CN117036639B - Multi-view geometric scene establishment method and device oriented to limited space - Google Patents
Multi-view geometric scene establishment method and device oriented to limited space Download PDFInfo
- Publication number
- CN117036639B CN117036639B CN202311058030.3A CN202311058030A CN117036639B CN 117036639 B CN117036639 B CN 117036639B CN 202311058030 A CN202311058030 A CN 202311058030A CN 117036639 B CN117036639 B CN 117036639B
- Authority
- CN
- China
- Prior art keywords
- voxel
- scale
- scene
- limited space
- voxel blocks
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 90
- 230000005855 radiation Effects 0.000 claims abstract description 53
- 210000005036 nerve Anatomy 0.000 claims abstract description 42
- 238000012549 training Methods 0.000 claims abstract description 37
- 238000013138 pruning Methods 0.000 claims abstract description 36
- 230000008569 process Effects 0.000 claims abstract description 24
- 239000013598 vector Substances 0.000 claims description 35
- 230000011218 segmentation Effects 0.000 claims description 22
- 230000002238 attenuated effect Effects 0.000 claims description 16
- 238000012804 iterative process Methods 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 10
- 230000007423 decrease Effects 0.000 claims description 9
- 230000000694 effects Effects 0.000 claims description 9
- 230000008030 elimination Effects 0.000 claims description 7
- 238000003379 elimination reaction Methods 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 5
- 238000013139 quantization Methods 0.000 claims description 4
- 238000012163 sequencing technique Methods 0.000 claims description 4
- 230000001537 neural effect Effects 0.000 abstract description 10
- 238000009776 industrial production Methods 0.000 abstract description 6
- 238000005065 mining Methods 0.000 abstract description 5
- 125000004122 cyclic group Chemical group 0.000 abstract 1
- 230000010354 integration Effects 0.000 abstract 1
- 238000007670 refining Methods 0.000 abstract 1
- 239000007787 solid Substances 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 10
- 238000005070 sampling Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/10—Constructive solid geometry [CSG] using solid primitives, e.g. cylinders, cubes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/05—Geographic models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T19/00—Manipulating 3D models or images for computer graphics
- G06T19/20—Editing of 3D images, e.g. changing shapes or colours, aligning objects or positioning parts
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2219/00—Indexing scheme for manipulating 3D models or images for computer graphics
- G06T2219/20—Indexing scheme for editing of 3D models
- G06T2219/2008—Assembling, disassembling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2219/00—Indexing scheme for manipulating 3D models or images for computer graphics
- G06T2219/20—Indexing scheme for editing of 3D models
- G06T2219/2012—Colour editing, changing, or manipulating; Use of colour codes
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- Geometry (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Graphics (AREA)
- Remote Sensing (AREA)
- Architecture (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Image Generation (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a multi-view geometric scene establishment method for a limited space, which is characterized in that a fixed or sliding rail type image acquisition device is used for acquiring images in the limited space with multiple views, and the space scene is quickly restored through iterative training of image coding pruning and a nerve radiation field model. Firstly, establishing multi-scale voxel blocks according to the size of a limited space to divide a scene in different layers. And then carrying out multi-scale coding on the position, the pose, the image attribute information and the like of the acquired image, wherein the multi-scale coding corresponds to the multi-scale voxel blocks. And then inputting the two neural radiation fields to carry out cyclic iteration of color value and voxel density integration so as to establish the solid geometry scene. And pruning and refining operations are carried out on the voxel blocks step by step in the iteration process, and the modeling of the detail area in the scene by the method is faster and more accurate as the voxel block scale becomes smaller gradually. Technical support is provided for modeling industrial production environments and equipment such as mine mining working surfaces in a limited space.
Description
Technical Field
The invention relates to the field of multi-view image processing and scene modeling, in particular to a multi-view geometric scene establishment method and device for a limited space.
Background
With the rapid development of computer vision and virtual reality technology, multi-view image processing and scene modeling are becoming hot spot areas for research. The above-described techniques have wide application in many fields including unmanned, architectural design, game development, augmented reality, and the like.
Traditional scene modeling methods are mainly based on single-view image or video data, but in limited spaces, such as narrow rooms or mine mining spaces, the single-view data cannot provide enough information to accurately reconstruct the geometry and features of the scene.
On the other hand, neural radiation fields are an emerging concept that utilizes neural network models to represent the radiation characteristics of an object or scene. The construction of neural radiation fields requires a large amount of data and computational resources and is often limited by scene complexity and computational complexity. However, in a confined space, the construction and application of the neural radiation field becomes more difficult due to the limitations of the viewing angle and the sensor position.
Aiming at the problem of establishing the geometric scene in the limited space, the prior art does not fully accelerate the training iteration process of the nerve radiation field on the premise of ensuring the establishment accuracy of the multi-view geometric scene. The partial method inputs a plurality of visual angle images or video data into the nerve radiation field directly in a pixel-level matrix form, and simultaneously carries out iterative training on the nerve radiation field by using large-scale graphic processing equipment so as to generate a scene model.
However, the hardware computing equipment required by the method has high requirements, and is difficult to apply to daily life of people. Although the method can achieve a certain scene modeling fineness, the multi-view geometric scene establishment speed is very slow, and the requirement of the current-stage social production on modeling instantaneity is far less met. Moreover, the modeling effect of the multi-view geometric scene in the limited space with higher requirements on fineness is poor, and the actual requirements of scenes such as industrial production cannot be met.
Disclosure of Invention
In view of the above problems, the present invention provides a multi-view geometric scene establishment method and a multi-view geometric scene establishment device for a limited space.
The embodiment of the invention provides a multi-view geometric scene establishment method facing a limited space, which comprises the following steps:
acquiring a target image group in a limited space meeting calculation requirements under multiple view angles through an image acquisition device;
The method comprises the steps of establishing voxel blocks with different dimensions for limited spaces with different dimensions, so as to divide the limited spaces with different dimensions into different layers, and facilitate the subsequent fine establishment of geometric scenes;
Integrating the spatial position information and the pose information of the image acquisition equipment when the target image group is acquired with the attribute information of the target image group to carry out multi-scale coding, wherein the multi-scale coding is in one-to-one correspondence with a plurality of voxel blocks segmented under the same scale size;
constructing a plurality of voxel blocks and multi-scale codes corresponding to the limited space into high-order feature vectors, inputting the high-order feature vectors into a nerve radiation field to integrate color values and voxel densities, and performing loop iteration to establish a geometric scene;
in the iteration process, removing redundancy from a plurality of voxel blocks corresponding to the limited space by utilizing pruning operation, so that the iteration speed of the nerve radiation field is improved;
In the iteration process, the sizes of a plurality of voxel blocks corresponding to the limited space are continuously attenuated, and the iteration loss value is continuously reduced, so that the multi-view geometric scene establishment method is more accurate and complete in the scene establishment of a detail area in the scene;
and establishing the required accuracy according to the actual geometric scene, judging the established geometric model, stopping iteration if the preset condition is met, and otherwise, continuing iteration operation until the preset condition is met, and stopping iteration.
Optionally, the image acquisition mode of the image acquisition device includes: adopting fixed image acquisition equipment to acquire images in a scene; or alternatively
Adopting a sliding rail type image acquisition device to acquire images in a scene;
wherein the image acquisition device is only used for acquiring images in the limited space, and the image acquisition device comprises: video camera, digital camera and terminal shooting equipment.
Optionally, establishing voxel blocks of different scale sizes for the limited spaces of different sizes includes:
determining the target scale size of a voxel block corresponding to a target limited space according to the size of the target limited space;
Dividing the target limited space into a plurality of voxel blocks with different numbers according to the target scale, so as to divide the target limited space into different layers and facilitate the fine establishment of a subsequent geometric scene, wherein the dividing method comprises the following steps: rectangular segmentation, hexagonal segmentation, and spherical segmentation.
Optionally, at the same scale size, the multi-scale coding has a one-to-one correspondence with the plurality of voxel blocks divided at the scale size, including:
Under the same scale, establishing a corresponding relation between each voxel block and the multi-scale code one by one according to the spatial position of each voxel block segmented under the scale;
the encoding mode for performing the multi-scale encoding comprises the following steps: bloom Filter coding, hash coding, and CRC coding.
Optionally, constructing the plurality of voxel blocks and the multi-scale codes corresponding to the limited space into high-order feature vectors, including:
Arranging and combining the plurality of voxel blocks and the multi-scale codes to construct the high-order feature vector;
wherein, the arrangement and combination modes comprise; and selecting the vertex information of each voxel block and the multi-scale coding bit-by-bit arrangement mode.
Optionally, removing redundancy from the plurality of voxel blocks corresponding to the limited space by using pruning operation includes:
Carrying out quantization sequencing on spatial information contained in each voxel block according to coding information contained in each voxel block in the plurality of voxel blocks to determine empty voxel blocks, wherein the empty voxel blocks are voxel blocks which do not contain or only contain a very small amount of spatial information and exist in the limited space;
And pruning operation is carried out on voxel blocks with preset proportion in the plurality of voxel blocks according to the actual refinement requirement of the scene, and the voxel density of the pruned voxel blocks is set to 0, namely the opacity of the pruned voxel blocks is 0, so that the voxel blocks do not participate in subsequent iterative training, and the aim of improving the iterative training speed of the nerve radiation field is fulfilled.
Optionally, in the iterative process, the sizes of the voxel blocks corresponding to the limited space are continuously attenuated, and as the iteration loss value is continuously reduced, the multi-view geometric scene establishment method enables the scene establishment of the detail area in the scene to be more accurate and complete, and the method comprises the following steps:
In the iterative process, after a certain number of iterations or the minimum loss value under the current scale is reached, the current spatial scale of the voxel blocks is attenuated downwards, so that the space of the current established geometric scene is continuously refined, the voxel blocks are iteratively trained, the multi-view geometric scene establishment method shows a better modeling effect in a detail area in the scene, and the scene establishment of the detail area is more accurate and complete.
The embodiment of the invention also provides a device for establishing the multi-view geometric scene facing the limited space, which comprises the following components:
the acquisition module is used for acquiring a target image group in a limited space meeting the calculation requirement under multiple view angles through the image acquisition equipment;
The voxel block module is used for establishing voxel blocks with different dimensions for the limited spaces with different dimensions, so that the limited spaces with different dimensions are divided into different layers, and the subsequent fine establishment of the geometric scene is facilitated;
The multi-scale coding module is used for integrating the spatial position information and the pose information of the image acquisition equipment when the target image group is acquired with the attribute information of the target image group to carry out multi-scale coding, wherein the multi-scale coding is in one-to-one correspondence with a plurality of voxel blocks segmented under the same scale;
The construction feature vector module is used for constructing a plurality of voxel blocks and multi-scale codes corresponding to the limited space into high-order feature vectors, inputting the high-order feature vectors into a nerve radiation field for integrating color values and voxel densities, and performing loop iteration to establish a geometric scene;
the redundancy elimination module is used for eliminating redundancy of the voxel blocks corresponding to the limited space by utilizing pruning operation so as to improve the iteration speed of the nerve radiation field;
the iteration attenuation module is used for continuously attenuating the sizes of a plurality of voxel blocks corresponding to the limited space in the iteration process, and the multi-view geometric scene establishment method is more accurate and complete in the scene establishment of the detail area in the scene along with the continuous decline of the iteration loss value;
And the judging module is used for judging the established geometric model according to the required accuracy of the actual geometric scene establishment, stopping iteration if the preset condition is met, and continuing the iteration operation until the preset condition is met, and stopping iteration.
Optionally, the voxel block building module is specifically configured to:
determining the target scale size of a voxel block corresponding to a target limited space according to the size of the target limited space;
Dividing the target limited space into a plurality of voxel blocks with different numbers according to the target scale, so as to divide the target limited space into different layers and facilitate the fine establishment of a subsequent geometric scene, wherein the dividing method comprises the following steps: rectangular segmentation, hexagonal segmentation, and spherical segmentation.
Optionally, the method for realizing that the multi-scale coding and the plurality of voxel blocks divided under the same scale size are in one-to-one correspondence by the multi-scale coding module comprises the following steps:
Under the same scale, establishing a corresponding relation between each voxel block and the multi-scale code one by one according to the spatial position of each voxel block segmented under the scale;
the encoding mode for performing the multi-scale encoding comprises the following steps: bloom Filter coding, hash coding, and CRC coding.
Optionally, the building feature vector module is specifically configured to:
Arranging and combining the plurality of voxel blocks and the multi-scale codes to construct the high-order feature vector;
wherein, the arrangement and combination modes comprise; and selecting the vertex information of each voxel block and the multi-scale coding bit-by-bit arrangement mode.
Optionally, the redundancy elimination module is specifically configured to:
Carrying out quantization sequencing on spatial information contained in each voxel block according to coding information contained in each voxel block in the plurality of voxel blocks to determine empty voxel blocks, wherein the empty voxel blocks are voxel blocks which do not contain or only contain a very small amount of spatial information and exist in the limited space;
And pruning operation is carried out on voxel blocks with preset proportion in the plurality of voxel blocks according to the actual refinement requirement of the scene, and the voxel density of the pruned voxel blocks is set to 0, namely the opacity of the pruned voxel blocks is 0, so that the voxel blocks do not participate in subsequent iterative training, and the aim of improving the iterative training speed of the nerve radiation field is fulfilled.
Optionally, the iterative attenuation module is specifically configured to:
In the iterative process, after a certain number of iterations or the minimum loss value under the current scale is reached, the current spatial scale of the voxel blocks is attenuated downwards, so that the space of the current established geometric scene is continuously refined, the voxel blocks are iteratively trained, the multi-view geometric scene establishment method shows a better modeling effect in a detail area in the scene, and the scene establishment of the detail area is more accurate and complete.
According to the multi-view geometric scene establishment method for the limited space, provided by the invention, the image acquisition equipment is used for acquiring the target image group under the multi-view condition in the limited space meeting the calculation requirement, after the target image group in the limited space under the multi-view condition is acquired, voxel blocks with different dimensions are established for the limited space with different dimensions, so that the limited space with different dimensions is segmented in different layers, and the subsequent fine establishment of the geometric scene is facilitated.
Integrating the spatial position information and the pose information of the image acquisition equipment when the target image group is acquired with the attribute information of the target image group to carry out multi-scale coding, wherein the multi-scale coding is in one-to-one correspondence with a plurality of voxel blocks segmented under the same scale size; and constructing a plurality of voxel blocks and multi-scale codes corresponding to the limited space into high-order feature vectors, inputting the high-order feature vectors into a nerve radiation field to integrate color values and voxel densities, and performing loop iteration to establish a geometric scene.
In the iteration process, redundancy is removed from a plurality of voxel blocks corresponding to the limited space by utilizing pruning operation, so that the iteration speed of the nerve radiation field is improved; in the iterative process, the sizes of a plurality of voxel blocks corresponding to the limited space are continuously attenuated, and the multi-view geometric scene establishment method is more accurate and complete in the scene establishment of the detail area in the scene along with the continuous decline of the iteration loss value.
And (3) establishing the required accuracy according to the actual geometric scene, judging the established geometric model, stopping iteration if the preset condition is met, otherwise, continuing iteration operation until the preset condition is met, stopping iteration, and finally establishing the multi-view geometric scene.
The limited space is divided into multi-scale voxel blocks according to different scene requirements, and high-order vectors are synthesized by encoding multi-scale image information. The multi-scale segmentation can enable the geometric scene to realize hierarchical training in the recovery process, and further enable the geometric scene to realize fine modeling in the iterative training of the nerve radiation field. And simultaneously, pruning operation based on spatial information redundancy is performed on the multi-scale voxel blocks, and the voxel density of the empty voxel blocks containing less spatial information is set to 0, so that redundancy in the training process of the nerve radiation field is reduced, and further, the training of the nerve radiation field is greatly accelerated. In the training iteration process, the sampling rate of each voxel block is kept unchanged, and the size of the multi-scale voxel block is continuously attenuated. And (3) along with the continuous decline of the iteration loss value, establishing the required accuracy according to the actual geometric scene, judging the established geometric model, if the requirement is met, stopping iteration, otherwise, continuing to input the nerve radiation field to perform pruning and attenuation operation until the establishment of the geometric scene is completed.
The multi-view geometric scene establishment method provided by the invention greatly improves the training speed of the nerve radiation field, has higher scene modeling fineness, simultaneously has higher multi-view geometric scene establishment speed and lower requirements on hardware computing equipment, meets the requirement of current-stage social production on modeling instantaneity, and is better applied to daily life of people. The modeling effect of the multi-view geometric scene of the limited space is good, the actual requirements of the scenes such as industrial production are completely met, and good technical support is provided for modeling of industrial production environments and equipment such as mine mining working faces in the limited space.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to designate like parts throughout the figures. In the drawings:
FIG. 1 is a flow chart of a method for establishing a multi-view geometric scene oriented to a limited space according to an embodiment of the invention;
FIG. 2 is a schematic diagram of a pruning, attenuation method for a plurality of voxel blocks as exemplified in an embodiment of the present invention;
FIG. 3 is a schematic diagram of iterative training of neural radiation field geometry scene setup, as exemplified in an embodiment of the present invention.
Detailed Description
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Referring to fig. 1, a flowchart of a multi-view geometric scene establishment method for a limited space according to an embodiment of the present invention is shown, where the multi-view geometric scene establishment method includes:
Step 101: and acquiring a target image group in a limited space meeting the calculation requirement under multiple view angles through an image acquisition device.
For the establishment of multi-view geometric scenes of a limited space, particularly a high-risk area such as a mine mining working face, firstly, a target image group in multiple views in the limited space meeting calculation requirements needs to be acquired through an image acquisition device. For limited spaces with different sizes, each limited space needs to acquire a plurality of images under multiple visual angles to form a target image group. For example: for the limited space A, the image acquisition equipment is used for acquiring a target image group A 'under the multi-view angle in the limited space A meeting the calculation requirement, and for the limited space B, the image acquisition equipment is used for acquiring a target image group B' under the multi-view angle in the limited space B meeting the calculation requirement, wherein the size of the limited space A and the size of the limited space B can be the same or different.
In one possible embodiment, the image capturing device performs image capturing in a manner including: adopting fixed image acquisition equipment to acquire images in a scene in a limited space; or adopting a slide rail type image acquisition device to acquire images in a scene in a limited space; wherein the image acquisition device is only used for acquiring images in a limited space and is not used for other purposes. The image acquisition apparatus includes: video cameras, digital cameras, terminal shooting devices (e.g., cell phones, tablet computers), and the like.
Step 102: and establishing voxel blocks with different dimensions for the limited spaces with different dimensions, so as to divide the limited spaces with different dimensions into different layers, thereby facilitating the fine establishment of the subsequent geometric scene.
In the actual working process, the fact that the sizes of different limited spaces are the same or different is considered, so that voxel blocks with different sizes are required to be established for the limited spaces with different sizes, different layers of segmentation is carried out on the limited spaces with different sizes, and the follow-up fine establishment of the geometric scene is facilitated. In one possible embodiment, a method of creating a voxel block of different scale sizes for a limited space of different sizes comprises:
Firstly, determining the target scale size of a voxel block corresponding to a target limited space according to the size of the target limited space; and then dividing the target limited space into a plurality of voxel blocks with different numbers according to the size of the target scale, so as to divide the target limited space into different layers and facilitate the fine establishment of the subsequent geometric scene, wherein the dividing method comprises the following steps: rectangular segmentation, hexagonal segmentation, spherical segmentation, etc.
For example: the size of the limited space A is smaller than that of the limited space B, and the target scale size of the voxel block corresponding to the limited space A is determined to be 3, so that the limited space A can be divided into 27 voxel blocks; the target scale size of the voxel block corresponding to the limited space A is determined to be 4, and the limited space B can be divided into 64 voxel blocks. If the B-constrained space is segmented into 27 voxel blocks, the subsequent geometric scene may be less refined, and the iterative training may be slower due to more pruning and attenuation operations in the iterative process. It is therefore necessary to build voxel blocks of different scale sizes from limited spaces of different sizes.
Step 103: and integrating the spatial position information and the pose information of the image acquisition equipment when the target image group is acquired with the attribute information of the target image group to perform multi-scale coding, wherein the multi-scale coding is in one-to-one correspondence with a plurality of voxel blocks segmented under the same scale.
After obtaining a target image group corresponding to a target limited space and dividing voxel blocks of the target limited space, the invention creatively combines the related information of the image acquisition equipment with the attribute information of the target image group to carry out multi-scale coding, namely: and integrating the spatial position information and the pose information of the image acquisition equipment when the target image group is acquired with the attribute information of the target image group to perform multi-scale coding. The multi-scale coding is in one-to-one correspondence with the plurality of voxel blocks segmented under the same scale.
A preferred way is: under the same scale size, establishing a one-to-one correspondence between each voxel block and multi-scale codes according to the spatial position of each voxel block segmented under the scale size; the encoding mode for performing multi-scale encoding comprises the following steps: bit Width per Bit coding, hash coding, and CRC coding.
Taking Bit Width as an example: respectively carrying out 8bit 0/1 coding on spatial position information and pose information when the image in the current voxel block is acquired, wherein the spatial position information comprises the spatial position when the image acquisition device acquires the image, and the pose information comprises a pitch angle, a roll angle and a course angle when the image acquisition device acquires the image; the split bit coding of the attribute information of the image group, for example, the RGB information of the image is composed of 3 groups of numerical matrixes of 0-255, each voxel block can be composed of 9 bit codes of 3 groups of 3 bit 0/1 codes, and then the three codes are spliced with the codes of the position and posture information according to the bits to form the multi-scale code.
After multi-scale coding is established, the size of the size is taken as an interval according to the size of the voxel block segmented by the current space, and the vertex coordinates of the voxel block in the space correspond to the block area where the multi-scale coding is located, and the two are combined.
Step 104: and constructing a plurality of voxel blocks and multi-scale codes corresponding to the limited space into high-order feature vectors, inputting the high-order feature vectors into a nerve radiation field to integrate color values and voxel densities, and performing loop iteration to establish a geometric scene.
Considering the characteristics of the nerve radiation field, the input end needs to input the feature vector when the nerve radiation field is subjected to iterative training, so that a plurality of voxel blocks corresponding to a limited space and multi-scale codes are constructed into high-order feature vectors.
After the high-order feature vector is constructed, the high-order feature vector is input into a nerve radiation field to integrate the color value and the voxel density, and loop iteration is carried out to establish a geometric scene.
In one possible embodiment, a method of constructing a plurality of voxel blocks and multi-scale codes corresponding to a restricted space into a high-order feature vector includes:
A plurality of voxel blocks and multi-scale codes are arranged and combined to construct a high-order feature vector; wherein, the arrangement combination mode used comprises; the vertex information and the multi-scale codes of each voxel block are selected according to a bit arrangement mode.
Taking the example of constructing a high-order feature vector by a full connection layer: and linearly combining the encoded permutation and combination with the weights, introducing nonlinear transformation through an activation function, thereby extracting high-level abstract features and outputting a higher-level feature vector.
Step 105: in the iteration process, redundancy is removed from a plurality of voxel blocks corresponding to the limited space by utilizing pruning operation, so that the iteration speed of the nerve radiation field is improved.
Considering that there may be a large number of voxel blocks in the restricted space that do not contain or contain only a very small amount of spatial information, such voxel blocks are referred to as empty voxel blocks, a pruning operation may be used to de-redundant the multi-scale voxel blocks in the restricted space, so that the following neural radiation field training emphasis is placed on the gray voxel blocks with more spatial information, thereby greatly improving the iteration speed of the neural radiation field. In an iterative process, the spatial information contained within each voxel block of the plurality of voxel blocks is quantized and ordered according to the coding information contained therein to determine empty voxel blocks.
Removing redundancy for a plurality of voxel blocks corresponding to the limited space by utilizing pruning operation, namely: according to the actual refinement requirement of the scene, pruning operation is carried out on voxel blocks (namely empty voxel blocks) with preset proportion in a plurality of voxel blocks, and the voxel density of the pruned voxel blocks is set to 0, namely the opacity of the pruned voxel blocks is 0, so that the voxel blocks do not participate in subsequent iterative training, the aim of improving the iterative training speed of the nerve radiation field is fulfilled, and the iterative speed of the nerve radiation field is improved.
Step 106: in the iterative process, the sizes of a plurality of voxel blocks corresponding to the limited space are continuously attenuated, and the multi-view geometric scene establishment method is more accurate and complete in the scene establishment of the detail area in the scene along with the continuous decline of the iteration loss value.
In the iteration process, not only pruning operation is needed, but also the sizes of a plurality of voxel blocks corresponding to the limited space are required to be continuously attenuated, and the multi-view geometric scene establishment method is more accurate and complete in scene establishment of detail areas in the scene along with continuous decline of iteration loss values. Specific:
In the iterative process, after the minimum loss value under a certain iteration number or the current scale size is reached, the current space scale size of a plurality of voxel blocks is attenuated downwards, so that the space of the current established geometric scene is continuously and finely trained by the voxel blocks, the multi-view geometric scene establishment method shows a better modeling effect in a detail area in the scene, and the scene establishment of the detail area is more accurate and complete. It should be noted that, when the sizes of the plurality of voxel blocks are continuously attenuated, it is preferable to keep the sampling rate of each voxel block unchanged. For example: and (3) dividing the limited space A into 27 voxel blocks in the first iteration, and then dividing each voxel block left after pruning operation into 27 small voxel blocks in a certain iteration number to achieve a better iteration training effect.
For a better explanation and illustration of pruning and attenuation operations, reference is made to the schematic diagram of pruning and attenuation methods for a plurality of voxel blocks illustrated in the embodiment of the invention shown in fig. 2, where the pruning and attenuation operations for an empty voxel block are illustrated with the letter P (gray voxel block in fig. 2).
Since there may be a large number of empty voxel blocks in the restricted space that do not contain any spatial information or contain only a very small amount of spatial information, such as the white voxel blocks in fig. 2. Therefore, the pruning operation can be used for removing redundancy from a plurality of voxel blocks in the limited space, so that the training focus of the nerve radiation field is placed on gray voxel blocks with more space information, and the iteration speed of the nerve radiation field is greatly improved.
The leftmost graph in fig. 2 shows a plurality of voxel blocks corresponding to the high-order eigenvectors when the nerve radiation field is first input, and the graph adjacent to the voxel blocks shows a schematic result obtained by performing redundancy elimination on the plurality of voxel blocks by using a pruning operation. The middle graph shows the result schematic diagram after a certain number of iterations, re-attenuating each voxel block remaining after pruning operation into smaller voxel blocks, and further determining empty voxel blocks.
The right hand graph adjacent to the middle graph shows a schematic diagram of the result of re-redundancy elimination of the plurality of voxel blocks after further determination of empty voxel blocks, again using pruning operations. The rightmost graph represents a schematic diagram of successful establishment of the multi-view geometric scene after the iteration is completed.
Step 107: and (3) establishing the required accuracy according to the actual geometric scene, judging the established geometric model, stopping iteration if the preset condition is met, and otherwise, continuing the iteration operation until the preset condition is met, and stopping iteration.
In the process of the iteration process, the required accuracy is required to be established according to the actual geometric scene, the established geometric model is judged, if the preset condition is met, iteration is stopped, otherwise, the iteration operation is continued until the preset condition is met, and iteration is stopped. In the embodiment of the present invention, the preset condition refers to:
The geometric scene is established with required accuracy and required training time, wherein the accuracy refers to spatial resolution required by scene voxel segmentation, the training time refers to maximum training time acceptable by actual requirements, and when the two conditions are simultaneously met in the training process, iteration is stopped.
The above process may be better understood in connection with creating an iterative training diagram of a neural radiation field geometry scenario as exemplified in the embodiment of the present invention shown in fig. 3. A specific geometric scene creation iteration procedure is presented in fig. 3, the leftmost example showing a certain restricted space being segmented into 27 voxel blocks. When the initial scale voxel block is subjected to voxel block position coding and voxel block information coding (namely multi-scale coding), a nerve radiation field is input.
After the neural radiation field training iteration generates a loss function (the loss function is represented by a neural radiation field block diagram and a block diagram between a pruning operation and a scale attenuation block diagram in fig. 3), the set fineness index is used for judging whether the pruning operation and the scale attenuation are needed. In the iterative process, the sampling rate of each voxel block is kept unchanged, the sizes of a plurality of voxel blocks are continuously attenuated (the sizes of 27 voxel blocks are represented by a plurality of voxel blocks below in an exemplary way in fig. 3) and the iteration loss value is continuously reduced, so that the multi-view geometric scene establishment method facing the limited space can enable the scene establishment of a detail area in the scene to be more accurate and complete until the geometric scene is finally established, namely the multi-view geometric scene is established.
Based on the above method for establishing the multi-view geometric scene facing the limited space, the embodiment of the invention also provides a device for establishing the multi-view geometric scene facing the limited space, which comprises the following steps:
the acquisition module is used for acquiring a target image group in a limited space meeting the calculation requirement under multiple view angles through the image acquisition equipment;
The voxel block module is used for establishing voxel blocks with different dimensions for the limited spaces with different dimensions, so that the limited spaces with different dimensions are divided into different layers, and the subsequent fine establishment of the geometric scene is facilitated;
The multi-scale coding module is used for integrating the spatial position information and the pose information of the image acquisition equipment when the target image group is acquired with the attribute information of the target image group to carry out multi-scale coding, wherein the multi-scale coding is in one-to-one correspondence with a plurality of voxel blocks segmented under the same scale;
The construction feature vector module is used for constructing a plurality of voxel blocks and multi-scale codes corresponding to the limited space into high-order feature vectors, inputting the high-order feature vectors into a nerve radiation field for integrating color values and voxel densities, and performing loop iteration to establish a geometric scene;
the redundancy elimination module is used for eliminating redundancy of the voxel blocks corresponding to the limited space by utilizing pruning operation so as to improve the iteration speed of the nerve radiation field;
the iteration attenuation module is used for continuously attenuating the sizes of a plurality of voxel blocks corresponding to the limited space in the iteration process, and the multi-view geometric scene establishment method is more accurate and complete in the scene establishment of the detail area in the scene along with the continuous decline of the iteration loss value;
And the judging module is used for judging the established geometric model according to the required accuracy of the actual geometric scene establishment, stopping iteration if the preset condition is met, and continuing the iteration operation until the preset condition is met, and stopping iteration.
Optionally, the voxel block building module is specifically configured to:
determining the target scale size of a voxel block corresponding to a target limited space according to the size of the target limited space;
Dividing the target limited space into a plurality of voxel blocks with different numbers according to the target scale, so as to divide the target limited space into different layers and facilitate the fine establishment of a subsequent geometric scene, wherein the dividing method comprises the following steps: rectangular segmentation, hexagonal segmentation, and spherical segmentation.
Optionally, the method for realizing that the multi-scale coding and the plurality of voxel blocks divided under the same scale size are in one-to-one correspondence by the multi-scale coding module comprises the following steps:
Under the same scale, establishing a corresponding relation between each voxel block and the multi-scale code one by one according to the spatial position of each voxel block segmented under the scale;
the encoding mode for performing the multi-scale encoding comprises the following steps: bloom Filter coding, hash coding, and CRC coding.
Optionally, the building feature vector module is specifically configured to:
Arranging and combining the plurality of voxel blocks and the multi-scale codes to construct the high-order feature vector;
wherein, the arrangement and combination modes comprise; and selecting the vertex information of each voxel block and the multi-scale coding bit-by-bit arrangement mode.
Optionally, the redundancy elimination module is specifically configured to:
Carrying out quantization sequencing on spatial information contained in each voxel block according to coding information contained in each voxel block in the plurality of voxel blocks to determine empty voxel blocks, wherein the empty voxel blocks are voxel blocks which do not contain or only contain a very small amount of spatial information and exist in the limited space;
And pruning operation is carried out on voxel blocks with preset proportion in the plurality of voxel blocks according to the actual refinement requirement of the scene, and the voxel density of the pruned voxel blocks is set to 0, namely the opacity of the pruned voxel blocks is 0, so that the voxel blocks do not participate in subsequent iterative training, and the aim of improving the iterative training speed of the nerve radiation field is fulfilled.
Optionally, the iterative attenuation module is specifically configured to:
In the iterative process, after a certain number of iterations or the minimum loss value under the current scale is reached, the current spatial scale of the voxel blocks is attenuated downwards, so that the space of the current established geometric scene is continuously refined, the voxel blocks are iteratively trained, the multi-view geometric scene establishment method shows a better modeling effect in a detail area in the scene, and the scene establishment of the detail area is more accurate and complete.
The method for establishing the multi-view geometric scene facing the limited space establishes voxel blocks on the multi-scale space after the image group in the limited space under the multi-view is acquired, encodes various information of the acquired image, inputs the two information into a nerve radiation field, performs pruning and scale attenuation operation and loop iteration, and finally establishes the geometric scene.
In summary, the method for establishing the multi-view geometric scene oriented to the limited space is characterized in that the limited space is divided into multi-scale voxel blocks according to different scene requirements, and high-order vectors are synthesized by encoding multi-scale image information. The multi-scale segmentation can enable the geometric scene to realize hierarchical training in the recovery process, and further enable the geometric scene to realize fine modeling in the iterative training of the nerve radiation field. And simultaneously, pruning operation based on spatial information redundancy is performed on the multi-scale voxel blocks, and the voxel density of the empty voxel blocks containing less spatial information is set to 0, so that redundancy in the training process of the nerve radiation field is reduced, and further, the training of the nerve radiation field is greatly accelerated. In the training iteration process, the sampling rate of each voxel block is kept unchanged, and the size of the multi-scale voxel block is continuously attenuated. And (3) along with the continuous decline of the iteration loss value, establishing the required accuracy according to the actual geometric scene, judging the established geometric model, if the requirement is met, stopping iteration, otherwise, continuing to input the nerve radiation field to perform pruning and attenuation operation until the establishment of the geometric scene is completed.
The multi-view geometric scene establishment method provided by the invention greatly improves the training speed of the nerve radiation field, has higher scene modeling fineness, simultaneously has higher multi-view geometric scene establishment speed and lower requirements on hardware computing equipment, meets the requirement of current-stage social production on modeling instantaneity, and is better applied to daily life of people. The modeling effect of the multi-view geometric scene of the limited space is good, the actual requirements of the scenes such as industrial production are completely met, and good technical support is provided for modeling of industrial production environments and equipment such as mine mining working faces in the limited space.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiment and all such alterations and modifications as fall within the scope of the embodiments of the invention.
Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or terminal device that comprises the element.
The embodiments of the present invention have been described above with reference to the accompanying drawings, but the present invention is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and many forms may be made by those having ordinary skill in the art without departing from the spirit of the present invention and the scope of the claims, which are to be protected by the present invention.
Claims (10)
1. The method for establishing the multi-view geometric scene facing the limited space is characterized by comprising the following steps of:
acquiring a target image group in a limited space meeting calculation requirements under multiple view angles through an image acquisition device;
The method comprises the steps of establishing voxel blocks with different dimensions for limited spaces with different dimensions, so as to divide the limited spaces with different dimensions into different layers, and facilitate the subsequent fine establishment of geometric scenes;
Integrating the spatial position information and the pose information of the image acquisition equipment when the target image group is acquired with the attribute information of the target image group to carry out multi-scale coding, wherein the multi-scale coding is in one-to-one correspondence with a plurality of voxel blocks segmented under the same scale size;
constructing a plurality of voxel blocks and multi-scale codes corresponding to the limited space into high-order feature vectors, inputting the high-order feature vectors into a nerve radiation field to integrate color values and voxel densities, and performing loop iteration to establish a geometric scene;
in the iteration process, removing redundancy from a plurality of voxel blocks corresponding to the limited space by utilizing pruning operation, so that the iteration speed of the nerve radiation field is improved;
In the iteration process, the sizes of a plurality of voxel blocks corresponding to the limited space are continuously attenuated, and the iteration loss value is continuously reduced, so that the multi-view geometric scene establishment method is more accurate and complete in the scene establishment of a detail area in the scene;
and establishing the required accuracy according to the actual geometric scene, judging the established geometric model, stopping iteration if the preset condition is met, and otherwise, continuing iteration operation until the preset condition is met, and stopping iteration.
2. The method for establishing a multi-view geometric scene according to claim 1, wherein the image acquisition device performs image acquisition in a manner comprising: adopting fixed image acquisition equipment to acquire images in a scene; or alternatively
Adopting a sliding rail type image acquisition device to acquire images in a scene;
wherein the image acquisition device is only used for acquiring images in the limited space, and the image acquisition device comprises: video camera, digital camera and terminal shooting equipment.
3. The method of claim 1, wherein creating voxel blocks of different scale sizes for different sizes of constrained space comprises:
determining the target scale size of a voxel block corresponding to a target limited space according to the size of the target limited space;
Dividing the target limited space into a plurality of voxel blocks with different numbers according to the target scale, so as to divide the target limited space into different layers and facilitate the fine establishment of a subsequent geometric scene, wherein the dividing method comprises the following steps: rectangular segmentation, hexagonal segmentation, and spherical segmentation.
4. The method for creating the multi-view geometric scene as defined in claim 1, wherein the multi-scale encoding has a one-to-one correspondence with the plurality of voxel blocks divided by the same scale, and the method comprises:
Under the same scale, establishing a corresponding relation between each voxel block and the multi-scale code one by one according to the spatial position of each voxel block segmented under the scale;
the encoding mode for performing the multi-scale encoding comprises the following steps: bloom Filter coding, hash coding, and CRC coding.
5. The multi-view geometry scene creation method according to claim 1, wherein constructing the plurality of voxel blocks and multi-scale codes corresponding to the constrained space into high-order feature vectors comprises:
Arranging and combining the plurality of voxel blocks and the multi-scale codes to construct the high-order feature vector;
wherein, the arrangement and combination modes comprise; and selecting the vertex information of each voxel block and the multi-scale coding bit-by-bit arrangement mode.
6. The multi-view geometry scene creation method according to claim 1, wherein the removing redundancy of the plurality of voxel blocks corresponding to the limited space by pruning operation comprises:
Carrying out quantization sequencing on spatial information contained in each voxel block according to coding information contained in each voxel block in the plurality of voxel blocks to determine empty voxel blocks, wherein the empty voxel blocks are voxel blocks which do not contain or only contain a very small amount of spatial information and exist in the limited space;
And pruning operation is carried out on voxel blocks with preset proportion in the plurality of voxel blocks according to the actual refinement requirement of the scene, and the voxel density of the pruned voxel blocks is set to 0, namely the opacity of the pruned voxel blocks is 0, so that the voxel blocks do not participate in subsequent iterative training, and the aim of improving the iterative training speed of the nerve radiation field is fulfilled.
7. The method according to claim 1, wherein the size of the voxel blocks corresponding to the limited space is continuously attenuated in the iterative process, and the multi-view geometric scene establishment method is more accurate and complete in the scene establishment of the detail area in the scene as the iteration loss value continuously decreases, and comprises:
In the iterative process, after a certain number of iterations or the minimum loss value under the current scale is reached, the current spatial scale of the voxel blocks is attenuated downwards, so that the space of the current established geometric scene is continuously refined, the voxel blocks are iteratively trained, the multi-view geometric scene establishment method shows a better modeling effect in a detail area in the scene, and the scene establishment of the detail area is more accurate and complete.
8. A multi-view geometric scene establishment device facing a limited space, characterized in that the multi-view geometric scene establishment device comprises:
the acquisition module is used for acquiring a target image group in a limited space meeting the calculation requirement under multiple view angles through the image acquisition equipment;
The voxel block module is used for establishing voxel blocks with different dimensions for the limited spaces with different dimensions, so that the limited spaces with different dimensions are divided into different layers, and the subsequent fine establishment of the geometric scene is facilitated;
The multi-scale coding module is used for integrating the spatial position information and the pose information of the image acquisition equipment when the target image group is acquired with the attribute information of the target image group to carry out multi-scale coding, wherein the multi-scale coding is in one-to-one correspondence with a plurality of voxel blocks segmented under the same scale;
The construction feature vector module is used for constructing a plurality of voxel blocks and multi-scale codes corresponding to the limited space into high-order feature vectors, inputting the high-order feature vectors into a nerve radiation field for integrating color values and voxel densities, and performing loop iteration to establish a geometric scene;
the redundancy elimination module is used for eliminating redundancy of the voxel blocks corresponding to the limited space by utilizing pruning operation so as to improve the iteration speed of the nerve radiation field;
the iteration attenuation module is used for continuously attenuating the sizes of a plurality of voxel blocks corresponding to the limited space in the iteration process, and the multi-view geometric scene establishment method is more accurate and complete in the scene establishment of the detail area in the scene along with the continuous decline of the iteration loss value;
And the judging module is used for judging the established geometric model according to the required accuracy of the actual geometric scene establishment, stopping iteration if the preset condition is met, and continuing the iteration operation until the preset condition is met, and stopping iteration.
9. The multi-view geometry scene creation device of claim 8, wherein the creation voxel block module is specifically configured to:
determining the target scale size of a voxel block corresponding to a target limited space according to the size of the target limited space;
Dividing the target limited space into a plurality of voxel blocks with different numbers according to the target scale, so as to divide the target limited space into different layers and facilitate the fine establishment of a subsequent geometric scene, wherein the dividing method comprises the following steps: rectangular segmentation, hexagonal segmentation, and spherical segmentation.
10. The multi-view geometric scene building device according to claim 8, wherein the multi-scale coding module is configured to implement a method for multi-scale coding with a same scale size in a one-to-one correspondence with a plurality of voxel blocks divided by the scale size, the method comprising:
Under the same scale, establishing a corresponding relation between each voxel block and the multi-scale code one by one according to the spatial position of each voxel block segmented under the scale;
the encoding mode for performing the multi-scale encoding comprises the following steps: bloom Filter coding, hash coding, and CRC coding.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311058030.3A CN117036639B (en) | 2023-08-21 | 2023-08-21 | Multi-view geometric scene establishment method and device oriented to limited space |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311058030.3A CN117036639B (en) | 2023-08-21 | 2023-08-21 | Multi-view geometric scene establishment method and device oriented to limited space |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117036639A CN117036639A (en) | 2023-11-10 |
| CN117036639B true CN117036639B (en) | 2024-04-30 |
Family
ID=88640976
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311058030.3A Active CN117036639B (en) | 2023-08-21 | 2023-08-21 | Multi-view geometric scene establishment method and device oriented to limited space |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117036639B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110458939A (en) * | 2019-07-24 | 2019-11-15 | 大连理工大学 | The indoor scene modeling method generated based on visual angle |
| CN114004941A (en) * | 2022-01-04 | 2022-02-01 | 苏州浪潮智能科技有限公司 | Indoor scene three-dimensional reconstruction system and method based on nerve radiation field |
| CN115359170A (en) * | 2022-10-19 | 2022-11-18 | 北京百度网讯科技有限公司 | Scene data generation method and device, electronic equipment and storage medium |
| CN115619951A (en) * | 2022-10-16 | 2023-01-17 | 浙江大学 | Dense synchronous positioning and mapping method based on voxel neural implicit surface |
| WO2023093186A1 (en) * | 2022-06-15 | 2023-06-01 | 之江实验室 | Neural radiation field-based method and apparatus for constructing pedestrian re-identification three-dimensional data set |
| CN116310095A (en) * | 2023-02-24 | 2023-06-23 | 电子科技大学 | Multi-view three-dimensional reconstruction method based on deep learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220335636A1 (en) * | 2021-04-15 | 2022-10-20 | Adobe Inc. | Scene reconstruction using geometry and reflectance volume representation of scene |
-
2023
- 2023-08-21 CN CN202311058030.3A patent/CN117036639B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110458939A (en) * | 2019-07-24 | 2019-11-15 | 大连理工大学 | The indoor scene modeling method generated based on visual angle |
| CN114004941A (en) * | 2022-01-04 | 2022-02-01 | 苏州浪潮智能科技有限公司 | Indoor scene three-dimensional reconstruction system and method based on nerve radiation field |
| WO2023093186A1 (en) * | 2022-06-15 | 2023-06-01 | 之江实验室 | Neural radiation field-based method and apparatus for constructing pedestrian re-identification three-dimensional data set |
| CN115619951A (en) * | 2022-10-16 | 2023-01-17 | 浙江大学 | Dense synchronous positioning and mapping method based on voxel neural implicit surface |
| CN115359170A (en) * | 2022-10-19 | 2022-11-18 | 北京百度网讯科技有限公司 | Scene data generation method and device, electronic equipment and storage medium |
| CN116310095A (en) * | 2023-02-24 | 2023-06-23 | 电子科技大学 | Multi-view three-dimensional reconstruction method based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| BungeeNeRF: Progressive Neural Radiance Field for Extreme Multi-scale Scene Rendering;Yuanbo Xiangli;Computer Vision – ECCV 2022;20221122;全文 * |
| 基于深度学习的视觉同时定位与建图研究进展;张耀;仪器仪表学报;20230731;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117036639A (en) | 2023-11-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111369681B (en) | Three-dimensional model reconstruction method, device, equipment and storage medium | |
| CN110570371B (en) | Image defogging method based on multi-scale residual error learning | |
| CN110503703B (en) | Method and apparatus for generating image | |
| Yuan et al. | Plug-and-play algorithms for video snapshot compressive imaging | |
| CN111325851B (en) | Image processing method and device, electronic equipment and computer readable storage medium | |
| CN112001914A (en) | Depth image completion method and device | |
| CN108898630A (en) | A kind of three-dimensional rebuilding method, device, equipment and storage medium | |
| CN115082639A (en) | Image generation method and device, electronic equipment and storage medium | |
| CN115690382B (en) | Training method of deep learning model, and method and device for generating panorama | |
| CA3137297C (en) | Adaptive convolutions in neural networks | |
| CN112419477B (en) | Face image style conversion method and device, storage medium and electronic equipment | |
| CN112381813B (en) | Panoramic view visual saliency detection method based on graph convolution neural network | |
| CN111833360B (en) | Image processing method, device, equipment and computer readable storage medium | |
| CN116958492B (en) | VR editing method for reconstructing three-dimensional base scene rendering based on NeRf | |
| CN116778045A (en) | Digital human generation method, system and device for nerve radiation field | |
| CN116630514A (en) | Image processing method, device, computer readable storage medium and electronic equipment | |
| CN110163095B (en) | Loop detection method, loop detection device and terminal equipment | |
| CN117094895B (en) | Image panorama stitching method and system | |
| Wang et al. | Unsupervised deep exemplar colorization via pyramid dual non-local attention | |
| CN117036639B (en) | Multi-view geometric scene establishment method and device oriented to limited space | |
| CN115908712A (en) | Three-dimensional reconstruction and model training method and equipment based on image and storage medium | |
| CN115512039A (en) | 3D face construction method | |
| US20200175746A1 (en) | Volumetric slicer | |
| CN110992474A (en) | Method for realizing time domain technology | |
| CN110910303A (en) | Image style migration method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |