Laser radar data processing method

Abstract

The invention provides a laser radar data processing method. The laser radar data processing method comprises the following steps: S1, acquiring an echo signal after a laser pulse emitted by a full-waveform laser radar acts on a target, and converting the echo signal into an analog electric signal; s2, converting the analog electric signal into a digital signal; s3, filtering the digital signal to obtain effective echo data; and S4, carrying out adaptive threshold Gaussian-Newton fitting processing on the effective echo data to realize fitting peak searching of the full-waveform laser radar echo data.

Description

Technical Field

[0001] The present disclosure relates to the technical field of lidar, and particularly to a method for processing full-waveform lidar data. Background Art

[0002] The full-waveform lidar technology is an advanced measurement method capable of acquiring continuous echo signals. Compared with traditional pulsed lidars, it can more comprehensively record the complete waveform information of the target echo. Therefore, it is widely used in fields such as topographic mapping, vegetation structure monitoring, and target recognition. Its core lies in the high-precision analysis and processing of echo waveform data. Compared with traditional lidars that only extract limited discrete echo points, the full-waveform lidar waveform is a continuous time-series data, which reflects the more detailed structure of the target object. It can not only measure the distance but also provide more detailed terrain and target information. To record the complete waveform, the full-waveform lidar has higher requirements for the speed of front-end data acquisition, and at the same time, more complex signal processing algorithms are required at the back end to extract useful echo characteristics. The full-waveform lidar echo data is usually stored in the form of a continuous waveform, containing complex information such as target surface reflection characteristics, noise interference, and multiple echoes. Accurately analyzing these data is the key to achieving high-precision measurement, and there are many problems in full-waveform data processing. First, the lidar echo data is often affected by instrument noise, environmental noise, and multipath effects. These noises may mask the echo characteristics of the target and reduce the data quality. Second, in complex scenarios, the laser beam may reflect on different target surfaces to form multiple echoes, and the waveform overlap problem increases the difficulty of analysis. In addition, real-time processing of full-waveform data poses high requirements on the computing performance of the embedded platform. In existing full-waveform lidar data processing methods, the Gauss-Newton algorithm is often used to fit and optimize the echo signal. However, the Gauss-Newton algorithm has low robustness to noise and outliers. When there is a large amount of noise, diverse reflector characteristics, or complex data distribution in the processing environment, traditional algorithms are prone to problems such as unstable convergence and decreased fitting accuracy. In addition, since the lidar echo signal in practical applications is often accompanied by dynamic scene changes (such as multiple targets and non-uniform noise distribution), how to remove outliers while retaining effective information has become a key challenge for improving processing accuracy and reliability. Summary of the Invention

[0003] In view of this, in order to at least partially solve at least one of the above-mentioned technical problems, the present disclosure provides a method for processing lidar data.

[0004] To achieve the above object, the technical solution of the present disclosure is as follows:

[0005] According to an embodiment of the present disclosure, a method for processing lidar data is provided, including: Operation S1: Collecting the echo signal after the laser pulse emitted by the full-waveform lidar acts on the target and converting it into an analog electrical signal; Operation S2: Converting the analog electrical signal into a digital signal; Operation S3: Filtering the digital signal to obtain effective echo data; and Operation S4: Performing adaptive threshold Gaussian-Newton fitting processing on the effective echo data to achieve fitting peak seeking of the full-waveform lidar echo data.

[0006] The lidar data processing method of the present disclosure is more robust to outliers. The traditional Gaussian-Newton algorithm treats all data points equally, and outliers and noise may have a serious impact on the fitting result. The present disclosure effectively eliminates outliers and gradually optimizes the threshold range in each iteration by dynamically adjusting the threshold, combining the standard deviation method and the exponential decay mechanism, thereby enhancing the robustness of the algorithm to noise and outliers. The lidar data processing method of the present disclosure combines exponential decay to achieve a balance between convergence speed and accuracy. When the standard Gaussian-Newton method processes data, it often cannot dynamically adjust the parameter elimination strategy, and fixed thresholds or static weights often cannot adapt to different data distributions in dynamic scenarios, resulting in inaccurate optimization results. The dynamic threshold update mechanism of the present disclosure enables the algorithm to adaptively adjust the threshold range according to the current residual distribution, accommodate more data points in the initial stage, and gradually converge to a high-precision fitting in the later stage, achieving a balance between convergence speed and accuracy. The lidar data processing method of the present disclosure dynamically screens and combines the Gaussian-Newton algorithm to enhance the fitting accuracy. Compared with methods such as the weighted least squares method (WLS), ordinary weighting strategies lack the dynamic adaptability to noise distribution when determining the initial weights and are difficult to process complex data with multiple peaks and high noise in full-waveform radar. The present disclosure adds a dynamic residual weighting module to the framework of the Gaussian-Newton algorithm, eliminates untrustworthy points by dynamically adjusting the screening threshold, and improves the limitations of traditional weighting methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, the above and other objects, features, and advantages of the present disclosure will become clearer. In the drawings:

[0008] Figure 1 It is a schematic flowchart of the lidar data processing method according to the embodiment of the present disclosure.

[0009] Figure 2 It is a schematic flowchart of the implementation of the lidar data processing method according to the embodiment of the present disclosure based on the ZYNQ platform.

[0010] Figure 3 It is a schematic flowchart of the process architecture of Operation S4 in the lidar data processing process according to the embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] The present disclosure provides a method for processing lidar data. First, Gaussian filtering is performed on the echo signal at the PL end of the ZYNQ embedded platform to remove high-frequency noise, smooth the waveform data, and obtain effective echo data. Then, the preprocessed data is transmitted to the PS part through the AXI4 bus for further processing. Secondly, based on the filtered echo data, an adaptive threshold Gaussian-Newton algorithm is proposed to accurately fit and find the peak of the full waveform at the PS end. The adaptive threshold mechanism enables the algorithm to dynamically adjust to adapt to changes in different waveforms and noise characteristics, thereby improving the peak fitting accuracy and positioning ability. Through the above process, the present disclosure significantly improves the processing accuracy of the full waveform data, ensures more reliable ranging results, and at the same time reduces the computational complexity to adapt to the hardware constraints of the embedded platform.

[0012] An important objective of the present disclosure is to improve the robustness to outliers and noise during the processing of full waveform lidar echo data. An adaptive threshold Gaussian-Newton algorithm combining the standard deviation method and exponential decay dynamic adjustment is proposed. By dynamically threshold screening and fitting optimization of the echo data, the robustness and accuracy of the algorithm in complex scenarios are improved. Specifically, the standard deviation method is used to calculate the initial threshold to quickly eliminate significant outliers in the initial data; during the iteration process, the threshold is dynamically adjusted using exponential decay and gradually converges to a stable value according to the residual distribution, thereby adaptively balancing the relationship between outlier elimination and effective data retention. The present disclosure overcomes the problem that the traditional Gaussian-Newton algorithm is sensitive to outliers, is applicable to multi-target recognition and precise echo analysis in high-noise environments, significantly improves the robustness and efficiency of full waveform lidar data processing, and is finally implemented on the embedded development platform. It provides an efficient and reliable solution for high-precision ranging and imaging applications in multi-target complex environments.

[0013] The echo signal of the full waveform lidar usually contains the reflection information of the target. By establishing a model to fit the echo data, tasks such as target extraction and distance estimation can be achieved. Assuming that the full waveform echo data can be fitted with a certain model, such as a Gaussian model, an exponential model, etc., the goal is to minimize the error between the echo data and the model.

[0014] The echo signals recorded by the full waveform lidar usually exhibit a Gaussian distribution or its superposition form, such as multi-target echoes (multi-peak waveforms). The parameters of these waveform signals (such as peak position, amplitude, and width) are difficult to accurately fit through a linear model. In order to accurately extract the target distance, intensity, and characteristics, these echoes must be non-linearly fitted.

[0015] The Gauss-Newton algorithm is a classic non-linear least squares fitting method that can achieve high-precision peak parameter estimation (such as peak position, width, amplitude) by iteratively optimizing the residual between the fitting function (such as Gaussian function or multi-peak Gaussian model) and the actual waveform data. This optimization method is very suitable for the non-linear waveform fitting of full-wave lidar.

[0016] Processing procedure based on the Gauss-Newton algorithm:

[0017] The echo signal of full-wave lidar can usually be fitted with a model function (such as Gaussian function) to extract the target reflection characteristics. The goal is to minimize the sum of the squares of the errors between the model predicted values and the observed values. Let the echo data be , and the goal is to optimize the parameters by minimizing the objective function :

[0018]

[0019] where, represents the objective function, is the time or spatial position of the -th sampling point in the echo signal; represents the observed value of the -th sampling point in the echo signal; represents the predicted value of the echo signal based on the model; are the model parameters to be fitted and optimized (such as the amplitude A of the echo waveform, the center position of the waveform, the width of the waveform, etc.).

[0020] (1) Initialize the parameters: Given the initial model parameters , and calculate the residual :

[0021] ;

[0022] (2) Calculate the Jacobian matrix: Take the partial derivative of the model with respect to the parameter to calculate the Jacobian matrix , and its element is:

[0023]

[0024] i represents the i-th sampling point and j represents the j-th model parameter to be optimized, represents the model parameters to be optimized (such as the amplitude A of the echo waveform, the center position of the waveform, the width of the waveform, etc.). For example, , or or 。

[0025] (3) Solve for the correction amount: According to the Gauss-Newton method formula, solve for the parameter update amount :

[0026]

[0027] represents the residual vector.

[0028] (4) Update the parameters: Update the model parameters using the correction amount: where is the number of the current iteration.

[0029] (5) Calculate the new residual: According to the new parameter calculate the new residual 。

[0030] (6) Check for convergence: If the change in the residual is less than the set threshold or the maximum number of iterations is reached, stop the iteration; otherwise, continue the iteration. The convergence condition is:

[0031] or ;

[0032] Finally, return the finally optimized parameter 。

[0033] However, the direct application of the traditional Gauss-Newton algorithm in lidar data processing also has deficiencies. As can be seen from the specific calculation process of the above Gauss-Newton algorithm, this algorithm requires a high-precision initial guess, otherwise it may converge to a locally sub-optimal solution. At the same time, since the algorithm does not have a threshold mechanism and an outlier rejection mechanism, it is vulnerable to noise and outliers, and has poor performance in complex noise scenarios, poor adaptability to complex full-wave lidar data (such as multi-peak noise and local outliers), and the accuracy will also decrease accordingly.

[0034] To make the objectives, technical solutions, and advantages of the present disclosure clearer and more understandable, the following further describes the present disclosure in detail with reference to specific embodiments and the accompanying drawings.

[0035] In an embodiment of the present disclosure, a lidar data processing method is provided. As shown in combination with Figures 1 to 3 the processing method includes:

[0036] Operation S1: Collect the echo signal after the laser pulse emitted by the full-wave lidar acts on the target and convert it into an analog electrical signal;

[0037] Operation S2: Convert the analog electrical signal into a digital signal;

[0038] Operation S3: Filter the digital signal to obtain effective echo data; and

[0039] Operation S4: Perform adaptive threshold Gaussian - Newton fitting processing on the effective echo data to achieve fitting peak finding of the full - waveform lidar echo data.

[0040] According to an embodiment of the present disclosure, in Operation S3, the digital signal is pre - processed by Gaussian filtering to obtain effective echo data.

[0041] According to an embodiment of the present disclosure, in Operation S4, a Gaussian - Newton objective function model introducing a dynamic threshold adjustment mechanism is constructed.

[0042] According to an embodiment of the present disclosure, the expression of the Gaussian - Newton objective function model is:

[0043]

[0044] Where represents the weight, represents the th observed value of the echo signal, is the time or spatial position, represents the predicted value of the echo signal based on the Gaussian - Newton objective function model; are the model parameters to be optimized by fitting.

[0045] According to an embodiment of the present disclosure, Operation S4 includes:

[0046] Operation S41: Set the initial parameters of the Gaussian - Newton objective function model;

[0047] Operation S42: Set the initial threshold;

[0048] Operation S43: Set the attenuation coefficient and the convergence condition;

[0049] Operation S44: Calculate the residual; and

[0050] Operation S45: Dynamically update the threshold until the convergence condition is met.

[0051] According to an embodiment of the present disclosure, in Operation S42, the initial threshold is determined by the standard deviation of the echo data residual.

[0052] According to an embodiment of the present disclosure, in Operation S44, let the residual be , then:

[0053] ;

[0054] is the time or spatial position of the th sampling point in the echo signal; represents the observed value of the th sampling point in the echo signal; represents the predicted value of the echo signal based on the model; are the model parameters to be fitted and optimized (such as the amplitude A of the echo waveform, the center position of the waveform , the width of the waveform , etc.).

[0055] According to an embodiment of the present disclosure, operation S45 includes dynamically adjusting weights, dynamically updating parameters, and dynamically updating thresholds.

[0056] According to an embodiment of the present disclosure, based on the residual and the threshold adjust the weights :

[0057]

[0058] is the dynamic threshold for the th iteration. Update the threshold by exponential decay, then:

[0059]

[0060] represents the scale factor, represents the standard deviation of the echo data residual, is the initial threshold; represents the minimum threshold; represents the decay coefficient;

[0061] Update the parameters using the weighted Gauss-Newton algorithm:

[0062] ;

[0063] J represents the Jacobian matrix of the objective function with respect to the parameters, W represents the weight diagonal matrix, r represents the residual vector, and k represents the iteration number.

[0064] According to an embodiment of the present disclosure, in operation S45, when or , terminate the iteration, where is the set threshold, is the update amount of the model parameters, is the residual at the kth iteration.

[0065] In summary, the objective of the present disclosure is to improve the robustness against outliers and noise in the processing of full-wave lidar echo data, and a lidar data processing method based on an adaptive threshold Gaussian-Newton algorithm is proposed. The calculation process of this processing method is as follows:

[0066] Construct a Gaussian-Newton objective function model introducing a dynamic threshold adjustment mechanism, and the expression is as follows:

[0067]

[0068] where, is the weight, defined as: , is the dynamic threshold for the th iteration, represents the observed value of the th echo signal, is the time or spatial position, represents the predicted value of the echo signal based on the Gaussian-Newton objective function model; are the model parameters to be fitted and optimized.

[0069] The above objective function explanation: The objective is the same as that of the ordinary Gaussian-Newton method, which is to minimize the model error, but the residual is weighted on the basis of the traditional Gaussian-Newton algorithm, and the optimization objective is re-constructed. In addition, a dynamic threshold screening mechanism is added to eliminate the influence of outliers, making the algorithm more robust to noise and outliers.

[0070] Specific calculation process:

[0071] (1) Initialization:

[0072] First, initialize the parameters , and the initial residual . Then calculate the initial standard deviation:

[0073] ,

[0074] Set the initial dynamic threshold as: , represents the scale factor.

[0075] Explanation of the initial dynamic threshold: The initial threshold is determined by the standard deviation of all data residuals, including the global characteristics of outliers. Using such a standard deviation method to dynamically estimate the initial threshold can adapt to full-wave data with different noise levels.

[0076] (2) Calculate the Jacobian matrix:

[0077] For the Find the partial derivative and calculate the Jacobian matrix , and this part has the same form as the traditional Gauss-Newton method.

[0078] (3)Screen out outliers:

[0079] Dynamically eliminate the abnormal residuals and update the weights according to the threshold : :

[0080] ;

[0081] Explanation of screening out outliers: Through the residual weighting strategy, the abnormal points are processed with zero weight.

[0082] (4)Solve for the correction:

[0083] Solve for the correction using the weighted residuals:

[0084] ;

[0085] where J represents the Jacobian matrix of the objective function with respect to the parameters, is the weight diagonal matrix; r represents the residual vector, .

[0086] Explanation of weighted residuals: Through the weighted matrix W, the effective information is retained, and the noise and outliers beyond the threshold are eliminated.

[0087] (5)Update the parameters:

[0088] Update the parameters ;

[0089] k represents the number of iterations, is the update amount of the model parameters.

[0090] (6)Dynamically adjust the threshold:

[0091] First, calculate the standard deviation of the echo data residuals in real time : ;

[0092] Then update the threshold using exponential decay: ;

[0093] where, represents the scaling factor, is the initial threshold; represents the minimum threshold; represents the decay coefficient.

[0094] Exponential Decay Update Threshold Explanation: This exponential decay method is used to dynamically adjust the threshold of the algorithm. The core idea is to adapt to the change of data distribution by gradually decreasing the threshold, eliminating more outliers and improving the fitting accuracy. In the initial stage: Due to the initial threshold being relatively large, most data points will be retained, including possible outliers. In the iterative stage: At each iteration, the threshold is dynamically updated according to the residual standard deviation . As the number of iterations increases, it will gradually decrease (because outliers are eliminated and the residual distribution becomes more concentrated). The exponential decay term controls the rate of decrease of the threshold, ensuring that the threshold gradually approaches the optimal range while avoiding the loss of valid data points due to too rapid a decrease. In the final stage: When converges to or a sufficiently small range, only the most important valid data points that meet the objective function are retained, and all outliers are eliminated, achieving the optimal state of fitting accuracy.

[0095] (7) Check Convergence:

[0096] If any of the following conditions is satisfied, terminate the iteration:

[0097] or ;

[0098] where, is the set threshold, is the update amount of the model parameter, is the residual at the k-th iteration.

[0099] At this point, return the finally optimized parameter .

[0100] According to the embodiments of the present disclosure, as Figure 2 shown, data acquisition and processing:

[0101] The digital signal obtained by ADC sampling is transmitted to the PL part in real time. In the PL, Gaussian filtering is performed on the echo signal to remove high-frequency noise and smooth the waveform data. In the above process, a synchronous clock is generated by a Clock Generator and a Clock Synchronizer to ensure the real-time nature of data acquisition and transmission. Finally, the preprocessed data is transmitted to the PS part through the AXI4 bus for further processing.

[0102] Processing Procedure Based on the Adaptive Threshold Gaussian-Newton Algorithm (PS Part):

[0103] First, initialize the threshold. Based on the input echo signal, calculate the residual standard deviation and initialize the threshold. Then, through the dynamic threshold adjustment mechanism, a mechanism combining the standard deviation method and the exponential decay method is adopted to dynamically adjust the threshold according to the residual distribution during the iteration process to enhance adaptability. Finally, weight the residuals with the dynamic threshold to reduce the influence of outliers and improve the fitting accuracy. Then, judge the convergence condition, end the iteration, and output the fitting parameters.

[0104] Data output and host computer display:

[0105] The PS part transmits the fitted result data to the host computer through the UART interface. The host computer displays and further analyzes the data to implement functions such as waveform reconstruction and precise ranging.

[0106] In summary, the disclosed example utilizes the hardware advantages of the ZYNQ platform. The PL part completes signal acquisition and preprocessing, and the PS part implements the adaptive threshold Gaussian-Newton algorithm, achieving efficient dynamic optimization fitting of full-wave lidar data. The results are transmitted to the host computer in real time through the UART, meeting the application requirements of high precision and real-time performance.

[0107] According to the embodiments of the present disclosure, as Figure 3 shown, the processing process based on the adaptive threshold Gaussian-Newton algorithm can be divided into three major stages: (1) is the data preprocessing stage; (2) to (4) are the initialization stages; (5) to (9) are the loop iteration optimization stages.

[0108] Specifically:

[0109] (1) Data preprocessing: Filter the lidar echo data (such as Gaussian filtering) to reduce noise.

[0110] (2) Set initial parameters (such as the distance of the target).

[0111] (3) Set the initial threshold , and calculate the initial threshold using the standard deviation of the waveform data. , is the scaling factor, which controls the initial threshold range.

[0112] (4) Set the attenuation coefficient , usually taking 0.8 - 0.95. Set the convergence condition (such as setting the threshold or reaching the maximum number of iterations, etc.).

[0113] (5) Calculate the residual ;

[0114] (6) Dynamically adjust the weight: According to the current residual and the threshold Adjust weights:

[0115] ;

[0116] (7) Update parameters: Update the parameters using the weighted Gauss - Newton method . Where J is the Jacobian matrix of the objective function with respect to the parameters, W is the diagonal weight matrix, and r is the residual vector.

[0117] (8) Dynamically update the threshold .

[0118] (1) Judge the loop condition: If or or the maximum number of iterations is reached, then exit the loop. If not satisfied, return to step (5) to continue the iteration.

[0119] The full - waveform lidar data processing method of the present disclosure combines a dynamic threshold adjustment mechanism of standard deviation and exponential decay:

[0120] Dynamically estimate the initial threshold using the standard deviation method to adapt to full - waveform data with different noise levels.

[0121] Introduce an exponential decay mechanism to gradually reduce the threshold range during the iteration process, so that the algorithm can not only effectively eliminate outliers but also ensure the convergence accuracy in the later stage.

[0122] The threshold adjustment mechanism is seamlessly integrated with the Gauss - Newton optimization process to dynamically adapt to changes in data characteristics.

[0123] An adaptive residual screening and elimination method is adopted:

[0124] Screen the residuals based on the dynamic threshold, retain the valid information through the weight matrix W, and eliminate the noise and outliers beyond the threshold.

[0125] Provide an efficient and robust way to process residuals, which performs particularly well in scenarios with complex noise distributions or dense outliers.

[0126] An optimized Gauss - Newton algorithm architecture (the combination of the Gauss - Newton algorithm and the adaptive mechanism) is adopted:

[0127] Weight the residuals based on the Gauss - Newton algorithm and reconstruct the optimization objective, making the algorithm more robust to noise and outliers.

[0128] Combined with the adjustment of dynamic thresholds, the convergence speed and stability of the algorithm are improved, and the fitting deviation caused by outliers is reduced.

[0129] The disclosed laser radar data processing method is particularly suitable for full-waveform laser radar multi-target and multi-peak echo data processing, and can effectively extract target information in complex environments. Similarly, in other scenes with multiple overlapping targets and significant noise interference, high-precision waveform fitting and optimization can still be achieved.

[0130] The disclosed laser radar data processing method is based on the initial method of residual distribution calculation threshold (standard deviation calculation) and the method of dynamic adjustment in iteration combined with exponential decay. The global noise level is determined by the standard deviation method, the dynamic convergence mechanism of the exponential decay method, and the combination logic of the two when dynamically adjusting the threshold. The residuals are screened by the dynamic threshold, the abnormal residual points exceeding the threshold are eliminated, and the remaining residual points are weighted to participate in the fitting optimization. It has a better residual weighting strategy (zero weight processing of abnormal points) and its dynamic update mechanism.

[0131] Specifically:

[0132] The residuals are screened based on the dynamic threshold, and the residual calculation method of the optimization target is adjusted through the weighting matrix W.

[0133] Dynamic filtering rules The specific implementation logic.

[0134] The dynamic residual screening module is introduced into the Gauss-Newton algorithm framework to complete the coordinated optimization of parameter fitting and residual elimination. The complete calculation process formed by combining the Gauss-Newton method with the dynamic threshold adjustment strategy includes steps such as dynamic weight assignment and correction update. Specifically:

[0135] The combination of dynamic threshold and weighted optimization forms an improved method for the traditional Gauss-Newton algorithm. The optimization equation can achieve high-precision fitting and convergence.

[0136] The disclosed laser radar data processing method solves the problem of unstable fitting caused by noise and outliers in full-waveform laser radar data processing through a dynamic threshold adjustment mechanism and an optimized Gauss-Newton algorithm architecture. The core content to be protected by the patent includes the calculation method of dynamic threshold adjustment, residual screening rules, and the improved structure of the Gauss-Newton algorithm.

[0137] So far, the embodiments of the present disclosure have been described in detail with reference to the accompanying drawings. It should be noted that, in the accompanying drawings or the main text of the specification, the implementation manners that are not illustrated or described are all forms known to those of ordinary skill in the art and are not described in detail. In addition, the definitions of the above elements and methods are not limited to the specific structures, shapes or manners mentioned in the embodiments, and those of ordinary skill in the art can make simple changes or substitutions thereto.

[0138] In this document, unless otherwise specified, the so-called feature A "or" (or) or "and / or" (and / or) feature B means that A exists alone, B exists alone, or A and B exist simultaneously; the so-called feature A "and" (and) or "and" (and) or "and" (and) feature B means that A and B exist simultaneously; the so-called "including", "comprising", "having", "containing" means including but not limited to this.

[0139] In addition, unless otherwise specifically described or the steps must occur in sequence, the order of the above steps is not limited to those listed above and can be changed or rearranged according to the required design. And based on considerations of design and reliability, the above embodiments can be used in combination with each other or in combination with other embodiments, that is, the technical features in different embodiments can be freely combined to form more embodiments.

[0140] The specific embodiments described above further elaborate on the purpose, technical solutions and beneficial effects of the present disclosure. It should be understood that the above are only specific embodiments of the present disclosure and are not used to limit the present disclosure. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present disclosure shall be included within the protection scope of the present disclosure.

Claims

1. A laser radar data processing method, comprising: Operation S1: collecting the echo signal after the laser pulse emitted by the full-waveform laser radar acts on the target and converting it into an analog electrical signal; Operation S2: converting the analog electrical signal into a digital signal; Operation S3: filtering the digital signal to obtain valid echo data; and Operation S4: Perform adaptive threshold Gauss-Newton fitting processing on the effective echo data to achieve peak fitting of the full-waveform lidar echo data.

2. According to the laser radar data processing method according to claim 1, in operation S3, the digital signal is pre-processed by Gaussian filtering to obtain valid echo data.

3. According to the laser radar data processing method according to claim 1, a Gauss-Newton objective function model introducing a dynamic threshold adjustment mechanism is constructed in operation S4.

4. According to the laser radar data processing method according to any one of claims 1 to 3, the Gauss-Newton objective function model expression is: in represents the weight, Indicates the echo signal The observed value of the sampling points, The echo signal The temporal or spatial position of the sampling points, represents the predicted value of the echo signal based on the Gauss-Newton objective function model; are the model parameters to be optimized.

5. The laser radar data processing method according to claim 4, wherein operation S4 comprises: Operation S41: setting initial parameters of the Gauss-Newton objective function model; Operation S42: setting an initial threshold; Operation S43: setting the attenuation coefficient and convergence condition; Operation S44: Calculating residuals; and Operation S45: Dynamically update the threshold until the convergence condition is met.

6. According to the laser radar data processing method according to claim 5, in operation S42, the initial threshold is determined by the standard deviation of the echo data residual.

7. According to the laser radar data processing method of claim 5, in operation S44, the residual is set to ,but: ； Indicates the echo signal The observed value of the sampling points, The echo signal The temporal or spatial position of the sampling points, represents the predicted value of the echo signal based on the Gauss-Newton objective function model; are the model parameters to be optimized.

8. According to the laser radar data processing method according to claim 5, operation S45 includes dynamically adjusting weights, dynamically updating parameters, and dynamically updating thresholds.

9. The laser radar data processing method according to claim 5, and threshold Adjust weights : For the The dynamic threshold of the iteration is updated by exponential decay, then: represents the scale factor, represents the standard deviation of the echo data residuals, is the initial threshold; represents the minimum threshold; represents the attenuation coefficient; Update the parameters using the weighted Gauss-Newton algorithm: ； J represents the Jacobian matrix of the objective function with respect to the parameters, W represents the weight diagonal matrix, r represents the residual vector, and k represents the number of iterations.

10. The laser radar data processing method according to claim 5, wherein in operation S45, or , terminate the iteration, where To set the threshold, is the model parameter update amount, is the residual of the kth iteration.

Images