Single-photon laser radar fog-penetrating three-dimensional imaging method cooperating with photon processing

Abstract

The invention discloses a single-photon laser radar fog-penetrating three-dimensional imaging method cooperating with photon processing, and relates to the field of single-photon laser radar three-dimensional imaging. The problem of how to reconstruct a high-quality target three-dimensional image by using a small number of statistical frames under the interference of smoke and strong noise in the prior art is solved. The method comprises the following steps: completing data preprocessing through a pixel-by-pixel signal accumulation compensation and global noise suppression method, and generating a guide image by adopting a multi-scale super-pixel three-dimensional imaging algorithm; performing block processing on the de-noised data, and realizing accurate extraction of signals through space-time-frequency collaborative photon processing by using a depth image guided photon processing algorithm and a pure noise band noise isolation method of multi-dimensional Fourier transform; and performing multi-scale partitioning processing on the de-noised data to obtain depth and intensity images reconstructed under different scales, fusing the obtained images, and outputting a three-dimensional image reconstruction result of smoke interference suppression. The method is suitable for the field of single-photon laser radar three-dimensional imaging under cloud, fog and other environmental conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of single-photon lidar three-dimensional imaging, and particularly relates to a method for single-photon lidar fog-penetrating three-dimensional imaging with collaborative photon processing. Background Art

[0002] Chinese Patent Document No. CN113406594A discloses a single-photon laser fog-penetrating method based on a dual-parameter estimation method. This method estimates the echo signal at the signal level, mainly estimating two parameters of the smoke distribution model (Gamma distribution) to complete the estimation of the smoke signal, and finally realizing the three-dimensional imaging of the single-photon lidar through smoke. This technology only estimates the echo data at the signal level and requires a large number of statistical frames of data. It takes nearly 20,000 frames of data to complete the signal estimation.

[0003] Chinese Patent Document No. CN115097484A discloses a single-photon lidar fog-penetrating imaging method based on dual-Gamma estimation. This method first performs the first Gamma estimation for the stacking effect caused by strong noise interference to complete stacking compensation and signal correction. Secondly, the Gamma estimation method is used to estimate the smoke noise, and finally, noise suppression and three-dimensional imaging are completed. This technology first estimates the echo data at the signal level, and then combines the image space information for noise suppression, image compensation and optimization. However, this technology still requires a large number of statistical frames of data. When performing three-dimensional imaging of a 1.4 km target at a visibility of 1.7 km, the required number of data frames is 20,000.

[0004] In summary, although multi-frame statistical data can effectively suppress the interference of the increasing attenuation length on the three-dimensional reconstruction of the single-photon lidar, multi-frame statistical data has problems such as large memory requirements, long data acquisition time, and being unsuitable for dynamic target perception. And the single-frame statistical data and the strong backscattering interference of smoke lead to extremely scarce signal photons.

[0005] Therefore, in the presence of smoke and strong noise interference, how to reconstruct a high-quality three-dimensional image of the target with a small number of statistical frames is still an important scientific problem faced by the current single-photon lidar. Summary of the Invention

[0006] The present invention aims to solve the important scientific problem currently faced by the single-photon lidar: how to reconstruct a high-quality three-dimensional image of the target with a small number of statistical frames in the presence of smoke and strong noise interference.

[0007] To solve the above technical problems, the present invention proposes a three-dimensional imaging method for smoke noise suppression that can accurately estimate the target position and significantly improve the target integrity with a small number of statistical frames and extremely scarce signal photons, and is realized through the following technical solutions:

[0008] Solution 1. The present invention proposes a method for fog-penetrating three-dimensional imaging of a single-photon lidar with collaborative photon processing. The method for fog-penetrating three-dimensional imaging of the single-photon lidar includes the following steps:

[0009] Step 1. Complete the preprocessing of the array GM-APD lidar data through a per-pixel signal stacking compensation and global noise suppression method, and generate a guidance image using a multi-scale superpixel three-dimensional imaging algorithm;

[0010] Step 2. Perform block processing on the preprocessed array GM-APD lidar data output in Step 1, and use a depth image-guided photon processing algorithm and a pure noise band noise isolation method based on multi-dimensional Fourier transform to accurately extract signals through spatio-temporal-frequency collaborative photon processing;

[0011] Step 3. Perform multi-scale block processing on the data output after block processing in Step 2 to obtain reconstructed depth images and intensity images at different scales, fuse the obtained images, and output the three-dimensional image reconstruction result with smoke interference suppression.

[0012] Furthermore, a preferred implementation is provided. The per-pixel signal stacking compensation method used in Step 1 is as follows:

[0013]

[0014] where S is the echo photon rate function under the condition of a single laser pulse, H is the observed histogram distribution, N is the number of laser pulse periods, and i represents the i-th time slot interval.

[0015] Furthermore, a preferred implementation is provided. The method for accurately extracting signals through spatio-temporal-frequency collaborative photon processing in Step 2 includes steps of noise isolation based on 1D denoised data, noise isolation based on 3D local data, search for non-locally correlated data, and noise isolation based on 4D non-local data.

[0016] Furthermore, a preferred implementation is provided. The method for noise isolation based on 1D data is: perform a per-pixel Fourier transform on the echo signal with global gating noise suppression, use a pure noise frequency band B n threshold to achieve noise suppression in the frequency domain, and obtain 1D denoised data through inverse Fourier transform,

[0017] where

[0018]

[0019] where τ p is the laser pulse width.

[0020] Further, a preferred implementation is provided. The method for noise isolation based on 3D local data is as follows: perform local spatial partitioning on the 1D denoised data, perform high-dimensional Fourier transform on the partitioned data, and use the pure noise frequency band B n threshold to separate noise and signal in the frequency domain.

[0021] Further, a preferred implementation is provided. The search for non-local correlated data is achieved by determining the correlation degree between local and non-local data, that is, using the structural similarity SSIM index to calculate the similarity S d_i as follows: The calculation expression is as follows:

[0022] S d_i = αSSIM(P dR , P d ) + βSSIM(P iR , P i ) (6)

[0023] where α and β are weight coefficients, satisfying the formula α + β = 1.

[0024] Sort the calculated similarity set from largest to smallest, and take the first N sim numbers as the number of similarities. Then, the data cubes corresponding to the first N sim partitioned blocks with the highest similarity rankings are defined as non-local correlated data.

[0025] Further, a preferred implementation is provided. The method for noise isolation of 4D non-local data is as follows: perform local spatial partitioning on the 3D denoised data, and at the same time, perform high-dimensional Fourier transform in combination with non-local correlation characteristics, and use the pure noise frequency band B n threshold to separate noise and signal in the frequency domain.

[0026] Further, a preferred implementation is provided. In step 3, the reconstructed depth image and intensity image at different scales are obtained. The method for fusing the obtained images and outputting the three-dimensional image reconstruction result of smoke interference suppression is as follows:

[0027] The fusion based on the depth image and intensity image includes three steps: intensity image fusion, depth image fusion, and denoising of the depth fusion image;

[0028] The intensity image fusion is to stack the generated intensity images at multiple scales at the pixel level;

[0029] The depth image fusion is to fill the generated depth images at multiple scales at the pixel level with the mode to obtain the fused depth image;

[0030] The fused depth image denoising is to perform threshold segmentation on the superimposed intensity image, denoise the fused depth image according to the pixel distribution after segmentation, and through the processing of the above steps, obtain the image fusion output result of multi-scale collaborative photon processing.

[0031] Solution 3: A computer device includes a memory and a processor. A computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes the method described in any one of Solution 1.

[0032] Solution 4: A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the steps of the method described in any one of Solution 1 are implemented.

[0033] The advantages of the present invention are as follows:

[0034] The single-photon lidar through-fog three-dimensional imaging method with collaborative photon processing described in the present invention solves the problems of large ranging errors, low restoration degree, and high demand for the number of statistical frames in image reconstruction in the single-photon lidar three-dimensional imaging technology in a smoke environment due to the strong scattering and high attenuation characteristics of smoke, achieving the purpose of improving the three-dimensional imaging performance of the single-photon lidar in a smoke environment, realizing accurate estimation of the target position under a small number of statistical frames and extremely scarce signal photons, and significantly improving the smoke noise suppression three-dimensional imaging method of target integrity.

[0035] The method described in the present invention is based on the spatio-temporal-frequency multi-dimensional characteristics of the echo signal, conducts research on the three-dimensional reconstruction method of multi-scale collaborative photon processing, and uses the time-domain characteristics, intensity image characteristics, distance image characteristics, data local characteristics, data non-local characteristics, and data frequency-domain characteristics of the echo signal to further reduce the demand for the number of data statistical frames in single-image reconstruction.

[0036] The present invention is also applicable to the field of single-photon lidar three-dimensional imaging under environmental conditions such as clouds, rain, fog, and haze. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flowchart of a single-photon lidar through-fog three-dimensional imaging method with collaborative photon processing described in Embodiment 1.

[0038] Figure 2 It is a flowchart of data preprocessing and guiding image generation described in Embodiment 1.

[0039] Figure 3 It is a flowchart of a signal extraction algorithm for spatio-temporal-frequency collaborative photon processing described in Embodiment 1.

[0040] Figure 4The flowchart of the depth image guidance described in Embodiment 1.

[0041] Figure 5 The flowchart of the spatio-temporal-frequency collaborative photon processing algorithm described in Embodiment 1.

[0042] Figure 6 The flowchart of the image fusion output algorithm for multi-scale collaborative photon processing described in Embodiment 1.

[0043] Figure 7 The schematic diagram for comparing the reconstruction results of different target depth images described in Embodiment 11. Specific embodiments

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application.

[0045] Embodiment 1. This embodiment provides a single-photon lidar through-fog three-dimensional imaging method for collaborative photon processing. The single-photon lidar through-fog three-dimensional imaging method includes the following steps:

[0046] Step 1: Complete the preprocessing of the array GM-APD lidar data through a pixel-by-pixel signal stacking compensation and global noise suppression method, and generate a guidance image using a multi-scale superpixel three-dimensional imaging algorithm;

[0047] Step 2: Perform block processing on the preprocessed array GM-APD lidar data output in Step 1, and use a depth image-guided photon processing algorithm and a pure noise band noise isolation method based on multi-dimensional Fourier transform to achieve precise signal extraction through spatio-temporal-frequency collaborative photon processing;

[0048] Step 3: Perform multi-scale block processing on the data output after block processing in Step 2 to obtain reconstructed depth images and intensity images at different scales, fuse the obtained images, and output the three-dimensional image reconstruction result with smoke interference suppression.

[0049] Embodiment 2. This embodiment further limits the single-photon lidar through-fog three-dimensional imaging method for collaborative photon processing described in Embodiment 1. The pixel-by-pixel signal stacking compensation method adopted in Step 1 is:

[0050]

[0051] In the formula, S is the echo photon rate function under the condition of a single laser pulse, H is the observed histogram distribution, N is the number of laser pulse periods, and i represents the i-th time slot interval.

[0052] Embodiment 3. This embodiment further limits the single-photon lidar fog-penetrating three-dimensional imaging method with collaborative photon processing described in Embodiment 1. The method for accurately extracting signals through spatio-temporal-frequency collaborative photon processing in step 2 includes steps of noise isolation based on 1D denoised data, noise isolation based on 3D local data, search for non-locally correlated data, and noise isolation based on 4D non-local data.

[0053] Embodiment 4. This embodiment further limits the single-photon lidar fog-penetrating three-dimensional imaging method with collaborative photon processing described in Embodiment 3. The method for noise isolation based on 1D data is as follows: perform pixel-by-pixel Fourier transform on the echo signal with global gating noise suppression, and use the pure noise frequency band B n threshold to achieve noise suppression in the frequency domain, and obtain 1D denoised data through inverse Fourier transform.

[0054] where

[0055]

[0056] In the formula, τ p is the laser pulse width.

[0057] Embodiment 5. This embodiment further limits the single-photon lidar fog-penetrating three-dimensional imaging method with collaborative photon processing described in Embodiment 3. The method for noise isolation based on 3D local data is as follows: perform local spatial block processing on the 1D denoised data, perform high-dimensional Fourier transform on the block data, and use the pure noise frequency band B n threshold to separate noise and signal in the frequency domain.

[0058] Embodiment 6. This embodiment further limits the single-photon lidar fog-penetrating three-dimensional imaging method with collaborative photon processing described in Embodiment 3. The search for non-locally correlated data is achieved by determining the correlation degree between local and non-local data, that is, using the structural similarity SSIM index to calculate the similarity S d_i The calculation expression is as follows:

[0059] S d_i =αSSIM(P dR , P d ) + βSSIM(P iR , P i ) (6)

[0060] In the formula, α and β are weight coefficients, satisfying the formula α + β = 1.

[0061] Sort the calculated similarity set from largest to smallest, and take the top N simIf the number is used as the number of similarities, then the top N in the similarity ranking sim The data cubes corresponding to the N blocks are defined as non-locally correlated data.

[0062] Embodiment Seven: This embodiment further limits the single-photon lidar fog-penetrating three-dimensional imaging method with collaborative photon processing described in Embodiment Three. The method for noise isolation of 4D non-local data is as follows: Perform local spatial block processing on the 3D denoised data, and at the same time, perform high-dimensional Fourier transform in combination with non-local correlation characteristics, and use the pure noise frequency band B n Threshold to separate noise and signal in the frequency domain.

[0063] Embodiment Eight: This embodiment further limits the single-photon lidar fog-penetrating three-dimensional imaging method with collaborative photon processing described in Embodiment Seven. The method for fusing the reconstructed depth images and intensity images obtained at different scales in step 3 and outputting the three-dimensional image reconstruction result with smoke interference suppression is as follows:

[0064] The fusion based on the depth image and the intensity image outputs includes three steps: intensity image fusion, depth image fusion, and denoising of the fused depth image;

[0065] The intensity image fusion is to stack the generated multi-scale intensity images at the pixel level;

[0066] The depth image fusion is to fill the generated multi-scale depth images at the pixel level with the mode to obtain the fused depth image;

[0067] The denoising of the fused depth image is to perform threshold segmentation on the stacked intensity image, and denoise the fused depth image according to the pixel distribution of the segmentation. Through the processing of the above steps, the image fusion output result of multi-scale collaborative photon processing is obtained.

[0068] Embodiment Nine: This embodiment proposes a computer device, including a memory and a processor. A computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes the method described in any one of Embodiments One to Eight.

[0069] Embodiment Ten: This embodiment proposes a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the steps of the method described in any one of Embodiments One to Eight are implemented.

[0070] Embodiment Eleven: This embodiment proposes an example, and the example is used to explain the above Embodiments One to Eight. The specific example is as follows:

[0071] SeeFigures 1 to 7 For this embodiment, the method described in this embodiment includes three steps: data preprocessing and guiding image generation, signal extraction through spatio-temporal-frequency collaborative photon processing, and image fusion output through multi-scale collaborative photon processing. The input end of the algorithm is the array GM-APD lidar data and lidar system parameters, and the output end is the three-dimensional image reconstruction result with smoke interference suppression.

[0072] First, through pixel-by-pixel signal stacking compensation and global noise suppression, data preprocessing is completed, and a multi-scale superpixel three-dimensional imaging algorithm is used to generate a guiding image.

[0073] Secondly, the denoised data is block-processed, and the depth image-guided photon processing algorithm and the pure noise band noise isolation method based on multi-dimensional Fourier transform are used to achieve precise signal extraction through spatio-temporal-frequency collaborative photon processing.

[0074] Finally, the denoised data is multi-scale block-processed using the processing flow of the second step to obtain the reconstructed depth image and intensity image at different scales, and the obtained images are fused to output the three-dimensional image reconstruction result with smoke interference suppression.

[0075] Step 1: Data preprocessing and guiding image generation

[0076] The input data of the algorithm is the array single-photon lidar data cube S, and the lidar system parameters include the FWHM obtained by Gaussian fitting. To further reduce the requirement for the number of data statistical frames during single-image reconstruction and avoid the situation where the model estimation algorithm fails in smoke signal estimation due to sparsely distributed echo photons, this section uses a global gating noise suppression method to achieve preliminary separation of signal photons and noise photons. The specific steps of data preprocessing and guiding image generation are as follows:

[0077] (1) Use formula (1) for pixel-by-pixel stacking compensation, stack the echo photons of all pixels after stacking compensation in the time domain, and perform smooth fitting on the stacked echo signal yall to obtain the curve ys, and remove the burrs and mutated data points.

[0078]

[0079] In the formula, S is the echo photon rate function under a single laser pulse condition, H is the observed histogram distribution, N is the number of laser pulse periods, and i represents the i-th time slot interval.

[0080] (2) According to the smoke distribution scenario, use an applicable estimation algorithm (such as the dual-parameter estimation method or the e-exponential fitting estimation) to estimate the noise signal and distribution interval to obtain the curve yf.

[0081] (3) Subtract the smoothed curve ys from the estimated result yf to obtain the signal photon distribution curve yt, where yt = max(ys - yf, 0), and perform a sliding window process on yt with the window size of GL to search for the distribution interval of the target signal.

[0082] (4) According to the determined signal distribution interval, intercept the signal on the curve ys and calculate the mean according to the number of pixels. Intercept each pixel signal after stack compensation according to the distribution interval and calculate the difference from the mean result to obtain the echo signal with global gating noise suppression. Subsequently, this echo signal is mainly used for data processing.

[0083] (5) Use the sliding window operation shown in formula (2) to stack the denoised data within the window W to obtain the histogram set Yl with enhanced signal photons. Use the logarithmic matched filtering algorithm to estimate the target distance for each histogram in the set Yl, and expand the estimated result to obtain the distance image Zl at scale l.

[0084]

[0085] In the formula, O is the window size.

[0086] (6) Statistically analyze the distance images Zl at different scales numerically by pixel, and output the fused depth image in the form of the mode. This image will be used for image guidance in the subsequent processing steps.

[0087] Step 2: Signal extraction of spatio-temporal-frequency collaborative photon processing

[0088] The signal extraction algorithm of spatio-temporal-frequency collaborative photon processing is divided into four steps, namely noise isolation based on 1D data, noise isolation based on 3D local data, search for non-locally correlated data, and noise isolation based on 4D non-local data.

[0089] The core idea of the noise isolation process based on 1D data is to perform a pixel-by-pixel Fourier transform on the echo signal with global gating noise suppression, use the pure noise frequency band Bn threshold to achieve noise suppression in the frequency domain, and obtain the 1D denoised data through the inverse Fourier transform.

[0090] Since the GM-APD lidar uses Gaussian pulsed laser, the target echo signal is approximately Gaussian distributed. In the case of a Gaussian-shaped pulse, the pure noise frequency band Bn is defined as the frequency band greater than three standard deviations of the Gaussian spectrum. The expression of Bn is as follows:

[0091]

[0092] In the formula, τ p is the laser pulse width.

[0093] Noise isolation processing based on 3D data. The core idea is to perform local spatial block processing on 1D denoised data, perform high-dimensional Fourier transform on the block data, and use the pure noise band Bn threshold to separate noise and signal in the frequency domain.

[0094] To obtain 3D data, windows of different scale sizes are used to block-process the 1D denoised data. In the block-processing process, according to formula (1), a sliding window operation is used to separate the local pixel data within window W. High-dimensional Fourier transform is performed on the separated local pixel data, and the pure noise band Bn threshold is used to separate the noise component and the signal component in the frequency domain. The mean value E(|Bin|2) of the Fourier coefficient energy of the signal component within the pure noise band Bn and the mean value E(|Bout|2) of the Fourier coefficient energy of the noise component outside the pure noise band Bn are calculated. At the same time, a sliding window of the same scale is used to block-process the normalized depth image for guidance, obtaining the same Fourier amplitude spatial distribution as the 3D block data. The spatial distribution is multiplied pointwise with the separated signal component, and the product result is subjected to inverse Fourier transform to obtain the estimated echo signal.

[0095] Due to the block and sliding window processing, when estimating the echo signals of all pixels, there is a situation of pixel overlap, that is, there is a problem of multiple superpositions when estimating the echo signal of each pixel. Therefore, these repeatedly superposed echo signals can be solved by weighted averaging, and the echo signal intensity φ of a single pixel is calculated. The calculation expression is as follows:

[0096]

[0097] In the formula, n is the number of all overlapping blocks on this pixel, φi is the estimated result of the echo signal of the i-th overlapping block on this pixel, and ωi corresponds to the weight of the i-th overlapping block. ωi is inversely proportional to the noise component of the i-th overlapping block, and the calculation expression is as follows:

[0098]

[0099] In the formula, E(|Biout|2) is the mean value of the Fourier coefficient energy of the noise component of the i-th overlapping block. Through the above processing, 3D denoised data and the estimated intensity image and depth image can be obtained.

[0100] To achieve noise isolation based on 4D non-local data, it is first necessary to find the correlation between local and non-local data cubes. To find similar data cubes, the search space can be determined according to the intensity image and depth image estimated from 3D denoised data, without relying on the overall data cube. Taking a pixel as an example, first define a depth reference block PdR (Lx×Ly) and an intensity reference block PiR (Lx×Ly) in the same area of the depth image and intensity image respectively, so that the size of the reference block is consistent with the sliding window size of 3D data acquisition, and at the same time make the pixel located at the upper left corner of the reference block (synchronized with the data cube). Then, with the reference block as the center, determine a search window with a scale of Ssearch (Ssearch > Lx), and search for depth image blocks Pd (Lx×Ly) and intensity image blocks Pi (Lx×Ly) similar to the reference block within the search window range of the depth image and intensity image respectively. The structural similarity SSIM index is used to calculate the similarity Sd_i, and the calculation expression is as follows:

[0101] S d_i = αSSIM(P dR , P d ) + βSSIM(P iR , P i ) (6)

[0102] In the formula, α and β are weight coefficients, satisfying the formula α + β = 1.

[0103] By sorting the calculated similarity set from large to small, and taking the first Nsim numbers as the number of similarities, the data cubes corresponding to the first Nsim blocks in the similarity sorting can be defined as non-local related data.

[0104] The core idea of noise isolation processing based on 4D data is to perform local spatial block processing on 3D denoised data, and at the same time combine non-local related characteristics to perform high-dimensional Fourier transform, and use the pure noise frequency band Bn threshold to separate noise and signals in the frequency domain.

[0105] In the process of guiding photon processing, the depth image estimated from 3D denoised data is used as the guiding image, and the echo signal is estimated based on the spatial frequency correlation between the guiding image and 3D denoised data. Finally, the echo signal of a single pixel is obtained by weighted average, and the estimated intensity image and depth image are output.

[0106] Step 3: Image fusion output of multi-scale collaborative photon processing

[0107] Perform local and non-local correlation processing on the 1D denoised data using different block scales, and achieve high-quality three-dimensional imaging of targets in a smoke environment through the image fusion output of multi-scale collaborative photon processing. For the image fusion output of multi-scale collaborative photon processing, first use the spatio-temporal-frequency collaborative photon processing signal extraction algorithm to extract signals from the 1D denoised data at different block scales respectively, and output the depth image and intensity image estimated based on different block scales.

[0108] The fusion output based on the depth image and intensity image includes three steps: intensity image fusion, depth image fusion, and denoising of the depth fusion image. Intensity image fusion is to superimpose the generated multi-scale intensity images at the pixel level; depth image fusion is to fill the generated multi-scale depth images at the pixel level with the mode to obtain the depth fusion image; denoising of the depth fusion image is to perform threshold segmentation on the superimposed intensity image and denoise the depth fusion image according to the pixel distribution of the segmentation. Through the above processing, the image fusion output result of multi-scale collaborative photon processing is obtained.

[0109] The present invention uses two indicators, the signal-to-background ratio (SBR) and the photons per pixel (PPP), to evaluate the three-dimensional imaging ability of the current single-photon lidar through atmospheric obscurants such as smoke. The results are shown in Table 1. Compared with the research results in Table 1, the algorithm proposed in the present invention reduces the PPP by 18.5% and the SBR by 98.0%.

[0110] Table 1 Comparison of experimental results of single-photon imaging through atmospheric obscurants

[0111]

[0112] Figure 1Any process or method description described in the flowchart or otherwise described herein can be understood to represent a module, segment, or portion of code that includes one or more executable instructions for implementing a customized logical function or process. The scope of the preferred embodiments of the present invention includes additional implementations in which functions may be performed in a substantially simultaneous manner or in a reverse order according to the functions involved, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of the present invention pertain. The logic and / or steps represented in the flowchart or otherwise described herein illustrate the possible architectures, functions, and operations of the apparatus and methods according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, program segment, or portion of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in a reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, and combinations of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or by a combination of dedicated hardware and computer instructions. For example, a sequenced list of executable instructions that can be considered to implement a logical function can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device.

[0113] Those skilled in the art can understand that the above description is only the preferred embodiment of the present invention. The features described in various embodiments and / or claims of the present disclosure can 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 foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

[0114] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn the basic inventive concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art can make various changes and variations 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 equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A single-photon laser radar three-dimensional imaging method through fog with coordinated photon processing, characterized in that: The single-photon laser radar fog-penetrating three-dimensional imaging method comprises the following steps: Step 1: Preprocess the array GM-APD lidar data through pixel-by-pixel signal accumulation compensation and global noise suppression methods, and generate a guide image using a multi-scale super-pixel three-dimensional imaging algorithm; Step 2, the pre-processed array GM-APD lidar data output in step 1 is processed in blocks, and the signal is accurately extracted through time-space-frequency coordinated photon processing using a deep image guided photon processing algorithm and a pure noise band noise isolation method based on multi-dimensional Fourier transform; Step 3: Perform multi-scale block processing on the block-processed data output in step 2 to obtain depth images and intensity images reconstructed at different scales, fuse the obtained images, and output a three-dimensional image reconstruction result with smoke interference suppressed.

2. The method for three-dimensional imaging through fog using a single-photon laser radar with coordinated photon processing according to claim 1, characterized in that: The pixel-by-pixel signal accumulation compensation method used in step 1 is: Where S is the echo photon rate function under a single laser pulse condition, H is the observed histogram distribution, N is the number of laser pulse cycles, and i represents the i-th time slot interval.

3. The method for three-dimensional imaging through fog using single-photon laser radar with coordinated photon processing according to claim 1, characterized in that: The method for accurately extracting signals by time-space-frequency coordinated photon processing in step 2 includes the steps of noise isolation based on 1D denoising data, noise isolation based on 3D local data, search for non-local correlation data and noise isolation based on 4D non-local data.

4. The method for three-dimensional imaging through fog using a single-photon laser radar with coordinated photon processing according to claim 3, characterized in that: The noise isolation method based on 1D data is as follows: the echo signal with global gated noise suppression is subjected to pixel-by-pixel Fourier transform, and the pure noise band B is used. n The threshold realizes noise suppression in the frequency domain, and the 1D denoised data is obtained through the inverse Fourier transform. in, Where τ p is the laser pulse width.

5. The method for three-dimensional imaging through fog using single-photon laser radar with coordinated photon processing according to claim 3, characterized in that: The noise isolation method based on 3D local data is as follows: the 1D denoised data is processed in local space blocks, the block data is subjected to high-dimensional Fourier transform, and the pure noise frequency band B is used. n The threshold separates the noise from the signal in the frequency domain.

6. The method for three-dimensional imaging through fog using single-photon laser radar with coordinated photon processing according to claim 3, characterized in that: The search for non-local related data is achieved by determining the correlation between local and non-local data, that is, using the structural similarity SSIM index to compare the similarity S d_i The calculation expression is as follows: Where α and β are weight coefficients, satisfying the formula α+β=1. Sort the calculated similarity sets from large to small and take the top N sim The number of similarities is taken as the number of similarities, and the similarity is ranked before N sim The data cube corresponding to each block is defined as non-locally correlated data.

7. The method for three-dimensional imaging through fog using single-photon laser radar with coordinated photon processing according to claim 3, characterized in that: The method of noise isolation of 4D non-local data is: local space block processing of 3D denoising data, and high-dimensional Fourier transform combined with non-local correlation characteristics, using pure noise band B n The threshold separates the noise from the signal in the frequency domain.

8. The method for three-dimensional imaging through fog using single-photon laser radar with coordinated photon processing according to claim 3, characterized in that: In step 3, the depth image and intensity image reconstructed at different scales are obtained, and the obtained images are fused to output the three-dimensional image reconstruction result with smoke interference suppressed. The method is: The fusion output based on the depth image and the intensity image includes three steps: intensity image fusion, depth image fusion and depth fusion image denoising; The intensity image fusion is to superimpose the generated intensity images of multiple scales at the pixel level; The depth image fusion is to fill the generated depth images of multiple scales with the majority at the pixel level to obtain a fused depth image; The fused depth image denoising is to perform threshold segmentation on the superimposed intensity image, and denoise the fused depth image according to the segmented pixel distribution. Through the above steps, the image fusion output result of multi-scale collaborative photon processing is obtained.

9. A computer device, characterized in that: The method comprises 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 implements the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method described in any one of claims 1 to 8.

Images