A laser radar scanning ranging sensor suitable for complex working conditions

Abstract

The application belongs to the technical field of laser radars, and particularly relates to a laser radar scanning ranging sensor suitable for complex working conditions, which carries out defect identification processing after obtaining original point cloud data, identifies light straight-line propagation failure areas and reflection characteristic abnormal areas, marks out false points, missing points and ambiguous points, attributes and classifies point cloud defects from a physical root source, realizes accurate tracing of point cloud abnormal root sources, actively carries out compensation optimization based on the defect identification result to generate a compensation data point set, obtains optimized point clouds through fusion processing, significantly improves the integrity and geometric accuracy of final point cloud data, and finally carries out closed-loop linkage of back-end point cloud defect features and front-end scanning parameter adjustment, carries out defect cause diagnosis and directly drives laser radar scanning parameter adjustment, which is helpful to reduce the point cloud defect occurrence rate, so that the optimal sensing performance can be continuously maintained in a complex and changeable environment.

Description

Technical Field

[0001] This invention belongs to the field of lidar technology, specifically a lidar scanning and ranging sensor suitable for complex working conditions. Background Technology

[0002] LiDAR scanning range sensors are active sensing devices that acquire three-dimensional environmental information by calculating the time of flight of a laser beam. Their working principle is based on two fundamental physical assumptions: laser light maintains its straight-line propagation characteristic in a homogeneous medium, and ideal diffuse reflection occurs on the target surface, with some energy returning along the original path. Traditional LiDAR systems built upon these assumptions can obtain accurate point cloud data under ideal conditions such as clear weather and targets with standard reflectivity.

[0003] However, in practical applications, when encountering complex weather conditions such as rain, snow, fog, and dust, or when facing special surfaces such as mirrors or transparent materials, the above-mentioned basic physical assumptions will systematically fail. Specifically, the scattering and absorption effects of the laser beam caused by atmospheric suspended particles disrupt the integrity of the straight-line propagation path, resulting in large-area holes in the point cloud; while the reflective surface of the mirror causes the laser beam to deviate from the receiving field of view, forming a detection blind zone with a clear geometric boundary; and transparent media will cause multiple reflections of the beam after penetration, producing ambiguous point clouds with distorted spatial positions.

[0004] To address data anomalies under these complex operating conditions, existing technologies primarily employ post-processing algorithms such as statistical filtering and radius filtering for noise suppression, or interpolation algorithms to fill in missing data. However, these methods have significant limitations: First, traditional methods only address anomalies at the data level, failing to establish a connection with the underlying physical mechanisms causing the anomalies, resulting in a lack of targeted processing strategies.

[0005] Secondly, existing systems use fixed scanning parameters and cannot dynamically adjust the perception strategy according to real-time operating conditions, making it difficult to maintain stable perception performance in continuously changing and complex environments.

[0006] Most importantly, when the assumptions of rectilinear propagation and reflection of light partially fail under non-ideal conditions, existing technologies lack effective mechanisms to ensure the reliability and completeness of point cloud data, thus posing potential risks to downstream applications that rely on such data. Summary of the Invention

[0007] To overcome the shortcomings of the prior art, the present invention provides a lidar scanning ranging sensor suitable for complex working conditions, which can effectively solve the problems involved in the prior art.

[0008] The objective of this invention can be achieved through the following technical solution: a lidar scanning ranging sensor suitable for complex working conditions, comprising: a data acquisition module, a defect identification module, a point cloud optimization module, and a parameter adjustment module.

[0009] The data acquisition module is connected to the defect identification module, the defect identification module is connected to the point cloud optimization module, and the point cloud optimization module is connected to the parameter adjustment module.

[0010] The data acquisition module acquires raw point cloud data of the lidar under complex working conditions. The raw point cloud data includes the spatial location information and intensity information of each point cloud.

[0011] The defect identification module performs defect identification processing on the original point cloud data, identifies regions where the linear propagation assumption of light fails and regions with abnormal reflection characteristics, and marks false points, missing points, and ambiguous points in the regions.

[0012] The point cloud optimization module, based on the defect identification results, performs compensation optimization on the marked points in the region to generate a set of compensation data points, and then merges the set of compensation data points with the original point cloud data to obtain the optimized point cloud data set.

[0013] The parameter adjustment module performs defect cause diagnosis based on the proportion and spatial distribution of various defect markers, and drives the adjustment of lidar scanning parameters based on the diagnosis results. The scanning parameters include at least the angular resolution of the emitted beam and the time window width of the received signal.

[0014] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention identifies the failure area of ​​the linear propagation hypothesis of light and the abnormal area of ​​reflection characteristics, and accurately marks false points, missing points and ambiguous points in this macroscopic area, thereby attributing and classifying point cloud defects from the physical root cause, realizing the accurate tracing of the root cause of point cloud anomalies, thus providing a data basis for the subsequent implementation of differentiated and targeted compensation and adjustment strategies.

[0015] (2) Based on the defect identification results, the present invention actively performs compensation optimization to generate a set of compensation data points, and obtains an optimized point cloud through fusion processing. It can actively repair rather than passively filter defective point cloud data. At the same time, it fully considers the characteristics of different defect types during the repair process, which significantly improves the integrity and geometric accuracy of the final point cloud data.

[0016] (3) This invention links the back-end point cloud defect features with the front-end scanning parameter adjustment in a closed loop. By analyzing the proportion and spatial distribution of various defect marker points, it can carry out defect cause diagnosis and directly drive the adjustment of lidar scanning parameters. This helps to reduce the point cloud defect rate and thus maintain the best perception performance in a complex and ever-changing environment. Attached Figure Description

[0017] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the module connection of the present invention.

[0019] Figure 2 This is a logical schematic diagram of the point cloud optimization module of the present invention.

[0020] Figure 3 This is a schematic diagram of the parameter adjustment module of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0022] Reference Figure 1 As shown, the present invention provides a lidar scanning ranging sensor suitable for complex working conditions, including: a data acquisition module, a defect identification module, a point cloud optimization module, and a parameter adjustment module.

[0023] The data acquisition module is connected to the defect identification module, the defect identification module is connected to the point cloud optimization module, and the point cloud optimization module is connected to the parameter adjustment module.

[0024] The data acquisition module acquires raw point cloud data of the lidar under complex working conditions. The raw point cloud data includes the spatial location information and intensity information of each point cloud.

[0025] The defect identification module performs defect identification processing on the original point cloud data, identifies regions where the linear propagation assumption of light fails and regions with abnormal reflection characteristics, and marks false points, missing points, and ambiguous points in the regions.

[0026] Given that the failure of the rectilinear propagation assumption of light mainly manifests in two ways: propagation path obstruction failure and propagation path penetration failure.

[0027] Propagation path obstruction failures originate from energy attenuation caused by the scattering or absorption of suspended particles such as rain, snow, fog, and dust during beam propagation. This physical process manifests as a significant decrease in local point cloud density in point cloud data. Based on this, spatial density analysis is used as an effective criterion for identifying this type of failure. By detecting continuous spatial regions with point cloud density below normal levels, the detection blind zone caused by energy attenuation can be located.

[0028] Propagation path penetration failure is caused by the beam hitting an object behind it after passing through a transparent or semi-transparent medium. This phenomenon violates the fundamental principle of the first echo in lidar ranging. To address this characteristic, ray tracing consistency analysis is used as the decisive criterion for identifying this type of failure. By constructing a virtual ray and identifying the logical conflict sequence between the radar origin, the non-point cloud space, and the effective point cloud, the abnormal detection area caused by the beam penetrating the interface can be accurately detected.

[0029] Based on the application of the above two physical mechanisms and their corresponding criteria, in a preferred embodiment of the present invention, the process of identifying the failure area of ​​the linear propagation assumption of light includes: delineating the beam propagation space between the lidar and the distant effective point cloud as the area to be inspected; the distant effective point cloud is identified from the original point cloud data through the following steps: first, collecting multiple frames of historical point cloud data collected by the lidar in a static calibration scenario, statistically analyzing the distance distribution between all point clouds and the lidar origin, and plotting a distance-point count curve; in the curve, establishing the distance value corresponding to the first trough in the distance distribution as the distance threshold, filtering out point cloud data in the original point cloud data that are lower than the distance threshold, so as to filter out dynamic or invalid near point clouds around the lidar caused by noise, and retaining the set of point cloud data of the corresponding medium and long distances.

[0030] Subsequently, spatial clustering based on Euclidean distance is performed on the point cloud datasets of the medium and long distances to exclude isolated points, resulting in a point cloud dataset representing continuous physical entities.

[0031] Finally, by statistically analyzing the mean and standard deviation of the reflection intensity of typical ground features in the calibrated scene, and based on the principle of adding or subtracting several times the standard deviation from the mean, points in the point cloud set whose reflection intensity is not within this range are filtered out, thus obtaining the far-end effective point cloud set that constitutes the continuous physical surface.

[0032] The area to be inspected is divided into three-dimensional voxels, and the point cloud density of each voxel is calculated.

[0033] Based on the effective point cloud density around the area to be inspected, a density threshold is dynamically set, and a continuous set of voxels with a point cloud density lower than the density threshold is marked as a propagation path occlusion type missing region.

[0034] As an example, the density threshold dynamic setting process is as follows: taking the area to be inspected as a reference, expand it outward by a specific width to form an outer area, calculate the average point cloud density and standard deviation of all voxels in the outer area, and subtract the product of the standard deviation and the preset sensitivity coefficient from the average point cloud density to obtain the density threshold.

[0035] The logic of this example is to set the threshold as the lower limit of the ambient background density. Continuous areas with a density lower than this can be regarded as abnormal missing areas. The preset sensitivity coefficient can be set to 3 based on the principle of 3 times the standard deviation.

[0036] Virtual rays are emitted from the origin of the lidar to each distant effective point cloud. The system detects whether there are physical contradictions in the ray path, such as a spatial segment without point clouds at the beginning but a spatial segment connecting to the distant effective point clouds at the end. All such spatial segments detected are marked as propagation path penetration-type missing regions.

[0037] The set of the aforementioned occlusion-type and penetration-type missing regions is collectively defined as the region where the straight-line propagation assumption of light fails.

[0038] Furthermore, the identification of regions with abnormal reflection characteristics relies on the abnormal deviation of the physical characteristics of the reflected waves. Specifically, abnormally high-intensity discrete point cloud clusters reveal non-target backscattering caused by suspended particles encountered by the laser in the transmission path. This type of reflection interference is identified and located through intensity threshold segmentation and spatial clustering analysis. Simultaneously, closed low-intensity regions with clear geometric contours appearing on macroscopically continuous surfaces characterize the loss of effective echo due to the beam deviating from the receiving path due to specular reflection. The precise definition of these abnormal regions is achieved through boundary extraction and contour analysis of the low-intensity regions.

[0039] Based on the collaborative judgment logic of the above two types of reflected wave characteristic anomalies, in a preferred embodiment of the present invention, the process of identifying the reflected characteristic anomaly region includes: extracting echo intensity features from the original point cloud data and generating an intensity distribution histogram.

[0040] Based on the intensity distribution histogram, the point cloud set of high intensity region and the point cloud set of low intensity region are separated. The specific separation process is as follows: the intensity distribution histogram is subjected to kurtosis detection, and the number of peaks is counted.

[0041] When the histogram is detected to exhibit a bimodal or multimodal distribution, the intensity value corresponding to the lowest valley between the global maximum peak and the local maximum peak is determined as the separation boundary value.

[0042] When the histogram is detected to show only a single-peak distribution, the mean and standard deviation of all point cloud intensities are calculated. The separation threshold is determined based on the principle of adding or subtracting a certain number of standard deviations from the mean.

[0043] Point clouds with an intensity higher than this separation threshold are classified as high-intensity point clouds, and point clouds with an intensity lower than or equal to this separation threshold are classified as low-intensity point clouds.

[0044] Spatial clustering analysis is performed on the high-intensity regional point cloud set to identify isolated point cloud clusters and mark the areas they occupy as discrete reflection interference regions.

[0045] Region growing segmentation based on normal vector consistency is performed on the original point cloud data: the normal vector and curvature of each point in the original point cloud data are calculated, and the point with the smallest curvature is used as the initial seed point. The curvature represents the flatness of the surface of the point cloud neighborhood.

[0046] Centered on the seed point, check all unassigned points in its spatial neighborhood. If the angle between the normal vector of the neighboring point and the normal vector of the current seed point, and the distance from the neighboring point to the local fitting plane where the current seed point is located are both less than the corresponding preset tolerance upper limit, then the seed point is assigned to the current growth region. The newly assigned point is used as a new seed point, and the above growth process is repeated until the current region can no longer be expanded.

[0047] From the remaining point cloud, select the point with the smallest curvature as the next seed point, and repeat the process to generate new growth regions until all points have been processed.

[0048] All generated growth regions are filtered out, and regions containing more than a certain number of points are identified as macroscopic continuous surfaces.

[0049] The low-intensity region point cloud set is mapped onto the identified macroscopic continuous surface.

[0050] Spatial connectivity analysis is performed on low-intensity point clouds mapped onto the same macroscopic continuous surface to identify regions whose boundaries can form closed contours, and these regions are marked as the specular reflection-type missing regions.

[0051] The set of discrete reflection interference regions and specular reflection-type missing regions are collectively defined as regions with abnormal reflection characteristics.

[0052] In a preferred embodiment of the present invention, the process of marking false points, missing points, and ambiguous points includes: marking all point clouds within the discrete reflection interference area as false points.

[0053] The geometric center point of each three-dimensional voxel in the region where the rectilinear propagation of light fails and the region of specular reflection-type missing points are marked as missing points.

[0054] The valid point cloud located at the end of the spatial segment retrieved in the propagation path penetration-type missing region is marked as an ambiguous point.

[0055] Specifically, the labeling criteria for different defect point types are as follows: Discrete reflection interference regions are isolated point cloud clusters identified through spatial clustering analysis of high-intensity point clouds. These clusters lack geometric continuity with the surrounding environment. The point clouds within these regions exhibit abnormally high echo intensity, but their spatial distribution characteristics indicate that they do not constitute any solid obstacles with continuous surfaces. Physically, this is backscattering caused by discrete suspended particles such as raindrops and dust intercepting the laser beam along its propagation path. These point clouds do not represent real navigation obstacles and constitute ineffective noise interference to the environmental model; therefore, they must be labeled as false points for filtering.

[0056] The regions where the rectilinear propagation assumption of light fails and the specular reflection-type missing regions are continuous three-dimensional spatial ranges that should contain point cloud echoes but are actually missing, determined by methods such as ray tracing void analysis and intensity gradient boundary analysis. The geometric center points of voxels within these ranges are marked to reconstruct and explicitly represent the physical spatial missingness within the point cloud data structure. These points are not actual measurement points, but rather defined placeholders indicating where data should exist but is not, thus providing clear and actionable processing targets for subsequent data compensation and 3D reconstruction modules. This is a crucial preprocessing step for achieving point cloud integrity restoration.

[0057] Ambiguous points are valid measurement points, but the reliability of their spatial location is reasonably questionable. They are located at the end of logically conflicting paths and may be caused by a laser penetrating a transparent medium and hitting an object behind it.

[0058] This invention identifies regions where the linear propagation hypothesis of light fails and regions with abnormal reflection characteristics. Within these macroscopic regions, false points, missing points, and ambiguous points are precisely marked. This allows for the attribution and classification of point cloud defects from a physical perspective, enabling precise tracing of the root causes of point cloud anomalies. This provides a data foundation for implementing differentiated and targeted compensation and adjustment strategies.

[0059] Reference Figure 2 As shown, the point cloud optimization module, based on the defect identification results, performs compensation optimization on the region marker points to generate a set of compensation data points, and then fuses the set of compensation data points with the original point cloud data to obtain the optimized point cloud data set.

[0060] In a preferred embodiment of the present invention, the process of generating the compensation data point set includes: performing spatial interpolation processing on the missing points based on environmental geometric priors to generate compensation data points that replace the missing points.

[0061] Specifically, the spatial interpolation processing based on environmental geometric priors is based on the local environmental geometric contour constructed from the effective point cloud around the missing point, and uses linear interpolation or surface fitting algorithms to generate compensation data points to replace the missing points.

[0062] Taking the surface fitting algorithm as an example, an appropriate mathematical model is selected for the local environmental geometric contour. Specifically, a template library containing a variety of basic geometric shapes is predefined. Each template in the library is uniquely represented by its normalized feature spectrum. The basic geometric shape types include, but are not limited to, plane, cylindrical surface, sphere, and torus. The feature spectrum includes at least flatness and curvature.

[0063] Simultaneously calculate the flatness and curvature of the current local environment after geometric contour normalization to form the feature spectrum to be measured.

[0064] The similarity between the feature spectrum to be tested and each template feature spectrum in the template library is calculated using cosine similarity or the reciprocal of Euclidean distance. The geometric shape corresponding to the template with the highest similarity is selected as the determination result of the geometric contour of the current local environment.

[0065] Based on the matched geometric shape, a corresponding fitting model is selected from predefined mapping relationships. If a planar template is matched, a first-order polynomial surface model is used; if a cylindrical or spherical template is matched, a quadratic surface model is used; and for other templates, a surface model based on radial basis functions or a higher-order polynomial surface model is used for fitting.

[0066] Since the mathematical models given above are all existing technologies, they will not be elaborated upon here.

[0067] The selected model is fitted to the local environment geometry using the least squares method to reconstruct the continuous geometric surface of the region.

[0068] Substitute the geometric center coordinates of the voxel containing the missing point into the mathematical model, calculate its corresponding three-dimensional coordinates, and insert the coordinate point as a compensation data point into the original point cloud.

[0069] For ambiguous points, a local point cloud neighborhood centered on the ambiguous point is constructed. The local surface normal vector is estimated using principal component analysis. Based on the spatial distribution trend of the point cloud in the neighborhood, the spatial coordinates of the ambiguous point are corrected to generate corrected compensation data points.

[0070] It should be noted that the specific process of correcting the spatial coordinates of the ambiguous points is as follows: with the ambiguous point as the center, a search radius is set, and all valid point clouds located within the spherical space are extracted from the original point cloud to form a local neighborhood point set.

[0071] Principal component analysis is performed on the local neighborhood point set. Specifically, the covariance matrix of the point set is calculated, the covariance matrix is ​​decomposed into eigenvalues, the eigenvector corresponding to the smallest eigenvalue is extracted, and the eigenvector is defined as the local surface normal vector, with its direction pointing towards the lidar sensor.

[0072] Calculate the three-dimensional centroid of the local neighborhood point set. This centroid represents the distribution center of the effective point cloud in space and defines the spatial position trend of the local surface.

[0073] Ambiguity Along the normal vector The direction of the projection onto the centroid of the point set. The projection point is on the local reference plane defined by the normal vector. The calculation formula is: .

[0074] in, It is a three-dimensional vector term that points from the ambiguous point to the centroid of its local neighborhood point cloud. The centroid is the average position of all valid points on the local surface, representing the most reliable collective consensus spatial position in the region. Therefore, this three-dimensional vector intuitively represents the overall offset direction and magnitude of the ambiguous point relative to its surrounding real surface.

[0075] It is a scalar distance term used to calculate the signed projection length of the above three-dimensional vector term in the direction of the unit normal vector, to measure the perpendicular distance of the ambiguous point from the surface normal direction to the local reference plane, and to indicate the precise normal distance of the ambiguous point that needs to be corrected.

[0076] If the dot product is positive, it means the point of ambiguity is on the same side as the normal vector. If the dot product is negative, it means the point of ambiguity is on the opposite side of the normal vector.

[0077] This involves multiplying the scalar distance term above by the unit normal vector to obtain a new vector. This new vector is parallel to the normal vector and its magnitude is equal to the aforementioned normal distance. This operation constructs a specific correction vector, transforming a purely numerical distance into a displacement with direction and magnitude in three-dimensional space.

[0078] This formula corrects ambiguous points from their potential offset locations to a locally smooth surface formed by the neighboring point cloud, thus completing the spatial coordinate correction.

[0079] All generated compensation data points are organized according to their original spatial location index to form a structured set of compensation data points.

[0080] Each data point in the set of compensation data points is assigned a compensation type identifier to indicate that the data point originates from missing point interpolation or ambiguous point correction.

[0081] In a preferred embodiment of the present invention, the process of fusing the compensation data point set with the original point cloud data includes: removing all data points marked as false points from the original point cloud data and retaining the unmarked data points as a valid subset of the original point cloud.

[0082] Insert the data points in the compensation data point set into the corresponding positions in the valid original point cloud subset according to their spatial location index.

[0083] Spatial deduplication is performed on the merged point cloud data, and the data is recombined according to a unified data format to generate an optimized point cloud data set.

[0084] Each point in the optimized point cloud dataset contains spatial coordinates, intensity information, and compensation type identifier fields.

[0085] Based on the defect identification results, the embodiments of the present invention actively perform compensation optimization to generate a set of compensation data points, and obtain an optimized point cloud through fusion processing. This can actively repair rather than passively filter defective point cloud data. At the same time, the characteristics of different defect types are fully considered during the repair process, which significantly improves the integrity and geometric accuracy of the final point cloud data.

[0086] Reference Figure 3 As shown, the parameter adjustment module performs defect cause diagnosis based on the proportion and spatial distribution of various defect marker points, and drives the adjustment of lidar scanning parameters based on the diagnosis results. The scanning parameters include at least the angular resolution of the emitted beam and the time window width of the received signal.

[0087] In a preferred embodiment of the present invention, the defect cause diagnosis process includes: when the proportion of a certain type of defect point is higher than the sum of the proportions of the other two types of defect points, the defect point of that type is regarded as the dominant defect point.

[0088] Furthermore, when no single type of defect has a higher percentage than the sum of the other two types, the defect type with the highest percentage is determined as the dominant defect.

[0089] Based on the type and spatial distribution characteristics of the dominant defect point, a corresponding defect cause diagnosis is performed: if the dominant defect point is a spurious point, and its spatial distribution shows a clustering characteristic along a specific scanning angle, then the current defect is diagnosed as multipath reflection interference. The diagnostic basis is that multipath reflection occurs before or after the laser beam reaches the target surface, and multiple reflections occur between multiple reflecting surfaces. This process generates a spurious point in an optical sense that does not actually exist as a physical entity.

[0090] Because the scanning mirror of a lidar oscillates periodically, its scanning angle is continuous. Therefore, spurious points generated by multipath interference that exists stably at the same mirror angle will naturally cluster along that specific scanning line in space, forming a unique angle-point cloud distribution pattern.

[0091] If the dominant defect point is a missing point, and its spatial distribution exhibits continuous linear or planar characteristics, then the current defect is diagnosed as signal attenuation caused by obstruction or transmission along the propagation path. The diagnostic basis is as follows: when a laser beam is completely blocked by an entity along its propagation path, or is severely scattered and absorbed by suspended particles such as rain, snow, or fog, the energy attenuates to the point where it cannot produce an effective echo, resulting in missing data. Similarly, when the beam completely penetrates a transparent medium such as glass, there will be no echo due to insufficient backscattering.

[0092] Whether it's physical occlusion or particle swarm scattering, the range of influence is continuous in three-dimensional space, rather than isolated points. Therefore, the resulting missing points will form continuous missing regions in space with certain geometric shapes, such as lines or surfaces.

[0093] If the dominant defect point is an ambiguous point, and its spatial distribution highly overlaps with high-curvature surface regions in the scene, then the current defect is diagnosed as echo distortion caused by complex surface scattering. The diagnostic basis is: on a high-curvature surface, the laser spot may simultaneously illuminate multiple micro-surfaces with different normal vectors. The mixed echoes returned by these surfaces will cause waveform distortion, resulting in distorted radar ranging and intensity calculations, generating an ambiguous point with unreliable position or intensity.

[0094] By calculating the curvature characteristics of each region in the point cloud, high-curvature areas in the scene can be identified. When it is found that ambiguous points are not randomly distributed in space, but densely appear in these identified high-curvature areas, a strong correlation between complex surface geometry and echo data ambiguity can be established, thereby confirming the cause.

[0095] The high curvature surface region is defined by the following process: calculating the local curvature feature value of effective points in the scene, identifying points whose curvature feature value is greater than the statistical quartile of all point curvature feature values ​​as high curvature points, performing spatial clustering on the high curvature points, and defining the range of the resulting point cluster containing more than the minimum point number threshold as the high curvature surface region.

[0096] The criteria for determining the high degree of overlap are as follows: calculate the proportion of points located inside the high curvature surface region among all ambiguous points, and when the proportion exceeds the upper quartile, it is determined that the spatial distribution is highly overlapped.

[0097] In the above embodiments of the present invention, the upper quartile is used as the preferred statistical threshold for both high curvature surface regions and high overlap regions. Those skilled in the art should understand that the upper quartile can be adaptively adjusted based on the principle of statistical significance, according to specific application scenarios and reliability requirements.

[0098] For example, in scenarios requiring extremely high accuracy, the threshold can be raised to the 95th percentile to filter out more extreme feature points, thereby improving the specificity of the diagnosis; in scenarios requiring higher sensitivity, it can be relaxed to the 70th percentile to capture a wider range of potential abnormal patterns.

[0099] Such adjustments based on the percentile system do not deviate from the core adaptive comparison logic of this invention; they are all equivalent substitutions or obvious specific implementations under the technical concept of this invention.

[0100] In a preferred embodiment of the present invention, adjusting the laser radar scanning parameters based on the diagnostic results includes: if multipath reflection interference is determined, generating an instruction to reduce the angular resolution of the emitted beam.

[0101] If signal attenuation is detected, an instruction is generated to increase the width of the received signal time window.

[0102] If echo distortion is detected, a command is generated to adjust the angular resolution and time window width to the preset anti-interference mode.

[0103] The preset anti-interference mode refers to a set of parameters that have been experimentally calibrated in advance. For example, the angular resolution is reduced to 70% of the standard value, while the time window width is increased to 150% of the standard value. The aim is to smooth the sampling of high curvature surfaces by reducing the angular resolution, while widening the window to capture potentially distorted echo signals.

[0104] The adjustment command is written into the lidar control register, driving it to perform data acquisition for the next scan cycle with the new parameter configuration.

[0105] In a preferred embodiment of the present invention, before generating compensation data points, the point cloud optimization module further includes a defect labeling result credibility verification: constructing a defect labeling credibility evaluation dataset, which at least includes the local point cloud density, echo intensity variance, and geometric consistency features with neighboring valid point clouds of the defect points.

[0106] Specifically, local point cloud density is used to verify whether the spatial distribution pattern of defect points conforms to the typical characteristics of its labeled type. The evaluation method is to take the target defect point as the center, calculate the number of point clouds in a spherical neighborhood with a preset radius, and normalize it into a density value.

[0107] Local point cloud density is used for verification logic targeting false and missing points.

[0108] For false points: if their local density is abnormally high, it is consistent with the cause of their discrete reflection interference and the marking is reliable; if they exist in isolation, they are not reliable and may be noise.

[0109] For missing points: if the density of the area where the missing point is located is lower than the average density of the environment, then it is consistent with the expectation of shading or transmission and the marking is reliable; otherwise, the marking is unreliable and may be a misjudgment.

[0110] Echo intensity variance is used to examine whether the reflection characteristics of a defect point conform to the physical mechanism of its labeling type. The evaluation method involves calculating the variance of the echo intensity of all points within the local neighborhood of the target defect point.

[0111] Echo intensity variance verification logic: For false points: if the variance of their echo intensity is higher than the variance of the intensity of the surrounding valid point cloud, then it meets the characteristics of high intensity noise generated by backscattering and is marked as reliable.

[0112] For missing regions caused by specular reflection: the valid points at their boundaries often exhibit extremely high intensity variance. Therefore, if the echo intensity variance of the missing point is higher than a preset multiple of the mean echo intensity variance of all point cloud data quantized, this feature can inversely support the credibility of the labeling of its neighboring missing regions.

[0113] For ambiguous points: if their intensity is significantly different from the expected intensity of the fitted local surface, then the data ambiguity is supported.

[0114] Geometric consistency features determine whether a defect point deviates from its surrounding physical environment from a three-dimensional geometric perspective. Specifically, the evaluation method involves fitting a local plane or surface of the effective point cloud near the defect point using principal component analysis, and calculating the distance from the defect point to this fitted geometry as the geometric consistency feature evaluation metric.

[0115] The geometric consistency feature is the verification logic for false points and ambiguous points: For false points: if the distance from the false point to the fitted plane exceeds a preset distance threshold, it indicates that the false point is suspended outside the real surface and is marked as reliable.

[0116] For ambiguous points: if the point can be successfully fused into the neighboring valid surface through the aforementioned projection correction algorithm, it proves that it is indeed a point with a positional deviation but which can be corrected, and the ambiguous point can be marked as trustworthy; otherwise, it is marked as untrustworthy.

[0117] Based on the evaluation dataset, the defect labeling results are filtered, and only defect points with a confidence level higher than the verification threshold are subjected to subsequent compensation and optimization processing.

[0118] In a preferred embodiment of the present invention, after the parameter adjustment module drives the LiDAR scanning parameters to be adjusted, it further includes validity verification: based on the newly acquired point cloud data after parameter adjustment, defect identification processing is performed again.

[0119] The defect identification results before and after parameter adjustment are compared and analyzed. The difference between the proportion of the number of dominant defect points before and after adjustment is compared with the proportion of the number of dominant defect points before adjustment. The result of this ratio calculation is used as the effectiveness evaluation index of parameter adjustment. If it is greater than 0, the adjustment is judged as an effective improvement; otherwise, it is judged as invalid.

[0120] Based on the effectiveness evaluation results, the logic of the defect cause diagnosis or parameter adjustment instruction is optimized through self-learning. Specifically, when the parameter adjustment triggered by a certain diagnostic rule is continuously judged as an effective improvement, the confidence weight of the diagnostic rule is increased.

[0121] Conversely, if a diagnostic rule is repeatedly deemed invalid, its confidence weight is reduced. If the confidence level falls below a preset confidence threshold, the rule is manually revised or deactivated. Additionally, alternative parameter adjustment instructions that are similar to the current defect characteristics but have previously resulted in effective improvements are automatically retrieved from the historical records and used as the preferred option for the next similar situation.

[0122] By iterating through the above process, the optimal parameter adjustment strategy for different complex working conditions can be gradually approximated.

[0123] This invention integrates point cloud defect features from the back end with front end scanning parameter adjustments in a closed loop. By analyzing the proportion and spatial distribution of various defect marker points, it diagnoses the causes of defects and directly drives the adjustment of LiDAR scanning parameters. This helps reduce the occurrence rate of point cloud defects, thereby maintaining optimal perception performance in complex and ever-changing environments.

[0124] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A lidar scanning ranging sensor suitable for complex working conditions, characterized in that, include: The data acquisition module acquires raw point cloud data of the lidar under complex working conditions. The raw point cloud data includes the spatial location information and intensity information of each point cloud. The defect identification module performs defect identification processing on the original point cloud data, identifies areas where the straight-line propagation assumption of light fails and areas with abnormal reflection characteristics, and marks false points, missing points, and ambiguous points in the areas. The point cloud optimization module, based on the defect identification results, performs compensation optimization on the marked points in the region to generate a set of compensation data points, and then fuses the set of compensation data points with the original point cloud data to obtain an optimized point cloud data set. The parameter adjustment module performs defect cause diagnosis based on the number and spatial distribution of various defect markers, obtains diagnostic results, and uses the diagnostic results to drive the adjustment of lidar scanning parameters. The scanning parameters include at least the angular resolution of the emitted beam and the time window width of the received signal. After adjusting the scanning parameters of the laser radar, the parameter adjustment module also includes validity verification: Based on the newly acquired point cloud data after parameter adjustment, defect identification processing is performed again; The defect identification results before and after parameter adjustment are compared and analyzed to evaluate the effectiveness of parameter adjustment; Based on the effectiveness evaluation results, the logic of the defect cause diagnosis or parameter adjustment instruction is optimized through self-learning. Specifically, when the parameter adjustment triggered by a certain diagnostic rule is continuously judged as an effective improvement, the confidence weight of the diagnostic rule is increased. Conversely, if a diagnostic rule is repeatedly deemed invalid, its confidence weight is reduced. If the confidence level falls below a preset confidence threshold, the rule is manually revised or deactivated. Additionally, alternative parameter adjustment instructions that are similar to the current defect characteristics but have previously resulted in effective improvements are automatically retrieved from the historical records and used as the preferred option for the next similar situation.

2. The lidar scanning ranging sensor suitable for complex working conditions according to claim 1, characterized in that, The process for identifying the failure area based on the assumption of rectilinear propagation of light includes: The beam propagation space between the lidar and the effective point cloud at a distance is defined as the area to be inspected. The area to be inspected is divided into three-dimensional voxels, and the point cloud density of each voxel is calculated. Based on the effective point cloud density of the area to be inspected, a density threshold is dynamically set, and a continuous set of voxels with a point cloud density lower than the density threshold is marked as a propagation path occlusion type missing region. Virtual rays are emitted from the origin of the lidar to each effective point cloud at a distance. The system detects whether there are physical contradictions in the ray path, such as a spatial segment without point clouds at the beginning but a spatial segment connecting to the effective point clouds at the end. All such spatial segments detected at the beginning are marked as regions with missing propagation paths. The set of the aforementioned occlusion-type and penetration-type missing regions is collectively defined as the region where the straight-line propagation assumption of light fails.

3. A lidar scanning ranging sensor suitable for complex working conditions according to claim 2, characterized in that, The process for identifying abnormal reflectance regions includes: Echo intensity features are extracted from the raw point cloud data to generate an intensity distribution histogram. Based on the intensity distribution histogram, the high-intensity region point cloud set and the low-intensity region point cloud set are separated. Spatial clustering analysis is performed on the high-intensity regional point cloud set to identify isolated point cloud clusters and mark the areas they occupy as discrete reflection interference regions; Boundary contours are extracted from the point cloud set of the low-intensity region to identify closed low-intensity regions on the macroscopic continuous surface, and these regions are marked as specular reflection type missing regions. The set of discrete reflection interference regions and specular reflection-type missing regions are collectively defined as regions with abnormal reflection characteristics.

4. A lidar scanning ranging sensor suitable for complex working conditions according to claim 3, characterized in that, The process of marking false points, missing points, and ambiguous points includes: Mark all point clouds within the discrete reflection interference area as false points; Mark the geometric center point of each three-dimensional voxel in the region where the rectilinear propagation assumption of light fails and in the region of specular reflection-type missing area as the missing point; The valid point cloud located at the end of the spatial segment retrieved in the propagation path penetration-type missing region is marked as an ambiguous point.

5. A lidar scanning ranging sensor suitable for complex working conditions according to claim 1, characterized in that, The process of generating the compensation data point set includes: The missing points are processed by spatial interpolation based on environmental geometric priors to generate compensation data points to replace the missing points. For ambiguous points, a local point cloud neighborhood centered on the ambiguous point is constructed. The local surface normal vector is estimated using the principal component analysis method. The spatial coordinates of the ambiguous point are corrected according to the spatial distribution trend of the point cloud in the neighborhood, and the corrected compensation data points are generated. All generated compensation data points are organized according to their original spatial location index to form a structured set of compensation data points; Each data point in the set of compensation data points is assigned a compensation type identifier to indicate that the data point originates from missing point interpolation or ambiguous point correction.

6. A lidar scanning ranging sensor suitable for complex working conditions according to claim 5, characterized in that, The process of fusing the compensation data point set with the original point cloud data includes: Remove all data points marked as false points from the original point cloud data, and retain the unmarked data points as a valid subset of the original point cloud; Insert the data points in the compensation data point set into the corresponding positions in the valid original point cloud subset according to their spatial location index; Spatial deduplication is performed on the merged point cloud data, and the data is recombined according to a unified data format to generate an optimized point cloud data set.

7. A lidar scanning ranging sensor suitable for complex working conditions according to claim 1, characterized in that, The defect cause diagnosis process includes: When the proportion of a certain type of defect is higher than the sum of the proportions of the other two types of defect, that type of defect is considered the dominant defect. Based on the type and spatial distribution characteristics of the dominant defect points, perform the corresponding defect cause diagnosis: If the dominant defect point is a false point and its spatial distribution shows the characteristic of clustering along a specific scanning angle, then the current defect is diagnosed as being caused by multipath reflection interference. If the dominant defect point is a missing point, and its spatial distribution shows continuous linear or planar features, then the current defect is diagnosed as being caused by signal attenuation due to propagation path obstruction or transmission. If the dominant defect point is an ambiguous point, and its spatial distribution highly overlaps with the high curvature surface region in the scene, then the current defect is diagnosed as echo distortion caused by complex surface scattering.

8. A lidar scanning ranging sensor suitable for complex working conditions according to claim 7, characterized in that, Adjusting the lidar scanning parameters based on the diagnostic results includes: If it is determined to be multipath reflection interference, then a command to reduce the angular resolution of the emitted beam is generated; If signal attenuation is detected, an instruction is generated to increase the width of the received signal time window; If the echo distortion is determined, an instruction is generated to adjust the angular resolution and time window width to the preset anti-interference mode. The parameter adjustment instruction is written into the lidar control register, driving it to perform data acquisition for the next scan cycle with the new parameter configuration.

9. A lidar scanning ranging sensor suitable for complex working conditions according to claim 1, characterized in that, Before generating compensation data points, the point cloud optimization module also includes verification of the reliability of the defect marking results: Construct a defect label credibility evaluation dataset, which includes at least the local point cloud density, echo intensity variance, and geometric consistency features with neighboring valid point clouds of the defect points; Based on the evaluation dataset, the defect labeling results are filtered, and only defect points with a confidence level higher than the verification threshold are subjected to subsequent compensation and optimization processing.

Images