Point cloud processing method, electronic equipment, vehicle and storage medium

Abstract

The invention provides a point cloud processing method, electronic equipment, a vehicle and a storage medium, and relates to the technical field of smart mines, unmanned driving and perception. The method comprises the steps of determining neighborhood points corresponding to points in a first point cloud according to position coordinates of the points in the first point cloud; according to the neighborhood points corresponding to the points in the first point cloud, geometric feature parameters of the points in the first point cloud are determined, and the geometric feature parameters are used for representing the distribution discrete degree of the neighborhood points corresponding to the points in the first point cloud; and filtering invalid points in the first point cloud according to the geometric feature parameters of the points in the first point cloud. The problem of random floating noisy points formed by equipment tail gas and hot gas in winter operation of the mine truck can be effectively solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent mines, unmanned driving and sensing, and particularly relates to a point cloud processing method, an electronic device, a vehicle and a storage medium. BACKGROUND

[0002] In recent years, point cloud data is collected by using a laser radar, and the collected point cloud data is used for mine area surveying, scheduling and navigation, which becomes a key to intelligent upgrading of coal mining. However, in winter operation, the device tail gas and hot gas form random floating noise points in the scanning range of the laser radar, which are mixed in the upper space and target point cloud, and produce dense, transient and unpredictable interference. SUMMARY

[0003] Therefore, the embodiments of the present application provide a point cloud processing method, an electronic device, a vehicle and a storage medium.

[0004] In a first aspect, the embodiments of the present application provide a point cloud processing method, comprising: determining, according to position coordinates of each point in a first point cloud, neighborhood points corresponding to each point in the first point cloud; determining, according to the neighborhood points corresponding to each point in the first point cloud, a geometric feature parameter of each point in the first point cloud, the geometric feature parameter being used to represent a distribution dispersion degree of the neighborhood points corresponding to each point in the first point cloud; and filtering out invalid points in the first point cloud according to the geometric feature parameter of each point in the first point cloud, to obtain a second point cloud, the invalid points including abnormal points caused by environmental interference.

[0005] In combination with the first aspect, in some implementation manners of the first aspect, the determining, according to the neighborhood points corresponding to each point in the first point cloud, of the geometric feature parameter of each point in the first point cloud comprises: calculating a covariance matrix of each point in the first point cloud according to the neighborhood points corresponding to each point in the first point cloud; performing eigenvalue decomposition on the covariance matrix of each point in the first point cloud respectively to obtain at least two eigenvalues corresponding to each point in the first point cloud, and determining the geometric feature parameter of each point in the first point cloud from the at least two eigenvalues, the at least two eigenvalues including a first eigenvalue and a second eigenvalue.

[0006] In combination with the first aspect, in some implementation manners of the first aspect, the filtering out of the invalid points in the first point cloud according to the geometric feature parameter of each point in the first point cloud comprises: filtering out scattered distribution points in the first point cloud.

[0007] In combination with the first aspect, in some implementation manners of the first aspect, the filtering out of the scattered distribution points in the first point cloud comprises: calculating a sum of the at least two eigenvalues corresponding to each point in the first point cloud based on the at least two eigenvalues; calculating a ratio of the second eigenvalue in the at least two eigenvalues to the sum of the at least two eigenvalues; and filtering out the point corresponding to the ratio if the ratio satisfies a target filtering condition.

[0008] With reference to the first aspect, in some implementations of the first aspect, the target filtering condition comprises a ratio less than a target ratio threshold.

[0009] With reference to the first aspect, in some implementations of the first aspect, filtering out the invalid points in the first point cloud according to the geometric feature parameter of each point in the first point cloud comprises: filtering out the invalid points in the first point cloud according to the geometric feature parameter of each point in the first point cloud and the reflection intensity of each point in the first point cloud.

[0010] With reference to the first aspect, in some implementations of the first aspect, the invalid points in the first point cloud satisfy the following conditions simultaneously: the geometric feature parameter satisfies an invalid point distribution condition; and the reflection intensity satisfies a target reflection intensity condition.

[0011] With reference to the first aspect, in some implementations of the first aspect, the target reflection intensity condition comprises a reflection intensity less than a target reflection intensity threshold.

[0012] With reference to the first aspect, in some implementations of the first aspect, after filtering out the invalid points in the first point cloud according to the geometric feature parameter of each point in the first point cloud to obtain the second point cloud, the method further comprises: determining the neighborhood points corresponding to each point in the second point cloud; determining the outlier points in the second point cloud according to the neighborhood points corresponding to each point in the second point cloud, the outlier points comprising points with a spatial distance greater than a target distance threshold from the neighborhood points and / or points with a number of neighborhood points less than a target number threshold; and filtering out the outlier points in the second point cloud to obtain a third point cloud.

[0013] With reference to the first aspect, in some implementations of the first aspect, after filtering out the outlier points in the second point cloud to obtain the third point cloud, the method further comprises: performing grid division on the third point cloud to obtain a grid point cloud; performing erosion on the grid point cloud to obtain a fourth point cloud; and performing dilation on the fourth point cloud to obtain a fifth point cloud.

[0014] With reference to the first aspect, in some implementations of the first aspect, determining the neighborhood points corresponding to each point in the first point cloud according to the position coordinates of each point in the first point cloud comprises: constructing a spatial index of the first point cloud according to the position coordinates of each point in the first point cloud, the spatial index comprising a KD tree; and determining the neighborhood points corresponding to each point in the first point cloud based on the spatial index.

[0015] In a second aspect, the embodiments of the present application further provide an electronic device, comprising: a data acquisition unit connected with at least one point cloud acquisition device at a vehicle end, configured to acquire environmental point cloud data at the vehicle end; a processor connected with the data acquisition unit, configured to filter the environmental point cloud data based on the point cloud processing method in the first aspect; and a memory connected with the processor, configured to store at least one of executable instructions of the processor, the environmental point cloud data before filtering and the environmental point cloud data after filtering.

[0016] In a third aspect, the embodiments of the present application further provide a vehicle, comprising: at least one point cloud acquisition device, configured to acquire environmental point cloud data of the vehicle; a controller connected with the point cloud acquisition device, configured to filter the environmental point cloud data based on the point cloud processing method in the first aspect; and a wireless communication unit, one end of which is connected with the point cloud acquisition device and the other end of which is connected with a remote server, and the remote server is configured to filter the environmental point cloud data based on the point cloud processing method in the first aspect.

[0017] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, which stores a computer program, and the computer program is configured to execute the point cloud processing method in the first aspect.

[0018] The technical scheme of the embodiments of the present application determines the neighborhood points corresponding to each point in the first point cloud according to the position coordinates of each point in the first point cloud; determines the geometric feature parameters of each point in the first point cloud according to the neighborhood points corresponding to each point in the first point cloud, and the geometric feature parameters are used to represent the distribution dispersion degree of the neighborhood points corresponding to each point in the first point cloud, so as to realize the local analysis of the distribution dispersion degree of each point based on the neighborhood points corresponding to each point in the point cloud. Then, the invalid points in the first point cloud are filtered out according to the local analysis result of the distribution dispersion degree, and the quality of the point cloud data is improved. The embodiments of the present application are particularly suitable for the point cloud processing demand in complex environments such as open-pit coal mines, and can effectively solve the problem of random floating noise points formed by equipment tail gas and hot gas in winter operation. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 Fig. 1 shows a flowchart of a point cloud processing method provided by an exemplary embodiment of the present application.

[0020] Figure 2 Fig. 2 shows a flowchart of a point cloud processing method provided by another exemplary embodiment of the present application.

[0021] Figure 3 Fig. 3 shows a flowchart of a point cloud processing method provided by another exemplary embodiment of the present application.

[0022] Figure 4Fig. 1 shows a flowchart of a point cloud processing method according to an example embodiment of the present application.

[0023] Figure 5 Fig. 2 shows a flowchart of a point cloud processing method according to another example embodiment of the present application.

[0024] Figure 6 Fig. 3 shows a flowchart of a point cloud processing method according to another example embodiment of the present application.

[0025] Figure 7 Fig. 4 shows a schematic diagram of original point cloud data according to an example embodiment of the present application.

[0026] Figure 8 Fig. 5 shows a schematic diagram of filtered point cloud based on second feature value and reflection intensity according to an example embodiment of the present application.

[0027] Figure 9 Fig. 6 shows a comparison between filtered point cloud based on second feature value and reflection intensity and filtered-out point cloud according to an example embodiment of the present application.

[0028] Figure 10 Fig. 7 shows a schematic diagram of filtered point cloud based on radius filtering and statistical filtering according to an example embodiment of the present application.

[0029] Figure 11 Fig. 8 shows a comparison between filtered point cloud based on radius filtering and statistical filtering and filtered-out point cloud according to an example embodiment of the present application.

[0030] Figure 12 Fig. 9 shows a schematic diagram of point cloud after erosion and dilation operations according to an example embodiment of the present application.

[0031] Figure 13 Fig. 10 shows a comparison between point cloud after erosion and dilation operations and point cloud before erosion and dilation operations according to an example embodiment of the present application.

[0032] Figure 14 Fig. 11 shows a comparison between final complete reserved point cloud and noise according to an example embodiment of the present application.

[0033] Figure 15 Fig. 12 shows a structural schematic diagram of a point cloud processing apparatus according to an example embodiment of the present application.

[0034] Figure 16 Fig. 13 shows a structural schematic diagram of a point cloud processing apparatus according to another example embodiment of the present application.

[0035] Figure 17 Fig. 14 shows a structural schematic diagram of a point cloud processing apparatus according to another example embodiment of the present application.

[0036] Figure 18 Fig. 1 shows a structural schematic diagram of a point cloud processing device according to an example embodiment of the present application.

[0037] Figure 19 Fig. 2 shows a structural schematic diagram of a point cloud processing device according to another example embodiment of the present application.

[0038] Figure 20 Fig. 3 shows a structural schematic diagram of a point cloud processing device according to another example embodiment of the present application.

[0039] Figure 21 Fig. 4 shows a structural schematic diagram of an electronic device according to an example embodiment of the present application.

[0040] Figure 22 Fig. 5 shows a structural schematic diagram of an electronic device according to another example embodiment of the present application.

[0041] Figure 23 Fig. 6 shows a structural schematic diagram of a point cloud processing system according to an example embodiment of the present application.

[0042] Figure 24 Fig. 7 shows a structural schematic diagram of a vehicle according to an example embodiment of the present application.

[0043] Figure 25 Fig. 8 shows a structural schematic diagram of a vehicle according to another example embodiment of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work, fall within the protection scope of the present application.

[0045] Point cloud data is a collection of a large number of discrete three-dimensional space points, each containing coordinate information (X, Y, Z) and can be attached with attributes such as reflectance intensity, color, or timestamp. By recording the spatial distribution of object surfaces, point cloud data truly restores the geometric structure of the environment and is an important form of three-dimensional digital expression. Point clouds are usually generated by LiDAR, RGB-D cameras, three-dimensional scanners, or millimeter wave / ultrasonic radars. LiDAR directly acquires high-precision three-dimensional point clouds by emitting laser beams and receiving reflected signals, and is the mainstream device in the field of autonomous driving (such as mechanical LiDAR and solid-state LiDAR). RGB-D cameras combine infrared structured light or time-of-flight (ToF) technology to output point clouds with depth information. Three-dimensional scanners are used for high-precision modeling of static scenes (such as industrial detection and cultural relic digitization). Millimeter wave / ultrasonic radars generate sparse point clouds and are mainly used for close-range obstacle detection. The main application scenarios of point cloud data include autonomous driving, surveying and remote sensing, industry and robotics, virtual reality (VR / AR), etc. With its rich three-dimensional information, point cloud data has become the core medium for intelligent systems to perceive the physical world.

[0046] In autonomous driving systems, point clouds record the spatial distribution of surrounding objects in three-dimensional coordinates, accurately representing the geometric characteristics of targets such as vehicles, pedestrians, and road structures, making up for the shortcomings of cameras in distance measurement and low-light environments. Through point cloud segmentation, target detection, and semantic understanding, autonomous vehicles can construct high-precision environment models in real time, supporting key functions such as positioning, path planning, and obstacle avoidance. For example, point cloud-based SLAM (Simultaneous Localization and Mapping) technology can achieve centimeter-level vehicle positioning, and multi-sensor fusion further enhances the robustness of perception in complex scenarios.

[0047] With the continuous development of open-pit coal mine unmanned stripping and coal mining technology, using LiDAR to collect environmental point cloud data to realize mine terrain mapping, intelligent scheduling, and vehicle autonomous navigation, etc. is gradually becoming a key link in the intelligent upgrading of the coal mining field. However, a large amount of hot gas and equipment exhaust emitted during winter operations will form randomly distributed floating noise points within the scanning range of the LiDAR. These point cloud noise points generated by hot gas and exhaust not only frequently appear above mine pits, dump sites, and other places, but also mix with target point clouds such as terrain, equipment, and ore, forming dense, transient, and unpredictable interference points, which significantly reduce the accuracy and efficiency of subsequent algorithms.

[0048] Therefore, point cloud filtering has become an indispensable link in preprocessing. Some point cloud filtering methods mainly rely on geometric distribution, statistical characteristics or echo intensity analysis, which are difficult to take into account the complexity and instantaneity of dynamic and static noise points in the special environment of open-pit coal mines. Therefore, when facing local high-temperature air flow disturbance and diversified operation scenes in the mine area, these filtering algorithms often appear incomplete filtering or over-filtering, which significantly affects the accuracy of key target detection and three-dimensional reconstruction.

[0049] Based on this, the embodiment of the application provides a point cloud processing method, comprising: determining the neighborhood points corresponding to each point in the first point cloud according to the position coordinates of each point in the first point cloud; determining the geometric feature parameters of each point in the first point cloud according to the neighborhood points corresponding to each point in the first point cloud, the geometric feature parameters being used to represent the distribution dispersion degree of the neighborhood points corresponding to each point in the first point cloud; and filtering out invalid points in the first point cloud according to the geometric feature parameters of each point in the first point cloud, to obtain a second point cloud, the invalid points including abnormal points caused by environmental interference. The method distinguishes and filters out hot gas and tail gas noise points in the point cloud by deeply mining the local structural features of the point cloud, and provides more accurate and robust point cloud data support for unmanned operation in the open-pit coal mine in winter conditions.

[0050] The technical solution of the embodiment of the application can be applied to a controller at the vehicle end or a server at the remote end. The controller at the vehicle end is arranged on a vehicle. The server at the remote end is in communication connection with a target vehicle, for example, through a wired or wireless network connection.

[0051] The vehicle is equipped with a point cloud acquisition device, which is exemplarily a laser radar, a millimeter wave / ultrasonic wave radar, etc., for acquiring environmental point cloud data around the vehicle to identify obstacles in the environment, which include but are not limited to other vehicles, pedestrians, etc. The number of point cloud acquisition devices is not limited in the embodiment of the application, and one or more can be set according to needs.

[0052] In the embodiment of the application, the controller at the vehicle end or the remote server adopts the provided point cloud processing method to perform filtering operation on the acquired environmental point cloud data, with the purpose of removing invalid points in the data, thereby laying a data foundation for subsequent application.

[0053] Optionally, the controller at the vehicle end or the server at the remote end is connected with a display, and at least one of the following information is displayed through the display: point cloud data before filtering, point cloud data after filtering, and information of obstacles identified based on the point cloud data after filtering, etc.

[0054] The server can be an interworking server or a background server between multiple heterogeneous systems, can also be a standalone physical server, can also be a server cluster or a distributed system composed of multiple physical servers, can also be a cloud server providing basic cloud computing services such as big data and artificial intelligence platforms, and the like.

[0055] Figure 1 A flowchart of a point cloud processing method provided by an example embodiment of the application is shown. As shown in the figure, Figure 1 The point cloud processing method provided by the embodiment of the application includes the following steps.

[0056] S10, determining the neighborhood points corresponding to each point of the first point cloud according to the position coordinates of each point in the first point cloud.

[0057] The first point cloud is a point cloud to be processed, and each point in the first point cloud contains position coordinate data.

[0058] The neighborhood points refer to other points within a preset number or a preset radius range closest to the current point in a three-dimensional space. The determination of the neighborhood points can be realized by constructing a spatial index structure such as a KD tree, an octree, and the like. These data structures can efficiently perform nearest neighbor search. In the determination of the neighborhood points, a fixed number of K-nearest neighbors or a neighborhood point set based on radius search can be selected according to a specific application scenario.

[0059] Exemplarily, the determination of the neighborhood points corresponding to each point of the first point cloud according to the position coordinates of each point in the first point cloud includes: constructing a spatial index of the first point cloud according to the position coordinates of each point in the first point cloud, the spatial index including a KD tree; and determining the neighborhood points corresponding to each point in the first point cloud based on the spatial index. This kind of implementation mode quickly locates the neighborhood point set of each point through an efficient spatial index structure, and lays a foundation for subsequent geometric feature analysis. The KD tree, as a binary tree structure, can recursively divide a three-dimensional space into multiple hyper-rectangular regions, and significantly improves the efficiency of neighborhood search. In actual application, the construction parameters of the KD tree can be dynamically adjusted according to the point cloud density, so as to balance the calculation accuracy and performance overhead.

[0060] For large-scale point cloud data, a parallel computing framework can also be used to accelerate the spatial index construction process. For example, the parallel computing capability of a GPU is used to process the point cloud data in blocks and construct multiple local KD trees at the same time, and then combine them into a global index structure. This distributed processing method can effectively meet the real-time requirements in high-density point cloud scenarios.

[0061] In the neighborhood point determination process, special attention should be paid to the processing of boundary points. For points located at the edge or sparse area of the point cloud, the number of neighborhood points may be insufficient. At this time, an adaptive radius search strategy can be used to dynamically adjust the search range until the minimum number of neighborhood points is met. At the same time, to avoid noise interference, a maximum search radius limit can be set to prevent misjudging points that are too far away as neighborhood points.

[0062] The maintenance of spatial index is also an important link. When the point cloud data needs to be dynamically updated, the incremental index update algorithm can avoid rebuilding the entire data structure and only make local adjustments to the affected area. This mechanism is particularly suitable for continuous acquisition of streaming point cloud processing scenarios, such as real-time environment perception systems in autonomous driving.

[0063] S20, according to the neighborhood points corresponding to each point in the first point cloud, determining the geometric feature parameters of each point in the first point cloud.

[0064] The geometric feature parameters are used to represent the distribution dispersion degree of the neighborhood points corresponding to each point in the first point cloud. The geometric feature parameters can include curvature, normal vector, planeness and other indicators, which can reflect the shape features and distribution characteristics of the local area of the point cloud.

[0065] Optionally, the geometric feature parameter can also be a feature value obtained by calculating the covariance matrix of the neighborhood points based on the PCA (Principal Component Analysis Algorithm) and performing eigenvalue decomposition. For example, the geometric feature parameter can be at least one of the first eigenvalue, the second eigenvalue and the third eigenvalue. The physical meaning of the first eigenvalue is the extension degree of the energy points in the first principal axis (the longest direction) in the maximum variance direction. The physical meaning of the second eigenvalue is the energy in the second principal axis (the medium direction). The physical meaning of the third eigenvalue is the energy in the third principal axis (the shortest direction). The discrimination based on the first eigenvalue, the second eigenvalue and the third eigenvalue includes: the point cloud of linear structure is in the form of elongated strip, the first eigenvalue of the point cloud of linear structure is much larger than the second eigenvalue, and the second eigenvalue is approximately equal to the third eigenvalue; the point cloud of planar structure is in the form of planar sheet, the first eigenvalue of the point cloud of planar structure is approximately equal to the second eigenvalue, and the second eigenvalue is much larger than the third eigenvalue; the point cloud of scattered or volumetric structure is relatively uniform in three-dimensional direction, and the first eigenvalue of the point cloud of scattered or volumetric structure is approximately equal to the second eigenvalue and the third eigenvalue. The selection of the geometric feature parameter can be determined according to different application scenarios. For example, if the target noise point is a linear structure, the first eigenvalue can be selected as the geometric feature parameter based on the distribution characteristics of the eigenvalues of the point cloud of the linear structure.

[0066] S30, filtering out invalid points in the first point cloud according to the geometric feature parameters of each point in the first point cloud, to obtain a second point cloud.

[0067] In the field of point cloud processing, invalid points generally refer to point cloud data that does not meet the expected quality requirements or cannot be used for effective analysis. The specific invalid points can be set according to the actual needs of the application scenario. Exemplarily, the invalid points can be point clouds of linear structures, point clouds of planar structures, or scattered points. These invalid points can include various types, such as noise points, outlier points, low-confidence points, etc. In terms of spatial characteristics, noise points are completely randomly distributed, outlier points are locally deviated but may be clustered, and low-confidence points are related to object attributes. Taking the open-pit coal mine unmanned application scenario as an example, the invalid points can be abnormal points caused by environmental interference. Exemplarily, the invalid points can be caused by various reasons, such as visual interference caused by vehicle exhaust emission, or thermal gas interference caused by heat sources, which will form invalid points during data collection.

[0068] The second point cloud is the remaining points after filtering out the invalid points based on the geometric feature parameters in the first point cloud.

[0069] An invalid point discrimination criterion is established according to the geometric feature parameters. The invalid points in the first point cloud are filtered out using the invalid point discrimination criterion. Taking the example of the geometric feature parameters being the eigenvalues obtained by calculating the covariance matrix of the neighborhood points based on PCA (Principal Component Analysis), for random invalid points formed by thermal gas or exhaust gas, the neighborhood points of the random invalid points usually exhibit disordered distribution, which is manifested as three eigenvalues being similar and having small values. By setting a reasonable threshold condition, these invalid points can be accurately identified and filtered out.

[0070] The technical solution of the embodiments of the present application calculates the geometric feature parameters of each point in the point cloud based on the neighborhood points of each point in the point cloud. These parameters can reflect the dense or sparse distribution of the neighborhood points, so the geometric feature parameters of each point can be used to analyze the local distribution dispersion of each point in the point cloud. By using these geometric feature parameters, the invalid points in the first point cloud are filtered out, thereby improving the quality of the point cloud data. The embodiments of the present application are particularly suitable for point cloud filtering processing in complex environments such as open-pit coal mines, and effectively solve the problem of random floating noise points caused by equipment exhaust and thermal gas in winter operations.

[0071] Meanwhile, combining the reflection intensity information of the point cloud can further improve the accuracy of filtering and avoid misjudging low-reflectivity real targets as noise.

[0072] Exemplarily, filtering out the invalid points in the first point cloud according to the geometric feature parameters of each point in the first point cloud includes: filtering out the invalid points in the first point cloud according to the geometric feature parameters of each point in the first point cloud and the reflection intensity of each point in the first point cloud.

[0073] The invalid points in the first point cloud simultaneously satisfy the following conditions: the geometric feature parameters satisfy the invalid point distribution condition; and the reflection intensity satisfies the target reflection intensity condition.

[0074] The invalid point distribution condition is used to describe the local distribution characteristics of the invalid points. The invalid point distribution condition is determined by the determination of the geometric feature parameters and the filtering logic of the invalid points based on the geometric feature parameters, which will not be described herein.

[0075] The target reflection intensity condition includes that the reflection intensity is less than a target reflection intensity threshold. The target reflection intensity threshold is the maximum value of the reflection intensity of the invalid points in the case where the distribution of the invalid points meets the invalid point distribution condition.

[0076] Figure 2 Fig. 2 shows a flowchart of a point cloud processing method provided by another example embodiment of the present application. As shown in Fig. 2, the point cloud processing method provided by the embodiment of the present application includes the following steps. Figure 2 As shown in Fig. 2, in the point cloud processing method provided by the embodiment of the present application, S20 includes the following steps.

[0077] S21, calculating the covariance matrix of each point in the first point cloud according to the neighborhood points corresponding to each point in the first point cloud.

[0078] It can be understood that each point has its corresponding neighborhood points in the first point cloud, and the covariance matrix of each point can be determined through the neighborhood points. The number of neighborhood points is at least one, and when the number of neighborhood points is too small, the subsequent steps can not be performed, and therefore the points meeting the subsequent calculation requirements can be selected according to the number of neighborhood points to perform the subsequent steps.

[0079] S22, performing eigenvalue decomposition on the covariance matrix of each point in the first point cloud respectively to obtain at least two eigenvalues corresponding to each point in the first point cloud, and determining the geometric feature parameters of each point in the first point cloud from the at least two eigenvalues.

[0080] The at least two eigenvalues include a first eigenvalue and a second eigenvalue, the first eigenvalue representing the main direction of the local distribution of the position of the point, and the second eigenvalue representing the secondary direction of the local distribution of the position of the point, and usually the first eigenvalue is greater than the second eigenvalue.

[0081] Optionally, the at least two eigenvalues can further include a third eigenvalue. The physical meaning of each eigenvalue is the same as described in the above embodiment, which will not be described herein. For example, at least one of the first eigenvalue, the second eigenvalue and the third eigenvalue can be selected as the geometric feature parameters of each point in the first point cloud.

[0082] The technical scheme of the embodiment of the present application calculates the covariance matrix of each point in the first point cloud, respectively performs eigenvalue decomposition on the covariance matrix of each point in the first point cloud, selects at least one of the first eigenvalue, the second eigenvalue and the third eigenvalue as the geometric feature parameter of each point in the first point cloud, and thus realizes accurate classification and noise filtering of point cloud data. The eigenvalue analysis effectively distinguishes real target points and environmental interference points, and has a significant advantage in particular for transient noise formed by thermal gas disturbance in an open coal mine scene.

[0083] In the eigenvalue decomposition process, the eigenvalue combination can be flexibly selected according to different application requirements. For example, in a scene where the linear structure feature needs to be retained, the ratio relationship between the first eigenvalue and the second eigenvalue can be mainly referred to. In the specific implementation process, the eigenvalue threshold can be dynamically adjusted to adapt to the changes of different environmental conditions. For example, when operating in a foggy day with low visibility, the filtering condition of the second eigenvalue can be appropriately relaxed to avoid excessive filtering of real point cloud data. At the same time, the method supports fusion judgment with multi-dimensional information such as reflection intensity and echo times, and further improves the processing robustness in complex working conditions.

[0084] Figure 3 Fig. 2 shows a flowchart of a point cloud processing method provided by another exemplary embodiment of the present application. As shown in Fig. 2, Figure 3 Figure 2 On the basis of the point cloud processing method shown in Fig. 1, S30 includes the following steps.

[0085] S31, filtering out the scattered distribution points in the first point cloud.

[0086] The scattered distribution points refer to the point cloud data that is irregularly and non-uniformly distributed in space, and the spatial arrangement lacks obvious geometric structure or topological continuity. In contrast, the non-scattered distribution points can include point clouds in linear structures in the form of slender strips and point clouds in planar structures in the form of planar sheets. Alternatively, there are many ways to filter out the scattered distribution points in the first point cloud, for example, the points with the first eigenvalue approximately equal to the second eigenvalue and the second eigenvalue approximately equal to the third eigenvalue can be filtered out as the scattered distribution points according to the characteristic that the point cloud in the scattered structure is relatively uniform in three-dimensional direction.

[0087] Continuing to refer to Fig. 2, Figure 3 To improve the accuracy of determining the scattered distribution points, filtering out the scattered distribution points in the first point cloud includes:

[0088] S311, calculating the sum of the at least two eigenvalues corresponding to each point in the first point cloud based on the at least two eigenvalues.

[0089] S312, calculating the ratio of the second eigenvalue in the at least two eigenvalues to the sum of the at least two eigenvalues. ​

[0090] S313, if the ratio meets a target filtering condition, filtering out the point corresponding to the ratio.

[0091] The target filtering condition includes that the ratio is less than a target ratio threshold.

[0092] The technical solution of the embodiment of the application filters out invalid points through the ratio of the second feature value to the sum of feature values, and the ratio of the second feature value to the sum of feature values can effectively represent the anisotropy degree of the local structure of the point cloud. By setting a scientific target ratio threshold, accurate identification and filtering of invalid points can be achieved. In actual application, the target ratio threshold can be dynamically adjusted according to specific scenes. For example, in a mining area operating environment, the threshold can be appropriately reduced to retain more potential valid points; and in a structured environment such as a highway, the threshold standard can be increased to obtain a cleaner filtering effect.

[0093] Figure 4 Fig. 2 shows a flowchart of a point cloud processing method provided by another exemplary embodiment of the application. As shown in Fig. 2, based on the point cloud processing method provided by the above embodiment, after S30, the point cloud processing method provided by the embodiment of the application further includes the following steps. Figure 4

[0094] S40, determining the neighborhood points corresponding to each point in the second point cloud.

[0095] S50, determining the outlier points in the second point cloud according to the neighborhood points corresponding to each point in the second point cloud.

[0096] The outlier points include points with a spatial distance greater than a target distance threshold from the neighborhood points, and / or points with a number of neighborhood points less than a target number threshold.

[0097] S60, filtering out the outlier points in the second point cloud to obtain a third point cloud.

[0098] Optionally, the outlier points in the second point cloud are filtered out based on radius filtering and / or statistical filtering. Radius filtering is a point cloud denoising algorithm based on spatial density, which is used to remove discrete outlier points. The core idea of radius filtering is that if the number of neighborhood points of a point within a given radius neighborhood is less than a threshold, the point is determined to be noise and is deleted. The embodiment of the application further excludes noise points with a low number of neighborhood points within a local radius through radius filtering.

[0099] Statistical filtering is a denoising algorithm based on the local statistical characteristics of the point cloud, which removes outlier points deviating from the main distribution by analyzing the distance distribution of the neighborhood of each point. The core idea of statistical filtering is that if the average distance of a point to its neighborhood points exceeds a global reasonable range, the point is determined to be noise. The embodiment of the application calculates the dispersion degree of each point relative to the neighborhood through statistical filtering, and removes outlier values deviating from the surrounding environment. ​

[0100] To improve the filtering accuracy of the point cloud, two-stage filtering can be performed on the second point cloud based on radius filtering and statistical filtering. Specifically, the execution order of each filtering algorithm in the two-stage filtering is not limited in the embodiments of the present application. For example, the radius filtering is performed before the statistical filtering.

[0101] For parameter setting of the radius filtering and the statistical filtering, the system automatically adjusts the target distance threshold and the number of neighbor points threshold according to the point cloud density. A smaller search radius is used in a dense area of the point cloud, and the search range is appropriately expanded in a sparse area, so that consistent filtering effects can be obtained in different density areas. Meanwhile, the system records historical filtering parameters, and continuously optimizes the parameter selection strategy through a machine learning algorithm to improve the adaptive ability.

[0102] In specific implementation, the system also supports combined application of multiple filtering algorithms. For example, filtering based on geometric features can be combined with filtering based on reflection intensity, or statistical filtering can be combined with conditional filtering to form a more flexible point cloud processing flow.

[0103] The technical solution provided in the embodiments of the present application realizes segmented double filtering by further removing outliers after filtering based on geometric feature parameters. In the process of performing segmented double filtering, the system first removes obvious invalid points through geometric feature analysis, and then detects potential outliers through spatial distribution characteristics. This phased processing method can effectively improve the filtering accuracy and avoid the problem of missing invalid points caused by single filtering.

[0104] Figure 5 Fig. 2 shows a flowchart of a point cloud processing method provided by another exemplary embodiment of the present application. As shown in Fig. 2, the point cloud processing method provided by the embodiment of the present application includes the following steps. Figure 5 As shown in Fig. 2, on the basis of the point cloud processing method provided in the above embodiment, the point cloud processing method provided by the embodiment of the present application further includes the following steps after S60.

[0105] S70, performing grid division on the third point cloud to obtain a grid point cloud.

[0106] S80, performing erosion on the grid point cloud to obtain a fourth point cloud.

[0107] S90, performing dilation on the fourth point cloud to obtain a fifth point cloud.

[0108] In the specific implementation of the erosion and dilation operations, the adaptive grid size strategy is adopted in the embodiments of the present application to dynamically adjust the grid unit size according to the point cloud density. A smaller grid is used for a high-density area to retain detailed features, and the grid size is appropriately increased in a low-density area to ensure processing efficiency. The erosion operation preferentially removes isolated small-scale noise groups and shrinks thin areas, and the dilation operation can effectively repair missing real point clouds caused by excessive filtering.

[0109] The technical solution of this application employs gridded point cloud erosion and dilation techniques. Based on multi-level filtering, it utilizes grid-based erosion and dilation operations to prevent over-filtering of edge or sparse regions and appropriately expand weak or broken areas, thereby improving the overall detail and continuity of the point cloud. Furthermore, by combining local geometric feature analysis with multi-dimensional information fusion, the technical solution of this application effectively addresses dynamic noise interference in the complex environment of open-pit coal mines. Through adaptive threshold settings and hierarchical processing strategies, it maximizes the retention of useful information while ensuring filtering effectiveness, significantly improving the reliability of point cloud data in applications such as autonomous driving and intelligent scheduling.

[0110] Figure 6 The diagram shown is a flowchart illustrating a point cloud processing method provided in another exemplary embodiment of this application. Figure 6 As shown, taking the following application scenario as an example, in the unmanned operation scenario of open-pit coal mines, the large amount of hot air released by the mine trucks and the exhaust gas emitted by the equipment during winter operations will generate randomly distributed floating noise points in the scanning area of ​​the lidar. These point cloud noise points caused by hot air and exhaust gas often appear above the mine pit and spoil heap, and mix with the point cloud of targets such as terrain, equipment, and ore, forming dense, transient, and unpredictable interference points. The steps of the point cloud processing method provided in this application embodiment are as follows.

[0111] S100 reads the point cloud and verifies the number of points.

[0112] After acquiring raw point cloud data from LiDAR or other sensors, the size of the raw point cloud is first determined. If the number of points in the raw point cloud is less than a first threshold, it is considered an abnormal point cloud and no further processing is performed. The raw point cloud that passes the verification is used as the first point cloud. The first threshold can range from 500 to 2000. For example, the first threshold can be 1500. This is only an example, and the value of the first threshold is not limited; it can be determined according to actual needs. Figure 7 The image shown is a schematic diagram of the original point cloud data provided in an embodiment of this application. Figure 7 As shown, the blue area represents the acquired point cloud data, and the direction indicated by the arrow represents the data at the noise points caused by hot air or exhaust fumes.

[0113] S200: Construct a Kd-tree for the first point cloud that has passed the test, and search for its neighborhood points within a set radius for each point in the first point cloud based on the Kd-tree.

[0114] Based on the constructed Kd-tree, neighborhood points within a certain radius around each point can be quickly found to perform local spatial feature analysis and provide efficient support for subsequent neighborhood search.

[0115] S300, filter out the points in the first point cloud whose number of neighborhood points is less than the second number threshold, calculate the covariance matrix of the neighborhood points of each point in the remaining first point cloud by PCA, and perform eigenvalue decomposition.

[0116] The second number threshold is much smaller than the first number threshold. The value range of the second number threshold can be 2 to 10. Exemplarily, the second number threshold can be 5. Similarly, the value of the second number threshold is only an example, and is not limited, and can be determined according to actual needs.

[0117] S400, determine whether each point in the first point cloud satisfies the filtering condition in combination with the second eigenvalue and the reflection intensity.

[0118] The filtering condition includes that the ratio of the second eigenvalue to the total sum of eigenvalues is less than a target ratio threshold, and the reflection intensity of the corresponding point is less than a target reflection intensity threshold.

[0119] When the ratio of the second eigenvalue to the total sum of eigenvalues of each point in the first point cloud is less than the target ratio threshold, and the reflection intensity of the corresponding point is less than the target reflection intensity threshold, the corresponding point can be regarded as a noise point with large interference of hot gas or tail gas. Figure 8 The figure shows a point cloud filtered based on the second eigenvalue and the reflection intensity provided by the embodiments of the present application. The blue area is the point cloud filtered based on the second eigenvalue and the reflection intensity. Figure 9 The figure shows a comparison diagram of the point cloud filtered based on the second eigenvalue and the reflection intensity and the filtered point cloud provided by the embodiments of the present application. The blue area is the point cloud filtered based on the second eigenvalue and the reflection intensity, and the white area is the filtered point cloud.

[0120] S500, perform two-stage filtering on the second point cloud retained after filtering based on radius filtering and statistical filtering.

[0121] The second point cloud is the point cloud retained after the first point cloud is filtered based on the second eigenvalue and the reflection intensity. Figure 10 The figure shows a point cloud filtered based on radius filtering and statistical filtering provided by the embodiments of the present application. The blue area is the point cloud filtered based on radius filtering and statistical filtering. Figure 11 The figure shows a comparison diagram of the point cloud filtered based on radius filtering and statistical filtering and the filtered point cloud provided by the embodiments of the present application. The blue area is the point cloud filtered based on radius filtering and statistical filtering, and the white area is the filtered point cloud.

[0122] S600, perform grid division based erosion and expansion operations on the retained third point cloud.

[0123] The third point cloud is the point cloud reserved after the first point cloud is filtered by the radius filtering and the statistical filtering. After the processing of S500, the reserved third point cloud may have the problem of over-segmentation. The embodiments of the present application perform the grid division-based erosion and dilation operations on the third point cloud to avoid over-filtering of edge or special region points in the aforementioned multi-stage filtering.

[0124] The embodiments of the present application remove scattered points by erosion, shrink thin areas, and restore edges or special regions by re-including some points that may be excessively removed by dilation. After the point cloud is subjected to these operations, the drift noise points generated due to the interference of hot gas and exhaust gas can be greatly reduced. Figure 12 Fig. 6 shows a schematic diagram of the point cloud after the erosion and dilation operations provided by the embodiments of the present application. The blue region is the point cloud after the erosion and dilation operations. Figure 13 Fig. 7 shows a comparative schematic diagram of the point cloud after the erosion and dilation operations and the point cloud before the erosion and dilation operations provided by the embodiments of the present application. The blue region is the point cloud after the erosion and dilation operations, and the white region is the point cloud before the erosion and dilation operations. Figure 14 Fig. 8 shows a comparative schematic diagram of the final complete reserved point cloud and noise points provided by the embodiments of the present application. The blue region is the final complete reserved point cloud, and the white region is the noise point cloud removed by the aforementioned multiple rounds of processing.

[0125] The technical scheme of the embodiments of the present application can achieve the following technical effects. The local high-frequency interference such as hot gas and exhaust gas is effectively suppressed: by local analysis based on eigenvalues, noise points subjected to strong hot gas and exhaust gas interference are identified and removed, and the damage to the point cloud precision in typical industrial scenes or outdoors is reduced. The multi-filtering mechanism guarantees the data integrity: the combination strategy integrating PCA, radius filtering and statistical filtering can, on the premise of removing outliers and low-density interference, reserve effective feature points as much as possible, and avoid excessive removal leading to loss of key regions. The additive / subtractive operation balances the edge fidelity: by means of the grid processing of erosion and dilation, the problems such as loss of edge points and local fracture after multi-stage filtering are solved, and further repair and optimization of key structures are achieved. The accuracy and reliability of the final point cloud are improved: by identifying and separating various noise points, cleaner and more coherent final point cloud data are output, and the accuracy and stability of subsequent positioning, mapping, target recognition and other links are significantly improved.

[0126] Overall, the embodiments of the present application achieve good results in the balance between reducing noise and reserving effective information, so that the point cloud data can still maintain high integrity and precision under complex conditions, and the reliability of the point cloud in practical applications is enhanced.

[0127] Figure 15 Fig. 9 shows a structural schematic diagram of a point cloud processing device provided by an exemplary embodiment of the present application. As shown in Fig. 9, the point cloud processing device includes a processor 901 and a memory 902. Figure 15As shown, the point cloud processing apparatus 150 provided by the embodiments of the present application comprises a first determining module 151, a second determining module 152 and a filtering module 153.

[0128] The first determining module 151 is configured to determine, according to the position coordinates of each point in the first point cloud, the neighborhood points corresponding to each point in the first point cloud. The second determining module 152 is configured to determine, according to the neighborhood points corresponding to each point in the first point cloud, the geometric feature parameter of each point in the first point cloud, the geometric feature parameter being used to represent the distribution dispersion degree of the neighborhood points corresponding to each point in the first point cloud. The filtering module 153 is configured to filter out the invalid points in the first point cloud according to the geometric feature parameter of each point in the first point cloud, and obtain a second point cloud.

[0129] Figure 16 As shown in the structural schematic diagram of the point cloud processing apparatus provided by another exemplary embodiment of the present application. Figure 16 As shown in the structural schematic diagram of the point cloud processing apparatus provided by another exemplary embodiment of the present application. Figure 15 The second determining module 152 comprises a matrix calculation unit 1521 and an eigenvalue decomposition unit 1522 on the basis of the point cloud processing apparatus.

[0130] The matrix calculation unit 1521 is configured to calculate the covariance matrix of each point in the first point cloud according to the neighborhood points corresponding to each point in the first point cloud. The eigenvalue decomposition unit 1522 is configured to perform eigenvalue decomposition on the covariance matrix of each point in the first point cloud respectively, and obtain at least two eigenvalues corresponding to each point in the first point cloud, the at least two eigenvalues comprising a first eigenvalue and a second eigenvalue, and determine the geometric feature parameter of each point in the first point cloud from the at least two eigenvalues.

[0131] As shown in the structural schematic diagram of the point cloud processing apparatus provided by another exemplary embodiment of the present application. Figure 16 As shown in the structural schematic diagram of the point cloud processing apparatus provided by another exemplary embodiment of the present application. In some implementations, the filtering module 153 is configured to filter out the scattered distribution points in the first point cloud, and at this time, the filtering module 153 can comprise a sum calculation unit 1531 configured to calculate the sum of the at least two eigenvalues corresponding to each point in the first point cloud based on the at least two eigenvalues, a ratio calculation unit 1532 configured to calculate the ratio of the second eigenvalue in the at least two eigenvalues to the sum of the at least two eigenvalues, and a first filtering unit 1533 configured to filter out the point corresponding to the ratio if the ratio satisfies a target filtering condition. In some implementations, the target filtering condition comprises that the ratio is less than a target ratio threshold.

[0132] Figure 17 As shown in the structural schematic diagram of the point cloud processing apparatus provided by another exemplary embodiment of the present application. Figure 17 As shown in the structural schematic diagram of the point cloud processing apparatus provided by another exemplary embodiment of the present application. Figure 15On the basis of the point cloud processing apparatus, in some implementations, the filtering module 153 includes a second filtering unit 1534. The second filtering unit 1534 is configured to filter out invalid points in the first point cloud according to the geometric feature parameter of each point in the first point cloud and the reflection intensity of each point in the first point cloud.

[0133] In some implementations, the invalid points in the first point cloud satisfy the following conditions simultaneously: the geometric feature parameter satisfies the invalid point distribution condition; and the reflection intensity satisfies the target reflection intensity condition. In some implementations, the target reflection intensity condition includes that the reflection intensity is less than a target reflection intensity threshold.

[0134] Figure 18 Fig. 6 shows a structural schematic diagram of a point cloud processing apparatus provided by another example embodiment of the present application. As shown in Fig. 6, the point cloud processing apparatus includes a filtering module 153. Figure 18 As shown in Fig. 6, the filtering module 153 is configured to filter out invalid points in the first point cloud according to the geometric feature parameter of each point in the first point cloud and the reflection intensity of each point in the first point cloud. Figure 15 On the basis of the point cloud processing apparatus, the point cloud processing apparatus 150 provided by the example embodiment of the present application further includes a neighborhood point determination module 154, an outlier point determination module 155, and an outlier point filtering module 156.

[0135] The neighborhood point determination module 154 is configured to, after filtering out the invalid points in the first point cloud according to the geometric feature parameter of each point in the first point cloud and obtaining the second point cloud, determine the neighborhood points corresponding to each point in the second point cloud. The outlier point determination module 155 is configured to determine the outlier points in the second point cloud according to the neighborhood points corresponding to each point in the second point cloud. The outlier points include points whose spatial distance from the neighborhood points is greater than a target distance threshold and / or points whose number of neighborhood points is less than a target number threshold. The outlier point filtering module 156 is configured to filter out the outlier points in the second point cloud to obtain a third point cloud.

[0136] Figure 19 Fig. 7 shows a structural schematic diagram of a point cloud processing apparatus provided by another example embodiment of the present application. As shown in Fig. 7, the point cloud processing apparatus includes a filtering module 153. Figure 19 As shown in Fig. 7, the filtering module 153 is configured to filter out invalid points in the first point cloud according to the geometric feature parameter of each point in the first point cloud and the reflection intensity of each point in the first point cloud. Figure 17 On the basis of the point cloud processing apparatus, the point cloud processing apparatus 150 provided by the example embodiment of the present application further includes a gridization module 157, an erosion module 158, and an inflation module 159.

[0137] The gridization module 157 is configured to, after filtering out the outlier points in the second point cloud to obtain the third point cloud, perform grid division on the third point cloud to obtain a gridized point cloud. The erosion module 158 is configured to perform erosion on the gridized point cloud to obtain a fourth point cloud. The inflation module 159 is configured to perform inflation on the fourth point cloud to obtain a fifth point cloud.

[0138] Figure 20 Fig. 8 shows a structural schematic diagram of a point cloud processing apparatus provided by another example embodiment of the present application. As shown in Fig. 8, the point cloud processing apparatus includes a filtering module 153. Figure 20 As shown in Fig. 8, the filtering module 153 is configured to filter out invalid points in the first point cloud according to the geometric feature parameter of each point in the first point cloud and the reflection intensity of each point in the first point cloud. Figure 15The point cloud processing apparatus comprises a first determining module 151.

[0139] The KD tree construction unit 1511 is configured to construct a spatial index of the first point cloud according to position coordinates of each point in the first point cloud, the spatial index comprising a KD tree.

[0140] The technical scheme of the embodiment of the application can achieve the technical effects of the point cloud processing method provided by the embodiment of the application, which will not be described here.

[0141] Figure 21 Fig. 1 shows a structural schematic diagram of an electronic device provided by an exemplary embodiment of the application. As shown in Fig. 1, the electronic device 200 comprises a data acquisition unit 201, a processor 202 and a memory 203. Figure 21 The electronic device 200 can be a vehicle-end controller arranged on a vehicle, or a remote server in communication connection with the vehicle.

[0142] The data acquisition unit 201 is connected with at least one point cloud acquisition device of the vehicle end, and is configured to acquire environment point cloud data of the vehicle end. The point cloud acquisition device is a perception system installed on the vehicle, and can be a laser radar installed on the target vehicle. Exemplarily, if the electronic device 200 is a vehicle-end controller on the vehicle, the data acquisition unit can be a vehicle bus directly connected with at least one point cloud acquisition device of the vehicle end; if the electronic device 200 is a remote server in communication connection with the vehicle, the data acquisition unit can be a wireless communication unit wirelessly connected with at least one point cloud acquisition device of the vehicle end.

[0143] The processor 202 is connected with the data acquisition unit 201, and is configured to filter the environment point cloud data based on the point cloud processing method described above.

[0144] The memory 203 is connected with the processor 202, and is configured to store at least one of executable instructions of the processor 202, environment point cloud data before filtering and environment point cloud data after filtering. The memory 203 can comprise one or more computer program products, which can comprise various forms of computer readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory can comprise, for example, a random access memory (RAM), a cache memory and / or the like. The non-volatile memory can comprise, for example, a read-only memory (ROM), a hard disk, a flash memory and / or the like. One or more computer program instructions can be stored on the computer readable storage medium, and the processor 202 can run the program instructions to implement the point cloud processing method of each embodiment of the application described above and / or other desired functions.

[0145] In some implementations, the processor 202 is connected with a display, and controls the display to display at least one of the following information: the point cloud data before filtering, the point cloud data after filtering, and information of the obstacle identified based on the point cloud data after filtering.

[0146] The display can be independent of the electronic device 200. For example, the electronic device 200 is a vehicle terminal controller on a vehicle, and the display is a vehicle display on the vehicle.

[0147] Figure 22 Fig. 1 shows a structural schematic diagram of an electronic device provided by another example embodiment of the present application. As shown in the figure, Figure 22 In some embodiments, the electronic device 200 can further include an input device 204 and an output device 205, which are interconnected through a bus system and / or other forms of connection mechanism (not shown).

[0148] The input device 204 can include, for example, a touch screen, a keyboard, a mouse, etc. The output device 205 can output the point cloud data after filtering and the like to the outside. The output device 205 can include a display, a vehicle terminal, a communication network and a remote output device connected thereto, etc.

[0149] Of course, for simplicity, Figure 21 and Figure 22 Only some of the components in the electronic device 200 related to the present application are shown in Figs. 1 and 2, and components such as buses, input / output interfaces, etc. are omitted. In addition, the electronic device 200 can include any other appropriate components according to specific application cases.

[0150] Figure 23 Fig. 3 shows a structural schematic diagram of a point cloud processing system provided by an example embodiment of the present application. As shown in the figure, Figure 23 As shown in the figure, the point cloud processing system 220 provided by the embodiments of the present application includes at least one point cloud acquisition device 221 and the electronic device 200 described above. The point cloud acquisition device 221 and the electronic device 200 are connected through wired or wireless mode. The point cloud acquisition device 221 is configured to acquire the environmental point cloud data of the vehicle.

[0151] Figure 24 Fig. 4 shows a structural schematic diagram of a vehicle provided by an example embodiment of the present application. As shown in the figure, Figure 24 As shown in the figure, the vehicle 230 provided by the embodiments of the present application includes at least one point cloud acquisition device 221 and a controller 231.

[0152] The controller 231 is connected with the point cloud acquisition device 221, and is configured to filter the environmental point cloud data based on the point cloud processing method described above.

[0153] Figure 25 Fig. 1 shows a schematic diagram of a vehicle according to an example embodiment of the present application. Figure 25 Fig. 1 shows a schematic diagram of a vehicle according to an example embodiment of the present application.

[0154] The wireless communication unit 241 is connected to the point cloud acquisition device 221 at one end and to the remote server 242 at the other end, and performs filtering on the environmental point cloud data based on the point cloud processing method described above.

[0155] In addition to the method and device described above, an embodiment of the present application can also be a computer program product, which includes computer program instructions that, when executed by a processor, cause the processor to perform the steps of the point cloud processing method according to various embodiments of the present application described above in the specification.

[0156] The computer program product can be written in any combination of one or more programming languages, including an object-oriented programming language such as Java, C++, and conventional procedural programming languages such as "C" language or similar programming languages, to perform the operations of the embodiments of the present application. The program code can be executed entirely on a user computing device, partially on a user device, as an independent software package, partially on a user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0157] In addition, an embodiment of the present application can also be a computer readable storage medium, which stores computer program instructions, which, when executed by a processor, cause the processor to perform the steps of the point cloud processing method according to various embodiments of the present application described above in the specification.

[0158] The computer readable storage medium can use any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium may, for example, include but is not limited to an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device or instrument, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: electrical connections with one or more conductive wires, portable disks, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), optical fibers, portable compact disk read only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0159] The above describes the basic principles of the present application in combination with specific embodiments, but it needs to be pointed out that the advantages, benefits, effects and the like mentioned in the present application are only examples and are not limiting, and these advantages, benefits, effects and the like cannot be considered as necessary for each embodiment of the present application. In addition, the above specific details disclosed are only for the purpose of example and understanding, and are not limiting, and the above details do not limit the present application to be necessarily implemented with the above specific details.

[0160] The block diagrams of the devices, apparatuses, equipment, systems involved in the present application are only illustrative examples and are not intended to require or imply the connection, arrangement, configuration shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, equipment, systems can be connected, arranged, configured in any manner. Words such as "include", "contain", "have" and the like are open-ended words, mean "including but not limited to", and can be used interchangeably. The words "or" and "and" used herein mean the word "and / or", and can be used interchangeably unless the context clearly indicates otherwise. The word "such as" used herein means the phrase "such as but not limited to", and can be used interchangeably.

[0161] It also needs to be pointed out that in the devices, equipment and methods of the present application, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions of the present application.

[0162] The above description of the disclosed aspects is provided so that any person skilled in the art can make or use the present application. Various modifications to these aspects will be apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but is intended to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0163] The above description has been given for the purpose of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A method of point cloud processing, the method comprising: The method comprises the following steps: According to the position coordinates of each point in the first point cloud, the neighborhood points corresponding to each point in the first point cloud are determined; According to the neighborhood points corresponding to each point in the first point cloud, the geometric feature parameters of each point in the first point cloud are determined, which are used to represent the distribution dispersion degree of the neighborhood points corresponding to each point in the first point cloud; According to the geometric feature parameters of each point in the first point cloud, the invalid points in the first point cloud are filtered out to obtain a second point cloud.

2. The method of claim 1, wherein, The method comprises the following steps: According to the neighborhood points corresponding to each point in the first point cloud, the covariance matrix of each point in the first point cloud is calculated; The covariance matrix of each point in the first point cloud is respectively subjected to eigenvalue decomposition to obtain at least two eigenvalues corresponding to each point in the first point cloud, and the geometric feature parameters of each point in the first point cloud are determined from the at least two eigenvalues, wherein the at least two eigenvalues include a first eigenvalue and a second eigenvalue.

3. The method of claim 2, wherein, The method comprises the following steps: The method comprises the following steps: Preferably, the method comprises the following steps: Based on the at least two eigenvalues corresponding to each point in the first point cloud, the sum of the at least two eigenvalues is calculated; The ratio of the second eigenvalue in the at least two eigenvalues to the sum of the at least two eigenvalues is calculated; If the ratio meets the target filtering condition, the point corresponding to the ratio is filtered out; Preferably, the target filtering condition comprises that the ratio is less than a target ratio threshold.

4. The method of claim 1, wherein, The method comprises the following steps: According to the geometric feature parameters of each point in the first point cloud and the reflection intensity of each point in the first point cloud, the invalid points in the first point cloud are filtered out; Preferably, the invalid points in the first point cloud meet the following conditions simultaneously: The geometric feature parameter meets the invalid point distribution condition; The reflection intensity meets the target reflection intensity condition; Preferably, the target reflection intensity condition comprises that the reflection intensity is less than a target reflection intensity threshold.

5. The method of claim 1, wherein, After the method of filtering out the invalid points in the first point cloud according to the geometric feature parameters of each point in the first point cloud to obtain a second point cloud, the method further comprises the following steps: The neighborhood points corresponding to each point in the second point cloud are determined; According to the neighborhood points corresponding to each point in the second point cloud, the outlier points in the second point cloud are determined, wherein the outlier points include points with a spatial distance greater than a target distance threshold from the neighborhood points and / or points with a number of neighborhood points less than a target number threshold; The outlier points in the second point cloud are filtered out to obtain a third point cloud.

6. The method of claim 5, wherein, After the method of filtering out the outlier points in the second point cloud to obtain a third point cloud, the method further comprises the following steps: The third point cloud is subjected to grid division to obtain a grid point cloud; The grid point cloud is subjected to erosion to obtain a fourth point cloud; The fourth point cloud is subjected to inflation to obtain a fifth point cloud.

7. The method of claim 1, wherein, The determining of the neighborhood points corresponding to each point in the first point cloud according to the position coordinates of each point in the first point cloud comprises: constructing a spatial index of the first point cloud according to the position coordinates of each point in the first point cloud, the spatial index comprising a KD tree; determining the neighborhood points corresponding to each point in the first point cloud based on the spatial index.

8. An electronic device, comprising: Comprise: a data acquisition unit connected with at least one point cloud acquisition device at the vehicle end, configured to acquire environment point cloud data of the vehicle end; a processor connected with the data acquisition unit, configured to filter the environment point cloud data based on the point cloud processing method in any one of claims 1 to 7; a memory connected with the processor, configured to store at least one of executable instructions of the processor, environment point cloud data before filtering and environment point cloud data after filtering.

9. A vehicle characterized by comprising: Comprise: at least one point cloud acquisition device configured to acquire environment point cloud data of the vehicle; a controller or a wireless communication unit, the controller is connected with the point cloud acquisition device, and is configured to filter the environment point cloud data based on the point cloud processing method in any one of claims 1 to 7, the wireless communication unit is connected with the point cloud acquisition device at one end and connected with a remote server at the other end, and the remote server is configured to filter the environment point cloud data based on the point cloud processing method in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and the computer program is configured to execute the point cloud processing method in any one of claims 1 to 7.

Images