Point cloud data completion method and device based on deep belief convolutional network
Through a hierarchical collaborative network based on a deep belief convolutional network and a 3D deconvolutional network, the problems of structural distortion and low computational efficiency in point cloud data completion are solved, and efficient and complete point cloud data reconstruction is achieved.
Patent Information
- Application Number
- CN202510923969.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-04
- Publication Date
- 2025-10-17
AI Technical Summary
Existing technologies have problems with structural distortion, detail loss, and low computational efficiency in the process of point cloud data completion. In particular, the point cloud data collected under sensor occlusion, object surface reflection characteristics, and environmental noise interference has a high defect rate, resulting in low application accuracy.
A hierarchical collaborative network based on deep belief convolutional network is used to complete point cloud data. Through voxel processing and 3D deconvolution network reconstruction, the deep belief convolutional network is used to learn the global distribution of point clouds. The adaptive learning mechanism and physical constraint optimization loss function are combined to achieve high-fidelity reconstruction of point cloud data.
It improves the completeness and efficiency of point cloud data completion, enabling high-fidelity reconstruction of complete and detailed 3D voxel meshes, thus improving reconstruction efficiency and ensuring the integrity of point cloud data.
Smart Images
Figure CN120807361A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of image processing, and particularly relates to a point cloud data completion method and device based on a deep belief convolutional network. BACKGROUND
[0002] Point cloud data, as a digital expression of a three-dimensional physical world, has core value in the fields of autonomous driving, industrial detection, robot navigation, etc. However, due to sensor occlusion (such as rain and fog), object surface reflection characteristics (such as deep hole parts), and environmental noise interference, the actual collected point cloud data has a high missing rate, which further reduces the application accuracy of the actual collected point cloud data. Therefore, the actual collected data needs to be processed for point cloud completion. At present, the K nearest neighbor (KNN) algorithm is usually used for missing point cloud interpolation, the traditional deep learning network PointNet++ network is used for point cloud completion, and the generator network is used to synthesize missing point cloud. However, the above existing technologies will cause structural distortion and detail loss of the completed point cloud data, and the integrity is low, and the completion efficiency is low due to the large calculation pressure in the point cloud completion calculation process. SUMMARY
[0003] The application provides a point cloud data completion method and device based on a deep belief convolutional network, which can improve the integrity and efficiency of point cloud data completion.
[0004] To achieve the above-mentioned purpose, the application provides a point cloud data completion method based on a deep belief convolutional network, which comprises:
[0005] Obtaining actual collected point cloud data, and voxelizing the point cloud data to obtain voxelized point cloud blocks;
[0006] Processing the voxelized point cloud blocks by using a hierarchical cooperative network to obtain target voxelized point cloud blocks;
[0007] Completing the point cloud by using a 3D deconvolution network in the hierarchical cooperative network according to the target voxelized point cloud blocks to obtain complete point cloud data.
[0008] Optionally, the hierarchical cooperative network comprises a first 3D convolution network, a deep belief convolutional network and a second 3D convolution network connected in sequence.
[0009] Optionally, the deep belief convolutional network adopts an implicit layer to learn the global distribution of the point cloud, and the global distribution features output by the implicit layer are transmitted to the second 3D convolution network as structural prior information.
[0010] Optionally, the first 3D convolution network is used to extract multi-scale spatial features in the voxelized point cloud blocks, comprising:
[0011] The preset scale convolution kernel is used to extract shallow detail features in the voxelized point cloud block with a first channel number;
[0012] According to the shallow detail features, a preset scale convolution kernel is used for convolution processing to obtain middle structure features with a second channel number;
[0013] According to the middle structure features, a preset scale convolution kernel is used for convolution processing to obtain global semantic features with a third channel number;
[0014] The channel alignment processing is performed on the shallow detail features, the middle structure features and the global semantic features by using the channel-by-channel convolution kernel to obtain an aligned feature map.
[0015] Optionally, before the voxelized point cloud block is processed by using the hierarchical cooperative network, an adaptive learning mechanism is further used to adjust a dynamic learning rate of the hierarchical cooperative network.
[0016] Optionally, the adaptive learning mechanism is used to adjust the dynamic learning rate of the hierarchical cooperative network, including:
[0017] The dynamic learning rate η of the hierarchical cooperative network is adjusted by using the following formula t :
[0018]
[0019] wherein, η min is a minimum learning rate, η max is a maximum learning rate, Tmax is a total iteration number, and Tcur is a current iteration number.
[0020] Optionally, before the voxelized point cloud block is processed by using the hierarchical cooperative network, a physical constraint optimization loss function is further used to train the hierarchical cooperative network.
[0021] To solve the above problems, the application further provides a point cloud data completion device based on a deep belief convolution network, which comprises:
[0022] A voxelization module is configured to acquire actual collected point cloud data and voxelize the point cloud data to obtain a voxelized point cloud block.
[0023] A voxel block processing module is configured to process the voxelized point cloud block by using a hierarchical cooperative network to obtain a target voxelized point cloud block.
[0024] A point cloud data reconstruction module is configured to perform point cloud completion on the target voxelized point cloud block by using a 3D deconvolution network in the hierarchical cooperative network to obtain complete point cloud data.
[0025] To solve the above problems, the application further provides an electronic device, which comprises:
[0026] at least one processor; and
[0027] a memory connected with the at least one processor in communication; wherein
[0028] the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the deep belief convolutional network-based point cloud data completion method described above.
[0029] To solve the above problems, the application further provides a computer readable storage medium, which stores at least one computer program, and the at least one computer program is executed by a processor in an electronic device to implement the deep belief convolutional network-based point cloud data completion method described above.
[0030] The application obtains the actually collected point cloud data, voxelizes the point cloud data to obtain a voxelized point cloud block, can convert irregular, sparse and unstructured original point cloud data into regular, dense and structured three-dimensional grid data, and further processes the voxelized point cloud block by using a hierarchical cooperative network to obtain a target voxelized point cloud block. The deep geometric semantic information and context relationship of the point cloud block can be effectively learned by the hierarchical cooperative network, and the structure features and potential complete morphology of the missing area can be accurately identified and encoded. In addition, the hierarchical cooperative network is used to perform point cloud completion on the target voxelized point cloud block by using a 3D deconvolution network to obtain complete point cloud data, and the complete and detailed three-dimensional voxel grid can be reconstructed with high fidelity, the reconstruction efficiency is improved, and the integrity of the reconstructed point cloud data is ensured. BRIEF DESCRIPTION OF DRAWINGS
[0031] Figure 1 A flowchart of a deep belief convolutional network-based point cloud data completion method provided by an embodiment of the application is shown in the figure.
[0032] Figure 2 A functional module diagram of a deep belief convolutional network-based point cloud data completion device provided by an embodiment of the application is shown in the figure.
[0033] Figure 3 A structural diagram of an electronic device for implementing the deep belief convolutional network-based point cloud data completion method provided by an embodiment of the application is shown in the figure.
[0034] The implementation, functional features and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION
[0035] It should be understood that the specific embodiments described herein are only used to explain the application, and are not used to limit the application.
[0036] The embodiment of the present application provides a point cloud data completion method based on a deep belief convolutional network. The execution subject of the point cloud data completion method based on the deep belief convolutional network includes but is not limited to at least one of electronic devices such as a server, a terminal and the like which can be configured to execute the method provided by the embodiment of the present application. In other words, the point cloud data completion method based on the deep belief convolutional network can be executed by software or hardware installed in a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to a single server, a server cluster, a cloud server or a cloud server cluster and the like. The server can be a stand-alone server, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (CDN), and basic cloud computing services such as big data and artificial intelligence platforms.
[0037] Referring to Figure 1 Fig. 1 shows a flowchart of the point cloud data completion method based on the deep belief convolutional network provided by an embodiment of the present application. In the embodiment, the point cloud data completion method based on the deep belief convolutional network includes:
[0038] S1, acquiring actual collected point cloud data, and voxelizing the point cloud data to obtain a voxelized point cloud block.
[0039] In the embodiment of the present application, the point cloud data refers to a data structure composed of a discrete point set in a three-dimensional space collected by a sensor in an industrial field, each point containing spatial coordinates (x, y, z), and the point cloud data can extend to store attributes such as color (RGB), normal vector and intensity.
[0040] As an embodiment of the present application, the point cloud data is voxelized to obtain a voxelized point cloud block, which includes:
[0041] The point cloud data is converted into grid data of a preset size, and the grid data is converted into a one-dimensional voxel tensor to obtain the voxelized point cloud block.
[0042] In the embodiment of the present application, the preset size can be set to 32*32*32.
[0043] S2, processing the voxelized point cloud block by using a hierarchical cooperative network to obtain a target voxelized point cloud block.
[0044] As an embodiment of the present application, the hierarchical cooperative network includes a first 3D convolutional network, a deep belief convolutional network and a second 3D convolutional network connected in sequence.
[0045] In the embodiment of the present application, the deep belief convolutional network refers to a multi-level implicit structure, realizes feature abstraction by stacking multiple nonlinear units, models the potential probability distribution of data, and the core is to learn the joint probability distribution of input data through the implicit variable, and realize probability modeling between convolutional layers.
[0046] In the embodiment of the present application, the deep belief convolutional network further comprises a deep belief unit (DBN).
[0047] Further, the deep belief convolutional network learns the global distribution of point clouds in the implicit layer, and transmits the output result of the implicit layer as structural prior information to the second 3D convolutional network.
[0048] In the embodiment of the present application, the structural prior information refers to a point cloud global probability distribution model learned by a deep belief unit (DBN), and the mathematical representation is a 512-dimensional feature vector output by the implicit layer.
[0049] Further, the first 3D convolutional network is used to extract multi-scale spatial features in the voxelized point cloud block, including:
[0050] The preset scale of the convolution kernel is used to extract shallow detail features in the voxelized point cloud block with the first channel number;
[0051] According to the shallow detail features, the preset scale of the convolution kernel is used to perform convolution processing with the second channel number to obtain middle structure features;
[0052] According to the middle structure features, the preset scale of the convolution kernel is used to perform convolution processing with the third channel number to obtain global semantic features;
[0053] The channel alignment processing of the shallow detail features, the middle structure features and the global semantic features is performed by using the channel-by-channel convolution kernel to obtain an aligned feature map.
[0054] In the embodiment of the present application, the preset scale of the convolution kernel can be set to 3*3*3, the first channel number can be set to 64, the second channel number can be set to 128, and the third channel number is 256.
[0055] As an embodiment of the present application, before the voxelized point cloud block is processed by using the hierarchical cooperative network, an adaptive learning mechanism is further used to adjust the dynamic learning rate of the hierarchical cooperative network.
[0056] Further, the adaptive learning mechanism is used to adjust the dynamic learning rate of the hierarchical cooperative network, including:
[0057] The dynamic learning rate η of the hierarchical cooperative network is adjusted by using the following formula t :
[0058]
[0059] wherein, η min is the minimum learning rate, η max is the maximum learning rate, Tmax is the total number of iterations, and Tcur is the current number of iterations.
[0060] As an embodiment of the present application, before the hierarchical cooperative network is used to process the voxelized point cloud block, network training of the hierarchical cooperative network is further performed by using a physical constraint optimization loss function.
[0061] In the embodiment of the present application, the physical constraint optimization loss function is a fusion loss function, and the fusion loss function includes a curvature consistency loss function and a normal vector constraint function, wherein the curvature consistency loss function adopts the following formula: curv wherein, L i pred is the curvature consistency loss value, N is the total number of point clouds in the point cloud data, P i gt is the coordinate of the i th point in the predicted point cloud, P normal is the coordinate of the i th point in the real point cloud, and k(·) is a point cloud curvature operator, is the square L2 norm, and the normal vector constraint function L normal adopts the following formula: wherein, is the normal vector of the i th point in the predicted point cloud,
[0062] S3, using the 3D deconvolution network in the hierarchical cooperative network to complete the point cloud according to the target voxelized point cloud block to obtain complete point cloud data.
[0063] The present application obtains the actually collected point cloud data, voxelizes the point cloud data to obtain a voxelized point cloud block, can convert irregular, sparse and unstructured original point cloud into regular, dense and structured three-dimensional grid data, and uses the hierarchical cooperative network to process the voxelized point cloud block to obtain a target voxelized point cloud block, which can effectively learn the deep geometric semantic information and context relationship of the point cloud block through the hierarchical cooperative network, accurately identify and encode the structural features and potential complete morphology of the damaged area, and further use the 3D deconvolution network in the hierarchical cooperative network to complete the point cloud according to the target voxelized point cloud block to obtain complete point cloud data, which can reconstruct a complete and detailed three-dimensional voxel grid with high fidelity, improve the reconstruction efficiency and ensure the integrity of the reconstructed point cloud data.
[0064] As Figure 2As shown is a functional module diagram of a point cloud data completion device based on a deep belief convolutional network according to an embodiment of the present application.
[0065] The point cloud data completion device 100 based on the deep belief convolutional network can be installed in an electronic device. According to the functions implemented, the point cloud data completion device 100 based on the deep belief convolutional network can include a voxelization module 101, a voxel block processing module 102, and a point cloud data reconstruction module 103.
[0066] The modules described in the present application can also be referred to as units, which refer to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, and are stored in the memory of the electronic device.
[0067] In the present embodiment, the functions of each module / unit are as follows:
[0068] The voxelization module 101 is configured to obtain actual collected point cloud data and voxelize the point cloud data to obtain a voxelized point cloud block.
[0069] In the embodiment of the present application, point cloud data refers to a data structure composed of a discrete point set in a three-dimensional space collected by a sensor in the industrial field, each point containing spatial coordinates (x, y, z), and the point cloud data can extend to store attributes such as color (RGB), normal vector, and intensity.
[0070] As an embodiment of the present application, the point cloud data is voxelized to obtain a voxelized point cloud block, which includes:
[0071] The point cloud data is converted into grid data of a preset size, and the grid data is converted into a one-dimensional voxel tensor to obtain the voxelized point cloud block.
[0072] In the embodiment of the present application, the preset size can be set to 32*32*32.
[0073] The voxel block processing module 102 is configured to process the voxelized point cloud block using a hierarchical cooperative network to obtain a target voxelized point cloud block.
[0074] As an embodiment of the present application, the hierarchical cooperative network includes a first 3D convolutional network, a deep belief convolutional network, and a second 3D convolutional network connected in sequence.
[0075] In the embodiment of the present application, the deep belief convolutional network refers to a multi-level hidden structure, which realizes feature abstraction by stacking multiple nonlinear units, models the probability distribution of data, and its core is to learn the joint probability distribution of input data through hidden variables and realize probability modeling between convolutional layers.
[0076] In the embodiment of the present application, the deep belief convolutional network further comprises a deep belief unit (DBN).
[0077] Further, the deep belief convolutional network learns the global distribution of the point cloud in the hidden layer and transmits the output result of the hidden layer as structural prior information to the second 3D convolutional network.
[0078] In the embodiment of the present application, the structural prior information refers to a global probability distribution model of the point cloud learned by the deep belief unit (DBN), which is mathematically represented as a 512-dimensional feature vector output by the hidden layer.
[0079] Further, the first 3D convolutional network is used to extract multi-scale spatial features in the voxelized point cloud block, including:
[0080] The preset scale of the convolution kernel is used to extract shallow detail features in the voxelized point cloud block with the first channel number;
[0081] According to the shallow detail features, the preset scale of the convolution kernel is used to perform convolution processing with the second channel number to obtain middle layer structure features;
[0082] According to the middle layer structure features, the preset scale of the convolution kernel is used to perform convolution processing with the third channel number to obtain global semantic features;
[0083] The channel alignment processing is performed on the shallow detail features, the middle layer structure features and the global semantic features by using the convolution kernel to obtain an aligned feature map.
[0084] In the embodiment of the present application, the preset scale of the convolution kernel can be set to 3*3*3, the first channel number can be set to 64, the second channel number can be set to 128, and the third channel number can be set to 256.
[0085] As an embodiment of the present application, before the voxelized point cloud block is processed by using the hierarchical cooperative network, an adaptive learning mechanism is used to adjust the dynamic learning rate of the hierarchical cooperative network.
[0086] Further, the adaptive learning mechanism is used to adjust the dynamic learning rate of the hierarchical cooperative network, including:
[0087] The dynamic learning rate η of the hierarchical cooperative network is adjusted by using the following formula t :
[0088]
[0089] wherein, η min is the minimum learning rate, η max is the maximum learning rate, Tmax is the total number of iterations, and Tcur is the current number of iterations.
[0090] As an embodiment of the present application, before the hierarchical cooperative network is used to process the voxelized point cloud block, network training of the hierarchical cooperative network is further included by using a physical constraint optimization loss function.
[0091] In the embodiment of the present application, the physical constraint optimization loss function is a fusion loss function, and the fusion loss function includes a curvature consistency loss function and a normal vector constraint function, wherein the curvature consistency loss function adopts the following formula: Wherein, L curv is the curvature consistency loss value, N is the total number of point clouds in the point cloud data, P i pred is the coordinate of the i th point in the predicted point cloud, P i gt is the coordinate of the i th point in the real point cloud, k(·) is a point cloud curvature operator, is the square L2 norm, and the normal vector constraint function L normal adopts the following formula: Wherein, is the normal vector of the i th point in the predicted point cloud, is the normal vector of the i th point in the real point cloud, and arccos(·) is the inverse cosine function.
[0092] The point cloud data reconstruction module 103 is configured to perform point cloud completion on the target voxelized point cloud block by using the 3D deconvolution network in the hierarchical cooperative network to obtain complete point cloud data.
[0093] As Figure 3 shown, it is a structural schematic diagram of an electronic device for implementing a point cloud data completion method based on a deep belief convolutional network according to an embodiment of the present application.
[0094] The electronic device can include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and can further include a computer program stored in the memory 11 and executable on the processor 10, such as a point cloud data completion method program based on a deep belief convolutional network.
[0095] The processor 10 may, in some embodiments, be composed of integrated circuits, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits of the same or different functions, including one or more combinations of central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connects various components of the electronic device through various interfaces and lines, executes programs or modules stored in the memory 11 (for example, a program of a point cloud data completion method based on a deep belief convolutional network), and calls data stored in the memory 11 to perform various functions and process data of the electronic device.
[0096] The memory 11 includes at least one type of readable storage medium, including a flash memory, a mobile hard disk, a multimedia card, a card-type memory (for example, an SD or DX memory), a magnetic memory, a magnetic disk, an optical disk, and the like. The memory 11 may, in some embodiments, be an internal storage unit of the electronic device, for example, a mobile hard disk of the electronic device. The memory 11 may, in other embodiments, also be an external storage device of the electronic device, for example, a plug-in mobile hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, and the like. Further, the memory 11 may include both an internal storage unit and an external storage device of the electronic device. The memory 11 can be used to store application software and various data installed in the electronic device, for example, a code of a program of a point cloud data completion method based on a deep belief convolutional network, and can also be used to temporarily store data that has been or will be output.
[0097] The communication bus 12 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, and the like. The bus is configured to enable connection and communication between the memory 11, the processor 10, and the like.
[0098] The communication interface 13 is used for communication between the electronic device and other devices, including a network interface and a user interface. Optionally, the network interface can include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is usually used to establish a communication connection between the electronic device and other electronic devices. The user interface can be a display (Display), an input unit (such as a keyboard (Keyboard)), and optionally, the user interface can also be a standard wired interface, a wireless interface. Optionally, in some embodiments, the display can be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode) touch, etc. Among them, the display can also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device and to display a visualized user interface.
[0099] Figure 3 Only the electronic device with components is shown, and those skilled in the art can understand that, Figure 3 The structure shown does not constitute a limitation on the electronic device, and can include fewer or more components than shown, or combine certain components, or different component arrangements.
[0100] For example, although not shown, the electronic device can also include a power supply (such as a battery) for powering each component. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management device, so that functions such as charge management, discharge management, and power consumption management can be realized through the power management device. The power supply can also include one or more direct current or alternating current power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and any other components. The electronic device can also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which are not described here.
[0101] It should be understood that the embodiments are only for illustration and are not limited in the scope of the patent application by this structure.
[0102] The point cloud data completion method program based on a deep belief convolutional network stored in the memory 11 in the electronic device is a combination of multiple instructions, which, when running in the processor 10, can realize:
[0103] Obtaining the actually collected point cloud data, and voxelizing the point cloud data to obtain a voxelized point cloud block;
[0104] Processing the voxelized point cloud block using a hierarchical cooperative network to obtain a target voxelized point cloud block;
[0105] According to the target voxelized point cloud block, a 3D deconvolution network in the hierarchical cooperative network is used for point cloud completion to obtain complete point cloud data.
[0106] Specifically, the specific implementation method of the processor 10 to the above instructions can refer to the description of the related steps in the corresponding embodiment of the drawings, and will not be repeated here.
[0107] Further, the modules / units integrated in the electronic device 1, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. The computer readable storage medium can be volatile or non-volatile. For example, the computer readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a U disk, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM).
[0108] The application further provides a computer readable storage medium, which stores a computer program, and the computer program can realize the following when executed by a processor of an electronic device:
[0109] Obtaining the actually collected point cloud data, and voxelizing the point cloud data to obtain a voxelized point cloud block;
[0110] Processing the voxelized point cloud block by using a hierarchical cooperative network to obtain a target voxelized point cloud block;
[0111] According to the target voxelized point cloud block, a 3D deconvolution network in the hierarchical cooperative network is used for point cloud completion to obtain complete point cloud data.
[0112] In several embodiments provided in the application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the above-described apparatus embodiment is only schematic, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner.
[0113] The modules described as separate components can or can not be physically separated, and the components displayed as modules can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. According to actual needs, part or all of the modules can be selected to achieve the purpose of the embodiment scheme.
[0114] In addition, each functional module in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware, or in the form of hardware plus software function module.
[0115] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be realized in other specific forms without departing from the spirit or essential characteristics of the present application.
[0116] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application. Any additional reference signs in the claims should not be considered as limiting the claims to which they relate.
[0117] The blockchain referred to in the present application is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithm. Blockchain, in essence, is a decentralized database, which is a series of data blocks associated using cryptographic methods, each data block containing information of a batch of network transactions, for verifying the validity (anti-fake) of the information and generating the next block. The blockchain can include a blockchain underlying platform, a platform product service layer, and an application service layer, etc.
[0118] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. Artificial intelligence (AI) is the use of digital computers or computer-controlled machines to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results.
[0119] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The plurality of units or devices stated in the system claims can also be realized by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, not any particular order.
[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application.
Claims
1. A point cloud data completion method based on deep belief convolutional network, characterized by: The method comprises: Obtain the actually collected point cloud data, and voxelize the point cloud data to obtain voxelized point cloud blocks; Use the hierarchical collaborative network to process the voxelized point cloud block to obtain the target voxelized point cloud block; According to the target voxelized point cloud block, the 3D deconvolution network in the hierarchical collaborative network is used to complete the point cloud to obtain complete point cloud data.
2. The point cloud data completion method based on deep belief convolutional network according to claim 1, characterized in that: The hierarchical collaborative network includes a first 3D convolutional network, a deep belief convolutional network, and a second 3D convolutional network connected in sequence.
3. The point cloud data completion method based on deep belief convolutional network according to claim 2, characterized in that: The deep belief convolutional network adopts a hidden layer to learn the global distribution of point clouds, and passes the global distribution features output by the hidden layer as structural prior information to the second 3D convolutional network.
4. The point cloud data completion method based on deep belief convolutional network according to claim 2, characterized in that: The first 3D convolutional network is used to extract multi-scale spatial features in voxelized point cloud blocks, including: Use the convolution kernel of preset scale to extract shallow detail features in the voxelized point cloud block with the first number of channels; According to the shallow detail features, the convolution kernel of the preset scale is used to perform convolution processing with the second number of channels to obtain the middle-level structural features; According to the middle-level structural features, a convolution kernel of a preset scale is used to perform convolution processing with the third number of channels to obtain the global semantic features; The channel-by-channel convolution kernel is used to perform channel alignment processing on shallow detail features, mid-level structural features and global semantic features to obtain an aligned feature map.
5. The point cloud data completion method based on deep belief convolutional network according to claim 1, characterized in that: Before using the hierarchical collaborative network to process the voxelized point cloud blocks, the method also includes using an adaptive learning mechanism to adjust the dynamic learning rate of the hierarchical collaborative network.
6. The point cloud data completion method based on deep belief convolutional network according to claim 5, characterized in that: The method of adjusting the dynamic learning rate of the hierarchical collaborative network by using an adaptive learning mechanism includes: The following formula is used to adjust the dynamic learning rate η of the hierarchical collaborative network t : Among them, η min is the minimum learning rate, η max is the maximum learning rate, Tmax is the total number of iterations, and Tcur is the current number of iterations.
7. The point cloud data completion method based on deep belief convolutional network according to claim 1, characterized in that: Before using the hierarchical collaborative network to process the voxelized point cloud blocks, the method also includes using a physical constraint optimization loss function to perform network training on the hierarchical collaborative network.
8. A point cloud data completion device based on deep belief convolutional network, characterized in that: The device is used to implement the point cloud data completion method based on a deep belief convolutional network according to any one of claims 1 to 7, and the device includes: The voxelization module is used to obtain the actually collected point cloud data and voxelize the point cloud data to obtain voxelized point cloud blocks; A voxel block processing module is used to process the voxelized point cloud block using a hierarchical collaborative network to obtain a target voxelized point cloud block; The point cloud data reconstruction module is used to complete the point cloud based on the target voxelized point cloud block using the 3D deconvolution network in the hierarchical collaborative network to obtain complete point cloud data.
9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the point cloud data completion method based on the deep belief convolutional network as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the point cloud data completion method based on a deep belief convolutional network is implemented.