Weak single photon signal preprocessing and three-dimensional imaging method and computer readable storage medium

Abstract

The invention provides a weak single photon signal preprocessing and three-dimensional imaging method and a computer readable storage medium, and solves the technical problems of poor scene adaptability, insufficient data compression rate, non-ideal imaging quality and the like in the prior art. The method comprises the steps of obtaining an original data matrix; performing pretreatment; reconstructing to generate a three-dimensional image; the preprocessing method comprises the following steps of: selecting photon event data of any pixel in an original photon event data matrix to obtain a corresponding basic histogram, and then forming a data block by using the photon event data of the pixel and all pixels adjacent to the pixel to obtain a data block basic histogram; constructing a macro histogram according to the basic histogram of the data block; searching in the macro histogram to obtain a peak value interval of the maximum count value; performing adaptive filtering on the photon event data by using the peak interval of the maximum count value to obtain a noise reduction matrix; and finally, generating a three-dimensional image by adopting a reconstruction algorithm. According to the method, the imaging quality and the imaging precision of the weak single-photon signal are improved.

Description

Technical Field

[0001] The present invention relates to a single-photon laser radar imaging method, and in particular to a weak single-photon signal preprocessing and three-dimensional imaging method, and a computer-readable storage medium. Background Art

[0002] Single-photon lidar (LIDAR) is widely used in three-dimensional (3D) imaging due to its exceptional sensitivity and high resolution. Even under low-photon flux conditions, it can achieve effective 3D imaging, with an average of only one photon per pixel. Currently, this technology offers new opportunities for applications in remote sensing, autonomous driving, environmental monitoring, and other fields. However, when the weak echo photon signal is interfered with by strong background noise, accurately determining the target distance becomes extremely difficult, directly affecting the accuracy of 3D imaging and reducing the detection capability of the single-photon lidar.

[0003] To overcome this shortcoming, researchers have used the horizontal and vertical structural information of natural scenes to effectively suppress noise such as background light and dark counts, thereby improving imaging accuracy under weak single-photon conditions. Furthermore, by applying a short-time threshold to each pixel to distinguish between signals and noise, the depth estimation can be improved by combining adjacent pixel data when the number of detections is insufficient. For example, the Chinese invention patent with publication number CN119335556 A discloses an underwater single-photon lidar imaging method and system based on a gated SPAD array. This imaging method can effectively filter out a large amount of noise and achieve millimeter-level reconstruction accuracy. However, it relies on the prior of the water body scattering model and has poor scene adaptability. At the same time, pixel-by-pixel gating leads to data redundancy and fails to effectively utilize spatial correlation, resulting in insufficient data compression rate.

[0004] In addition, the Chinese invention patent with publication number CN117572381 A discloses a single-photon lidar data processing method based on spatiotemporal correlation. This method is based on the temporal correlation of single-photon lidar data, and selects multiple depth layers of lidar three-dimensional point cloud data in the time domain. It can identify target scenes at different depth layers and improve the signal-to-noise ratio of three-dimensional point cloud data. However, this method assumes that the noise obeys a static distribution, while the actual scene noise has time-varying characteristics (such as dynamic background light and pulse interference). Therefore, the scene adaptability of this method is poor. At the same time, because this method does not introduce prior constraints, artifacts are prone to appear in the edge area of ​​the reconstructed image, resulting in less than ideal imaging quality and imaging accuracy.

[0005] Chinese invention patent publication number CN119511292 A discloses a method for extracting long-range single-photon lidar signals. This method uses a dual-phase histogram for coarse target detection and a secondary subdivision histogram for fine target detection. While ensuring highly reliable target detection, it achieves high-precision ranging of the target. However, it ignores the spatial continuity between pixels, resulting in a discontinuity in depth estimation under signal sparsity. As a result, the imaging quality is still less than ideal.

[0006] In summary, the above methods can improve the imaging accuracy under weak single-photon conditions to a certain extent, but they also have problems such as poor scene adaptability, insufficient data compression rate or unsatisfactory imaging quality. Summary of the Invention

[0007] The purpose of the present invention is to solve the technical problems of the prior art such as poor scene adaptability, insufficient data compression rate, and unsatisfactory imaging quality, and to provide a weak single-photon signal preprocessing and three-dimensional imaging method and a computer-readable storage medium.

[0008] To achieve the above objectives, the technical solutions provided by the present invention are:

[0009] A method for preprocessing and three-dimensional imaging of weak single-photon signals is characterized in that it includes the following steps:

[0010] Step 1: Get the original data matrix

[0011] Obtaining a matrix of raw photon event data collected by a sampling system; the sampling system is a single-photon lidar system;

[0012] Step 2: Preprocessing

[0013] 2.1. Select the photon event data of any pixel in the original photon event data matrix, perform photon counting statistics on it, and obtain the corresponding basic histogram. Then, the photon event data of this pixel and all adjacent pixels are formed into a data block, and then perform pixel-by-pixel photon counting statistics on this data block to obtain the data block basic histogram;

[0014] 2.2. Construct a corresponding macrohistogram based on the basic histogram of the data block; the time resolution of the macrohistogram is 100 times that of the basic histogram of the data block;

[0015] 2.3. Search pixel by pixel in the macro histogram to obtain the peak interval of the maximum count value;

[0016] 2.4. Using the peak interval of the maximum count value, adaptively filter the photon event data of the pixel selected in step 2.1 to obtain a noise reduction matrix;

[0017] 2.5. According to the methods of steps 2.1 to 2.4, the photon event data of other pixels in the original photon event data matrix are adaptively filtered pixel by pixel to obtain corresponding noise reduction matrices;

[0018] Step 3: Reconstruct and generate a 3D image

[0019] The denoising matrix obtained in steps 2.4 and 2.5 is processed using a reconstruction algorithm to generate a three-dimensional image, thereby completing the preprocessing and three-dimensional imaging of weak single-photon signals.

[0020] Furthermore, step 3 is specifically as follows:

[0021] 3.1. Construct an observation model based on the fact that single-photon detection signals conform to the Poisson distribution.

[0022] 3.2. Based on the observation model of step 3.1 and the noise reduction matrix obtained in steps 2.4 and 2.5, the echo photon statistical histogram matrix Z is constructed;

[0023] 3.3. Based on the echo photon statistical histogram matrix, the objective function is constructed and the prior constraints are introduced to obtain the optimized objective function The prior constraints include total variation regularization constraints and strength constraints;

[0024] 3.4. Perform deconvolution on the optimization objective function to generate a three-dimensional image, thereby achieving preprocessing and three-dimensional imaging of weak single-photon signals.

[0025] Furthermore, in step 3.1, the observation model is:

[0026]

[0027] Where: z i,j,t represents the photon count at the tth time in the (i, j)th pixel, (i, j) represents the pixel in the i-th row and j-th column, t∈[0,T r ] represents the sampling time range of each detection cycle of the sampling system, T r represents the pulse period of the laser; d i,j represents the actual distance of the target point mapped by the (i, j)th pixel; r i,j Indicates the reflectivity of the target point mapped by the (i, j)th pixel; b i,j represents the noise level, which includes b si,j and b d , b si,j represents the number of photons in the ambient noise; b d represents the dark count of the detector; P indicates that the observation model conforms to the Poisson distribution; h ij It represents the spatial nucleus generated by the laser passing through the transmission medium; ht represents the time kernel generated when the laser passes through the transmission medium; c is the speed of light; Δt is the calibration error of the sampling system.

[0028] Furthermore, in step 3.2, the echo photon statistical histogram matrix Z obeys the following Poisson distribution:

[0029] Z~Poisson(H*Y+B)

[0030] Where: H represents the spatiotemporal kernel matrix; * represents the convolution operation, Y represents the three-dimensional matrix of the target scene; B represents the noise matrix.

[0031] Furthermore, in step 3.3, the optimization objective function It is expressed by the following formula:

[0032]

[0033] Among them, L Y (Y; Z, H, B) represents the negative log-likelihood function; represents the strength constraint; λ and γ represent the weights of the control constraint terms, respectively; Φ(Y) represents an additional prior term with a total variation regularization constraint.

[0034] Furthermore, in step 3.4, the deconvolution solution is solved by the SPIRAL-TAP three-dimensional deconvolution method.

[0035] At the same time, the present invention also provides a computer-readable storage medium on which a computer program or instruction is stored. The special feature of the computer program or instruction is that when the computer program or instruction is executed by the processor, the steps of the above-mentioned weak single-photon signal preprocessing and three-dimensional imaging method are implemented.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. In the process of preprocessing the original photon event data matrix, the present invention successively obtains a noise reduction matrix by statistically analyzing the basic histogram, constructing a macrohistogram, searching for the peak interval of the maximum count value, and performing adaptive filtering using the peak interval as the filtering interval. In this process, the method based on the peak interval extraction of the macrohistogram and the combination of adaptive filtering improves the single-photon data preprocessing denoising performance, breaks through the limitations of traditional pixel-by-pixel or static noise assumptions, and ultimately improves the imaging quality and imaging accuracy of weak single-photon signals.

[0038] 2. The present invention reduces the uncertainty of distance without distance prior by using the adaptive filtering method, thereby significantly reducing the amount of original data and further improving the reconstruction speed of the later image.

[0039] 3. During the image reconstruction process, the present invention abstracts the imaging modeling problem into a Poisson convolution problem. Finally, the SPIRAL-TAP three-dimensional deconvolution method can be used to directly solve the three-dimensional matrix and obtain a three-dimensional image of the target object, avoiding pixel-by-pixel error accumulation and further improving imaging accuracy.

[0040] 4. This invention introduces both intensity constraints and total variation regularization constraints during image reconstruction. The intensity constraint suppresses overexposure noise, while the total variation regularization constraint maintains structural continuity under conditions of signal sparsity. By combining physical and data-driven approaches, this invention addresses the pathological issues of sparse signal reconstruction in high-noise environments, thereby improving imaging accuracy in complex scenarios (such as point targets with strong noise).

[0041] 5. Based on the weak single-photon signal preprocessing and three-dimensional imaging method of the present invention, data noise reduction and compression efficiency are significantly improved, imaging accuracy and edge continuity are optimized, and robustness in complex scenes is enhanced.

[0042] 6. The present invention systematically solves the problem of single-photon signal reconstruction under high noise and low signal-to-noise ratio, providing high-precision imaging capabilities for long-distance dynamic scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flow chart of an embodiment of a method for preprocessing weak single-photon signals and three-dimensional imaging according to the present invention;

[0044] Figure 2 This is the original data matrix diagram obtained in step 1 of the embodiment of the present invention;

[0045] Figure 3 is the basic histogram obtained in step 2.1 of the embodiment of the present invention;

[0046] Figure 4 The macrohistogram constructed in step 2.2 of the embodiment of the present invention;

[0047] Figure 5 This is the noise reduction matrix diagram obtained in step 2.4 of the embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the objects, advantages and features of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and specific examples. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0049] like Figure 1 As shown, this embodiment provides a weak single-photon signal preprocessing and three-dimensional imaging method, including the following steps:

[0050] Step 1: Get the original data matrix

[0051] Obtain the original photon event data matrix collected by the sampling system. The sampling system of this embodiment selects a single-photon lidar system for scanning multi-depth scenes. The system can collect photon detection signals at a basic resolution to ensure the integrity of data collection, while recording the time information of the photon arrival.

[0052] Step 2: Preprocessing

[0053] 2.1. Select the photon event data of any pixel in the original photon event data matrix, perform photon counting statistics on it, and obtain the corresponding basic histogram. Then, the photon event data of this pixel and all adjacent pixels are formed into a data block ( Figure 2 As shown), the data block is then subjected to pixel-by-pixel photon counting statistics, and the photon events are spatially correlated to obtain the basic histogram of the data block, as shown in Figure 3 As shown, the graph can intuitively feedback the photon distribution information in the original photon event data.

[0054] 2.2, Construct the corresponding macro histogram based on the basic histogram of the data block ( Figure 4 The time resolution of the macro histogram is 100 times that of the basic histogram of the data block. The construction of the macro histogram is used to improve the overall resolvability of the data.

[0055] 2.3. In the macro histogram, search each interval one by one to obtain the peak interval of the maximum count value. The search of this peak interval is used to extract the main photon signal. Based on this peak interval, the upper and lower limits of the interval can be determined and used as the basis for subsequent filtering processing.

[0056] 2.4. Using the peak interval of the maximum count value as the filtering interval, perform adaptive filtering on the photon event data of the pixel selected in step 2.1 to remove noise and retain the effective signal, and obtain the corresponding noise reduction matrix ( Figure 5 ), providing more accurate data for subsequent depth estimation.

[0057] 2.5. According to the methods of steps 2.1 to 2.4, the photon event data of other pixels in the original photon event data matrix are adaptively filtered pixel by pixel to obtain corresponding noise reduction matrices.

[0058] Step 3: Reconstruct and generate a 3D image

[0059] 3.1. Construct an observation model based on the fact that single-photon detection signals conform to the Poisson distribution.

[0060] The observation model expression is:

[0061]

[0062] Where: z i,j,t represents the photon count at the t-th time in the (i, j)-th pixel, (i, j) represents the pixel in the i-th row and j-th column, t∈[0,T r ] represents the sampling time range of each detection cycle of the sampling system, T r represents the pulse period of the laser; d i,j represents the actual distance of the target point mapped by the (i, j)th pixel; r i,j represents the reflectivity of the target point mapped by the (i, j)th pixel; b i,j represents the noise level, which includes b si,j and b d , b si,j represents the number of photons in the ambient noise; b d represents the dark count of the detector; P indicates that the observation model conforms to the Poisson distribution; h ij It represents the spatial nucleus generated by the laser passing through the transmission medium; h t represents the time kernel generated when the laser passes through the transmission medium; c is the speed of light; Δt is the calibration error of the sampling system.

[0063] 3.2. Based on the observation model of step 3.1 and the noise reduction matrix obtained in steps 2.4 and 2.5, the echo photon statistical histogram matrix is ​​constructed.

[0064] In actual operation, the time-to-digital converter converts the time range T r Discretize to n t (n t =T r / Δ) time windows to record the discrete histogram of echo photon statistics, where Δ represents the time resolution of the time-to-digital converter in the sampling system. Assume that the target scene consists of n x ×n y pixels, n x ×n y Represents the number of pixels of the acquired three-dimensional graphics, n x is the number of pixel rows, n y is the number of pixel columns, from which a three-dimensional matrix Y of the target scene can be constructed, whose size is n x ×n y ×n t The (i, j) element of the three-dimensional matrix Y is a vector containing only one non-zero value, corresponding to a specific (depth, reflectivity) pair describing the target scene. In addition, the spatiotemporal kernel matrix H is defined, whose dimension is k x ×k y ×k t , which is the spatial kernel h in the observation model ij and time kernel h tThe outer product of contains the characteristics of spatial and temporal degradation of the sampling system. The noise matrix B is a dimension of n. x ×n y ×n t The matrix is ​​used to describe the background noise. Since some noise is not removed in the noise reduction matrix obtained after preprocessing, this part of the noise is represented by the noise matrix B.

[0065] On this basis, the echo photon statistical histogram matrix Z obeys the following Poisson distribution:

[0066] Z~Poisson(H*Y+B)

[0067] Where: H represents the spatiotemporal kernel matrix; * represents the convolution operation, Y represents the three-dimensional matrix of the target scene; B represents the noise matrix, and the dimension of Z is also n x ×n y ×n t .

[0068] 3.3. Based on the echo photon statistical histogram matrix, an objective function is constructed and prior constraints are introduced to obtain an optimized objective function; the prior constraints include total variation regularization constraints and intensity constraints.

[0069] In order to accurately recover the depth estimation matrix from the noise-contaminated observation data, this embodiment defines the following optimization objective function

[0070]

[0071] Among them, L Y (Y; Z, H, B) represents the negative log-likelihood function, which is used to measure the deviation between the observed data and the estimated matrix; the constraint condition Determined by the non-negativity of photon reflectivity to ensure physical rationality; represents an intensity constraint, which helps better model the distribution of noise and further limit its impact, especially when the noise intensity is correlated with the signal strength. λ and γ respectively denote the weights of the control constraint terms, which are used to optimize the signal recovery quality in high-noise environments. Φ(Y) represents an additional prior term with a total variation (TV) regularization constraint. Using this regularization method, we can exploit the spatial correlation between pixels to improve the reconstruction quality.

[0072] 3.4. Solving the optimization objective function using the SPIRAL-TAP 3D deconvolution method can improve the robustness and accuracy of the reconstruction algorithm, thereby expanding the image from 2D to 3D, more effectively deconvolving the measurement data affected by noise, and ultimately achieving preprocessing of weak single-photon signals and 3D imaging.

[0073] In addition, this embodiment also provides a computer-readable storage medium on which a computer program or instruction is stored. When the computer program or instruction is executed by a processor, the steps of the weak single-photon signal preprocessing and three-dimensional imaging method described in this embodiment are implemented.

[0074] This paper proposes a method for preprocessing and high-precision reconstruction of weak single-photon signals. This method effectively improves the signal-to-noise ratio (SNR) of single-photon detection, enhances signal quality, and improves the accuracy of depth map recovery. Based on spatial correlation analysis, histogram screening, multi-stage noise reduction, and an optimized reconstruction algorithm, combined with a 3D deconvolution solution strategy, this method addresses the need for high-precision 3D single-photon imaging in high-noise environments.

[0075] Compared with the existing methods of processing noise and signals in a single dimension (time or space) or in a local range, the present invention achieves the coordinated optimization of noise suppression and signal reconstruction through global modeling (three-dimensional convolution), multi-level constraints (TV+intensity) and adaptive filtering, providing a systematic solution for high-precision imaging of single-photon lidar in strong dynamic noise scenarios.

[0076] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present invention.

Claims

1. A method for preprocessing weak single-photon signals and three-dimensional imaging, characterized in that: The following steps are involved: Step 1: Get the original data matrix Obtaining a matrix of raw photon event data collected by a sampling system; the sampling system is a single-photon lidar system; Step 2: Preprocessing 2.

1. Select the photon event data of any pixel in the original photon event data matrix, perform photon counting statistics on it, and obtain the corresponding basic histogram. Then, the photon event data of this pixel and all adjacent pixels are formed into a data block, and then perform pixel-by-pixel photon counting statistics on this data block to obtain the data block basic histogram; 2.

2. Construct a corresponding macrohistogram based on the basic histogram of the data block; the time resolution of the macrohistogram is 100 times that of the basic histogram of the data block; 2.

3. Search pixel by pixel in the macro histogram to obtain the peak interval of the maximum count value; 2.

4. Using the peak interval of the maximum count value, adaptively filter the photon event data of the pixel selected in step 2.1 to obtain a noise reduction matrix; 2.

5. According to the methods of steps 2.1 to 2.4, the photon event data of other pixels in the original photon event data matrix are adaptively filtered pixel by pixel to obtain corresponding noise reduction matrices; Step 3: Reconstruct and generate a 3D image The denoising matrix obtained in steps 2.4 and 2.5 is processed using a reconstruction algorithm to generate a three-dimensional image, thereby completing the preprocessing and three-dimensional imaging of weak single-photon signals.

2. The weak single-photon signal preprocessing and three-dimensional imaging method according to claim 1, characterized in that: Step 3 is as follows: 3.

1. Construct an observation model based on the fact that single-photon detection signals conform to the Poisson distribution. 3.

2. Based on the observation model of step 3.1 and the noise reduction matrix obtained in steps 2.4 and 2.5, the echo photon statistical histogram matrix Z is constructed; 3.

3. Based on the echo photon statistical histogram matrix, the objective function is constructed and the prior constraints are introduced to obtain the optimized objective function The prior constraints include total variation regularization constraints and strength constraints; 3.

4. Perform deconvolution on the optimization objective function to generate a three-dimensional image, thereby achieving preprocessing and three-dimensional imaging of weak single-photon signals.

3. The weak single-photon signal preprocessing and three-dimensional imaging method according to claim 2, characterized in that: In step 3.1, the observation model is: Where: z i,j,t represents the photon count at the tth time in the (i, j)th pixel, (i, j) represents the pixel in the i-th row and j-th column, t∈[0,T r ] represents the sampling time range of each detection cycle of the sampling system, T r represents the pulse period of the laser; d i,j represents the actual distance of the target point mapped by the (i, j)th pixel; r i,j Indicates the reflectivity of the target point mapped by the (i, j)th pixel; b i,j Represents the noise level, which includes and b d , represents the number of photons in the ambient noise; b d represents the dark count of the detector; P indicates that the observation model conforms to the Poisson distribution; h ij It represents the spatial nucleus generated by the laser passing through the transmission medium; h t represents the time kernel generated when the laser passes through the transmission medium; c is the speed of light; Δt is the calibration error of the sampling system.

4. The weak single-photon signal preprocessing and three-dimensional imaging method according to claim 3, characterized in that: In step 3.2, the echo photon statistical histogram matrix Z obeys the following Poisson distribution: Z~Poisson(H*Y+B) Where: H represents the spatiotemporal kernel matrix; * represents the convolution operation, Y represents the three-dimensional matrix of the target scene; B represents the noise matrix.

5. The weak single-photon signal preprocessing and three-dimensional imaging method according to claim 4, characterized in that: In step 3.3, the optimization objective function It is expressed by the following formula: Among them, L Y (Y; Z, h, B) represents the negative log-likelihood function; represents the strength constraint; λ and γ represent the weights of the control constraint terms, respectively; Φ(Y) represents an additional prior term with a total variation regularization constraint.

6. The weak single-photon signal preprocessing and three-dimensional imaging method according to claim 5, characterized in that: In step 3.4, the deconvolution solution is solved by the SPIRAL-TAP three-dimensional deconvolution method.

7. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instructions are executed by a processor, the steps of the weak single-photon signal preprocessing and three-dimensional imaging method according to any one of claims 1 to 6 are implemented.

Images