A method for multi-view single-frame phase demodulation of a structured light field and related components

Abstract

The present invention discloses a method for multi-view single-frame phase demodulation of a structured light field and related components. The method includes constructing an LFDNet neural network; collecting a data set through a structured light field system and training the LFDNet neural network to optimize the LFDNet neural network; inputting the multi-view fringe image to be predicted into the optimized LFDNet neural network, outputting the numerator term and the denominator term of multiple views, and calculating the numerator term and the denominator term of multiple views through an arctangent function to obtain the wrapped phase of multiple views. The present invention combines deep learning with the characteristics of multi-view imaging of a structured light field system, and uses a neural network to learn the mapping relationship between a multi-view fringe image and a numerator term and a denominator term, and has the advantage of realizing single-frame high-precision phase demodulation of multiple views of a light field through one prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of light field imaging, and in particular to a structured light field multi-view single-frame phase demodulation method and related components. Background Art

[0002] Light Field Imaging can record the intensity information and direction information of light rays simultaneously through single-frame exposure, and has the ability to computationally invert the three-dimensional topography of the scene.

[0003] Passive light field depth estimation technology does not require projection of active illumination. Only a single exposure imaging of the scene by the light field imaging system is needed, and the depth information of the scene can be recovered from the light field data by using an algorithm. It has the advantages of flexibility, high efficiency, and applicability to dynamic scenes, but its disadvantages are low accuracy and poor stability. To address this shortcoming, researchers introduced phase-encoded structured light technology into light field imaging to achieve active light field imaging, which can significantly improve the accuracy of light field depth estimation.

[0004] Active Structured LightField performs phase encoding on space and recovers the phase information modulated by the scene depth from the collected fringe images through calculation. This process requires phase demodulation of the collected fringe images. In the prior art, the Fourier transform method is usually used for single-frame fringe phase demodulation. This method obtains the wrapped phase by performing Fourier transform, window filtering, and inverse Fourier transform on the fringe image, and belongs to the spatial phase demodulation method.

[0005] In addition, researchers have developed a single-frame fringe phase demodulation method based on deep learning for ordinary cameras, but such methods only perform single-frame demodulation for a single view, and for a light field camera with the characteristics of multi-view imaging, the demodulation efficiency of such methods is low. Summary of the Invention

[0006] The purpose of the present invention is to provide a structured light field multi-view single-frame phase demodulation method and related components, aiming to solve the problem of low single-frame demodulation efficiency of the existing single-frame fringe phase demodulation method based on deep learning for multi-view imaging.

[0007] In a first aspect, an embodiment of the present invention provides a structured light field multi-view single-frame phase demodulation method, including:

[0008] Construct an LFDNet neural network;

[0009] Collect a data set through a structured light field system and train the LFDNet neural network to optimize the LFDNet neural network;

[0010] Input the multi - perspective fringe image to be predicted into the optimized LFDNet neural network, output the multi - perspective numerator and denominator terms, and calculate the multi - perspective numerator and denominator terms through the arctangent function to obtain the multi - perspective wrapped phase.

[0011] In a second aspect, an embodiment of the present invention provides a structured light field multi - perspective single - frame phase demodulation device, including:

[0012] A construction unit, configured to construct an LFDNet neural network;

[0013] A training unit, configured to collect a data set through a structured light field system and train the LFDNet neural network to optimize the LFDNet neural network;

[0014] A calculation unit, configured to input the multi - perspective fringe image to be predicted into the optimized LFDNet neural network, output the multi - perspective numerator and denominator terms, and calculate the multi - perspective numerator and denominator terms through the arctangent function to obtain the multi - perspective wrapped phase.

[0015] In a third aspect, an embodiment of the present invention provides a computer device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the structured light field multi - perspective single - frame phase demodulation method described in the first aspect above.

[0016] In a fourth aspect, an embodiment of the present invention provides a computer - readable storage medium, in which the computer - readable storage medium stores a computer program. When the computer program is executed by a processor, the processor is caused to execute the structured light field multi - perspective single - frame phase demodulation method described in the first aspect above.

[0017] An embodiment of the present invention discloses a structured light field multi - perspective single - frame phase demodulation method and related components. The method includes constructing an LFDNet neural network; collecting a data set through a structured light field system and training the LFDNet neural network to optimize the LFDNet neural network; inputting the multi - perspective fringe image to be predicted into the optimized LFDNet neural network, outputting the multi - perspective numerator and denominator terms, and calculating the multi - perspective numerator and denominator terms through the arctangent function to obtain the multi - perspective wrapped phase. The embodiment of the present invention combines deep learning with the characteristics of multi - perspective imaging of a structured light field system, and uses a neural network to learn the mapping relationship from a multi - perspective fringe image to a numerator term and a denominator term, having the advantage of achieving single - frame high - precision phase demodulation of multiple perspectives of a light field through one prediction. Description of the Drawings

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0019] Figure 1 It is a schematic flowchart of the structured light field multi-view single-frame phase demodulation method provided by the embodiments of the present invention;

[0020] Figure 2 It is a schematic sub-flowchart of the structured light field multi-view single-frame phase demodulation method provided by the embodiments of the present invention;

[0021] Figure 3 It is another schematic sub-flowchart of the structured light field multi-view single-frame phase demodulation method provided by the embodiments of the present invention;

[0022] Figure 4 It is another schematic sub-flowchart of the structured light field multi-view single-frame phase demodulation method provided by the embodiments of the present invention;

[0023] Figure 5 It is a schematic block diagram of the LFDNet neural network provided by the embodiments of the present invention;

[0024] Figure 6 It is a schematic block diagram of the structured light field multi-view single-frame phase demodulation device provided by the embodiments of the present invention;

[0025] Figure 7 It is a schematic block diagram of the computer device provided by the embodiments of the present invention. Detailed implementation manners

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0027] It should be understood that when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0028] It should also be understood that the terms used in the specification of the present invention are merely for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0029] It should be further understood that the term "and / or" used in the specification of the present invention and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0030] Please refer to Figure 1 , Figure 1 which is a schematic flow chart of the structured light field multi-view single-frame phase demodulation method provided by the embodiment of the present invention;

[0031] As Figure 1 shown, the method includes steps S101 to S103.

[0032] S101. Construct an LFDNet neural network;

[0033] S102. Collect a data set through a structured light field system and train the LFDNet neural network to optimize the LFDNet neural network;

[0034] In this step, the LFDNet neural network is optimized through the data set to improve the accuracy of network prediction.

[0035] S103. Input the multi-view fringe image to be predicted into the optimized LFDNet neural network, output the numerator and denominator terms of multiple views, and calculate the numerator and denominator terms of multiple views through the arctangent function to obtain the wrapped phase of multiple views.

[0036] Based on the optimized LFDNet neural network, this step calculates the mechanical energy arctangent function of the numerator and denominator terms of the output view, and a high-precision wrapped phase of multiple views can be obtained.

[0037] This embodiment combines deep learning with the characteristics of multi-view imaging of a structured light field system, and uses a neural network to learn the mapping relationship between multi-view fringe images and numerator and denominator terms. It has the advantage of realizing single-frame high-precision phase demodulation of multiple views of the light field through one prediction. At the same time, through experimental comparison, the phase demodulation method provided by the present invention reduces the demodulation error by 70% compared with the Fourier transform method used in the prior art, and the efficiency is increased by V times compared with the single-frame fringe phase demodulation method based on deep learning for ordinary cameras, where V is the number of light field views to be demodulated.

[0038] In one embodiment, step S101 includes:

[0039] S201. Obtain the tensor feature H×W×V of the multi-view fringe image and use it as the input tensor of the LFDNet neural network, where H represents the tensor height, W represents the tensor width, and V represents the number of tensor channels;

[0040] S202. Perform multiple convolutional processing, downsampling processing, and upsampling processing on the input tensor to obtain the output tensor H×W×2V of the LFDNet neural network.

[0041] In this embodiment, the tensor feature H×W×V of the multi-view fringe image obtained in the light field is used as the input, and the output tensor H×W×2V is output, and this output corresponds to the numerator term and the denominator term of each of the V input viewpoints.

[0042] Specifically, as Figure 5 shown, the specific processes of the multiple convolutional processing, downsampling processing, and upsampling processing in step S202 include:

[0043] Perform convolutional processing on the input tensor through the first dense convolutional block 101 to obtain the feature tensor H×W×4V;

[0044] Perform downsampling processing on the feature tensor H×W×4V through the first downsampling block 106 to obtain the feature tensor

[0045] Perform convolutional processing on the feature tensor through the second dense convolutional block 102 to obtain the feature tensor

[0046] Perform downsampling processing on the feature tensor through the second downsampling block 107 to obtain the feature tensor

[0047] Perform convolutional processing on the feature tensor through the third dense convolutional block 103 to obtain the feature tensor

[0048] Perform upsampling processing on the feature tensor through the first upsampling block 108 to obtain the feature tensor

[0049] Perform splicing in the channel dimension on the feature tensor output by the first upsampling block and the feature tensor output by the second dense convolutional block to obtain the first spliced tensor;

[0050] Perform convolutional processing on the first spliced tensor through the fourth dense convolutional block 104 to obtain the feature tensor

[0051] The feature tensor is upsampled by the second upsampling block 109 to obtain a feature tensor of H×W×4V;

[0052] The feature tensor of H×W×4V output by the second upsampling block and the feature tensor of H×W×4V output by the first dense convolution block are concatenated in the channel dimension through the second skip connection 111 to obtain a second concatenated tensor;

[0053] The second concatenated tensor is convolved by the fifth dense convolution block 105 to obtain a feature tensor of H×W×11V;

[0054] The feature tensor of H×W×11V is convolved by the output convolution block 110 to obtain an output tensor of H×W×2V of the LFDNet neural network, where the output tensor corresponds to the numerator and denominator terms of each of the V multi-view fringe images.

[0055] In the LFDNet neural network proposed in this embodiment, the details of the modules used are as follows:

[0056] In the dense convolution block, the input tensor passes through three convolution blocks in sequence, and the input of each convolution block is obtained by concatenating the outputs of all previous layers in the dense convolution block and the input of the dense convolution block.

[0057] In the downsampling block, the input tensor passes through a 1×1 convolution to change the number of feature channels to 4V, and passes through a 3×3 convolution with a stride of 2 to reduce the resolution of the feature tensor by half.

[0058] In the upsampling block, the input tensor passes through a 1×1 convolution to change the number of feature channels to 4V, and passes through a transposed 3×3 convolution with a stride of 2 to double the resolution of the feature tensor.

[0059] In the output convolution block, the input passes through two consecutive 1×1 convolutions and 3×3 convolutions, and then passes through a 3×3 convolution to obtain the output.

[0060] In one embodiment, as Figure 3 shown, in step S102, a data set is collected through the structured light field system, including:

[0061] S301. Measure S different scenes through the structured light field system, and perform 12-step phase-shifted fringe projection on each scene and collect 12-step phase-shifted fringe images;

[0062] In this step, the structured light field system includes a projection optical machine and a light field camera; a single fringe image is projected onto the scene to be measured through the projection optical machine; the deformed fringes in the scene to be measured are collected through the light field camera to obtain 12-step phase-shifted fringe images from multiple perspectives.

[0063] S302. Calculate the numerator term and denominator term for the 12-step phase-shifted fringe images of each perspective in each scene according to the following formula:

[0064]

[0065]

[0066] where Nu represents the numerator term, De represents the denominator term, N is the number of phase-shifting steps, I n is the phase-shifted fringe pattern, and δ n represents the phase-shift amount;

[0067] In this step, substitute N, I n and δ n into the above calculation formula to obtain the numerator term and denominator term of the 12-step phase-shifted fringe images of each perspective.

[0068] S303. Concatenate the numerator term Nu and the denominator term De in the channel dimension to obtain the output tensor H×W×2V, so that the data of each scene can be expressed as {I→Nu,De};

[0069] S304. Based on the measured S different scenes, obtain the data set {I s →Nu s ,De s |s = 1, 2, …, S};

[0070] In this embodiment, a data set of S different scenes is collected through a structured light field system to optimize and train the LFDNet neural network.

[0071] Specific optimization training is as Figure 4 shown and may include:

[0072] S401. Divide the data set into a training set, a validation set, and a test set;

[0073] S402. Use the training set to train the LFDNet neural network, and during the training process, use the validation set for prediction and calculate the prediction result error to optimize the LFDNet neural network;

[0074] S403. Use the test set to predict the optimized LFDNet neural network and calculate the prediction result error to verify the network effect and the accuracy of the phase demodulation method.

[0075] Based on the training of S401 - S403, an LFDNet neural network with high-precision prediction results can be obtained.

[0076] In one embodiment, step S103 includes:

[0077] Calculate the wrapped phase of multiple perspectives according to the following formula:

[0078]

[0079] Where atan() represents the arctangent function, and v represents the number of perspectives.

[0080] In this embodiment, the multi-perspective fringe image to be predicted is input into the optimized and trained LFDNet neural network. The output tensor H×W×2V is obtained through the prediction of the LFDNet neural network. The first V channels in the tensor are the numerator terms {Nu v , v = 1, 2, … V} of V perspectives, and the last V channels in the tensor are the denominator terms {De v , v = 1, 2, … V} of V perspectives. Substitute the obtained numerator term Nu v and the denominator term De v into the above calculation formula and calculate through the arctangent function to obtain the wrapped phase of multiple perspectives.

[0081] An embodiment of the present invention further provides a structured light field multi-perspective single-frame phase demodulation device. This structured light field multi-perspective single-frame phase demodulation device is used to execute any embodiment of the foregoing structured light field multi-perspective single-frame phase demodulation method. Specifically, please refer to Figure 6 , Figure 6 which is a schematic block diagram of the structured light field multi-perspective single-frame phase demodulation device provided by the embodiment of the present invention.

[0082] As Figure 6 shown, the structured light field multi-perspective single-frame phase demodulation device 600 includes: a construction unit 601, a training unit 602, and a calculation unit 603.

[0083] The construction unit 601 is used to construct the LFDNet neural network;

[0084] The training unit 602 is used to collect a data set through the structured light field system and train the LFDNet neural network to optimize the LFDNet neural network;

[0085] The calculation unit 603 is used to input the multi-perspective fringe image to be predicted into the optimized LFDNet neural network, output the numerator and denominator terms of multiple perspectives, and calculate the numerator and denominator terms of multiple perspectives through the arctangent function to obtain the wrapped phase of multiple perspectives.

[0086] This device combines deep learning with the characteristics of multi-view imaging in a structured light field system. It uses a neural network to learn the mapping relationship between multi-view fringe images and the numerator and denominator terms, and has the advantage of achieving high-precision single-frame phase demodulation for multiple views of the light field through a single prediction. At the same time, through experimental comparison, the phase demodulation method provided by the present invention reduces the demodulation error by 70% compared with the Fourier transform method used in the prior art, and improves the efficiency by V times compared with the deep learning-based single-frame fringe phase demodulation method for ordinary cameras, where V is the number of views of the light field to be demodulated.

[0087] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described device and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0088] The above-described structured light field multi-view single-frame phase demodulation device can be implemented in the form of a computer program, and this computer program can run on a computer device as shown in Figure 7 the following.

[0089] Please refer to Figure 7 , Figure 7 which is a schematic block diagram of the computer device provided by an embodiment of the present invention. This computer device 700 is a server, and the server can be an independent server or a server cluster composed of multiple servers.

[0090] Referring to Figure 7 , this computer device 700 includes a processor 702, a memory, and a network interface 705 connected through a system bus 701. Among them, the memory can include a non-volatile storage medium 703 and an internal memory 704.

[0091] The non-volatile storage medium 703 can store an operating system 7031 and a computer program 7032. When this computer program 7032 is executed, it can cause the processor 702 to execute the structured light field multi-view single-frame phase demodulation method.

[0092] The processor 702 is used to provide computing and control capabilities to support the operation of the entire computer device 700.

[0093] The internal memory 704 provides an environment for the operation of the computer program 7032 in the non-volatile storage medium 703. When this computer program 7032 is executed by the processor 702, it can cause the processor 702 to execute the structured light field multi-view single-frame phase demodulation method.

[0094] The network interface 705 is used for network communication, such as providing the transmission of data information, etc. Those skilled in the art can understand that Figure 7The structure shown is only a block diagram of some structures related to the solution of the present invention, and does not constitute a limitation on the computer device 700 to which the solution of the present invention is applied. Specifically, the computer device 700 may include more or fewer components than those shown in the figure, or combine some components, or have a different component arrangement.

[0095] Those skilled in the art can understand that Figure 7 the embodiments of the computer device shown do not constitute a limitation on the specific composition of the computer device. In other embodiments, the computer device may include more or fewer components than those shown in the figure, or combine some components, or have a different component arrangement. For example, in some embodiments, the computer device may only include a memory and a processor. In such an embodiment, the structures and functions of the memory and the processor are the same as those Figure 7 in the shown embodiment and will not be elaborated herein.

[0096] It should be understood that in the embodiments of the present invention, the processor 702 may be a central processing unit (CPU), and the processor 702 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0097] In another embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium may be a non-volatile computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for multi-view single-frame phase demodulation of structured light fields in the embodiments of the present invention is implemented.

[0098] The storage medium is a physical, non-transitory storage medium, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disc, etc., which are all physical storage media that can store program codes.

[0099] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described devices, apparatuses, and units may refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0100] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for multi-view single-frame phase demodulation of structured light fields, characterized in that, it includes: Construct an LFDNet neural network; Collect a data set through a structured light field system and train the LFDNet neural network to optimize the LFDNet neural network; wherein, the data set is: 12-step phase-shifted fringe images obtained by measuring S different scenes through the structured light field system and performing 12-step phase-shifted fringe projection on each scene; Input the multi-view fringe images to be predicted into the optimized LFDNet neural network, output the numerator and denominator terms of multiple views, and calculate the multi-view wrapped phase through the arctangent function for the numerator and denominator terms of the multiple views.

2. The method for multi-view single-frame phase demodulation of structured light fields according to claim 1, characterized in that, The construction of the LFDNet neural network includes: Obtain the tensor feature H×W×V of the multi-view fringe image and use it as the input tensor of the LFDNet neural network, where H represents the tensor height, W represents the tensor width, and V represents the number of tensor channels; Perform multiple convolutional processes, downsampling processes, and upsampling processes on the input tensor to obtain the output tensor H×W×2V of the LFDNet neural network.

3. The method for multi-view single-frame phase demodulation of structured light fields according to claim 2, characterized in that, The performing multiple convolutional processes, downsampling processes, and upsampling processes on the input tensor to obtain the output tensor H×W×2V of the LFDNet neural network includes: Perform convolutional processing on the input tensor through the first dense convolutional block to obtain a feature tensor H×W×4V; The feature tensor H×W×4V is downsampled by the first downsampling block to obtain a feature tensor Perform convolution processing on the feature tensor through the second dense convolution block to obtain a feature tensor Downsample the feature tensor through a second downsampling block to obtain a feature tensor The feature tensor is subjected to convolution processing through a third dense convolution block to obtain a feature tensor Perform upsampling processing on the feature tensor through the first upsampling block to obtain a feature tensor The feature tensor output by the first upsampling block is concatenated with the feature tensor output by the second dense convolution block in the channel dimension to obtain a first concatenated tensor; Performing convolution processing on the first concatenated tensor through a fourth dense convolution block to obtain a feature tensor Perform upsampling processing on the feature tensor through a second upsampling block to obtain a feature tensor of H×W×4V; Splice the feature tensor H×W×4V output by the second upsampling block and the feature tensor H×W×4V output by the first dense convolutional block in the channel dimension through the second skip connection to obtain a second spliced tensor; Perform convolutional processing on the second spliced tensor through the fifth dense convolutional block to obtain a feature tensor H×W×11V; Perform convolutional processing on the feature tensor H×W×11V through the output convolutional block to obtain the output tensor H×W×2V of the LFDNet neural network, where the output tensor corresponds to the numerator and denominator terms of each of the V-view fringe images.

4. The method for multi-view single-frame phase demodulation of structured light fields according to claim 1, characterized in that, The collecting the data set through the structured light field system includes: Measure S different scenes through the structured light field system, perform 12-step phase-shifted fringe projection on each scene, and collect 12-step phase-shifted fringe images; Calculate the 12-step phase-shifted fringe images of each view in each scene according to the following formula: Among them, Nu represents the numerator term, De represents the denominator term, N is the number of phase-shifted steps, and I n is the phase-shifted fringe pattern, and δ n represents the phase-shift amount; Splice the numerator term Nu and the denominator term De in the channel dimension to obtain an output tensor H×W×2V, so that the data of each scene can be expressed as {I→Nu,De}; Based on S different measurement scenarios, a data set {I s →Nu s ,De s | s = 1, 2, …, S} is obtained.

5. The method for multi-view single-frame phase demodulation of structured light fields according to claim 4, characterized in that, The collecting the data set through the structured light field system and training the LFDNet neural network to optimize the LFDNet neural network includes: Divide the said data set into a training set, a validation set, and a test set; Use the training set to train the LFDNet neural network, and during the training process, use the validation set for prediction and calculate the prediction result error to optimize the LFDNet neural network; Use the test set to predict the optimized LFDNet neural network and calculate the prediction result error to verify the network effect and the accuracy of the phase demodulation method.

6. The structured light field multi-view single-frame phase demodulation method according to claim 1, characterized in that, inputting the multi-view fringe image to be predicted into the optimized LFDNet neural network, outputting the multi-view numerator term and denominator term, and calculating the multi-view wrapped phase through the arctangent function for the multi-view numerator term and denominator term, includes: calculating the multi-view wrapped phase according to the following formula: where atan() represents the arctangent function, and v represents the number of views.

7. The structured light field multi-view single-frame phase demodulation method according to claim 4, characterized in that, measuring S different scenes through the structured light field system, and performing 12-step phase-shifted fringe projection on each scene and collecting 12-step phase-shifted fringe images, includes: projecting a single fringe image onto the scene to be measured through the projection optical machine in the structured light field system; collecting the deformed fringes in the scene to be measured through the light field camera in the structured light field system and obtaining 12-step phase-shifted fringe images of multiple views.

8. A structured light field multi-view single-frame phase demodulation device, characterized in that, comprises: a construction unit for constructing the LFDNet neural network; a training unit for collecting a data set through the structured light field system and training the LFDNet neural network to optimize the LFDNet neural network; wherein, the data set is: 12-step phase-shifted fringe images obtained by measuring S different scenes through the structured light field system and performing 12-step phase-shifted fringe projection on each scene; a calculation unit for inputting the multi-view fringe image to be predicted into the optimized LFDNet neural network, outputting the multi-view numerator term and denominator term, and calculating the multi-view wrapped phase through the arctangent function for the multi-view numerator term and denominator term.

9. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the structured light field multi-view single-frame phase demodulation method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor is caused to execute the structured light field multi-view single-frame phase demodulation method according to any one of claims 1 to 7.

Images