A method for deconvolution and symmetric reconstruction of laser triangulation displacement signals and a laser displacement detection device

Abstract

The application discloses a laser triangular displacement signal deconvolution and symmetric reconstruction method and a laser displacement detection device, acquires optical point spread functions of one-dimensional linear array photoelectric detectors in a measuring range of the device at different physical positions, and constructs a space time-varying optical point spread function reference library; a one-dimensional original light intensity signal of a measured surface is collected, corresponding optical point spread functions are called from the reference library after initial positioning of a light spot, iterative deconvolution is performed on the original light intensity signal, termination iteration is determined by a normalized residual change rate and a preset maximum iteration number, and a physical sharpened light spot signal is obtained; signal variance and statistical kurtosis of a main peak local area are calculated, and a dynamic pixel window width is adaptively adjusted; an asymmetric index is calculated in the dynamic pixel window, forced symmetric reconstruction is performed when the asymmetric index is greater than a preset truncation threshold, sub-pixel centroid coordinates are extracted based on a reconstructed symmetric waveform, and an actual displacement value is calculated.

Description

Technical Field

[0001] This invention relates to a method for deconvolution and symmetric reconstruction of laser triangular displacement signals and a laser displacement detection device. Background Technology

[0002] Displacement sensors based on laser triangulation are widely used in industrial automation inspection, semiconductor wafer measurement, and 3D contour scanning due to their advantages such as non-contact operation, high precision, and fast response. To maintain clear imaging over a large measurement range, high-end laser displacement sensors typically employ a tilted optical path architecture that satisfies Schahm's law. However, in this architecture with a tilted viewing angle, the light spot signal received by the one-dimensional linear photodetector is highly susceptible to asymmetric waveform distortion due to the deep influence of the complex optical properties of the measured surface (such as subsurface scattering and abrupt changes in reflectivity), posing a significant performance bottleneck for subsequent centroid extraction.

[0003] Currently, the mainstream algorithm for extracting sub-pixel center coordinates of light spots in industry is the traditional gray-scale centroid method and its various derivative algorithms. The underlying effectiveness of these algorithms relies on an extremely stringent physical assumption: "the energy distribution of the received light spot exhibits an ideal symmetrical Gaussian shape." However, in complex industrial measurement scenarios, this assumption is often broken, leading to two major insurmountable technical bottlenecks for existing technologies:

[0004] (1) The inherent contradiction between noise resistance stability and measurement resolution: To improve the measurement resolution of displacement sensors, the peak of the light spot must be sharpened as much as possible. Existing traditional signal processing links are usually trapped in a vicious cycle: if low-pass filters such as Gaussian filtering are used to suppress the speckle noise of the measured surface, it will inevitably lead to the broadening of the light spot peak and the decrease in measurement resolution; in order to compensate for the resolution, if mathematical difference or high-pass filtering is forcibly introduced for secondary sharpening, the high-frequency speckle noise will be amplified many times over. This forms a technical deadlock of "smoothing loses resolution, sharpening loses stability". Once the measured surface becomes rough, the extracted centroid will experience high-frequency and violent jitter.

[0005] (2) Energy truncation leads to severe "black-and-white error" and contour distortion: When the sensor scans across black-and-white color transition areas, reflectivity step edges, or measures translucent materials (such as subsurface scattering phenomena caused by resin and plastic), the energy of the light spot received by the one-dimensional linear array photodetector will undergo severe asymmetric truncation or long tail dragging (for example, strong absorption in black areas leads to one side being dark, and high reflectivity in white areas leads to one side being extremely bright). Since the existing centroid method performs gray-scale weighted integration on the entire fixed window, this severe energy asymmetry will act like a "magnet," forcibly pulling the calculated centroid coordinates towards abnormally bright areas or trailing areas. This causes the measurement results to produce false jump pits or bumps of tens of micrometers or even larger on the originally flat surface, completely destroying the fidelity of the true three-dimensional contour.

[0006] In summary, the existing laser triangulation signal processing architecture lacks a dynamic sensing and adaptive correction mechanism for waveform distortion, and cannot eliminate cross-boundary jump errors caused by energy asymmetry truncation from the mathematical level while eliminating high-frequency speckle interference. Summary of the Invention

[0007] The present invention provides a method for deconvolution and symmetric reconstruction of laser triangular displacement signals and a laser displacement detection device to solve the problems existing in the prior art.

[0008] The technical solutions adopted in this invention are as follows:

[0009] A method for deconvolution and symmetric reconstruction of laser triangular displacement signals, applied to a laser displacement detection device based on Scherm's law architecture, includes the following steps:

[0010] S1: Obtain the optical point spread function of the one-dimensional linear array photodetector in the laser displacement detection device at different physical positions within the measurement range, and construct a spatial time-varying optical point spread function reference library;

[0011] S2: The one-dimensional original light intensity signal of the surface under test is acquired by the one-dimensional linear array photodetector. The one-dimensional original light intensity signal is initially located to obtain the initial location of the light spot. According to the initial location of the light spot, the corresponding optical point spread function is called from the spatial time-varying optical point spread function reference library. Iterative deconvolution operation is performed on the one-dimensional original light intensity signal. When the normalized residual change rate is less than the preset residual threshold, or the number of iterations reaches the preset maximum number of iterations, the iteration is terminated to obtain the physically sharpened light spot signal.

[0012] S3: Calculate the signal variance and statistical kurtosis of the physically sharpened spot signal in the local region of the main peak, and adaptively adjust the dynamic pixel window width used for subsequent centroid calculation based on the joint evaluation results of the signal variance and statistical kurtosis.

[0013] S4: Within the dynamic pixel window, calculate the energy integral on both sides of the main peak and obtain the asymmetry index. When the asymmetry index is greater than the preset truncation threshold, perform forced symmetric reconstruction on the physically sharpened spot signal, extract the sub-pixel centroid coordinates based on the reconstructed symmetric waveform, and calculate the actual displacement value based on the sub-pixel centroid coordinates.

[0014] Furthermore, in S2, before performing initial spot localization on the one-dimensional original light intensity signal, median filtering preprocessing is also performed on the one-dimensional original light intensity signal.

[0015] The initial spot localization is achieved by extracting the approximate pixel position of the main peak from the filtered signal using the extreme value method.

[0016] Furthermore, in S2, the rate of change of the normalized residual is the rate of change of the normalized residual obtained in two adjacent iterations; when the rate of change of the normalized residual is less than the preset residual threshold, it is determined that the deconvolution has converged and the iteration is forcibly terminated.

[0017] Furthermore, in S3, the dynamic pixel window width adaptive adjustment strategy is as follows:

[0018] When the signal variance is greater than a preset variance threshold and the statistical kurtosis is less than a preset kurtosis threshold, the surface under test is determined to be a rough surface, and the width of the dynamic pixel window used for subsequent centroid calculation is expanded to a preset maximum width.

[0019] When the signal variance is less than or equal to a preset variance threshold and the statistical kurtosis is greater than or equal to a preset kurtosis threshold, the surface under test is determined to be a smooth surface, and the width of the dynamic pixel window used for subsequent centroid calculation is narrowed to a preset minimum width.

[0020] In other cases, the measured surface is determined to be in a mixed transition state, and the width of the dynamic pixel window used for subsequent centroid calculation is adjusted to a preset medium width.

[0021] Furthermore, in S4, the asymmetric index is calculated in the following manner:

[0022] The asymmetry index is equal to the absolute value of the difference between the energy integral on the left and the energy integral on the right of the main wave peak, divided by the sum of the two.

[0023] When the asymmetry index is greater than the preset cutoff threshold, if the energy integral on the left side of the main peak is greater than the energy integral on the right side, the right waveform data is discarded, and the left waveform data is mathematically mirrored and filled to the right side with the peak pixel as the axis of symmetry, thus completing the forced symmetry reconstruction.

[0024] If the energy integral on the right side of the main peak is greater than the energy integral on the left side, the waveform data on the left side is discarded. Using the peak pixel as the axis of symmetry, the waveform data on the right side is mathematically mirrored and flipped and then filled to the left side, thus completing the forced symmetry reconstruction.

[0025] Furthermore, in S4, sub-pixel centroid coordinates are extracted based on the reconstructed symmetrical waveform, using the square-weighted gray-level centroid method;

[0026] The square-weighted gray-scale centroid method uses the square of the reconstructed pixel light intensity as the weight for discrete integration, thereby exponentially amplifying the weight of the center peak.

[0027] Further, in S1, the optical point spread function of the one-dimensional linear array photodetector at different physical locations within the measurement range is obtained, including:

[0028] The measurement range is divided into multiple depth intervals, and each depth interval is back-mapped to the effective pixel area of ​​the one-dimensional linear photodetector using a geometric calibration matrix.

[0029] The corresponding optical point spread function is extracted at the center of each depth interval, and after normalization, the spatial time-varying optical point spread function reference library is constructed.

[0030] The present invention also discloses a laser displacement detection device, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the above-described method.

[0031] The present invention has the following beneficial effects:

[0032] (1) The present invention adopts iterative deconvolution based on dynamic termination of normalized residual change rate, uses the optical point spread function to recover the sharp peak in reverse, and suppresses the excessive amplification of high frequency speckle noise caused by fixed iteration number through residual monitoring logic from the bottom layer of the algorithm, so that it has higher stability when achieving high resolution measurement under complex surface conditions.

[0033] (2) This invention introduces a dual joint evaluation mechanism of signal variance and statistical kurtosis, enabling the sensor to adaptively adjust the dynamic pixel window width used for centroid calculation according to the microscopic morphology of the measured surface. The window is narrowed on smooth surfaces to make full use of the high resolution advantage brought by deconvolution, and the window is expanded on rough surfaces to absorb random fluctuations through macroscopic spatial averaging, thereby taking into account both peak feature extraction and noise reduction and smoothing capabilities, and improving the sensor's environmental adaptability to different surface roughness.

[0034] (3) This invention quantitatively captures the energy tearing phenomenon of light spot under reflectivity abrupt boundary or subsurface scattering conditions by asymmetric exponent. Combined with mirror symmetry reconstruction and high-order weighted centroid extraction, it blocks the pull of energy asymmetric distortion on the integral centroid, improves the three-dimensional contour fidelity when crossing complex material boundaries or measuring semi-transparent materials, and effectively suppresses the centroid offset error caused by energy asymmetric truncation. Attached Figure Description

[0035] Figure 1 This is a flowchart of the present invention.

[0036] Figure 2 This is a comparison image of the measurement contours generated when the traditional centroid method and the algorithm of this invention cross the black-and-white boundary. Detailed Implementation

[0037] The invention will now be further described with reference to the accompanying drawings.

[0038] This embodiment relies on a high-end laser displacement sensor hardware platform, whose hardware architecture specifically includes: a red semiconductor laser source with an emission wavelength of 660nm; an inclined receiving optical path arranged based on Schamm's law; and a 1536-pixel one-dimensional linear array CMOS as the core one-dimensional linear array photodetector, with a physical width of 5.5 micrometers for each pixel, and a built-in high-speed ADC and FPGA / DSP signal processing unit.

[0039] The specific optical parameters of this embodiment are configured as follows: the physical measurement range of the sensor is Z∈[65mm,135mm]. The reference is that when the target object is at a reference distance Z=100mm and perpendicular to the emission axis, the surface scattered light just enters the receiving lens perpendicularly. The midline base distance between the emission optical axis and the receiving main optical axis is B=20mm, and the back focal length of the receiving lens is f′=25mm.

[0040] The 660nm red laser used in this platform has a stronger material penetration capability compared to shorter wavelength blue light. When irradiating white plastic or translucent media, red light easily diffuses into the material and induces severe subsurface scattering, resulting in a blurred spot distortion with severe asymmetric tailing on a one-dimensional linear photodetector. Therefore, implementing the dynamic symmetry reconstruction and high-order weighting algorithm of this invention on such a high-penetration red light hardware platform can compensate for the distortion defects of physical optics using digital signal processing without increasing hardware costs.

[0041] like Figure 1The diagram shows the overall flowchart of a deconvolution and symmetric reconstruction method for laser triangular displacement signals provided in an embodiment of the present invention. The process mainly includes four core steps: multi-frame average calibration of a spatial time-varying point spread function (PSF) reference library; deconvolution processing based on lightweight pre-filtering initial localization and dynamic termination of residuals; adaptive window adjustment based on joint evaluation of local variance and kurtosis; and forced symmetric reconstruction and high-order centroid extraction based on asymmetric exponents.

[0042] S1: Reference library for multi-frame averaging calibration spatial time-varying point spread function PSF(x).

[0043] Using a precision micro-stage and a standard diffuse whiteboard, the spatially time-varying point spread function is obtained. Considering the nonlinear characteristics of optical magnification in laser triangulation, the system performs non-uniform spatial partitioning of the linear array detector based on the specific optical parameters of the aforementioned hardware environment.

[0044] Based on Scherrer's law and triangular geometric projection calculations, within the complete physical measurement range of 65mm to 135mm, the actual total physical span of the light spot movement on the linear CMOS array is approximately 5.69mm. Given that the width of a single pixel is 5.5μm, this effective measurement area actually covers approximately 1035 consecutive effective pixels, with the remaining area used as optical path assembly tolerance redundancy.

[0045] This embodiment employs an equal physical depth mapping strategy: the total physical range of 65mm to 135mm is divided into 16 depth intervals, each with a physical depth span of 4.375mm. These 16 physical intervals are then mapped back onto the aforementioned 1035 effective pixels using a geometric calibration matrix. Due to the influence of nonlinear magnification, this mapping results in a non-uniform pixel control array: specifically, calculations using tilted optical path projection show that the pixel sensitivity at the near end (Z=65mm) is approximately 2.8 times that at the far end (Z=135mm). The depth interval at the closest end (65mm to 65.375mm) is in a high optical magnification region, where the light spot will move approximately 0.57mm on the linear CMOS array, occupying approximately 104 consecutive pixels as its control area. Conversely, the depth interval at the farthest end (130.625mm to 135mm) experiences sensitivity attenuation, causing the light spot to move only approximately 0.22mm on the linear CMOS array, occupying approximately 39 consecutive pixels as its control area.

[0046] The corresponding optical point spread function is extracted at the center of each mapping interval. After normalization, a one-dimensional array with a length of 31 pixels is truncated as the local optical point spread function for that interval. These 16 groups The reference library of spatial time-varying optical point spread functions is constructed by storing the function as a static array in the sensor's Flash memory.

[0047] Step S2: Deconvolution processing based on lightweight pre-filtering initial localization and dynamic termination of residuals.

[0048] When the sensor is scanning at high speed, the one-dimensional linear photodetector outputs a one-dimensional grayscale sequence. DSP firstly A lightweight median filter with a window width of 3 pixels is performed as a preprocessing step to remove random high-frequency noise pulses caused by speckle effect, preventing the initial localization algorithm from being interfered with by isolated noise. Then, the extreme value method is used to find the approximate pixel position of the main peak in the filtered signal, determine which of the aforementioned non-uniform jurisdiction intervals it falls into, and retrieve the corresponding 31 pixels for that interval. As a local deconvolution kernel.

[0049] Subsequently, within the local grayscale sequence containing the main peak, the 31-pixel deconvolution kernel is used to enter the Richardson-Lucy iterative loop. After each iteration, the normalized residual Eres and the rate of change of the residual between adjacent iterations are calculated. Preset residual change threshold When judged At this point, the loop is forcibly exited, resulting in a sharpened signal where the waveform narrows and no noise fluctuations occur. Meanwhile, a preset maximum number of iterations is set to 50 as a safety fallback mechanism. When the number of iterations reaches 50, the iteration is forcibly terminated regardless of whether the residual change rate meets the threshold condition, in order to suppress the excessive amplification of high-frequency speckle noise in the deconvolution operation.

[0050] Step S3: Adaptive window adjustment based on joint evaluation of local variance and kurtosis.

[0051] In sharpening signal Around the main peak, i.e., within a fixed pixel local area at the coarse extraction center (in this embodiment, within a local area of ​​±20 pixels at the coarse extraction center), let the local pixel grayscale mean be μ, and the total number of pixels in the local area be N. The signal variance is calculated according to the following formula. and statistical kurtosis :

[0052] ,

[0053] .

[0054] It represents the grayscale value (signal strength) of the i-th pixel within the local region.

[0055] Set the first threshold, i.e., the variance threshold Th var =150, set the second threshold, i.e., the kurtosis threshold Th. ku =2.5.

[0056] Scenario 1: If the calculation yields... and This indicates that when irradiated on a rough surface such as brushed metal, the signal exhibits strong speckle noise and a flat waveform. Therefore, the system sets the dynamic pixel window W used for subsequent centroid calculation to 15 pixels on each side of the center. To smooth out speckles.

[0057] Scenario 2: If and This indicates that the light is illuminating a smooth mirror surface with extremely sharp peaks. Therefore, the dynamic pixel window W is narrowed to 5 pixels on each side of the center. Make full use of the sharpened center high-frequency energy.

[0058] Scenario 3: If other crossover scenarios are calculated, such as... and ,or and This represents a mixed transition state such as a semi-polished surface or an extremely dark matte surface. The system adaptively adjusts the dynamic pixel window W to a medium width, that is, 10 pixels on each side of the center. To achieve a balance between noise reduction and peak extraction.

[0059] Step S4: Forced symmetry reconstruction and higher-order centroid extraction based on asymmetric exponents.

[0060] Within the locked dynamic pixel window W, using the peak pixel of the peak as the dividing point, calculate the total gray-level integral E on the left side. L The sum of grayscale integrals on the right, E R Calculate the asymmetric exponent I. asy :

[0061] ,

[0062] in, It is the sum of the grayscale integrals to the left of the main peak-to-peak pixel within the dynamic pixel window; It is the sum of grayscale integrals to the right of the peak-to-peak pixel within the dynamic pixel window.

[0063] Set the edge truncation threshold Th edge =0.3, execute the corresponding reconstruction judgment branch based on the detection result:

[0064] Scenario 1: Abnormal truncation, perform reconstruction: If the current frame is detected as... This indicates that the light spot hits the black-and-white boundary line precisely or that severe subsurface scattering occurs. At this point, the system further compares the energy levels on both sides for directional reconstruction: if... For example, if the left side is a white high-reflectivity area and the right side is a black absorption area resulting in energy loss, then the low-energy waveform on the right is discarded, and the high-energy waveform data on the left is mirrored and copied to the right side, using the peak pixel as the axis of symmetry; conversely, if... If the low-energy waveform on the left is discarded, the waveform on the right is mirrored and copied to the left. Through the above orientation judgment, a physically forced symmetric reconstruction array is generated. .

[0065] Scenario 2, standard measurement, no reconstruction required: If the current frame is detected... This indicates that the current energy distribution of the light spot is relatively uniform and symmetrical, and has not been severely affected by physical edges or scattering interference. At this point, the system retains the original waveform and directly assigns the sharpening signal within window W to the computation array. .

[0066] Finally, the processed array Sub-pixel centroids are extracted using the square-weighted gray-level centroid method. The specific calculation formula is as follows: .

[0067] in, The coordinates of the i-th pixel within the dynamic pixel window; It represents the reconstructed light intensity value (grayscale value) of the i-th pixel within the dynamic pixel window.

[0068] The quadratic operation further exponentially amplifies the weight of the centrally highlighted real physical reflection point. Finally, the extracted... Substitute the factory-pre-calibrated high-order polynomial model middle( , , …are preset high-order polynomial calibration coefficients), and the mapped output is the actual sensor physical displacement in millimeters. .

[0069] like Figure 2 The image shows a comparison of the measured contours when crossing the black-and-white boundary region using the traditional gray-scale centroid method and the reconstruction algorithm of this invention. It can be seen that the traditional centroid method, due to the pull of the residual high-reflectivity tail, produces a cliff-like jump peak of approximately 80 μm. However, by employing the forced mirror reconstruction and high-order weighting algorithm of this invention, the pull caused by energy distortion is effectively eliminated, successfully blocking the jump error, greatly smoothing the measurement trajectory, and ensuring the fidelity of the three-dimensional contour.

[0070] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A method for deconvolution and symmetric reconstruction of laser triangular displacement signals, the method being applied to a laser displacement detection device based on the Schahm's law architecture, characterized in that: Includes the following steps: S1: Obtain the optical point spread function of the one-dimensional linear array photodetector in the laser displacement detection device at different physical positions within the measurement range, and construct a spatial time-varying optical point spread function reference library; S2: The one-dimensional original light intensity signal of the surface under test is acquired by the one-dimensional linear array photodetector. The one-dimensional original light intensity signal is initially located to obtain the initial location of the light spot. According to the initial location of the light spot, the corresponding optical point spread function is called from the spatial time-varying optical point spread function reference library. Iterative deconvolution operation is performed on the one-dimensional original light intensity signal. When the normalized residual change rate is less than the preset residual threshold, or the number of iterations reaches the preset maximum number of iterations, the iteration is terminated to obtain the physically sharpened light spot signal. S3: Calculate the signal variance and statistical kurtosis of the physically sharpened spot signal in the local region of the main peak, and adaptively adjust the dynamic pixel window width used for subsequent centroid calculation based on the joint evaluation results of the signal variance and statistical kurtosis. S4: Within the dynamic pixel window, calculate the energy integral on both sides of the main peak and obtain the asymmetry index. When the asymmetry index is greater than the preset truncation threshold, perform forced symmetric reconstruction on the physically sharpened spot signal, extract the sub-pixel centroid coordinates based on the reconstructed symmetric waveform, and calculate the actual displacement value based on the sub-pixel centroid coordinates.

2. The method for deconvolution and symmetric reconstruction of laser triangular displacement signals as described in claim 1, characterized in that: In S2, before performing initial spot localization on the one-dimensional original light intensity signal, the method further includes performing median filtering preprocessing on the one-dimensional original light intensity signal. The initial spot localization is achieved by extracting the approximate pixel position of the main peak from the filtered signal using the extreme value method.

3. The method for deconvolution and symmetric reconstruction of laser triangular displacement signals as described in claim 1, characterized in that: In S2, the rate of change of the normalized residual is the rate of change of the normalized residual obtained in two adjacent iterations; when the rate of change of the normalized residual is less than the preset residual threshold, it is determined that the deconvolution has converged and the iteration is forcibly terminated.

4. The method for deconvolution and symmetric reconstruction of laser triangular displacement signals as described in claim 1, characterized in that: In S3, the dynamic pixel window width adaptive adjustment strategy is as follows: When the signal variance is greater than a preset variance threshold and the statistical kurtosis is less than a preset kurtosis threshold, the surface under test is determined to be a rough surface, and the width of the dynamic pixel window used for subsequent centroid calculation is expanded to a preset maximum width. When the signal variance is less than or equal to a preset variance threshold and the statistical kurtosis is greater than or equal to a preset kurtosis threshold, the surface under test is determined to be a smooth surface, and the width of the dynamic pixel window used for subsequent centroid calculation is narrowed to a preset minimum width. In other cases, the measured surface is determined to be in a mixed transition state, and the width of the dynamic pixel window used for subsequent centroid calculation is adjusted to a preset medium width.

5. The method for deconvolution and symmetric reconstruction of laser triangular displacement signals as described in claim 1, characterized in that: In S4, the asymmetric index is calculated in the following way: The asymmetry index is equal to the absolute value of the difference between the energy integral on the left and the energy integral on the right of the main wave peak, divided by the sum of the two. When the asymmetry index is greater than the preset cutoff threshold, if the energy integral on the left side of the main peak is greater than the energy integral on the right side, the right waveform data is discarded, and the left waveform data is mathematically mirrored and filled to the right side with the peak pixel as the axis of symmetry, thus completing the forced symmetry reconstruction. If the energy integral on the right side of the main peak is greater than the energy integral on the left side, the waveform data on the left side is discarded. Using the peak pixel as the axis of symmetry, the waveform data on the right side is mathematically mirrored and flipped and then filled to the left side, thus completing the forced symmetry reconstruction.

6. The method for deconvolution and symmetric reconstruction of laser triangular displacement signals as described in claim 1 or 5, characterized in that: In S4, sub-pixel centroid coordinates are extracted based on the reconstructed symmetrical waveform, using the square-weighted gray-level centroid method. The square-weighted gray-scale centroid method uses the square of the reconstructed pixel light intensity as the weight for discrete integration, thereby exponentially amplifying the weight of the center peak.

7. The method for deconvolution and symmetric reconstruction of laser triangular displacement signals as described in claim 1, characterized in that: In S1, the optical point spread function of the one-dimensional linear array photodetector at different physical positions within the measurement range is obtained, including: The measurement range is divided into multiple depth intervals, and each depth interval is back-mapped to the effective pixel area of ​​the one-dimensional linear photodetector using a geometric calibration matrix. The corresponding optical point spread function is extracted at the center of each depth interval, and after normalization, the spatial time-varying optical point spread function reference library is constructed.

8. A laser displacement detection device, comprising a memory and a processor, characterized in that: The memory stores a computer program, which, when executed by the processor, causes the processor to perform the method according to any one of claims 1 to 7.

Images