An autocorrelational 3D reconstruction method and system for targets obscured by lidar
By processing the data of lidar-masked targets using an autocorrelation 3D reconstruction method, the problems of completeness and accuracy of 3D reconstruction of masked targets are solved, achieving efficient point cloud data processing and feature extraction, and adapting to complex environments.
Patent Information
- Application Number
- CN202611320361.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-28
- Publication Date
- 2026-11-17
AI Technical Summary
Existing laser 3D imaging technology cannot effectively construct a full-range stereo model in scenarios where targets are obscured, resulting in limited completeness and accuracy of 3D reconstruction. Furthermore, Geiger APD lidar faces issues such as redundancy of point cloud data, difficulty in detecting obscured targets, and insufficient algorithm robustness.
An autocorrelation-based 3D reconstruction method for LiDAR-masked targets is adopted. The original acquired data is standardized and outlier is processed to construct a pixel-level histogram and a global total histogram. A sliding window with adaptive width is used for peak extraction and temporal correlation verification to generate 2D point cloud data with reflectivity information. Compressed sensing encoding is then performed to optimize the reconstruction process.
It effectively removes invalid data, improves point cloud utilization, reduces computational load, meets the needs of real-time applications, solves the balance problem between high-precision feature extraction and lightweight data, and adapts to more environments.
Smart Images

Figure CN123118906A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of lidar 3D reconstruction technology, specifically relating to an autocorrelation 3D reconstruction method and system based on lidar-masked targets. Background Technology
[0002] Laser 3D imaging technology, a core area of modern sensing, works by emitting a laser beam to scan a target scene, receiving the reflected echo signal, and analyzing key information such as time to accurately reconstruct the target's spatial position and shape. This technology, with its high resolution and strong environmental adaptability, has been widely applied in various fields. The core principle of laser 3D imaging utilizes the rectilinear propagation characteristic of lasers. By precisely measuring the time difference between laser emission and echo reception (time of flight), the distance from the target point to the radar is calculated. Combined with the scanning angle of the laser beam, the 3D coordinates of the target point can be determined. Finally, a large number of target points are combined to generate a point cloud image, thus presenting the spatial morphology of the target.
[0003] Essentially, laser 3D imaging only detects the surface contours of a target along the line of sight. In other words, only the target area directly illuminated by the laser beam can be effectively captured and have corresponding point cloud data generated. For target structures outside the line of sight, such as areas obscured by obstacles, the laser beam cannot penetrate the obstruction to reach that area, thus preventing the acquisition of relevant distance and contour information. This makes it impossible to directly construct a comprehensive 3D model of the target. This technological limitation is particularly evident in scenarios involving the detection of obscured targets, severely impacting the completeness and accuracy of 3D reconstruction. When a laser pulse penetrates the obstruction, some photons directly illuminate the target surface and are reflected back to the detector; even if the laser pulse is blocked, a small number of photons still indirectly affect the target after diffuse reflection through the mesh. By collecting and extracting these weak echo signals, and then based on the principles of time-of-flight (ToF) and single-photon response statistical correlation imaging, a 3D image of the target is reconstructed.
[0004] As an important branch of lidar technology, Geiger APD lidar uses avalanche photodiodes operating in Geiger mode as its core detection device. Its detection sensitivity reaches the single-photon level, effectively capturing weak laser echo signals. This characteristic gives it significant advantages in complex scenarios such as long-range detection, low-light environments, and detection of obscured targets. For example, it can still effectively detect targets in low-visibility environments such as foggy days and nighttime. However, high sensitivity also brings a series of challenges: on the one hand, the generated point cloud data is extremely large and contains a large amount of geometric redundancy, which puts enormous pressure on data transmission and real-time processing; on the other hand, the nature of laser 3D imaging dictates that it can only detect the surface contours of targets in the line-of-sight direction. For target structures outside the line-of-sight view, effective capture is difficult because the laser beam cannot directly reach them, making it challenging to construct a complete 3D model.
[0005] Geiger APD lidar 3D reconstruction technology faces three major challenges. First, point cloud data redundancy is a significant issue; the massive amount of data leads to low transmission and processing efficiency, making it difficult to meet the needs of real-time application scenarios. Second, detecting occluded targets is challenging; the echo signals from occluded objects and the occluded targets overlap, resulting in insufficient accuracy in target feature extraction and an inability to accurately distinguish the contour information of occluded objects and the occluded targets. Third, the robustness of the algorithm needs improvement; in complex environments such as strong noise and the coexistence of targets at multiple depths, reconstruction distortion can easily occur, affecting the accuracy of subsequent target recognition and analysis.
[0006] A new 3D reconstruction method is needed to solve the above-mentioned technical problems. Summary of the Invention
[0007] The purpose of this invention is to provide an autocorrelation-based 3D reconstruction method for masked targets based on lidar, which solves the technical problems of low accuracy and slow speed of lidar in 3D reconstruction of masked targets.
[0008] The present invention also aims to provide a reconstruction system employing an autocorrelation-based three-dimensional reconstruction method for targets obscured by lidar.
[0009] The technical solution of this invention to solve its technical problem is as follows:
[0010] An autocorrelation-based 3D reconstruction method for targets masked by lidar includes the following steps:
[0011] S1: Acquire the raw data collected by the LiDAR, decompose the raw data into a two-dimensional data matrix and perform standardization and outlier processing; construct a pixel-level histogram for each pixel based on the processed two-dimensional data matrix, accumulate the pixel-level histograms to obtain the global total histogram and smooth the global total histogram.
[0012] S2: An adaptive-width sliding window is used to traverse the smoothed global total histogram. Candidate peaks are obtained by calculating the mean and variance within the window and combining them with neighborhood comparison. The candidate peaks are deduplicated and verified based on the echo count ratio to obtain the effective peaks and their corresponding time indices. The multi-target depth information of each pixel is obtained by using the temporal correlation of the target echo signals.
[0013] S3: Use an adaptive width sliding window to traverse the cell-level histogram to obtain the initial peak value, perform time correlation verification on the initial peak value to obtain the target peak value, and use the multi-target depth information of the cell to process the target peak value to obtain the final target peak value.
[0014] S4: After filtering the target time and eliminating noise for the final target peak, calculate the reflectance of each pixel and perform reflectance normalization. Reconstruct the reflectance vector after normalization into a reflectance image. Combine the reflectance of each pixel with its corresponding pixel multi-target depth information to generate two-dimensional point cloud data with reflectance information.
[0015] S5: Compress and encode the two-dimensional point cloud data to obtain measurement values, use optimization algorithms to reconstruct the original point cloud from the measurement values, and use a reconstruction accuracy analysis model to evaluate and optimize the reconstruction process.
[0016] Preferably, step S1 specifically includes:
[0017] S1.1: The raw acquisition data of the lidar will be acquired, and the raw acquisition data will be converted from three-dimensional data into a two-dimensional data matrix. The two-dimensional data matrix will be normalized, and outlier values in the two-dimensional data will be removed using an outlier detection algorithm. Linear interpolation will be used to fill in the missing data positions of outlier values to obtain the processed two-dimensional data matrix.
[0018] S1.2: Construct a cell-level histogram based on each column of the processed two-dimensional data matrix; and sum the obtained cell-level histograms according to the vertical axis to obtain a global histogram, and then smooth the global histogram.
[0019] Preferably, the normalization process for the two-dimensional data matrix specifically involves:
[0020] The echo count of each pixel in the converted two-dimensional data matrix is converted into a relative value in the range of 0-1.
[0021] Preferably, the outlier removal algorithm in the two-dimensional data specifically involves:
[0022] Calculate the mean and standard deviation of each cell in each column of the two-dimensional data, determine whether each data point exceeds the normal range, if so, identify it as an outlier and remove it, if not, identify it as a normal value and save it.
[0023] Preferably, the step of using linear interpolation to fill in the missing data positions of outliers specifically involves:
[0024] For the saved normal value column data, linear interpolation is used to fill in the missing data positions of outliers.
[0025] Preferably, the smoothing process specifically involves using a 5-point moving average filtering algorithm to supplement data at the boundary position of the time index using a mirror expansion method.
[0026] Preferably, step S3 specifically comprises:
[0027] S3.1: Use an adaptive-width sliding window to traverse the pixel-level histogram to capture the peak signals of targets at different depths and obtain the initial peak values;
[0028] S3.2: Calculate the echo data of each pixel and statistically analyze the temporal distribution of the initial peak in different frames, and use the set time deviation to filter out the target peak;
[0029] S3.3: Enhance the weak peaks in the target peaks and merge the neighboring target peaks. Based on the depth difference between the peaks, cluster, separate, extract and evaluate the quality of the multi-target signals to obtain the final target peak.
[0030] Preferably, step S4 specifically comprises:
[0031] S4.1: Filter the target time for the final target peak, retain the pixel response corresponding to the target time, and construct the target time response matrix. The specific expression is as follows:
[0032] ;
[0033] in, The target time response matrix, The frame number of the target signal; The pixel number of the target signal;
[0034] S4.2: Count multiple frames of photons from the detector at a certain moment before the echo signal reaches the receiving system, construct the dark field response matrix and the dark field equivalent bucket signal sequence, calculate the second-order correlation function of the dark field, and perform noise cancellation processing; the specific expression is:
[0035] ;
[0036] ;
[0037] in, For the first The second-order correlation function of the dark field in the nth frame at the pixel index of each detector; For the target echo signal in the nth frame, For the first The nth frame at the detector pixel index is paired with the pixel. The target echo is counted in conjunction with photons; d represents the dark field; n is the dark field data frame number; r represents the target echo signal; s is the target echo signal containing noise. Let k be the dark-field target echo signal in the nth frame, and k be the detector pixel number. For paired pixels; The second-order correlation function of the original noisy signal; The target second-order correlation function after removing dark noise;
[0038] S4.3: Map the second-order correlation function of the original noisy signal to the [0,1] interval to obtain the target reflectivity, the specific expression of which is:
[0039] ;
[0040] in, For target reflectivity;
[0041] S4.4: Reconstruct the reflectance vector of the target into a two-dimensional array by row-wise inverse reshaping to obtain a reconstructed reflectance image. Combine the reflectance of each pixel with the multi-target depth information of its corresponding pixel to generate two-dimensional point cloud data with reflectance information.
[0042] A reconstruction system employing an autocorrelation-based 3D reconstruction method for targets obscured by lidar includes:
[0043] The lidar data acquisition module is used to acquire the raw acquisition data of lidar, decompose the raw acquisition data into a two-dimensional data matrix and perform standardization and outlier processing; based on the processed two-dimensional data matrix, a pixel-level histogram is constructed for each pixel, the pixel-level histograms are accumulated to obtain the global total histogram and the global total histogram is smoothed.
[0044] The autocorrelation feature analysis unit is used to traverse the smoothed global total histogram using an adaptive-width sliding window, calculate the mean and variance within the window and combine them with neighborhood comparison to obtain candidate peaks, perform deduplication and verification based on echo count ratio on the candidate peaks to obtain the effective peaks and their corresponding time indices; utilize the temporal correlation of the target echo signals to extract multi-target depth information from the data of each pixel; use an adaptive-width sliding window to traverse the pixel-level histogram to obtain the initial peak, verify the temporal correlation of the initial peak to obtain the target peak, and process the target peak using the pixel multi-target depth information to obtain the final target peak;
[0045] The two-dimensional point cloud data acquisition unit is used to filter the target time and remove noise from the final target peak, calculate the reflectance of each pixel and perform reflectance normalization, reconstruct the reflectance vector after normalization, and combine the reflectance of each pixel with its corresponding pixel multi-target depth information to generate two-dimensional point cloud data with reflectance information.
[0046] The evaluation and optimization unit is used to compress and encode the two-dimensional point cloud data to obtain measurement values, reconstruct the original point cloud from the measurement values using optimization algorithms, and evaluate and optimize the reconstruction process using a reconstruction accuracy analysis model.
[0047] Preferably, the outlier processing includes using an outlier detection algorithm to remove outliers from the two-dimensional data and using linear interpolation to fill in the missing data positions of the outliers.
[0048] The beneficial effects of this invention are as follows: By processing the raw acquisition data from the lidar, pixel-level histograms and global total histograms are obtained; autocorrelation feature analysis is performed on the pixel-level and global total histograms to obtain the final target peak value. Autocorrelation feature analysis effectively removes invalid data, improving point cloud utilization. The final target peak value is processed to obtain two-dimensional point cloud data with reflectivity information; the two-dimensional point cloud data is then compressed using sensing encoding to obtain measurement values. Compressed sensing performs lightweight processing based on feature points, significantly reducing computational load, improving reconstruction, meeting real-time application requirements, and solving the balance problem between "high-precision feature extraction" and "lightweight data". An optimization algorithm is used to reconstruct the original point cloud from the measurement values, and a reconstruction accuracy analysis model is used to evaluate and optimize the reconstruction process, making this reconstruction method adaptable to more environments. Attached Figure Description
[0049] Figure 1 This is a flowchart illustrating the reconstruction method of the present invention;
[0050] Figure 2 This is a schematic diagram of the scene layout scheme of Embodiment 1 of the present invention;
[0051] Figure 3 This is a schematic diagram of the reconstruction effect of Embodiment 1 of the present invention. Detailed Implementation
[0052] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
[0053] like Figure 1 As shown, this invention discloses an autocorrelation-based 3D reconstruction method for targets obscured by lidar, comprising the following steps:
[0054] S1: Acquire the raw data collected by the LiDAR, decompose the raw data into a two-dimensional data matrix and perform standardization and outlier processing; construct a pixel-level histogram for each pixel based on the processed two-dimensional data matrix, accumulate the pixel-level histograms to obtain the global total histogram and smooth the global total histogram.
[0055] Step S1 specifically involves:
[0056] S1.1: The raw acquisition data of the lidar will be acquired, and the raw acquisition data will be converted from three-dimensional data into a two-dimensional data matrix. The two-dimensional data matrix will be normalized, and outlier values in the two-dimensional data will be removed using an outlier detection algorithm. Linear interpolation will be used to fill in the missing data positions of outlier values to obtain the processed two-dimensional data matrix.
[0057] The normalization process for the two-dimensional data matrix specifically involves converting the echo count of each pixel in the transformed two-dimensional data matrix into a relative value within the 0-1 range. The outlier detection algorithm for removing outliers from the two-dimensional data specifically involves calculating the mean and standard deviation of each individual pixel in each column of the two-dimensional data, determining whether each data point exceeds the normal range, identifying it as an outlier and removing it if so, and identifying it as a normal value and saving it if not.
[0058] The method of using linear interpolation to fill in the missing data positions of outliers specifically involves using linear interpolation to fill in the missing data positions of outliers for the saved normal value column data.
[0059] S1.2: Construct a cell-level histogram based on each column of the processed two-dimensional data matrix; and sum the obtained cell-level histograms according to the vertical axis to obtain a global histogram, and then smooth the global histogram.
[0060] The smoothing process specifically involves using a 5-point moving average filtering algorithm to supplement data at the boundary positions of the time index using a mirror expansion method.
[0061] S2: An adaptive-width sliding window is used to traverse the smoothed global total histogram. Candidate peaks are obtained by calculating the mean and variance within the window and combining them with neighborhood comparison. The candidate peaks are deduplicated and verified based on the echo count ratio to obtain the effective peaks and their corresponding time indices. The multi-target depth information of each pixel is obtained by using the temporal correlation of the target echo signals.
[0062] S3: Use an adaptive width sliding window to traverse the cell-level histogram to obtain the initial peak value, perform time correlation verification on the initial peak value to obtain the target peak value, and use the multi-target depth information of the cell to process the target peak value to obtain the final target peak value.
[0063] Step S3 specifically involves:
[0064] S3.1: Use an adaptive-width sliding window to traverse the pixel-level histogram to capture the peak signals of targets at different depths and obtain the initial peak values;
[0065] S3.2: Calculate the echo data of each pixel and statistically analyze the temporal distribution of the initial peak in different frames, and use the set time deviation to filter out the target peak;
[0066] S3.3: Enhance the weak peaks in the target peaks and merge the neighboring target peaks. Based on the depth difference between the peaks, cluster, separate, extract and evaluate the quality of the multi-target signals to obtain the final target peak.
[0067] S4: After filtering the target time and eliminating noise for the final target peak, calculate the reflectance of each pixel and perform reflectance normalization. Reconstruct the reflectance vector after normalization into a reflectance image. Combine the reflectance of each pixel with its corresponding pixel multi-target depth information to generate two-dimensional point cloud data with reflectance information.
[0068] Step S4 specifically involves:
[0069] S4.1: Filter the target time for the final target peak, retain the pixel response corresponding to the target time, and construct the target time response matrix. The specific expression is as follows:
[0070] ;
[0071] in, The target time response matrix, The frame number of the target signal; The pixel number of the target signal;
[0072] S4.2: Count multiple frames of photons from the detector at a certain moment before the echo signal reaches the receiving system, construct the dark field response matrix and the dark field equivalent bucket signal sequence, calculate the second-order correlation function of the dark field, and perform noise cancellation processing; the specific expression is:
[0073] ;
[0074] ;
[0075] in, For the first The second-order correlation function of the dark field in the nth frame at the pixel index of each detector; For the target echo signal in the nth frame, For the first The nth frame at the detector pixel index is paired with the pixel. The target echo is counted in conjunction with photons; d represents the dark field; n is the dark field data frame number; r represents the target echo signal; s is the target echo signal containing noise. Let k be the dark-field target echo signal in the nth frame, and k be the detector pixel number. For paired pixels; The second-order correlation function of the original noisy signal; The target second-order correlation function after removing dark noise;
[0076] S4.3: Map the second-order correlation function of the original noisy signal to the [0,1] interval to obtain the target reflectivity, the specific expression of which is:
[0077] ;
[0078] in, For target reflectivity;
[0079] S4.4: Reconstruct the reflectance vector of the target into a two-dimensional array by row-wise inverse reshaping to obtain a reconstructed reflectance image. Combine the reflectance of each pixel with the multi-target depth information of its corresponding pixel to generate two-dimensional point cloud data with reflectance information.
[0080] S5: Compress and encode the two-dimensional point cloud data to obtain measurement values, use optimization algorithms to reconstruct the original point cloud from the measurement values, and use a reconstruction accuracy analysis model to evaluate and optimize the reconstruction process.
[0081] Example 1:
[0082] Take the following configuration as an example:
[0083] Geiger APD lidar: The pixel array adopts a 64×64 specification, the time resolution reaches 1ns, and the detection range is about 2km.
[0084] Industrial computer: The CPU is a high-performance Intel Core i7-12700H with 14 cores and 20 threads, and a clock speed of 2.7GHz, providing powerful computing capabilities.
[0085] Auxiliary equipment: vehicle target (4.2m van), transmission net (70% obstruction rate), laser rangefinder.
[0086] Layout plan as follows Figure 2 As shown, the net is fixed approximately 2km away from the lidar; a vehicle is parked behind the transmission net.
[0087] The specific workflow includes: preparation phase, operation phase, and termination phase, as follows:
[0088] 1. Preparation stage
[0089] After the lidar-masked target self-reconstruction software is powered on, the system first initializes and loads modules. After startup and the device control system are started, it enters the working state, establishes a communication link, and completes the initialization of each unit module. According to the business logic, each module independently completes its own task, and the communication module is responsible for the entire communication business process.
[0090] Call up the laser, detector, and servo settings to complete the initialization parameters, such as laser wavelength, laser type, and frequency.
[0091] 2. Operational Phase
[0092] The equipment control system sets up scenarios as needed and calls upon lasers, detectors, and servo information.
[0093] The prototype of heterogeneous correlation laser 3D imaging processes the detected laser data, determines whether there is a response, and when the response conditions are met, parses the information and packages the data to generate the original point cloud information.
[0094] Computer information processing integrates all collected output data, calculates the 3D imaging data of the obstruction and the target behind the obstruction, and generates reconstruction information from the data;
[0095] The equipment control system and computer information processing system simultaneously display process information.
[0096] 3. Conclusion
[0097] Store the calculation process and results data; release the resources of each software module and exit the system.
[0098] like Figure 3As shown in the diagram, the reconstruction effect illustration simultaneously displays a 3D image based on time-correlated single-photon counting. Both methods can effectively restore the target outline. In terms of algorithm flow, both involve key steps such as filtering and occlusion, locating the target's plane, target extraction, and noise filtering. The core difference lies in the target extraction stage: this scheme obtains a normalized binary image of the target by correlating photon count fluctuations on a specific plane where the target is located; the time-correlated single-photon counting method uses a sliding window multi-peak detection algorithm for each pixel to extract the target's spatial location by filtering and matching peak signals from different distance layers. From the perspective of data processing volume and efficiency, this scheme only needs to complete the correlation imaging operation within a single distance plane, while time-correlated single-photon counting requires multi-peak detection and comparison for each pixel channel within a certain distance range. Therefore, this scheme has a faster processing speed.
[0099] The raw data acquired by the LiDAR is processed to obtain pixel-level histograms and global total histograms. Autocorrelation feature analysis is then performed on these histograms to obtain the final target peak value. This autocorrelation feature analysis effectively removes invalid data, improving point cloud utilization. The final target peak value is then processed to obtain two-dimensional point cloud data with reflectivity information. Compressed sensing encoding of this two-dimensional point cloud data is used to obtain measurement values. Compressed sensing performs lightweight processing based on feature points, significantly reducing computational load, improving reconstruction, and meeting real-time application requirements, thus solving the balance problem between "high-precision feature extraction" and "lightweight data". An optimized algorithm is used to reconstruct the original point cloud from the measurement values, and the reconstruction process is evaluated and optimized using a reconstruction accuracy analysis model, making this reconstruction method adaptable to more environments.
[0100] A reconstruction system employing an autocorrelation-based 3D reconstruction method for targets obscured by lidar includes:
[0101] The lidar data acquisition module is used to acquire the raw acquisition data of the lidar, decompose the raw acquisition data into a two-dimensional data matrix, and perform standardization and outlier processing; based on the processed two-dimensional data matrix, a pixel-level histogram is constructed for each pixel, the pixel-level histograms are accumulated to obtain a global total histogram, and the global total histogram is smoothed; the outlier processing includes using an outlier detection algorithm to remove outliers in the two-dimensional data and using linear interpolation to fill in the data positions of outliers.
[0102] The autocorrelation feature analysis unit is used to traverse the smoothed global total histogram using an adaptive-width sliding window, calculate the mean and variance within the window and combine them with neighborhood comparison to obtain candidate peaks, perform deduplication and verification based on echo count ratio on the candidate peaks to obtain the effective peaks and their corresponding time indices; utilize the temporal correlation of the target echo signals to extract multi-target depth information from the data of each pixel; use an adaptive-width sliding window to traverse the pixel-level histogram to obtain the initial peak, verify the temporal correlation of the initial peak to obtain the target peak, and process the target peak using the pixel multi-target depth information to obtain the final target peak;
[0103] The two-dimensional point cloud data acquisition unit is used to filter the target time and remove noise from the final target peak, calculate the reflectance of each pixel and perform reflectance normalization, reconstruct the reflectance vector after normalization, and combine the reflectance of each pixel with its corresponding pixel multi-target depth information to generate two-dimensional point cloud data with reflectance information.
[0104] The evaluation and optimization unit is used to compress and encode the two-dimensional point cloud data to obtain measurement values, reconstruct the original point cloud from the measurement values using optimization algorithms, and evaluate and optimize the reconstruction process using a reconstruction accuracy analysis model.
[0105] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
Claims
1. An autocorrelation-based 3D reconstruction method for targets obscured by lidar, characterized in that: Includes the following steps: S1: Obtain the raw acquisition data from the lidar, decompose the raw acquisition data into a two-dimensional data matrix, and perform standardization and outlier processing; Based on the processed two-dimensional data matrix, a cell-level histogram is constructed for each cell. The cell-level histograms are accumulated to obtain the global total histogram, and the global total histogram is then smoothed. S2: An adaptive-width sliding window is used to traverse the smoothed global total histogram. Candidate peaks are obtained by calculating the mean and variance within the window and combining them with neighborhood comparison. The candidate peaks are deduplicated and verified based on the echo count ratio to obtain the effective peaks and their corresponding time indices. The multi-target depth information of each pixel is obtained by using the temporal correlation of the target echo signals. S3: Use an adaptive width sliding window to traverse the cell-level histogram to obtain the initial peak value, perform time correlation verification on the initial peak value to obtain the target peak value, and use the multi-target depth information of the cell to process the target peak value to obtain the final target peak value. S4: After filtering the target time and eliminating noise for the final target peak, calculate the reflectance of each pixel and perform reflectance normalization. Reconstruct the reflectance vector after normalization into a reflectance image. Combine the reflectance of each pixel with its corresponding pixel multi-target depth information to generate two-dimensional point cloud data with reflectance information. S5: Compress and encode the two-dimensional point cloud data to obtain measurement values, use optimization algorithms to reconstruct the original point cloud from the measurement values, and use a reconstruction accuracy analysis model to evaluate and optimize the reconstruction process.
2. The autocorrelation 3D reconstruction method for targets masked by lidar according to claim 1, characterized in that, Step S1 specifically involves: S1.1: The raw acquisition data of the lidar will be acquired, and the raw acquisition data will be converted from three-dimensional data into a two-dimensional data matrix. The two-dimensional data matrix will be normalized, and outlier values in the two-dimensional data will be removed using an outlier detection algorithm. Linear interpolation will be used to fill in the missing data positions of outlier values to obtain the processed two-dimensional data matrix. S1.2: Construct a cell-level histogram based on each column of the processed two-dimensional data matrix; The obtained cell-level histograms are then summed along the vertical axis to obtain a global histogram, which is then smoothed.
3. The autocorrelation 3D reconstruction method for targets masked by lidar according to claim 2, characterized in that, The normalization process for the two-dimensional data matrix specifically involves: The echo count of each pixel in the converted two-dimensional data matrix is converted into a relative value in the range of 0-1.
4. The autocorrelation 3D reconstruction method for targets masked by lidar according to claim 3, characterized in that, The specific steps for using an outlier detection algorithm to remove outliers from two-dimensional data are as follows: Calculate the mean and standard deviation of each cell in each column of the two-dimensional data, determine whether each data point exceeds the normal range, if so, identify it as an outlier and remove it, if not, identify it as a normal value and save it.
5. The autocorrelation 3D reconstruction method for targets masked by lidar according to claim 4, characterized in that, The specific steps for using linear interpolation to fill in the missing data positions of outliers are as follows: For the saved normal value column data, linear interpolation is used to fill in the missing data positions of outliers.
6. The autocorrelation 3D reconstruction method for targets masked by lidar according to claim 5, characterized in that, The smoothing process specifically involves using a 5-point moving average filtering algorithm to supplement data at the boundary positions of the time index using a mirror expansion method.
7. The autocorrelation 3D reconstruction method for targets masked by lidar according to claim 1, characterized in that, Step S3 specifically involves: S3.1: Use an adaptive-width sliding window to traverse the pixel-level histogram to capture the peak signals of targets at different depths and obtain the initial peak values; S3.2: Calculate the echo data of each pixel and statistically analyze the temporal distribution of the initial peak in different frames, and use the set time deviation to filter out the target peak; S3.3: Enhance the weak peaks in the target peaks and merge the neighboring target peaks. Based on the depth difference between the peaks, cluster, separate, extract and evaluate the quality of the multi-target signals to obtain the final target peak.
8. The autocorrelation 3D reconstruction method for targets masked by lidar according to claim 1, characterized in that, Step S4 specifically involves: S4.1: Filter the target time for the final target peak, retain the pixel response corresponding to the target time, and construct the target time response matrix. The specific expression is as follows: ; in, The target time response matrix, The frame number of the target signal; The pixel number of the target signal; S4.2: Count multiple frames of photons from the detector at a certain moment before the echo signal reaches the receiving system, construct the dark field response matrix and the dark field equivalent bucket signal sequence, calculate the second-order correlation function of the dark field, and perform noise cancellation processing; the specific expression is: ; ; in, For the first The second-order correlation function of the dark field in the nth frame at the pixel index of each detector; For the target echo signal in the nth frame, For the first The nth frame at the detector pixel index is paired with the pixel. The target echo is counted in conjunction with photons; d represents the dark field; n is the dark field data frame number; r represents the target echo signal; s is the target echo signal containing noise. Let k be the dark-field target echo signal in the nth frame, and k be the detector pixel number. For paired pixels; The second-order correlation function of the original noisy signal; The target second-order correlation function after removing dark noise; S4.3: Map the second-order correlation function of the original noisy signal to the [0,1] interval to obtain the target reflectivity, the specific expression of which is: ; in, For target reflectivity; S4.4: Reconstruct the reflectance vector of the target into a two-dimensional array by row-wise inverse reshaping to obtain a reconstructed reflectance image. Combine the reflectance of each pixel with the multi-target depth information of its corresponding pixel to generate two-dimensional point cloud data with reflectance information.
9. A reconstruction system employing the autocorrelation three-dimensional reconstruction method based on lidar-masked targets as described in any one of claims 1-8, characterized in that, include: The lidar data acquisition module is used to acquire the raw acquisition data of the lidar, decompose the raw acquisition data into a two-dimensional data matrix, and perform standardization and outlier processing. Based on the processed two-dimensional data matrix, a cell-level histogram is constructed for each cell. The cell-level histograms are accumulated to obtain the global total histogram, and the global total histogram is then smoothed. The autocorrelation feature analysis unit is used to traverse the smoothed global total histogram by using an adaptive width sliding window, calculate the mean and variance within the window and combine them with neighborhood comparison to obtain candidate peaks, perform deduplication on candidate peaks and verification based on echo count ratio, and obtain effective peaks and their corresponding time indices. By utilizing the temporal correlation of the target echo signal, multi-target extraction is performed on each pixel data to obtain pixel multi-target depth information; an adaptive-width sliding window is used to traverse the pixel-level histogram to obtain the initial peak value, and the temporal correlation of the initial peak value is verified to obtain the target peak value; the pixel multi-target depth information is used to process the target peak value to obtain the final target peak value. The two-dimensional point cloud data acquisition unit is used to filter the target time and remove noise from the final target peak, calculate the reflectance of each pixel and perform reflectance normalization, reconstruct the reflectance vector after normalization, and combine the reflectance of each pixel with its corresponding pixel multi-target depth information to generate two-dimensional point cloud data with reflectance information. The evaluation and optimization unit is used to compress and encode the two-dimensional point cloud data to obtain measurement values, reconstruct the original point cloud from the measurement values using optimization algorithms, and evaluate and optimize the reconstruction process using a reconstruction accuracy analysis model.
10. The reconstruction system employing the autocorrelation-based 3D reconstruction method for target masking using lidar as described in claim 9, characterized in that, The outlier processing includes using an outlier detection algorithm to remove outliers from the two-dimensional data and using linear interpolation to fill in the missing data positions of outliers.
