序或指令，程序或指令被处理器执行时能够实现如上文所述的基于语义级拓扑结构的点云特征提取方法、系统及设备中的步骤。In order to solve the above problems, the present invention also provides a computer-readable storage medium for storing computer-readable programs or instructions. When the program or instructions are executed by a processor, the steps in the point cloud feature extraction method, system and device based on semantic-level topological structure as described above can be implemented.
采用上述实施例的有益效果是：本发明提供一种基于语义级拓扑结构的点云特征提取方法、系统及设备，该方法通过先对点云数据局部聚类，实现了对点云数据的分块处理，得到了有相近关系的点集；通过对点集的特征向量和全局特征进行融合，保证了嵌入向量的完整性，避免了信息的过量损失；通过与邻接点集的邻接特征融合构建语义级别的点集的拓扑结构特征向量，实现了有效表征点集的高频率邻接关系；最终通过将嵌入向量和拓扑结构特征向量进行融合，不仅保证了目标点云特征的完整性，还实现了表征高频率邻接关系。The beneficial effects of adopting the above-mentioned embodiments are as follows: the present invention provides a point cloud feature extraction method, system and device based on semantic-level topological structure, which realizes block processing of point cloud data by first locally clustering the point cloud data, and obtains point sets with similar relationships; by fusing the feature vector and global features of the point set, the integrity of the embedded vector is ensured and excessive information loss is avoided; by fusing the adjacent features of the adjacent point set to construct the topological structure feature vector of the point set at the semantic level, the high-frequency adjacency relationship of the point set is effectively represented; finally, by fusing the embedded vector and the topological structure feature vector, not only the integrity of the target point cloud features is ensured, but also the representation of high-frequency adjacency relationships is achieved.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
图1为本发明提供的基于语义级拓扑结构的点云特征提取方法一实施例的流程示意图；FIG1 is a flow chart of an embodiment of a method for extracting point cloud features based on a semantic-level topological structure provided by the present invention;
图2为本发明提供的基于语义级拓扑结构的点云特征提取网络一实施例的结构示意图；FIG2 is a schematic diagram of the structure of an embodiment of a point cloud feature extraction network based on a semantic level topological structure provided by the present invention;
图3为本发明提供的得到点集的嵌入向量一实施例的流程示意图；FIG3 is a schematic diagram of a flow chart of an embodiment of obtaining an embedding vector of a point set provided by the present invention;
图4为本发明提供的构建语义级别的点集的拓扑结构特征向量一实施例的流程示意图；FIG4 is a schematic diagram of a flow chart of an embodiment of constructing a topological structure feature vector of a point set at a semantic level provided by the present invention;
图5为本发明提供的基于语义级拓扑结构的点云特征提取系统一实施例的结构框图；FIG5 is a structural block diagram of an embodiment of a point cloud feature extraction system based on a semantic level topological structure provided by the present invention;
图6为本发明提供的基于语义级拓扑结构的点云特征提取设备一实施例的结构框图。FIG6 is a structural block diagram of an embodiment of a point cloud feature extraction device based on a semantic-level topological structure provided by the present invention.
具体实施方式Detailed ways
下面结合附图来具体描述本发明的优选实施例，其中，附图构成本申请一部分，并与本发明的实施例一起用于阐释本发明的原理，并非用于限定本发明的范围。The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.
随着数字化、网络化、智能化的新型城市建设的大力发展，城市全要素数字化和虚拟化逐渐提上日程。而实现智慧城市的前提是利用数字信息表达现实的世界，使得各类空间结构更为数字化和透明化。在此过程中常用激光雷达等设备采集相关数据，比如点云数据。点云数据是指在一个三维坐标系统中的一组向量的集合。数据以点的形式记录，每一个点包含有三维坐标，并且可以携带有关该点属性的其他信息，例如颜色、反射率、强度等。点云数据的主要特点是具有高精度、高分辨率和高维度的几何信息，可以直观地表示空间中的物体形状、表面和纹理等信息。点云数据的处理和分析通常需要使用计算机视觉。With the vigorous development of new digital, networked and intelligent urban construction, the digitization and virtualization of all elements of the city are gradually on the agenda. The premise of realizing a smart city is to use digital information to express the real world, making various spatial structures more digital and transparent. In this process, laser radar and other devices are often used to collect relevant data, such as point cloud data. Point cloud data refers to a set of vectors in a three-dimensional coordinate system. The data is recorded in the form of points, each of which contains three-dimensional coordinates and can carry other information about the attributes of the point, such as color, reflectivity, intensity, etc. The main characteristics of point cloud data are high-precision, high-resolution and high-dimensional geometric information, which can intuitively represent the shape, surface and texture of objects in space. The processing and analysis of point cloud data usually requires the use of computer vision.
然而，现有采集到的点云数据同时往往具有离散无序性、非结构性、稀疏性等特点。在结构复杂、特征丰富、形态多样的各式各样的场景下，这类数据很难直接使用端到端的模型进行处理。点云数据的处理第一步是点云特征提取，点云数据的特征提取对目标检测、识别、分割等下游任务的开展起到至关重要的作用。另外，在处理点云数据时，涉及到的信息量太少，没有充分表示现实世界中临近物体间丰富的邻接关系。However, the existing point cloud data collected often has the characteristics of discrete disorder, unstructured, and sparsity. In a variety of scenarios with complex structures, rich features, and diverse forms, it is difficult to directly use end-to-end models to process such data. The first step in processing point cloud data is point cloud feature extraction. The feature extraction of point cloud data plays a vital role in the implementation of downstream tasks such as target detection, recognition, and segmentation. In addition, when processing point cloud data, the amount of information involved is too small, and the rich adjacency relationship between adjacent objects in the real world is not fully represented.
因此，现有技术中在进行点云数据处理的过程中，存在信息损失大、无法表征高频率邻接关系的问题。Therefore, in the process of processing point cloud data in the prior art, there are problems such as large information loss and inability to represent high-frequency adjacency relationships.
为了解决上述问题，本发明提供一种基于语义级拓扑结构的点云特征提取方法、系统及设备，以下分别进行详细说明。In order to solve the above problems, the present invention provides a point cloud feature extraction method, system and device based on semantic-level topological structure, which are described in detail below.
图1为本发明提供的基于语义级拓扑结构的点云特征提取方法一实施例的流程示意图，如图1所示，基于语义级拓扑结构的点云特征提取方法包括：FIG1 is a flow chart of an embodiment of a point cloud feature extraction method based on a semantic level topological structure provided by the present invention. As shown in FIG1 , the point cloud feature extraction method based on a semantic level topological structure includes:
步骤S101：获取点云数据，并对点云数据进行局部聚类，得到多个点集；Step S101: acquiring point cloud data, and performing local clustering on the point cloud data to obtain multiple point sets;
步骤S102：提取各点集的特征向量和全局特征，并对特征向量和全局特征进行融合，得到点集的嵌入向量；Step S102: extracting feature vectors and global features of each point set, and fusing the feature vectors and global features to obtain an embedding vector of the point set;
步骤S103：确定各点集和与之相邻的邻接点集的注意力系数，并根据注意力系数和邻接点集的嵌入向量，确定点集的拓扑结构特征向量；Step S103: determining the attention coefficient of each point set and its adjacent point set, and determining the topological structure feature vector of the point set according to the attention coefficient and the embedding vector of the adjacent point set;
步骤S104：对嵌入向量和拓扑结构特征向量进行融合，确定点集的目标点云特征。Step S104: Fusing the embedding vector and the topological structure feature vector to determine the target point cloud features of the point set.
本实施例中，首先，获取点云数据，并对点云数据进行局部聚类，得到多个点集；其次，提取各点集的特征向量和全局特征，并对特征向量和全局特征进行融合，得到点集的嵌入向量；然后，确定各点集和与之相邻的邻接点集的注意力系数，并根据注意力系数和邻接点集的嵌入向量，确定点集的拓扑结构特征向量；最后，对嵌入向量和拓扑结构特征向量进行融合，确定点集的目标点云特征。In this embodiment, first, point cloud data is acquired, and the point cloud data is locally clustered to obtain multiple point sets; secondly, the feature vector and global features of each point set are extracted, and the feature vector and the global features are fused to obtain the embedding vector of the point set; then, the attention coefficient of each point set and its adjacent point set is determined, and the topological structure feature vector of the point set is determined based on the attention coefficient and the embedding vector of the adjacent point set; finally, the embedding vector and the topological structure feature vector are fused to determine the target point cloud features of the point set.
本实施例中，通过先对点云数据局部聚类，实现了对点云数据的分块处理，得到了有相近关系的点集；通过对点集的特征向量和全局特征进行融合，保证了嵌入向量的完整性，避免了信息的过量损失；通过与邻接点集的邻接特征融合构建语义级别的点集的拓扑结构特征向量，实现了有效表征点集的高频率邻接关系；最终通过将嵌入向量和拓扑结构特征向量进行融合，不仅保证了目标点云特征的完整性，还实现了表征高频率邻接关系。In this embodiment, by first locally clustering the point cloud data, the point cloud data is processed in blocks, and a point set with similar relationships is obtained; by fusing the feature vector and global features of the point set, the integrity of the embedded vector is ensured and excessive information loss is avoided; by fusing the adjacency features of the adjacent point set, the topological structure feature vector of the point set at the semantic level is constructed, and the high-frequency adjacency relationship of the point set is effectively represented; finally, by fusing the embedded vector and the topological structure feature vector, not only the integrity of the target point cloud features is ensured, but also the representation of high-frequency adjacency relationships is achieved.
在一具体实施例中，如图2所示，图2为本发明提供的基于语义级拓扑结构的点云特征提取网络一实施例的结构示意图。In a specific embodiment, as shown in FIG2 , FIG2 is a schematic diagram of the structure of an embodiment of a point cloud feature extraction network based on a semantic-level topological structure provided by the present invention.
作为优选的实施例，在步骤S101中，点云数据是由激光雷达在一定空间范围内扫描得到的，点云数据是指在一个三维坐标系统中的一组向量的集合。点云数据以点的形式记录，每一个点包含有三维坐标，并且可以携带有关该点属性的其他信息。As a preferred embodiment, in step S101, point cloud data is obtained by scanning a laser radar within a certain spatial range. Point cloud data refers to a set of vectors in a three-dimensional coordinate system. Point cloud data is recorded in the form of points, each point contains three-dimensional coordinates, and can carry other information about the attributes of the point.
具体地，为了对点云数据局部聚类，具体根据属性信息对点云进行局部聚类，根据原始点云的颜色、纹理等划分为语义上同质的基础几何形状，称之为点集，每个点集是一个属性的点的集合。Specifically, in order to locally cluster point cloud data, point clouds are locally clustered based on attribute information, and are divided into semantically homogeneous basic geometric shapes based on color, texture, etc. of the original point cloud, which are called point sets, and each point set is a collection of points with one attribute.
作为优选的实施例，在步骤S102中，为了得到点集的嵌入向量，如图3所示，图3为本发明提供的得到点集的嵌入向量一实施例的流程示意图，包括：As a preferred embodiment, in step S102, in order to obtain the embedding vector of the point set, as shown in FIG3 , FIG3 is a flow chart of an embodiment of obtaining the embedding vector of the point set provided by the present invention, including:
步骤S121：将点集输入至点云网络模型，得到点集的特征向量；Step S121: input the point set into the point cloud network model to obtain a feature vector of the point set;
步骤S122：将点集输入至自注意力网络模型，得到点集的全局特征；Step S122: input the point set into the self-attention network model to obtain the global features of the point set;
步骤S123：将特征向量和全局特征进行特征拼接并输入至多层感知器，得到点集的嵌入向量。Step S123: concatenate the feature vector and the global feature and input them into a multi-layer perceptron to obtain an embedding vector of the point set.
本实施例中，首先，将点集输入至点云网络模型，得到点集的特征向量；然后，将点集输入至自注意力网络模型，得到点集的全局特征；最后，将特征向量和全局特征进行特征拼接并输入至多层感知器，得到点集的嵌入向量。In this embodiment, first, the point set is input into the point cloud network model to obtain the feature vector of the point set; then, the point set is input into the self-attention network model to obtain the global features of the point set; finally, the feature vector and the global features are concatenated and input into the multi-layer perceptron to obtain the embedding vector of the point set.
本实施例中，通过点云网络模型获取点集的特征向量，自注意力网络模型获取点集的全局特征，再由多层感知器将特征向量和全局特征进行特征拼接，从而确定点集的嵌入向量，保证了嵌入向量的完整性，避免了信息的过量损失。In this embodiment, the feature vector of the point set is obtained through the point cloud network model, the global features of the point set are obtained by the self-attention network model, and then the feature vector and the global features are concatenated by the multi-layer perceptron to determine the embedding vector of the point set, thereby ensuring the integrity of the embedding vector and avoiding excessive loss of information.
在一具体实施例中，根据点集内原始点云的特征和点集的密集程度等构造点集的特征，将N个点集的特征输入至Pointnet网络中，首先Pointnet网络中的T-Net进行校准，然后再用多层感知器对特征进行学习，接着再用T-Net进行对齐，得到点集的特征向量。In a specific embodiment, the features of the point set are constructed according to the features of the original point cloud in the point set and the density of the point set, and the features of N point sets are input into the Pointnet network. First, the T-Net in the Pointnet network is calibrated, and then the features are learned using a multi-layer perceptron, and then aligned using T-Net to obtain the feature vector of the point set.
由于常规的Pointnet在进行全局特征提取时，采用的是最大池化提取每个维度上最显著的特征，但仅用最大值很难表征点云的全局特征，因此本实施例中采用Transformer机制代替最大池化层在此的作用。Transformer机制可以学习到每两个点集间的注意力，将全局特征更新到每个点集的特征向量上，每个点集有差异性地学习到了全局特征信息，最后将全局特征与点集特征向量拼接后输入至多层感知器中，实现了融合局部特征（即原点集特征向量）与全局特征，得到点集的嵌入向量。表示式如下：Since the conventional Pointnet uses maximum pooling to extract the most significant features in each dimension when performing global feature extraction, but it is difficult to characterize the global features of the point cloud using only the maximum value, the Transformer mechanism is used in this embodiment to replace the role of the maximum pooling layer. The Transformer mechanism can learn the attention between every two point sets, update the global features to the feature vector of each point set, and each point set learns the global feature information differently. Finally, the global features are concatenated with the point set feature vector and input into the multi-layer perceptron, realizing the fusion of local features (i.e., the feature vector of the original point set) and global features to obtain the embedding vector of the point set. The expression is as follows:
其中，表示点集的特征向量，/>表示Transformer网络，/>表示多层感知器，/>为融合全局特征的点集的嵌入向量。in, The feature vector representing the point set, /> Represents the Transformer network, /> represents a multilayer perceptron, /> is the embedding vector of the point set that integrates the global features.
作为优选的实施例，在步骤S103中，为了确定各点集和与之相邻的邻接点集的注意力系数，并根据注意力系数和邻接点集的嵌入向量，确定点集的拓扑结构特征向量，如图4所示，图4为本发明提供的构建语义级别的点集的拓扑结构特征向量一实施例的流程示意图，包括：As a preferred embodiment, in step S103, in order to determine the attention coefficient of each point set and the adjacent point set adjacent thereto, and determine the topological structure feature vector of the point set according to the attention coefficient and the embedding vector of the adjacent point set, as shown in FIG4 , FIG4 is a flow chart of an embodiment of constructing a topological structure feature vector of a semantic-level point set provided by the present invention, including:
步骤S131：计算点集和与之相邻的邻接点集的注意力系数；Step S131: Calculate the attention coefficient of the point set and its adjacent point sets;
步骤S132：将注意力系数进行归一化处理，并和邻接点集的嵌入向量相乘，聚合更新点集的节点特征；Step S132: normalize the attention coefficient and multiply it with the embedding vector of the neighboring point set to aggregate and update the node features of the point set;
步骤S133：由激活函数激活节点特征，得到语义级别的点集的拓扑结构特征向量。Step S133: Activate the node features by the activation function to obtain the topological structure feature vector of the point set at the semantic level.
本实施例中，首先，计算点集和与之相邻的邻接点集的注意力系数；然后，将注意力系数进行归一化处理，并和邻接点集的嵌入向量相乘，聚合更新点集的节点特征；最后，由激活函数激活节点特征，得到语义级别的点集的拓扑结构特征向量。In this embodiment, first, the attention coefficients of the point set and its adjacent point sets are calculated; then, the attention coefficients are normalized and multiplied with the embedding vectors of the adjacent point sets to aggregate and update the node features of the point set; finally, the node features are activated by the activation function to obtain the topological structure feature vector of the point set at the semantic level.
本实施例中，通过专门计算点集和与之相邻的邻接点集的注意力系数，能够量化确定点集与邻接点集之间的相邻关系，通过将注意力系数与嵌入向量相乘进行聚合更新得到节点特征，最终得到语义级别的点集的拓扑结构特征向量，实现了有效表征点集的高频率邻接关系。In this embodiment, by specifically calculating the attention coefficient of the point set and its adjacent point set, the adjacency relationship between the point set and the adjacent point set can be quantified and determined, and the node features are obtained by multiplying the attention coefficient with the embedding vector for aggregation update, and finally the topological structure feature vector of the point set at the semantic level is obtained, thereby achieving effective characterization of the high-frequency adjacency relationship of the point set.
作为优选的实施例，在计算点集和与之相邻的邻接点集的注意力系数之前，还需要先确定点集的邻接点集，因此，首先，构建Delaunay三角网，得到点集的二值邻接矩阵；然后，根据二值邻接矩阵确定点集的邻接点集。As a preferred embodiment, before calculating the attention coefficient of a point set and its adjacent point set, it is also necessary to determine the adjacent point set of the point set. Therefore, first, a Delaunay triangulation is constructed to obtain a binary adjacency matrix of the point set; then, the adjacent point set of the point set is determined based on the binary adjacency matrix.
在一具体实施例中，通过点集的特征向量，能够设计出表征邻接关系的邻接特征，例如：点集对之间的相对位置、法向量夹角、表面积之比、粗糙程度差异等。In a specific embodiment, adjacency features that characterize adjacency relationships can be designed through feature vectors of point sets, such as relative positions between point set pairs, normal vector angles, surface area ratios, roughness differences, and the like.
然而，虽然这里的邻接特征是高维特征向量，但仍是低频的，无法表征现实世界中复杂的语义。例如：窗户与墙面之间的嵌入关系，楼梯与上层地板下层天花板之间的贯穿关系。这些关系都是高频的，无法仅仅用上述的高维特征向量表示。而这种高频率的拓扑结构对点云的语义分割等下游任务意义重大，极大程度地避免了由于关系语义的缺失而造成的错误，例如：房梁只可能出现在天花板之下，门框和窗户只可能嵌入在墙壁的内部。However, although the adjacency features here are high-dimensional feature vectors, they are still low-frequency and cannot represent the complex semantics in the real world. For example: the embedded relationship between windows and walls, and the through relationship between stairs and upper floors and lower ceilings. These relationships are high-frequency and cannot be represented by the above high-dimensional feature vectors alone. This high-frequency topological structure is of great significance to downstream tasks such as semantic segmentation of point clouds, and it greatly avoids errors caused by the lack of relationship semantics, for example: beams can only appear under the ceiling, and door frames and windows can only be embedded inside the wall.
因此，需要对通过图自注意力机制融合邻接特征，构建语义级别的点集拓扑结构特征向量，图示中，可表示为：Therefore, it is necessary to fuse the adjacent features through the graph self-attention mechanism to construct the semantic-level point set topological structure feature vector, which can be expressed as:
其中，为当前点集的嵌入向量，/>( )表示图自注意力网络，/>为拓扑结构特征向量。in, is the embedding vector of the current point set, /> ( ) represents the graph self-attention network, /> is the topological structure feature vector.
作为优选的实施例，在步骤S131中，根据注意力系数计算公式，计算点集和与之相邻的邻接点集的注意力系数。As a preferred embodiment, in step S131, the attention coefficients of the point set and its adjacent point sets are calculated according to the attention coefficient calculation formula.
注意力系数计算公式为：The calculation formula of attention coefficient is:
其中，为注意力系数，/>为点集，/>为邻接点集，/>为共享参数，[/>||/>]表示对两个向量进行拼接，/>为点集与邻接点集的邻接特征，/>表示多层感知器，/>( )表示将高维特征映射到一个实数。in, is the attention coefficient, /> is a point set, /> is the set of adjacent points, /> For shared parameters, [/> ||/> ] means concatenating two vectors, /> is the adjacency feature between the point set and the adjacent point set, /> represents a multilayer perceptron, /> ( ) represents mapping a high-dimensional feature to a real number.
需要说明的是，注意力系数计算表达式的含义是：将点集与邻接点集均通过共享参数线性映射，然后对映射变换后的特征向量进行拼接，并与邻接特征通过多层感知器后的结果相乘，最后将拼接相乘后的高维特征映射到一个实数上，即注意力系数。It should be noted that the meaning of the attention coefficient calculation expression is: the point set and the adjacent point set are linearly mapped through shared parameters, and then the feature vectors after the mapping transformation are spliced and multiplied with the adjacent features after passing through the multilayer perceptron. Finally, the high-dimensional features after splicing and multiplication are mapped to a real number, that is, the attention coefficient.
作为优选的实施例，在步骤S132-133中，将注意力系数归一化后与邻接点集的嵌入向量加权求和后，得到当前点集的拓扑结构特征向量；As a preferred embodiment, in step S132-133, the attention coefficient is normalized and weighted summed with the embedding vector of the adjacent point set to obtain the topological structure feature vector of the current point set: ;
其中，为激活函数，/>为归一化操作，/>为当前点集的邻接点集集合。in, is the activation function, /> For normalization operation, /> is the set of adjacent point sets of the current point set.
作为优选的实施例，在步骤S104中，在确定了点集的嵌入向量和拓扑结构特征向量之后，为了确定点集的目标点云特征，需要将嵌入向量和拓扑结构特征向量拼接后输入至多层感知器，确定点集的目标点云特征。As a preferred embodiment, in step S104, after determining the embedding vector and topological structure feature vector of the point set, in order to determine the target point cloud features of the point set, it is necessary to concatenate the embedding vector and the topological structure feature vector and input them into a multilayer perceptron to determine the target point cloud features of the point set.
在一具体实施例中，拓扑结构特征向量富含语义信息，与点集的嵌入向量融合后得到点集的目标点云特征/>，即：In a specific embodiment, the topological structure feature vector Rich in semantic information, the target point cloud features of the point set are obtained after fusion with the embedding vector of the point set/> ,Right now:
上述公式的含义是：将拓扑结构特征向量与嵌入向量拼接后输入至多层感知器，多层感知器输出的即为点集的目标点云特征。The meaning of the above formula is: the topological structure feature vector and the embedding vector are concatenated and input into the multi-layer perceptron, and the output of the multi-layer perceptron is the target point cloud feature of the point set.
需要说明的是，如果将本实施例中将点云特征提取方法用于点云分割，多层感知器融合特征后充当全连接层，输出维度对应分割的类别数L。It should be noted that if the point cloud feature extraction method in this embodiment is used for point cloud segmentation, the multi-layer perceptron acts as a fully connected layer after fusing the features, and the output dimension corresponds to the number of segmented categories L.
通过上述方式，通过先对点云数据局部聚类，实现了对点云数据的分块处理，得到了有相近关系的点集；通过对点集的特征向量和全局特征进行融合，保证了嵌入向量的完整性，避免了信息的过量损失；通过与邻接点集的邻接特征融合构建语义级别的点集的拓扑结构特征向量，实现了有效表征点集的高频率邻接关系；最终通过将嵌入向量和拓扑结构特征向量进行融合，不仅保证了目标点云特征的完整性，还实现了表征高频率邻接关系。Through the above method, by first locally clustering the point cloud data, the block processing of the point cloud data is realized, and a point set with similar relationships is obtained; by fusing the feature vector and global features of the point set, the integrity of the embedded vector is guaranteed and excessive information loss is avoided; by fusing the adjacent features of the adjacent point set with the adjacent feature vector of the semantic level, the topological structure feature vector of the point set is constructed, and the high-frequency adjacency relationship of the point set is effectively represented; finally, by fusing the embedded vector and the topological structure feature vector, not only the integrity of the target point cloud features is guaranteed, but also the representation of high-frequency adjacency relationships is achieved.
为了解决上述问题，本发明还提供一种基于语义级拓扑结构的点云特征提取系统，如图5所示，图5为本发明提供的基于语义级拓扑结构的点云特征提取系统一实施例的结构框图，基于语义级拓扑结构的点云特征提取系统500包括：In order to solve the above problems, the present invention further provides a point cloud feature extraction system based on a semantic level topological structure, as shown in FIG5 , which is a structural block diagram of an embodiment of a point cloud feature extraction system based on a semantic level topological structure provided by the present invention. The point cloud feature extraction system based on a semantic level topological structure 500 includes:
点集获取模块501，用于获取点云数据，并对点云数据进行局部聚类，得到多个点集；The point set acquisition module 501 is used to acquire point cloud data and perform local clustering on the point cloud data to obtain multiple point sets;
嵌入向量获取模块502，用于提取各点集的特征向量和全局特征，并对特征向量和全局特征进行融合，得到点集的嵌入向量；The embedding vector acquisition module 502 is used to extract the feature vector and global features of each point set, and fuse the feature vector and the global features to obtain the embedding vector of the point set;
拓扑结构特征向量获取模块503，用于确定各点集和与之相邻的邻接点集的注意力系数，并根据注意力系数和邻接点集的嵌入向量，确定点集的拓扑结构特征向量；A topological structure feature vector acquisition module 503 is used to determine the attention coefficient of each point set and its adjacent point set, and determine the topological structure feature vector of the point set according to the attention coefficient and the embedding vector of the adjacent point set;
目标点云特征获取模块504，用于对嵌入向量和拓扑结构特征向量进行融合，确定点集的目标点云特征。The target point cloud feature acquisition module 504 is used to fuse the embedding vector and the topological structure feature vector to determine the target point cloud features of the point set.
本发明还相应提供了一种基于语义级拓扑结构的点云特征提取设备，如图6所示，图6为本发明提供的基于语义级拓扑结构的点云特征提取设备一实施例的结构框图。基于语义级拓扑结构的点云特征提取设备600可以是移动终端、桌上型计算机、笔记本、掌上电脑及服务器等计算设备。基于语义级拓扑结构的点云特征提取设备600包括处理器601以及存储器602，其中，存储器602上存储有基于语义级拓扑结构的点云特征提取程序603。The present invention also provides a point cloud feature extraction device based on a semantic level topological structure, as shown in FIG6 , which is a structural block diagram of an embodiment of a point cloud feature extraction device based on a semantic level topological structure provided by the present invention. The point cloud feature extraction device 600 based on a semantic level topological structure can be a computing device such as a mobile terminal, a desktop computer, a notebook, a PDA, and a server. The point cloud feature extraction device 600 based on a semantic level topological structure includes a processor 601 and a memory 602, wherein the memory 602 stores a point cloud feature extraction program 603 based on a semantic level topological structure.
存储器602在一些实施例中可以是计算机设备的内部存储单元，例如计算机设备的硬盘或内存。存储器602在另一些实施例中也可以是计算机设备的外部存储设备，例如计算机设备上配备的插接式硬盘，智能存储卡（Smart Media Card，SMC），安全数字（SecureDigital，SD）卡，闪存卡（Flash Card）等。进一步地，存储器602还可以既包括计算机设备的内部存储单元也包括外部存储设备。存储器602用于存储安装于计算机设备的应用软件及各类数据，例如安装计算机设备的程序代码等。存储器602还可以用于暂时的存储已经输出或者将要输出的数据。在一实施例中，基于语义级拓扑结构的点云特征提取程序603可被处理器601所执行，从而实现本发明各实施例的基于语义级拓扑结构的点云特征提取方法、系统及设备。In some embodiments, the memory 602 may be an internal storage unit of a computer device, such as a hard disk or memory of the computer device. In other embodiments, the memory 602 may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the computer device. Further, the memory 602 may also include both an internal storage unit of the computer device and an external storage device. The memory 602 is used to store application software and various types of data installed on the computer device, such as program codes installed on the computer device. The memory 602 can also be used to temporarily store data that has been output or is to be output. In one embodiment, the point cloud feature extraction program 603 based on the semantic level topological structure can be executed by the processor 601, thereby realizing the point cloud feature extraction method, system and device based on the semantic level topological structure of each embodiment of the present invention.
处理器601在一些实施例中可以是中央处理器（Central Processing Unit，CPU），微处理器或其他数据处理芯片，用于运行存储器602中存储的程序代码或处理数据，例如执行基于语义级拓扑结构的点云特征提取程序等。In some embodiments, the processor 601 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 602, such as executing a point cloud feature extraction program based on a semantic-level topological structure.
本实施例还提供了一种计算机可读存储介质，其上存储有基于语义级拓扑结构的点云特征提取程序，计算机该程序被处理器执行时，实现如上述所述的基于语义级拓扑结构的点云特征提取方法、系统及设备。This embodiment also provides a computer-readable storage medium, on which a point cloud feature extraction program based on a semantic-level topological structure is stored. When the program is executed by a computer processor, the point cloud feature extraction method, system and device based on a semantic-level topological structure as described above are implemented.
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法的实施例的流程。本申请所提供的各实施例中所使用的对存储器、存储、数据库或其他介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器（ROM）、可编程ROM（PROM）、电可编程ROM（EPROM）、电可擦除可编程ROM（EEPROM）或闪存。易失性存储器可包括随机存取存储器（RAM）或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM（SRAM）、动态RAM（DRAM）、同步DRAM（SDRAM）、双数据率SDRAM（DDRSDRAM）、增强型SDRAM（ESDRAM）、同步链路（Synchlink） DRAM（SLDRAM）、存储器总线（Rambus）直接RAM（RDRAM）、直接存储器总线动态RAM（DRDRAM），以及存储器总线动态RAM（RDRAM）等。Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and/or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
以上所述，仅为本发明较佳的具体实施方式，但本发明的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本发明揭露的技术范围内，可轻易想到的变化或替换，都应涵盖在本发明的保护范围之内。The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410121594.5ACN117649530B (en) | 2024-01-30 | 2024-01-30 | Point cloud feature extraction method, system and equipment based on semantic level topological structure |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410121594.5ACN117649530B (en) | 2024-01-30 | 2024-01-30 | Point cloud feature extraction method, system and equipment based on semantic level topological structure |
| Publication Number | Publication Date |
|---|---|
| CN117649530A CN117649530A (en) | 2024-03-05 |
| CN117649530Btrue CN117649530B (en) | 2024-04-23 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410121594.5AActiveCN117649530B (en) | 2024-01-30 | 2024-01-30 | Point cloud feature extraction method, system and equipment based on semantic level topological structure |
| Country | Link |
|---|---|
| CN (1) | CN117649530B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111814874A (en)* | 2020-07-08 | 2020-10-23 | 东华大学 | A multi-scale feature extraction enhancement method and module for point cloud deep learning |
| CN114241226A (en)* | 2021-12-07 | 2022-03-25 | 电子科技大学 | Three-dimensional point cloud semantic segmentation method based on multi-neighborhood characteristics of hybrid model |
| CN114612660A (en)* | 2022-03-01 | 2022-06-10 | 浙江工业大学 | Three-dimensional modeling method based on multi-feature fusion point cloud segmentation |
| CN115019043A (en)* | 2022-06-10 | 2022-09-06 | 华南理工大学 | 3D object detection method based on image point cloud fusion based on cross-attention mechanism |
| CN115294563A (en)* | 2022-07-21 | 2022-11-04 | 武汉大学 | 3D point cloud analysis method and device based on Transformer and capable of enhancing local semantic learning ability |
| CN115375877A (en)* | 2022-09-20 | 2022-11-22 | 广东工业大学 | Three-dimensional point cloud classification method and device based on channel attention mechanism |
| CN116052159A (en)* | 2023-02-17 | 2023-05-02 | 南京邮电大学 | 3D object recognition system and recognition method based on depth Hough voting |
| CN116258903A (en)* | 2023-03-09 | 2023-06-13 | 北京工业大学 | 3D Point Cloud Data Classification Method Based on Hierarchical Graph Convolutional Network and Spatial Attention Mechanism |
| CN116363439A (en)* | 2023-04-21 | 2023-06-30 | 北京石油化工学院 | A method, device and equipment for point cloud classification based on multi-head self-attention |
| CN116524197A (en)* | 2023-06-30 | 2023-08-01 | 厦门微亚智能科技有限公司 | Point cloud segmentation method, device and equipment combining edge points and depth network |
| CN116778156A (en)* | 2023-06-05 | 2023-09-19 | 苏州大学 | Three-dimensional measurement segmentation method based on deep learning |
| CN116824533A (en)* | 2023-07-14 | 2023-09-29 | 东北石油大学 | A feature enhancement method for distant and small target point cloud data based on attention mechanism |
| WO2023185494A1 (en)* | 2022-03-28 | 2023-10-05 | 苏州浪潮智能科技有限公司 | Point cloud data identification method and apparatus, electronic device, and storage medium |
| CN117218351A (en)* | 2023-09-20 | 2023-12-12 | 河北工业大学 | Three-dimensional point cloud semantic segmentation method based on local and global context awareness |
| CN117237643A (en)* | 2023-11-01 | 2023-12-15 | 重庆数字城市科技有限公司 | A point cloud semantic segmentation method and system |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US12013457B2 (en)* | 2020-03-05 | 2024-06-18 | Uatc, Llc | Systems and methods for integrating radar data for improved object detection in autonomous vehicles |
| US20220148311A1 (en)* | 2022-01-24 | 2022-05-12 | Intel Corporation | Segment fusion based robust semantic segmentation of scenes |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111814874A (en)* | 2020-07-08 | 2020-10-23 | 东华大学 | A multi-scale feature extraction enhancement method and module for point cloud deep learning |
| CN114241226A (en)* | 2021-12-07 | 2022-03-25 | 电子科技大学 | Three-dimensional point cloud semantic segmentation method based on multi-neighborhood characteristics of hybrid model |
| CN114612660A (en)* | 2022-03-01 | 2022-06-10 | 浙江工业大学 | Three-dimensional modeling method based on multi-feature fusion point cloud segmentation |
| WO2023185494A1 (en)* | 2022-03-28 | 2023-10-05 | 苏州浪潮智能科技有限公司 | Point cloud data identification method and apparatus, electronic device, and storage medium |
| CN115019043A (en)* | 2022-06-10 | 2022-09-06 | 华南理工大学 | 3D object detection method based on image point cloud fusion based on cross-attention mechanism |
| CN115294563A (en)* | 2022-07-21 | 2022-11-04 | 武汉大学 | 3D point cloud analysis method and device based on Transformer and capable of enhancing local semantic learning ability |
| CN115375877A (en)* | 2022-09-20 | 2022-11-22 | 广东工业大学 | Three-dimensional point cloud classification method and device based on channel attention mechanism |
| CN116052159A (en)* | 2023-02-17 | 2023-05-02 | 南京邮电大学 | 3D object recognition system and recognition method based on depth Hough voting |
| CN116258903A (en)* | 2023-03-09 | 2023-06-13 | 北京工业大学 | 3D Point Cloud Data Classification Method Based on Hierarchical Graph Convolutional Network and Spatial Attention Mechanism |
| CN116363439A (en)* | 2023-04-21 | 2023-06-30 | 北京石油化工学院 | A method, device and equipment for point cloud classification based on multi-head self-attention |
| CN116778156A (en)* | 2023-06-05 | 2023-09-19 | 苏州大学 | Three-dimensional measurement segmentation method based on deep learning |
| CN116524197A (en)* | 2023-06-30 | 2023-08-01 | 厦门微亚智能科技有限公司 | Point cloud segmentation method, device and equipment combining edge points and depth network |
| CN116824533A (en)* | 2023-07-14 | 2023-09-29 | 东北石油大学 | A feature enhancement method for distant and small target point cloud data based on attention mechanism |
| CN117218351A (en)* | 2023-09-20 | 2023-12-12 | 河北工业大学 | Three-dimensional point cloud semantic segmentation method based on local and global context awareness |
| CN117237643A (en)* | 2023-11-01 | 2023-12-15 | 重庆数字城市科技有限公司 | A point cloud semantic segmentation method and system |
| Title |
|---|
| Laplacian Mesh Transformer: Dual Attention and Topology Aware Network for 3D Mesh Classification and Segmentation;Li X J, et al;European Conference on Computer Vision. Cham: Springer Nature Switzerland;20221022;第541-560页* |
| MPCT: Multiscale Point Cloud Transformer With a Residual Network;Y. Wu, et al;IEEE Transactions on Multimedia;20231212;第3505-3516页* |
| Pct: Point cloud transformer;Guo M H, et al;Computational Visual Media;20210410;第187-199页* |
| Point transformer;Zhao H, et al;Proceedings of the IEEE/CVF international conference on computer vision;20211231;第16259-16268页* |
| 基于激光点云语义分割的同时定位与地图构建;王梓辰;大连理工大学;20220115;第1-78页* |
| Publication number | Publication date |
|---|---|
| CN117649530A (en) | 2024-03-05 |
| Publication | Publication Date | Title |
|---|---|---|
| JP7376233B2 (en) | Semantic segmentation of 2D floor plans using pixel-wise classifiers | |
| CN114764778A (en) | Target detection method, target detection model training method and related equipment | |
| US9824494B2 (en) | Hybrid surfaces for mesh repair | |
| WO2023164933A1 (en) | Building modeling method and related apparatus | |
| CN118456446B (en) | Intelligent reinforcement cage binding method, device, equipment and medium based on three-dimensional reconstruction | |
| CN117523134B (en) | Three-dimensional building model simplification method based on structure perception point cloud semantic segmentation | |
| CN119251443B (en) | A method, system, product and medium for seamless and rapid fusion of multivariate three-dimensional data | |
| CN117934846A (en) | Three-dimensional point cloud segmentation method and device based on local weighted curvature and two-point method | |
| CN119830423B (en) | Building engineering three-dimensional modeling method and system based on BIM | |
| CN120279026A (en) | Electromagnetic environment measurement method and device for complex terrain crossing area | |
| CN119991908A (en) | A method and system for quickly constructing a three-dimensional spatial model of a multi-level building | |
| CN112967333A (en) | Complex point cloud framework extraction method and system based on grade division | |
| CN118551584B (en) | A scene automatic generation method and system based on digital twin | |
| CN117649530B (en) | Point cloud feature extraction method, system and equipment based on semantic level topological structure | |
| CN118470268A (en) | A method, device and electronic equipment for constructing assembly features of a digital twin workshop | |
| CN113393568A (en) | Training method, device, equipment and medium for neck-edge linear deformation prediction model | |
| CN116824132B (en) | Plan view segmentation method and device and electronic equipment | |
| CN117808676A (en) | Drawing method and system for rapidly inheriting trans-scale land cover data | |
| CN116863428A (en) | Lane line detection method, device, electronic equipment and storage medium | |
| CN116912817A (en) | Three-dimensional scene model splitting method and device, electronic equipment and storage medium | |
| CN114299243A (en) | Method and device for point cloud feature enhancement based on multi-scale fusion | |
| CN114764860A (en) | Feature extraction method and device, computer equipment and storage medium | |
| JP2008533614A (en) | System and method for generating matched contour profiles | |
| CN110704897A (en) | Method for placing connecting node between wall keel model and bottom guide beam model and product | |
| CN119848318B (en) | Scientific research task recommendation method and device based on distributed AI kernel density |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |