128*256 array single-photon laser radar reconstruction method and system based on multi-domain stable feature fusion

Abstract

The invention discloses a 128 * 256 array single-photon laser radar reconstruction method and system based on multi-domain stable feature fusion, belongs to the technical field of laser imaging radars, and solves the problem that details are fuzzy due to the fact that a current array laser radar reconstruction method usually does not consider the difference between array pixels. And a traditional intensity threshold method has little influence on the non-uniformity of echo distribution, and a traditional range gating method is difficult to process a complex target with relatively large distance gradient change. Comprising the steps of constructing a multi-domain stable feature fusion judgment model, obtaining an initial depth image, pixel distribution and missing pixels of a target, and performing noise suppression on the initial depth image; carrying out iterative supplement on internal noise pixels and missing pixels of the target; and performing intensity image reconstruction by adopting a uniformity correction method. The method is suitable for echo data extraction and target reconstruction under the weak light condition of the laser imaging radar.

Description

Technical Field

[0001] The present invention belongs to the technical field of laser imaging radar, and in particular relates to a method for extracting echo data and reconstructing a target under weak light conditions of a laser imaging radar. Background Art

[0002] Array GM-APD laser imaging radar has single-photon detection capability and can quickly perform real-time imaging. It has been widely used in military and civilian fields.

[0003] However, when the array GM-APD lidar is imaging under weak light conditions, the random distribution of light and the sensitivity between array pixels reduce the correlation and difference between pixels, making it impossible for the same targets to enhance each other, and different target pixels and background pixels cannot be distinguished. Therefore, unstable light intensity information can be obtained, which significantly reduces the detection capability of the array.

[0004] The key to reconstruction under weak light conditions is signal enhancement and the distinction between the signal and the background. Current array GM-APD lidar reconstruction methods such as 3D deconvolution SPIRAL, Cross correlation method, Non-local MSA, etc., usually do not consider the differences between array pixels. The processing of weak signal pixels is to directly perform neighborhood superposition on each pixel of the entire array, resulting in blurred details. For the reconstructed range image, the current commonly used method using the traditional peak intensity threshold segmentation method cannot distinguish between weak trigger pixels and background light noise pixels, and it is difficult to use the traditional range gating method to process complex targets with large range gradient changes. Summary of the invention

[0005] The present invention proposes a 128×256 array single-photon lidar reconstruction method and system based on multi-domain stable feature fusion, which aims to solve the problems that the current array GM-APD lidar reconstruction method usually does not consider the differences between array pixels, and the processing of weak signal pixels is to directly perform neighborhood superposition on each pixel of the entire array, resulting in blurred details, and the traditional intensity threshold method cannot distinguish between weak trigger pixels and background light noise pixels, and the traditional distance gating method has difficulty in processing complex targets with large distance gradient changes.

[0006] The present invention proposes a 128×256 array single-photon lidar reconstruction method based on multi-domain stable feature fusion, comprising:

[0007] S1: construct a multi-domain stable feature fusion judgment model to obtain the initial depth image, pixel distribution and missing pixels of the target, and perform noise suppression on the initial depth image;

[0008] S2: iteratively supplementing the internal noise pixels and missing pixels of the target, wherein the missing pixels include internal missing pixels and edge missing pixels;

[0009] S3: Intensity image reconstruction using uniformity correction method.

[0010] Furthermore, a preferred solution is provided: the steps of constructing the multi-domain stable feature fusion judgment model include:

[0011] Decompose the initial depth image into different time domain layers using the time domain step size;

[0012] Classify and preprocess the pixels of each layer according to their spatial position relationship;

[0013] The time-domain distance features, echo waveform factor, trigger root mean square and other spatiotemporal features, and frequency-domain features such as centroid frequency are fused to obtain a multi-domain stable feature fusion judgment model.

[0014] Furthermore, a preferred solution is provided: the preprocessing includes: utilizing spatial correlation to remove scattered noise pixel interference.

[0015] Furthermore, a preferred solution is provided: S2 uses a multi-scale algorithm and a selective supplementation algorithm based on local features to iteratively supplement the internal noise pixels and missing pixels of the target.

[0016] Furthermore, a preferred solution is provided: S2 includes:

[0017] S2.1: Use a multi-scale algorithm to iteratively supplement the internal noise pixels, partial missing pixels and edge missing pixels of the target, and retain the remaining signal pixels judged as noise areas.

[0018] Furthermore, a preferred solution is provided: S2 further includes:

[0019] S2.2: performing spatial domain division on the internal missing pixels to obtain a plurality of area blocks;

[0020] S2.3: Perform statistics on each area block to obtain the characteristic distribution of the distance value in the area block;

[0021] S2.4: Use the multi-domain stable feature fusion judgment model to judge all pixels. If they are target pixels, they are retained. If they are missing pixels, perform S2.1.

[0022] S2.5: Input the edge neighborhood echo data of the edge missing pixel into the multi-domain stable feature fusion judgment model for judgment. If it is the target pixel, it is extended outward to supplement; otherwise, it remains unchanged.

[0023] Furthermore, a preferred solution is provided: S3 includes:

[0024] S3.1: Collect target pixels and background pixels as prior data;

[0025] S3.2: Count the total number of triggering times of each pixel, and obtain a histogram of the number of triggering times of background pixels and a histogram of the number of triggering times of target pixels;

[0026] S3.3: Select weak triggering pixels, supplement the weak triggering pixels, and obtain a preliminary correction result of the triggering times;

[0027] S3.4: Obtain the corrected relative intensity based on the remaining pixels;

[0028] S3.5: According to the principle of Poisson distribution of photons, the number of photons is estimated using the number of triggers, and an optimization equation is established. The number of photons is estimated using the sparse Poisson intensity reconstruction method, and the spatial gradient constraint regularization is used to solve the optimization equation to obtain the reconstructed intensity image.

[0029] The present invention also proposes a computer device, including a memory and a processor, wherein a computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes the 128×256 array single-photon lidar reconstruction method based on multi-domain stable feature fusion described in any one of the above-mentioned combinations of solutions.

[0030] The present invention also proposes a computer-readable storage medium, which is used to store a computer program, and the computer program executes the steps of the 128×256 array single-photon lidar reconstruction method based on multi-domain stable feature fusion described in any combination of the above-mentioned solutions.

[0031] The present invention also proposes a 128×256 array single-photon laser radar reconstruction system based on multi-domain stable feature fusion, the system comprising:

[0032] Noise suppression module: used to build a multi-domain stable feature fusion judgment model, obtain the initial depth image, pixel distribution and missing pixels of the target, and perform noise suppression on the initial depth image;

[0033] Iterative supplementation module: used for iteratively supplementing the internal noise pixels and missing pixels of the target, wherein the missing pixels include internal missing pixels and edge missing pixels;

[0034] Image reconstruction module: used to reconstruct intensity images using uniformity correction method.

[0035] Compared with the prior art, the advantages of the present invention are:

[0036] The method proposed in the present invention makes full use of the advantages of random noise distribution and target feature aggregation, proposes a distance layer decomposition method combined with stable feature fusion to suppress noise, and improves the Multi-Scale Algorithm (MSA) to reduce detail erosion.

[0037] In addition, the inhomogeneity between pixels is corrected to obtain clearer intensity information. Under the condition that the average number of photons per pixel is 0.0068 to 0.02 per pixel, this method has better reconstruction effect than other commonly used methods.

[0038] In summary, the method described in the present invention establishes a multi-domain stable feature fusion judgment model within the distance layer, which can effectively distinguish signal trigger pixels and noise pixels; selectively supplements missing information, which can improve the target restoration while maintaining the imaging resolution; and proposes an intensity image reconstruction algorithm based on uniformity correction to improve the quality of intensity image restoration.

[0039] The method can improve the signal-to-noise ratio of the GM-APD laser radar reconstructed image under low-flux photon conditions.

[0040] The invention is used for extracting echo data and reconstructing targets under weak light conditions of Geiger mode-avalanche photodiode (Gm-APD) laser imaging radar, and is specifically used for improving the image signal-to-noise ratio of a reconstructed three-dimensional image while maintaining the resolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 This is a flow chart of a 128×256 array single-photon laser radar reconstruction method based on multi-domain stable feature fusion described in Specific Embodiments 1 and 2 of the present invention;

[0043] Figure 2 This is a flow chart of the distance layer decomposition algorithm described in the first and second specific implementation modes of the present invention;

[0044] Figure 3 This is a flow chart of the improved multi-scale algorithm described in Specific Implementation Methods 1 and 2 of the present invention;

[0045] Figure 4This is a flow chart of a selective supplement algorithm based on local features as described in Specific Embodiments 1 and 2 of the present invention;

[0046] Figure 5 This is a flow chart of the intensity image reconstruction algorithm based on uniformity correction described in the first and second specific embodiments of the present invention;

[0047] Figure 6 The results of reconstruction using different methods described in the second specific implementation mode of the present invention, wherein (A) is a depth image reconstructed without a threshold, (B) is a depth image reconstructed after a threshold, (C) is a detail of a depth image, (D) is a reconstructed intensity image, (a) is a peak method, (b) is a matched filtering method, (c) is a three-dimensional deconvolution method, (d) is a non-local multi-scale method, (e) is a collaborative filtering method, and (f) is a 128×256 array single-photon lidar reconstruction method based on multi-domain stable feature fusion;

[0048] Figure 7 It is a schematic diagram of the intensity image reconstruction result described in the second specific implementation mode of the present invention. DETAILED DESCRIPTION

[0049] In the following description, specific details such as specific system structures and technologies are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0050] It should be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0051] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this application specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise.

[0052] The following is a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0053] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present application. Therefore, the present application is not limited to the specific implementation methods disclosed below.

[0054] Implementation method 1:

[0055] Reference Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 This embodiment will be described.

[0056] A 128×256 array single-photon lidar reconstruction method based on multi-domain stable feature fusion, comprising:

[0057] S1: construct a multi-domain stable feature fusion judgment model to obtain an initial depth image, pixel distribution and missing pixels of the target, and perform noise suppression on the initial depth image; the steps of constructing the multi-domain stable feature fusion judgment model include:

[0058] Decompose the initial depth image into different time domain layers using the time domain step size;

[0059] Classifying and preprocessing the pixels of each layer according to the spatial position relationship, the preprocessing includes: removing the interference of scattered noise pixels by using spatial correlation;

[0060] The time domain distance feature, echo waveform factor, trigger root mean square and other time and space features, and the frequency domain features such as center of gravity frequency are integrated to obtain a multi-domain stable feature fusion judgment model;

[0061] S2: using a multi-scale algorithm and a selective supplementation algorithm based on local features to iteratively supplement the internal noise pixels and missing pixels of the target, wherein the missing pixels include internal missing pixels and edge missing pixels;

[0062] S2.1: Use a multi-scale algorithm to iteratively supplement the internal noise pixels, partial missing pixels and edge missing pixels of the target, and retain the remaining signal pixels judged as noise areas;

[0063] S2.2: performing spatial domain division on the internal missing pixels to obtain a plurality of area blocks;

[0064] S2.3: Perform statistics on each area block to obtain the characteristic distribution of the distance value in the area block;

[0065] S2.4: Use the multi-domain stable feature fusion judgment model to judge all pixels. If they are target pixels, they are retained. If they are missing pixels, perform S2.1.

[0066] S2.5: Input the edge neighborhood echo data of the edge missing pixel into the multi-domain stable feature fusion judgment model for judgment. If it is the target pixel, it is extended outward to supplement; otherwise, it remains unchanged;

[0067] S3: Use uniformity correction method to reconstruct intensity image, including:

[0068] S3.1: Collect target pixels and background pixels as prior data;

[0069] S3.2: Count the total number of triggering times of each pixel, and obtain a histogram of the number of triggering times of background pixels and a histogram of the number of triggering times of target pixels;

[0070] S3.3: Select weak triggering pixels, supplement the weak triggering pixels, and obtain a preliminary correction result of the triggering times;

[0071] S3.4: Obtain the corrected relative intensity based on the remaining pixels;

[0072] S3.5: According to the principle of Poisson distribution of photons, the number of photons is estimated using the number of triggers, and an optimization equation is established. The number of photons is estimated using the sparse Poisson intensity reconstruction method, and the spatial gradient constraint regularization is used to solve the optimization equation to obtain the reconstructed intensity image.

[0073] Implementation method 2:

[0074] Reference Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 This embodiment will be described.

[0075] This embodiment is a further example of the 128×256 array single-photon lidar reconstruction method based on multi-domain stable feature fusion described in embodiment one.

[0076] Step 1: Distance layer decomposition noise suppression.

[0077] Step 1: Use the time domain step size (D step) The initial depth image (Rng int ) is decomposed into different time domain layers, and then the pixels of each layer are classified according to the spatial position relationship. step Usually 50ns is needed to obtain the Layer distribution.

[0078] Step 2: In each layer, the pixels are divided into different labels using spatial correlation, and the clustering characteristics of the labels are used to remove the interference of scattered noise pixels. The preprocessing process is to count the number of pixels in each label, with the peak value being P and the average value being M.

[0079] According to the fact that the noise pixels in the background are dispersed and the target pixels are concentrated, the preliminary judgment criteria are:

[0080] if P>α·M then th=α·M if P<α·M then th=β

[0081] Among them, α is a hyperparameter, β is the minimum object size in the image. Outliers in the label distribution are considered to be target pixels.

[0082] ∑[rng∈label]>th=[label∈target].

[0083] Step 3: Fuse the time-domain distance features, echo waveform factor, trigger root mean square and other time-space features, and frequency-domain features such as centroid frequency to achieve accurate pixel point judgment.

[0084] The time domain characteristic expression is:

[0085]

[0086] D(x) represents the variance of x. When the time domain feature D1 is less than θ1, the distance values ​​are relatively concentrated, and the label is more likely to be the target pixel, F1 = 1; the category with a larger variance has a higher probability of becoming a noise pixel, F1 = 0. When only deep features are used, there is a possibility of misidentifying targets with large gradient changes as noise. According to the characteristics of low sensitivity and high stability of the echo waveform factor, the echo waveform is found from the original data corresponding to the pixel, and W is calculated for each pixel. F feature:

[0087]

[0088]

[0089] W F is the waveform factor, echo is the echo distribution. When D2 is less than θ2, F2=1, otherwise F2=0.

[0090] Calculate the RMS:

[0091]

[0092] When D3 is greater than θ3, F3=1, otherwise F3=0.

[0093] According to the frequency domain distribution of signal and noise, the centroid frequency characteristics of the echo are calculated to describe the frequency components of the signal with larger components in the spectrum.

[0094] D4=∑p·f / ∑p,

[0095] Where p is the power spectrum and f is the frequency. When D4 is less than θ4, F4=1, otherwise F4=0.

[0096] The fusion feature model expression is:

[0097]

[0098] Among them, γ is the weight corresponding to different features.

[0099] When the echo data is sufficient, the frequency domain characteristic F4 feature difference of the center of gravity is large, and the weight adjustment amplitude becomes larger. When the target distance is flat and the echo is sparse, the distance is obvious and the weight adjustment becomes larger.

[0100] Using this feature model, we can get the initial target pixel distribution, initial depth image rng0 and initial missing pixel S0. Multi-feature fusion can reduce false alarms and missed detections, making the target more complete. The selection of these four features comes from laboratory experimental analysis. By performing multi-feature analysis on data with and without signals, we found that the expression of the four features in the signal and background light has stable differences.

[0101] Step 2: Iterative pixel supplementation based on multi-scale algorithm.

[0102] The internal noise pixels, missing parts and edge missing parts of the target are supplemented, and the remaining signal pixels judged as noise areas are retained. Only the missing points and noise points are processed.

[0103] D Missing =MSA(Raw,O),

[0104] D missing is the distance value of the lost pixel, and MSA is the parameter estimation result under scale O. missing The selection of missing points and noise points is determined by local features in the third step.

[0105] Step 3: Selective supplementation algorithm based on local features.

[0106] For internal missing points, the initial missing point distribution S is divided into different area blocks in the spatial domain.

[0107] label←bwlabel(Siter ),

[0108] bwlabel is the calculation function of the spatial connection domain.

[0109] Statistics are taken for each area block to obtain the characteristic distribution H of the distance value within the block.

[0110] H←histogram(rng int ∈label),

[0111] The multi-domain feature model in the first step is used again. The hyperparameters need to be adjusted at this time. The target points are retained after the feature model judgment, and the missing points are returned to the second step and supplemented using multi-scale.

[0112] For edge missing points, the edge neighborhood echo data of the target pixel distribution is input into the multi-domain feature model for secondary judgment. If it is judged as a target, it is extended outward for supplementation. If it is judged as background light, it remains unchanged. A more detailed distinction requires multi-scale supplementary D missing Pixels, detail pixels. Perform secondary screening on the internal parts judged as noise. In the case of cluttered pixel triggering, more pixels that may be targets are retained to ensure the details of the reconstructed image. The surrounding pixels are expanded to supplement the pixels that may be targets and reduce the edge expansion phenomenon caused by multi-scale.

[0113] Step 4: Intensity image reconstruction based on uniformity correction.

[0114] Collect prior data for the detector, correct the number of triggers, and correct the sensitivity non-uniformity.

[0115] Assuming the sky background light is relatively uniform, collect data Raw sky As a priori, correct for array uniformity.

[0116] The data collected by target detection is Raw. Count the total number of triggers of each pixel to get Counts sky and Counts, count Counts sky The histogram of the number of triggers:

[0117] H sky =His(Counts sky ),

[0118] Find the pixel points whose trigger times are less than the threshold and determine them as weak trigger pixels {Pixels Weak}, remove weak trigger pixels.

[0119] According to the central limit theorem, the remaining trigger counts obey the Gaussian distribution, and the trigger expected counts in the Gaussian distribution are obtained. skyThe weighted mean is modulated, and the corrected relative strength is obtained as follows:

[0120]

[0121] Counts new The weak trigger pixel is assigned a value of 0, waiting for the next step of empty point compensation. For the weak trigger pixel, the trigger frequency average of the adjacent pixels is used for preliminary supplementation to obtain the preliminary trigger frequency correction result.

[0122] The photon distribution of the laser can be approximated as a Poisson distribution with a stable optical power mean of S in each interval, expressed as:

[0123]

[0124] The total photon distribution is:

[0125] λ(t)=S(t)+B(t),

[0126] The photon-triggered detector has a binomial distribution with a quantum efficiency of η. Combined with the photon distribution model, when the noise photons are also approximately Poisson distributed, it can be obtained that the photoelectron number f model obeys the Poisson distribution:

[0127]

[0128] According to the principle of Poisson distribution of photons, the number of photons is estimated using the triggering times, an optimization equation is established, and the number of photons is estimated using the sparse Poisson intensity reconstruction method.

[0129]

[0130] The spatial gradient constrained Total Variation (TV) regularization and Fast Iterative Shrinkage / Thresholding Algorithm (FISTA) are used to solve the optimization equation and obtain the reconstructed intensity image.

[0131] The method described in this embodiment establishes a multi-domain stable feature fusion judgment model within the distance layer to effectively distinguish signal trigger pixels and noise pixels; selectively supplements missing information to improve target restoration while maintaining imaging resolution; and proposes an intensity image reconstruction algorithm based on uniformity correction to improve the quality of intensity image restoration.

[0132] The method can improve the signal-to-noise ratio of the GM-APD laser radar reconstructed image under low-flux photon conditions.

[0133] Implementation method three.

[0134] A 128×256 array single-photon laser radar reconstruction system based on multi-domain stable feature fusion, the system is implemented based on the 128×256 array single-photon laser radar reconstruction method based on multi-domain stable feature fusion described in any one of Implementation Methods 1 or 2, and the system includes:

[0135] Noise suppression module: used to build a multi-domain stable feature fusion judgment model, obtain the initial depth image, pixel distribution and missing pixels of the target, and suppress noise on the initial depth image;

[0136] Iterative supplementation module: used for iteratively supplementing the internal noise pixels and missing pixels of the target, wherein the missing pixels include internal missing pixels and edge missing pixels;

[0137] Image reconstruction module: Use uniformity correction method to reconstruct intensity images.

[0138] Those skilled in the art will appreciate that the above are only preferred embodiments of the present invention, and the various embodiments of the present disclosure and / or the features described in the claims may be combined or combined in various ways, even if such combinations or combinations are not explicitly described in the present disclosure. It is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art may still modify the technical solutions described in the aforementioned embodiments, or perform equivalent substitutions on some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

[0139] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0140] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

[0141] It will be appreciated by those skilled in the art that the embodiments of the present disclosure may be provided as methods, systems or computer program products. Therefore, the present disclosure may be provided in the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Furthermore, the present disclosure may be provided in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0142] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0143] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0144] Finally, it should be noted that the above implementation modes are only used to illustrate the technical solutions of the present disclosure rather than to limit its protection scope. Although the present disclosure has been described in detail with reference to the above implementation modes, ordinary technicians in the relevant field should understand that after reading the present disclosure, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation modes of the invention, but these changes, modifications or equivalent substitutions are all within the protection scope of the disclosed claims to be approved.

Claims

1. A 128×256 array single-photon lidar reconstruction method based on multi-domain stable feature fusion, characterized in that: The method comprises: S1: construct a multi-domain stable feature fusion judgment model to obtain the initial depth image, pixel distribution and missing pixels of the target, and perform noise suppression on the initial depth image; S2: iteratively supplementing the internal noise pixels and missing pixels of the target, wherein the missing pixels include internal missing pixels and edge missing pixels; S3: Intensity image reconstruction using uniformity correction method.

2. The 128×256 array single-photon laser radar reconstruction method based on multi-domain stable feature fusion according to claim 1 is characterized in that: The steps of constructing the multi-domain stable feature fusion judgment model include: Decompose the initial depth image into different time domain layers using the time domain step size; Classify and preprocess the pixels of each layer according to their spatial position relationship; The time-domain distance features, echo waveform factor, trigger root mean square and other spatiotemporal features, and frequency-domain features such as center of gravity frequency are fused to obtain a multi-domain stable feature fusion judgment model.

3. The 128×256 array single-photon laser radar reconstruction method based on multi-domain stable feature fusion according to claim 2 is characterized in that: The preprocessing includes: removing scattered noise pixel interference by utilizing spatial correlation.

4. The 128×256 array single-photon laser radar reconstruction method based on multi-domain stable feature fusion according to claim 1 is characterized in that: The S2 uses a multi-scale algorithm and a selective supplementation algorithm based on local features to iteratively supplement the internal noise pixels and missing pixels of the target.

5. The 128×256 array single-photon laser radar reconstruction method based on multi-domain stable feature fusion according to claim 4 is characterized in that: The S2 includes: S2.1: Use a multi-scale algorithm to iteratively supplement the internal noise pixels, partial missing pixels and edge missing pixels of the target, and retain the remaining signal pixels judged as noise areas.

6. The 128×256 array single-photon laser radar reconstruction method based on multi-domain stable feature fusion according to claim 5 is characterized in that: The S2 further includes: S2.2: performing spatial domain division on the internal missing pixels to obtain a plurality of area blocks; S2.3: Perform statistics on each area block to obtain the characteristic distribution of the distance value in the area block; S2.4: Use the multi-domain stable feature fusion judgment model to judge all pixels. If they are target pixels, they are retained. If they are missing pixels, perform S2.

1. S2.5: Input the edge neighborhood echo data of the edge missing pixel into the multi-domain stable feature fusion judgment model for judgment. If it is the target pixel, it is extended outward to supplement; otherwise, it remains unchanged.

7. The 128×256 array single-photon laser radar reconstruction method based on multi-domain stable feature fusion according to claim 1 is characterized in that: The S3 includes: S3.1: Collect target pixels and background pixels as prior data; S3.2: Count the total number of triggering times of each pixel, and obtain a histogram of the number of triggering times of background pixels and a histogram of the number of triggering times of target pixels; S3.3: Select weak triggering pixels, supplement the weak triggering pixels, and obtain a preliminary correction result of the triggering times; S3.4: Obtain the corrected relative intensity based on the remaining pixels; S3.5: According to the principle of Poisson distribution of photons, the number of photons is estimated using the number of triggers, and an optimization equation is established. The number of photons is estimated using the sparse Poisson intensity reconstruction method, and the spatial gradient constraint regularization is used to solve the optimization equation to obtain the reconstructed intensity image.

8. Computer device, characterized in that It includes a memory and a processor, wherein a computer program is stored in the memory, and when the processor runs the computer program stored in the memory, the processor executes the 128×256 array single-photon lidar reconstruction method based on multi-domain stable feature fusion according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store a computer program, which executes the steps of the 128×256 array single-photon lidar reconstruction method based on multi-domain stable feature fusion according to any one of claims 1 to 7.

10. A 128×256 array single-photon lidar reconstruction system based on multi-domain stable feature fusion, characterized in that: The system comprises: Noise suppression module: used to build a multi-domain stable feature fusion judgment model, obtain the initial depth image, pixel distribution and missing pixels of the target, and perform noise suppression on the initial depth image; Iterative supplementation module: used for iteratively supplementing the internal noise pixels and missing pixels of the target, wherein the missing pixels include internal missing pixels and edge missing pixels; Image reconstruction module: used to reconstruct intensity images using uniformity correction method.

Images