CN114648582A - Method, system, device and medium for constructing point cloud local coordinate system - Google Patents

Method, system, device and medium for constructing point cloud local coordinate system Download PDF

Info

Publication number
CN114648582A
CN114648582A CN202210338386.1A CN202210338386A CN114648582A CN 114648582 A CN114648582 A CN 114648582A CN 202210338386 A CN202210338386 A CN 202210338386A CN 114648582 A CN114648582 A CN 114648582A
Authority
CN
China
Prior art keywords
point cloud
local
coordinate system
constructing
point
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210338386.1A
Other languages
Chinese (zh)
Inventor
陈联
肖旭光
饶永生
邹宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangzhou University
Original Assignee
Guangzhou University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangzhou University filed Critical Guangzhou University
Priority to CN202210338386.1A priority Critical patent/CN114648582A/en
Publication of CN114648582A publication Critical patent/CN114648582A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
    • G06T7/75—Determining position or orientation of objects or cameras using feature-based methods involving models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10028—Range image; Depth image; 3D point clouds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a method, a system, a device and a medium for constructing a point cloud local coordinate system, wherein the method comprises the following steps: making 3D characteristics of the point cloud local block; inputting the 3D characteristics into an EE-Net network model to obtain a weight coefficient of each point pair of the predicted point cloud local block for constructing a local reference coordinate system; constructing a weight coefficient of a local reference coordinate system according to each point pair of the predicted point cloud local block to obtain a result of predicting the weight; and constructing a point cloud local reference coordinate system according to the result of the prediction weight. The construction method provided by the invention can ensure that the rigid motion of the point cloud local block is unchanged, so that a more robust and repeatable local reference coordinate system is designed, and the method is convenient and efficient and occupies less system resources.

Description

Method, system, device and medium for constructing point cloud local coordinate system
Technical Field
The invention relates to the technical field of computer data processing, in particular to a method, a system, a device and a medium for constructing a point cloud local coordinate system.
Background
Object recognition is an active area of research in the field of computer vision technology, and this problem has been extensively studied for many applications over the last several decades, including navigation, surveillance, automation, biometric recognition, surgery, and education. The goal of object recognition is to correctly recognize objects in a scene and to recover their pose (i.e., position and orientation). In addition to object recognition of 2D images, 3D object recognition has been extensively studied in the past two decades due to the availability of low cost lidar and high speed computing devices. Existing three-dimensional object recognition algorithms can be roughly classified into two categories, namely global feature-based algorithms and local feature-based algorithms. And constructing a group of features based on the algorithm of the global features, and coding the geometric characteristics of the whole three-dimensional object. Examples of such algorithms include geometric three-dimensional moments, shape distributions, and covariance functions. However, these algorithms require a complete 3D model and are therefore very sensitive to occlusion and clutter, and therefore they are more suitable for three-dimensional shape retrieval rather than object recognition and surface registration. In contrast, a set of features is defined by the algorithm based on local features, local neighborhood features of feature points are encoded, and point-to-point correspondence between curved surfaces is realized by calculating and comparing local feature descriptors of key points. The algorithm based on local features has stronger robustness to occlusion and clutter. Thus, the local approach is more efficient and robust to incomplete three-dimensional surface matching, especially in terms of target recognition and surface registration.
The LRF (local reference coordinate system) is independent of a global coordinate system, is a standard coordinate system established on a 3D local curved surface, is a useful geometric clue of a three-dimensional point cloud, and has three orthogonal axes, however, the existing LRF method is based on Covariance Analysis (CA) or Point Space Distribution (PSD), the established local coordinate system often has a symbol fuzzy problem, or has limited robustness to high noise and grid resolution change, so that the established local coordinate system has poor effect and cannot meet the requirements.
Disclosure of Invention
In view of the above problems, the present invention provides a method, system, device and medium for constructing a local coordinate system of a point cloud, which aims to solve at least one of the above problems.
A method for constructing a point cloud local coordinate system comprises the following steps:
making 3D characteristics of the point cloud local block;
inputting the 3D characteristics into an EE-Net network model to obtain a weight coefficient of each point pair of the predicted point cloud local block for constructing a local reference coordinate system;
constructing a weight coefficient of a local reference coordinate system according to each point pair of the predicted point cloud local block to obtain a result of predicting the weight;
and constructing a point cloud local reference coordinate system according to the result of the prediction weight.
A system for constructing a local coordinate system of a point cloud, comprising the following computer program modules in communication with each other:
the 3D characteristic making module is used for making 3D characteristics of the point cloud local block;
the weight coefficient acquisition module is used for inputting the 3D characteristics into an EE-Net network model to obtain a weight coefficient of each point pair of the predicted point cloud local block for constructing a local reference coordinate system;
the weight result acquisition module is used for constructing a weight coefficient of a local reference coordinate system according to each point pair of the predicted point cloud local block to obtain a result of predicting the weight;
and the coordinate system building module is used for building a point cloud local reference coordinate system according to the result of the prediction weight.
A computer arrangement comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements a method of constructing a local coordinate system of a point cloud as claimed in any one of the first aspect when executing the computer program.
A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out a method of constructing a local coordinate system of a point cloud of any one of the first aspects.
The invention has the beneficial effects that:
the method, the system, the device and the medium for constructing the point cloud local coordinate system directly take the three-dimensional data obtained by the processed three-dimensional point cloud as the input of the network, and simultaneously can ensure the rigid motion of the point cloud local block to be unchanged by designing a network structure, thereby designing a more robust and repeatable local reference coordinate system, being convenient and efficient, and occupying less system resources.
Drawings
The invention is further illustrated by means of the attached drawings, but the embodiments in the drawings do not constitute any limitation to the invention, and for a person skilled in the art, other drawings can be obtained on the basis of the following drawings without inventive effort.
FIG. 1 is a schematic flow chart of a method for constructing a local coordinate system of a point cloud according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of an EE-Net network model provided by an embodiment of the invention;
FIG. 3 is a schematic structural diagram of a twin neural network provided by an embodiment of the present invention;
FIG. 4 is a schematic diagram of a T-Net model according to an embodiment of the present invention;
fig. 5 is a schematic diagram of an attention mechanism SE module according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made with reference to the accompanying drawings. It should be noted that the description of the embodiments is provided to help understanding of the present invention, but the present invention is not limited thereto. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Referring to fig. 1, a flowchart of a method for constructing a local coordinate system of a point cloud according to an embodiment of the present invention includes the following steps:
s101, 3D characteristics of the point cloud local block are manufactured.
The present application used the following two classical and authoritative databases for the experiments:
a BR database: the model is complete point cloud data, and the scene is point cloud data after the model is rotated and translated;
in the UWAOR database, the model is complete point cloud data, and the scene is formed by rotating and mixing a plurality of models and has the characteristics of shielding, missing and the like.
The specific steps of making the 3D characteristics of the point cloud local block are as follows: firstly, generating a pair of model scene local block pairs, and specifically obtaining the cloud model scene local block pairs point to point according to the key characteristics corresponding to the model and the scene. And acquiring a local block which takes the key characteristic point as a center and has the size of 0.15 times of the radius of the whole model in the model. And then obtaining a rotation matrix, matching corresponding key characteristic points in a specific scene file according to the rotation matrix, if the same corresponding characteristic points exist in the scene file, similarly generating a point cloud local block with the size of 0.15 times of the size of the whole model radius by taking the key points as the center, and generating a pair of model scene local blocks by using the above rule.
And generating 3D features according to the point cloud local blocks. Specifically, a key characteristic point of a cloud local block is set to be p, and adjacent points of a certain key point are set to be qiFor each neighboring point qiCalculating to obtain a point q pointing to the neighboring point from the key characteristic point piAnd then normalizing the vector to obtain the desired 3D features. By the method, the data characteristics with distinction and no loss can be acquired.
S102, inputting the 3D features into an EE-Net network model to obtain a weight coefficient of each point pair of the predicted point cloud local block for constructing a local reference coordinate system.
And inputting the generated 3D characteristics as an EE-Net network model, and outputting the EE-Net network model as a weight for constructing a local reference coordinate system for each point pair of the predicted point cloud local block. The concrete structure of the EE-Net network model is realized by combining a twin network Simese-Net, T-Net and SE attention mechanism. The structure of the EE-Net network model is shown in figure 2.
Twin network: the twin neural network is a coupling framework established based on two artificial neural networks and comprises two sub-networks, wherein each sub-network receives an input, maps the input to a high-dimensional feature space and outputs a corresponding representation. By calculating the distance of two tokens, e.g. the euclidean distance, the degree of similarity of the two inputs can be compared. In the method, the similarity of two point cloud local blocks under a new local reference coordinate system is compared through a twin network structure and chamfer loss, so that the network parameters are iteratively optimized. The twin neural network structure is shown in fig. 3;
T-Net network: the learned characteristics have the characteristics of rigid motion invariance, and the specific realization principle is as follows: the point cloud data is processed according to a point cloud data rotation translation transformation principle, and a group of new point cloud data, namely the same point cloud object, can be obtained after the point cloud data is subjected to rotation translation, and different data can be represented after the point cloud data is subjected to rotation translation. The point cloud data rigid motion change characteristic can be processed in a way that firstly the pose information of the point cloud is learned to a DxD rotation matrix (D represents a characteristic dimension and represents 3) which is most beneficial to network classification or segmentation. The principle is to adjust the transformation matrix by controlling the final loss function loss, and the network structure does not care what transformation is actually done at last, as long as the final result is facilitated. Intuitively, it is understood that a rotation matrix is learned to rotate the input data to an angle more favorable for extracting the target feature, such as turning the object to the front.
Attention mechanism SE module is introduced:
see the schematic diagram of the attention mechanism SE module shown in the embodiment of fig. 5;
the SE module firstly compresses the feature graph obtained by convolution to obtain the global feature of a channel level, then performs excitation operation on the global feature to learn the relation among the channels, also obtains the weights of different channels, and finally multiplies the weights by the original feature graph to obtain the final feature. In essence, by paying attention to the relationship among channels, the model can be expected to automatically learn the importance degree of different channel characteristics, and the attention mechanism enables the model to pay more attention to the channel characteristics with the largest information quantity and restrain unimportant characteristics.
The compression operation refers to that since the convolution operates only in a local space, it is difficult to obtain enough information to extract the relationship between channels, which is more serious for the previous layers in the network because the receptive field is small. The compression operation encodes the entire spatial feature on a channel into a global feature, implemented using global tie pooling.
The excitation operation refers to that after the compression operation obtains the global description feature, another operation is needed to obtain the relationship between the channels, and this operation needs to satisfy two principles: firstly, it needs to be able to learn the nonlinear relationship between the channels; second, it is a learned relationship that is not mutually exclusive, since here a combination of multiple features is allowed, rather than a form of one-hot encoding. In order to reduce the complexity and generalization capability of the model, a structure comprising two fully-connected layers is adopted, wherein the first fully-connected layer plays a role in dimension reduction, then the ReLU activation function is adopted for processing, and the final fully-connected layer restores the original dimension.
And finally multiplying the learned activation value of each channel by the original characteristic through a proportional operation: and obtaining the weight coefficient of each channel, so that the model has better discrimination capability on the characteristics of each channel.
S103, constructing a weight coefficient of a local reference coordinate system according to each point pair of the predicted point cloud local block to obtain a result of predicting the weight.
The structure of the EE-Net network model is shown in figure 2. The input of the network model is the 3D characteristics of the point cloud local blocks, and the output is the weight of each point in the local blocks for constructing a local reference coordinate system. And (3) performing dimension increasing in the first hidden layer, firstly learning a rotation matrix of 3x3 for aligning the point cloud curved surface through a T-Net network, and then multiplying the input data by the rotation matrix to obtain the characteristic with rotation invariance. And then, one-dimensional convolution is used for increasing the dimension of one 2-dimensional feature to 64-dimensional feature, and richer feature information is obtained. Then, normalization processing is carried out on the obtained characteristics, and nonlinear mapping is carried out on the result.
The second layer hides the dimensionality in the layer and introduces the SE attention mechanism. Specifically, the dimension of 64-dimensional features is increased to 128-dimensional features through one-dimensional convolution to obtain richer feature information. Then, the obtained features are normalized, the result is subjected to nonlinear mapping, and then an SE attention mechanism is used to enable the model to pay more attention to the channel features with the largest information quantity and suppress unimportant channel features. And then continuously carrying out dimension raising on the 128-dimensional features to 1024-dimensional features, carrying out normalization processing, carrying out nonlinear mapping and carrying out SE attention mechanism processing. And then 1024 global features of the whole local block are obtained through a pooling layer.
And executing a splicing operation, and carrying out one splicing on 1024 global features of the local block and 64-dimensional features in the first hiding to obtain a matrix with the size of N multiplied by 1088.
And then, dimension reduction is carried out through a third hidden layer to obtain a prediction weight. Specifically, firstly, the 1088-dimensional features are compressed to 518-dimensional features through one-dimensional convolution, then normalization processing is carried out on the obtained features, nonlinear mapping is carried out on the normalization result, and the similar operation is repeated until the dimension reduction is 1-dimensional features, namely the Nx 1-dimensional features are obtained.
And finally, quantifying the advantages and disadvantages of the current network through the Hausdorff distance loss. And then updating parameters through back propagation, automatically optimizing the parameters of the network model, repeatedly iterating, and finally obtaining a weight prediction result by combining weight coefficients.
The parameters of the network model are set as follows: training the network by using batch size of 256 local curved surface pairs and setting the initial learning rate to 10 by using an adaptive learning rate adam optimizer-4The epoch size is set to 20 at 5% decay per epoch pass.
And S104, constructing a point cloud local reference coordinate system according to the result of the prediction weight.
And the z-axis calculation step comprises the following steps of setting the key feature point of the cloud local block as p, and calculating the tangent plane normal Z (p) of the key feature point p of the local block as the z-axis. Since the normal vector has sign ambiguity, the sign ambiguity can be removed by the following formula.
Figure BDA0003577589530000051
Wherein q isiI-th neighbor representing keypoint p
The x-axis calculation step is as follows, using step S101 as the input of an EE-Net network model in which the same architecture and basis weights are shared by twin networks, using the chamfer distance as a loss function, training and predicting to obtain each neighboring point qiWeight w to construct x-axisi. The x-axis is then constructed by means of a weighted sum. The method comprises the following specific steps: first, calculate the neighboring point qiThe projection vector projected onto the tangent plane s of the z-axis is given by the formula:
vi＝Pqi-(Pqi·Z(p))·Z(p)
in the formula, Z (p) represents a unit vector of a z axis, and then all projection vectors are integrated in a weighted vector sum mode, wherein the weight is the output of an EE-Net network model, and the formula is shown as follows:
Figure BDA0003577589530000052
obtaining an x-axis X (p) of a local coordinate system; y-axis is cross product of x-axis and z-axis
The embodiment of the invention directly uses the three-dimensional data obtained by the processed three-dimensional point cloud as the input of the network, and simultaneously can ensure the rigid motion of the local block of the point cloud to be unchanged by designing a network structure, thereby designing a more robust and repeatable local reference coordinate system.
A system for constructing a local coordinate system of a point cloud, the system comprising the following computer program modules in communication with each other: :
the 3D characteristic making module is used for making 3D characteristics of the point cloud local block;
the weight coefficient acquisition module is used for inputting the 3D characteristics into an EE-Net network model to obtain a weight coefficient of each point pair of the predicted point cloud local block for constructing a local reference coordinate system;
the weight result acquisition module is used for constructing a weight coefficient of a local reference coordinate system according to each point pair of the predicted point cloud local block to obtain a result of predicting the weight;
and the coordinate system building module is used for building a point cloud local reference coordinate system according to the result of the prediction weight.
A computer arrangement comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor, when executing the computer program, implements a method of constructing a local coordinate system of a point cloud as described in any of the above embodiments.
A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, performs a method for constructing a local coordinate system of a point cloud as in any of the above embodiments.
The method, the system, the device and the medium for constructing the point cloud local coordinate system directly take the three-dimensional data obtained by the processed three-dimensional point cloud as the input of the network, and simultaneously can ensure the rigid motion of the point cloud local block to be unchanged by designing a network structure, thereby designing a more robust and repeatable local reference coordinate system, being convenient and efficient, and occupying less system resources.
It should be noted that, in other embodiments of the present invention, different steps, parameters, modules and models are specifically selected within the scope of the above description to achieve the technical effects of the present invention, and therefore, they are not listed one by one.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the protection scope of the present invention, although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions can be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims (10)

1. A method for constructing a point cloud local coordinate system is characterized by comprising the following steps:
making 3D characteristics of the point cloud local block;
inputting the 3D characteristics into an EE-Net network model to obtain a weight coefficient of each point pair of the predicted point cloud local block for constructing a local reference coordinate system;
constructing a weight coefficient of a local reference coordinate system according to each point pair of the predicted point cloud local block to obtain a result of predicting the weight;
and constructing a point cloud local reference coordinate system according to the result of the prediction weight.
2. The method for constructing a local coordinate system of a point cloud according to claim 1, wherein the step of creating 3D features of the local blocks of the point cloud specifically comprises:
and generating a pair of model scene local blocks, and obtaining the 3D characteristics of the point cloud local blocks according to the model scene local blocks.
3. The method of claim 2, wherein the step of generating a pair of model scene local blocks specifically comprises:
firstly, according to key feature points corresponding to a model and a scene, obtaining a local block which takes the key feature points as the center and has the size of 0.15 time of the radius of the whole model from the model;
and then obtaining a rotation matrix, matching corresponding key characteristic points in a specific scene file according to the rotation matrix, and if the same corresponding characteristic points exist in the scene file, generating a point cloud local block with the size of 0.15 time of the radius of the whole model by taking the key points as the center.
4. The method of claim 3, wherein obtaining 3D features of the point cloud local blocks from the model scene local blocks comprises:
setting the key characteristic point of the point cloud local block as p and the adjacent point of a certain key point as qiFor each neighboring point qiCalculating to obtain a point q pointing to the adjacent point from the key feature point piThen normalizing the direction vector to obtain the 3D feature.
5. The method for constructing a point cloud local coordinate system of claim 1, wherein the EE-Net network model is implemented by combining twin networks Siamese-Net, T-Net, SE attention mechanism.
6. The method for constructing the point cloud local coordinate system of claim 5, wherein the inputting the 3D features into the EE-Net network model to obtain the weight coefficient of the point cloud local block to construct the local reference coordinate system comprises:
receiving the 3D characteristics through a twin network, and comparing the similarity of the two point cloud local blocks in a new local reference coordinate system, thereby iteratively optimizing network parameters;
then learning the pose information of the point cloud by a T-Net network to learn a DxD rotation matrix which is beneficial to network classification or segmentation;
and finally, by introducing an SE attention mechanism, inhibiting unimportant features in the rotation matrix, and learning through a designed network model, so as to obtain a weight coefficient of each point for constructing an LRF local coordinate system.
7. The method for constructing a point cloud local coordinate system according to claim 1, wherein the step of obtaining the result of predicting the weight according to the weight coefficient of the local reference coordinate system constructed by each point pair of the predicted point cloud local block comprises:
inputting the point cloud local block 3D characteristics into an EE-Net network model, obtaining a rotation matrix through T-Net network learning, and lifting 2-dimensional characteristics of the matrix to 64-dimensional characteristics in a first hidden layer by using one-dimensional convolution rising dimension;
in the second hidden layer, 64-dimensional features are raised to 128-dimensional features through one-dimensional convolution to obtain feature information, the obtained feature information is subjected to normalization processing, the result of the normalization processing is subjected to nonlinear mapping, then an attention mechanism is used for inhibiting unimportant channel features, the 128-dimensional features are raised to 1024-dimensional features, the normalization processing, the nonlinear mapping and the attention mechanism processing are continuously carried out, and then 1024 global features of the whole local block are obtained through a pooling layer;
executing a splicing operation, and carrying out one splicing on 1024 global features of the local block and 64-dimensional features in the first hidden layer to obtain a matrix with the size of N multiplied by 1088;
then, performing dimensionality reduction through a third hidden layer to obtain a prediction weight, specifically: firstly, the 1088-dimensional features are compressed to 518-dimensional features through one-dimensional convolution, then normalization processing is carried out on the obtained features, nonlinear mapping is carried out on the normalization result, and the similar operation is repeated until the dimension reduction is 1-dimensional features, namely the Nx 1-dimensional features are obtained;
and finally, quantifying the advantages and disadvantages of the current network through the Hausdorff distance loss, updating parameters through back propagation, automatically optimizing the parameters of the network model, repeatedly iterating, combining the weight coefficients, and finally obtaining the result of the predicted weight.
8. A system for constructing a local coordinate system of a point cloud, comprising the following computer program modules in communication with each other:
the 3D characteristic making module is used for making 3D characteristics of the point cloud local block;
the weight coefficient acquisition module is used for inputting the 3D characteristics into an EE-Net network model to obtain a weight coefficient of each point pair of the predicted point cloud local block for constructing a local reference coordinate system;
the weight result acquisition module is used for constructing a weight coefficient of a local reference coordinate system according to each point pair of the predicted point cloud local block to obtain a result of predicting the weight;
and the coordinate system building module is used for building a point cloud local reference coordinate system according to the result of the prediction weight.
9. A computer arrangement comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements a method of constructing a local coordinate system of a point cloud as claimed in any one of claims 1 to 7.
10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out a method of constructing a local coordinate system of a point cloud as claimed in any one of claims 1 to 7.
CN202210338386.1A 2022-04-01 2022-04-01 Method, system, device and medium for constructing point cloud local coordinate system Pending CN114648582A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210338386.1A CN114648582A (en) 2022-04-01 2022-04-01 Method, system, device and medium for constructing point cloud local coordinate system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210338386.1A CN114648582A (en) 2022-04-01 2022-04-01 Method, system, device and medium for constructing point cloud local coordinate system

Publications (1)

Publication Number Publication Date
CN114648582A true CN114648582A (en) 2022-06-21

Family

ID=81994725

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210338386.1A Pending CN114648582A (en) 2022-04-01 2022-04-01 Method, system, device and medium for constructing point cloud local coordinate system

Country Status (1)

Country Link
CN (1) CN114648582A (en)

Similar Documents

Publication Publication Date Title
Liu et al. Multi-modal clique-graph matching for view-based 3D model retrieval
CN109558862B (en) Crowd counting method and system based on attention thinning framework of space perception
US10210430B2 (en) System and a method for learning features on geometric domains
CN112288011B (en) Image matching method based on self-attention deep neural network
WO2019227479A1 (en) Method and apparatus for generating face rotation image
Sarkar et al. Computer perceptual organization in computer vision
US20090103786A1 (en) Apparatus and method for generating a three-dimensional representation from a two-dimensional image
Bhople et al. Point cloud based deep convolutional neural network for 3D face recognition
Sun et al. An efficient and compact 3D local descriptor based on the weighted height image
WO2023164933A1 (en) Building modeling method and related apparatus
CN114913552B (en) Three-dimensional human body density corresponding estimation method based on single-view-point cloud sequence
CN115830375A (en) Point cloud classification method and device
CN110348299B (en) Method for recognizing three-dimensional object
Hutchcroft et al. CoVisPose: Co-visibility pose transformer for wide-baseline relative pose estimation in 360∘ indoor panoramas
CN114612698A (en) Infrared and visible light image registration method and system based on hierarchical matching
CN112329662B (en) Multi-view saliency estimation method based on unsupervised learning
CN112115954B (en) Feature extraction method and device, machine readable medium and equipment
CN116912486A (en) Target segmentation method based on edge convolution and multidimensional feature fusion and electronic device
CN114648582A (en) Method, system, device and medium for constructing point cloud local coordinate system
CN115601408A (en) Point cloud registration method based on particle swarm optimization and topological graph
CN115018710A (en) Point cloud super-resolution method based on FPFH (field programmable gate array) features and deep neural network
CN115471636A (en) Virtual scene determination method and device
CN115272696A (en) Point cloud semantic segmentation method based on self-adaptive convolution and local geometric information
Dalara et al. Entity Recognition in Indian Sculpture using CLAHE and machine learning
Wang et al. A geometry feature aggregation method for point cloud classification and segmentation

Legal Events

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