Method and device for detecting cliffs in the middle of a road based on laser radar point cloud

Abstract

The present application provides a method and device for detecting cliffs in the middle of a road based on a laser radar point cloud, and relates to the field of radar detection technology. The method for detecting cliffs in the middle of a road based on a laser radar point cloud includes: extracting preliminary point cloud data of the target road based on the received laser radar point cloud data; extracting ground point cloud data from the preliminary point cloud data according to a preset ground segmentation method; splicing the ground point cloud data based on the received inertial navigation positioning data and preset calibration parameters to determine spliced ​​ground point cloud data; determining clustering results based on the spliced ​​ground point cloud data and a preset clustering method, and extracting the middle cliff data of the target road based on the clustering results. The present application combines laser radar point cloud data, inertial navigation positioning data and calibration parameters to analyze the situation of the target road, and determines the middle cliff data in combination with ground splicing, thereby improving the detection efficiency of the middle cliff.

Description

Technical Field

[0001] The present application relates to the field of radar detection technology, for example, to a method and device for detecting a cliff in the middle of a road based on a laser radar point cloud. Background Art

[0002] With the advancement of science and technology, the autonomous driving industry is also developing rapidly. In the field of autonomous driving, various modules such as positioning, perception, and regulation and control are indispensable, especially the perception module, which is the "eye" of the autonomous vehicle. It can sense road obstacles and provide reference for the regulation and control module. In outdoor scenes, it is very common to have elevation jumps in the middle of the road. When encountering a cliff scene with elevation jumps, it is very easy for the vehicle to roll over or other unexpected situations if it does not take measures such as alignment and deceleration. Therefore, it is very necessary to detect cliffs in the middle of the road.

[0003] At present, existing perception modules seem to perceive and identify road elements, pedestrians, vehicles, and static or dynamic targets. In related technologies, road edge detection and road geometry information extraction are generally based on laser point clouds, and there is a lack of detection methods for cliff scenes with elevation jumps in the middle of the road. Summary of the invention

[0004] The present application aims to provide a method and device for detecting cliffs in the middle of a road based on a laser radar point cloud.

[0005] According to one aspect of the present application, a method for detecting a cliff in the middle of a road based on a lidar point cloud is proposed, comprising: extracting preliminary point cloud data of a target road based on received lidar point cloud data; extracting ground point cloud data from the preliminary point cloud data according to a preset ground segmentation method; splicing the ground point cloud data based on received inertial navigation positioning data and preset calibration parameters to determine spliced ​​ground point cloud data; determining a clustering result based on the spliced ​​ground point cloud data and a preset clustering method, and extracting the middle cliff data of the target road based on the clustering result.

[0006] According to one aspect of the present application, a detection device for intermediate cliffs of a road based on a laser radar point cloud is proposed, comprising: a preliminary point cloud data determination module, used to extract preliminary point cloud data of a target road based on received laser radar point cloud data; a ground segmentation module, used to extract ground point cloud data from the preliminary point cloud data according to a preset ground segmentation method; a ground splicing module, used to splice the ground point cloud data based on the received inertial navigation positioning data and preset calibration parameters to determine spliced ​​ground point cloud data; an intermediate cliff data determination module, used to determine a clustering result based on the spliced ​​ground point cloud data and a preset clustering method, and extract the intermediate cliff data of the target road according to the clustering result.

[0007] According to one aspect of the present application, an electronic device is proposed, which includes: a processor; a memory storing a computer program, and when the computer program is executed by the processor, the processor executes the method for detecting a cliff in the middle of a road based on a laser radar point cloud as described above.

[0008] According to one aspect of the present application, a non-transitory computer-readable medium is provided, on which readable instructions are stored. When the instructions are executed by a processor, the processor executes the method for detecting mid-road cliffs based on a laser radar point cloud as described above.

[0009] It should be understood that the foregoing general description and the following detailed description are exemplary only and are not restrictive of the present application.

[0010] Beneficial effects:

[0011] Through the above-mentioned embodiments provided by the present application, by utilizing the preset ground segmentation method, purer ground point cloud data can be further extracted from the preliminary point cloud data, reducing the interference of non-ground points and improving the efficiency and accuracy of subsequent processing. Combined with the received inertial navigation positioning data and the preset calibration parameters, the ground point cloud data can be accurately spliced ​​to form coherent spliced ​​ground point cloud data, overcoming the problems of single-frame lidar point cloud data and limited scanning range, and achieving comprehensive perception of road features over a wide range. Based on the spliced ​​ground point cloud data and the preset clustering method, accurate clustering results can be determined. These clustering results reflect the characteristic distribution of the road and its surrounding environment, and improve the accuracy of the intermediate cliff data. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without exceeding the scope of protection required by the present application.

[0013] Figure 1 A schematic diagram of an application scenario provided for an embodiment of the present application;

[0014] Figure 2 A flowchart of a method for detecting a cliff in the middle of a road based on a laser radar point cloud provided in an embodiment of the present application;

[0015] Figure 3 A flowchart for implementing the European clustering provided in the embodiment of the present application;

[0016] Figure 4 A logical diagram of coordinate system conversion provided in an embodiment of the present application;

[0017] Figure 5 A schematic diagram of a preferred clustering result provided in an embodiment of the present application;

[0018] Figure 6 A schematic diagram of target edge points after fitting provided in an embodiment of the present application;

[0019] Figure 7 A schematic diagram of a cliff boundary box provided in an embodiment of the present application;

[0020] Figure 8 A block diagram of a device for detecting a cliff in the middle of a road based on a laser radar point cloud provided in an embodiment of the present application;

[0021] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this application will be comprehensive and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted.

[0023] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present application. However, those skilled in the art will appreciate that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, known methods, devices, realizations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0024] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0025] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0026] It should be understood that although the terms first, second, third, etc. may be used herein to describe various components, these components should not be limited by these terms. These terms are used to distinguish one component from another component. Therefore, the first component discussed below can be referred to as the second component without departing from the teachings of the concepts of the present application. As used herein, the term "and / or" includes any one of the associated listed items and all combinations of one or more.

[0027] Figure 1 Schematic diagram of application scenario provided for an embodiment of the present application. An unmanned vehicle can be pre-set to detect the intermediate cliff of the target road. The detection method of the present application can be built on an industrial computer, and the industrial computer, the laser radar and the inertial navigation device can be installed on the unmanned vehicle. When the unmanned vehicle moves on the target road, the laser radar and the inertial navigation device can respectively detect the laser radar point cloud data and the inertial navigation positioning data, and send them to the industrial computer. The industrial computer can perform data analysis based on the received data to determine whether there is an intermediate cliff on the target road. If so, it can also determine the location and size of the cliff, that is, the intermediate cliff data.

[0028] For specific implementation methods, please refer to the following embodiments.

[0029] Figure 2 This is a flow chart of a method for detecting a road cliff based on a laser radar point cloud provided in an embodiment of the present application. The method of this embodiment can be applied to Figure 1 The industrial computer in the application scenario shown. Figure 2 As shown, the method includes: step S20, step S21, step S22 and step S23.

[0030] In step S20, preliminary point cloud data of the target road is extracted based on the received laser radar point cloud data.

[0031] In this application, the lidar point cloud data can be monitored and sent by the lidar installed on the unmanned vehicle. In some implementations, the lidar can be a solid-state lidar, or it can be replaced by a fusion of other mechanical lidars, such as using a horizontally placed 32-line Hesai lidar and a 32-line Suteng blind spot lidar together.

[0032] According to the example embodiment, laser radar point cloud data is first received. Since the present application detects the middle cliff of the target road, the data range of the middle cliff of the target road can be determined based on the preset relevant information of the target road, such as the road width, the road edge shape, etc. The laser radar point cloud data is eliminated based on the data range to obtain preliminary point cloud data.

[0033] In some implementations, the lidar point cloud data may be preprocessed such as denoising, and then the ground points may be identified, and then a plate may be fitted based on the ground points to obtain preliminary point cloud data.

[0034] In step S21, ground point cloud data is extracted from the preliminary point cloud data according to a preset ground segmentation method.

[0035] In the present application, the preset ground segmentation method can be a segmentation method based on height, slope, grid division, plane fitting or machine learning.

[0036] In some implementations, a reasonable height range can be set in advance based on the terrain characteristics of the target road to determine that points within the height range may be ground points. Then, for each point, the slope of the points adjacent to it, that is, the angle between the normal and the horizontal plane, can be calculated based on the preliminary point cloud data. Then, based on the preset slope threshold, points with excessive slopes are excluded to obtain ground points, and then the ground point cloud data is obtained.

[0037] In other implementations, a plane fitting method, such as a random sampling consistency algorithm, may be used to fit a plane. The distance from each point to the fitted plane may then be calculated, and based on a preset distance threshold range, points within the distance threshold range may be extracted as ground points to obtain ground point cloud data.

[0038] In step S22, the ground point cloud data is spliced ​​based on the received inertial navigation positioning data and preset calibration parameters to determine spliced ​​ground point cloud data.

[0039] In this application, the inertial navigation positioning data is collected and sent by the inertial navigation device built on the unmanned vehicle. The inertial navigation device may include an accelerometer, a gyroscope, etc. The inertial navigation positioning data may include acceleration, angular velocity, etc. The translation and rotation parameters of the laser radar to the inertial navigation device may be pre-calibrated, i.e., preset calibration parameters.

[0040] According to an example embodiment, a data stitching model may be preset, and stitched ground point cloud data may be obtained by inputting inertial navigation positioning data, preset calibration parameters, and ground point cloud data into the data stitching model.

[0041] In some implementations, the ground point cloud data can be transformed into a world coordinate system, spliced ​​based on the point cloud data in the world coordinate system, and then inversely transformed back to the lidar coordinate system to obtain spliced ​​ground point cloud data.

[0042] In step S23, a clustering result is determined based on the spliced ​​ground point cloud data and a preset clustering method, and the middle cliff data of the target road is extracted based on the clustering result.

[0043] In this application, the preset clustering method can be Euclidean clustering, which uses the Euclidean distance between points in the spliced ​​ground point cloud data for clustering. When the Euclidean distance between points is less than the set threshold, they are considered as one class. Euclidean clustering is accelerated based on the nearest neighbor query algorithm of KD-Tree. KD-Tree is an index structure evolved from the binary search tree for large-scale high-dimensional data search scenarios, mainly used for nearest neighbor search and approximate nearest neighbor search. KD-Tree divides the entire space into a binary tree, which is convenient and effective for finding the nearest neighbor.

[0044] The specific process of Euclidean clustering can be referred to Figure 3 ,refer to Figure 3 For a point P in space, the k points closest to P are found through the nearest neighbor search algorithm. Among these points, those whose distance is less than the set threshold are clustered into the set Q. If the number of elements in Q no longer increases, the entire clustering process ends; otherwise, points other than P are selected from the set Q and the above process is repeated until the number of elements in Q no longer increases.

[0045] Based on the above process, multiple clustering results can be obtained. Different clustering results may correspond to different ground features, such as roads, vegetation on the road surface, and cliffs. Since cliffs have certain point cloud features, such as the height and slope changes of the point clouds on both sides, it is possible to find out whether there is a cliff in the middle in the clustering results.

[0046] This application uses a preset ground segmentation method to further extract purer ground point cloud data from the preliminary point cloud data, reducing interference from non-ground points and improving the efficiency and accuracy of subsequent processing. Combined with the received inertial navigation positioning data and the preset calibration parameters, the ground point cloud data can be accurately spliced ​​to form coherent spliced ​​ground point cloud data, overcoming the problem of limited number and scanning range of single-frame lidar point cloud data, and achieving comprehensive perception of large-scale road features. Based on the spliced ​​ground point cloud data and the preset clustering method, accurate clustering results can be determined. These clustering results reflect the characteristic distribution of the road and its surrounding environment, and improve the accuracy of the intermediate cliff data.

[0047] According to some embodiments, outliers in the laser radar point cloud data may be filtered, which may be specifically achieved through the following steps: acquiring the laser radar point cloud data; filtering preset outliers in the laser radar point cloud data to determine preliminary point cloud data.

[0048] In this application, preset abnormal points can be used to indicate points that are too close or too far from the unmanned vehicle, reflected noise points, non-ground points, and distortion points caused by equipment or environmental factors, etc., where too close and too far can be pre-set based on the vehicle parameters of the unmanned vehicle, the equipment performance of the lidar used, etc.

[0049] According to the example embodiment, the laser radar point cloud data is first obtained, and the pre-set point cloud features corresponding to the preset abnormal points are compared with the laser radar point cloud data to determine the abnormal points and remove them to obtain preliminary point cloud data.

[0050] This application can significantly reduce noise points, non-ground points, distorted points due to equipment or environmental factors, and points at inappropriate distances by filtering preset abnormal points in the lidar point cloud data, effectively purifying the point cloud data and improving the accuracy and reliability of the data.

[0051] According to some embodiments, ground point cloud data can be extracted through region segmentation, ground fitting and further ground estimation, which can be specifically achieved through the following steps: according to the ground segmentation method, determine the region segmentation model, the ground fitting method and the ground estimation method; based on the region segmentation model, perform polar coordinate grid division on the preliminary point cloud data to determine the grid ground point cloud data; according to the ground fitting method, perform plane fitting on the grid ground point cloud data to determine the ground fitting plane; according to the ground estimation method and the ground fitting plane, determine the ground point cloud data in the preliminary point cloud data.

[0052] In this application, there are two problems with uniform grids: First, in autonomous driving, most point clouds are near the LiDAR. If a uniform grid is used, the distant points will be too sparse in the grid (too few points in the grid), resulting in failure to fit the plane. Second, if the grid is too close to the LiDAR, the grid will be too small, which will cause the plane normal vector estimation to fail or be wrong.

[0053] Therefore, the regional segmentation model used in the ground segmentation method of the present application can be a concentric zone model (CZM). CAM can be understood as a uniform or multi-resolution regional segmentation model; most multi-plane-based methods are based on the assumption that the observable world may not be flat. Therefore, the ground plane estimation should be overturned by assuming that the possible non-flat world has small blocks or bins, and the ground can indeed be flat in this local area. The number of points in each bin is different depending on the distance and angle. Compared with the uniform grid model, the CZM model proposed in Patchwork can reduce the number of grids and better represent the point cloud.

[0054] The ground fitting method in this application can be region-wise ground plane fitting (R-GPF), and the ground is estimated for each bin through R-GPF, and then the ground is merged. The ground segmentation method of this application uses principal component analysis, which is faster than RANSAC (Random Sample Consensus) for plane fitting, and the ground segmentation method uses adaptive initial seed selection to prevent R-GPF from converging to a local minimum.

[0055] The ground estimation method of the present application may be Adaptive Ground Likelihood Estimation (A-GLE). A-GLE is a regional probability test for binary classification, which obtains features such as verticality, height, and flatness of a point to determine whether the point cloud is the actual ground.

[0056] According to an example embodiment, CZM can be used to divide the preliminary point cloud data to obtain multiple grids and point cloud data within the grids, which are used as grid ground point cloud data as a whole. Then, R-GPF can be used to perform plane fitting on the point cloud data within each grid to obtain a ground fitting plane. Then, A-GLE can be used to calculate the distance from each point in the preliminary point cloud data to the ground fitting plane, and determine whether the point is a ground point based on the distance, a preset distance threshold, and a preset likelihood threshold. The data finally determined to be a ground point is output as ground point cloud data.

[0057] The present application divides the preliminary point cloud data into multiple grids by polar coordinate grid division, and performs independent plane fitting and ground estimation on each grid. This parallel processing method significantly improves the data processing efficiency and reduces the calculation time. At the same time, polar coordinate grid division can also better adapt to the distribution characteristics of ground point cloud data and improve the accuracy of data processing. The preliminary point cloud data is divided into multiple grids by polar coordinate grid division, and independent plane fitting and ground estimation are performed on each grid. This parallel processing method significantly improves the data processing efficiency and reduces the calculation time. The preliminary point cloud data is divided into multiple grids by polar coordinate grid division, and independent plane fitting and ground estimation are performed on each grid. This parallel processing method significantly improves the data processing efficiency and reduces the calculation time. At the same time, polar coordinate grid division can also better adapt to the distribution characteristics of ground point cloud data and improve the accuracy of data processing.

[0058] According to some embodiments, ground point cloud data can be spliced ​​by converting the coordinate system, which can be specifically achieved by the following steps: preprocessing the ground point cloud data according to preset preprocessing requirements to determine pure point cloud data; wherein the pure point cloud data includes multiple frames; acquiring inertial navigation positioning data and calibration parameters; converting multiple frames of pure point cloud data to an inertial navigation coordinate system according to the calibration parameters to determine multiple frames of inertial navigation point cloud data; determining the inertial navigation positioning data corresponding to the timestamp according to the timestamp of the ground point corresponding to the multiple frames of inertial navigation point cloud data; converting the multiple frames of inertial navigation point cloud data to a world coordinate system according to the inertial navigation positioning data to determine multiple frames of world point cloud data, and splicing the multiple frames of world point cloud data; performing inverse coordinate conversion on the spliced ​​multiple frames of world point cloud data according to the last frame of positioning data in the inertial navigation positioning data to determine the inverse inertial navigation point cloud data of the inertial navigation coordinate system at the time corresponding to the last frame; performing inverse coordinate conversion on the inverse inertial navigation point cloud data according to the calibration parameters to determine the spliced ​​ground point cloud data of the lidar coordinate system at the time corresponding to the last frame.

[0059] In this application, after extracting the ground point cloud data for each frame of point cloud, some preprocessing can be performed on the extracted ground point cloud data, corresponding to the preset preprocessing requirements, including cropping and downsampling. Cropping is to only retain the point cloud data within the preset range of the front, back, left and right of the unmanned vehicle for processing. After cropping, downsampling is performed at a preset size to reduce the number of point clouds and reduce running time.

[0060] According to an exemplary embodiment, the ground point data is first preprocessed as described above to obtain multiple frames of clean point cloud data. Figure 4 , according to the calibration parameters of the laser radar to the inertial navigation (inertial navigation device), the multi-frame pure point cloud data is converted to the inertial navigation coordinate system. Then, according to the timestamp of the ground point in each frame in the multi-frame, the inertial navigation positioning data at the corresponding moment is found, and the timestamp is carried by the pure point cloud data. Then, according to the inertial navigation positioning data, the multi-frame inertial navigation point cloud data can be converted to the world coordinate system. Since the world coordinate system is an absolute coordinate system, continuous multi-frame inertial navigation point cloud data can be converted into multi-frame world point cloud data and spliced.

[0061] However, it is not convenient to calculate the angle and distance of multi-frame world point cloud data in the world coordinate system, so the spliced ​​multi-frame world point cloud data can be reversely converted back to the lidar coordinate system. In some implementations, the ground point cloud in the world coordinate system can be converted to the inertial navigation coordinate system at the last frame time according to the last frame positioning data in the inertial navigation positioning data, and then the ground point cloud in the inertial navigation coordinate system is converted to the lidar coordinate system corresponding to the last frame positioning data according to the calibration parameters, so as to obtain the spliced ​​ground point cloud data of the lidar coordinate system at the time corresponding to the last frame.

[0062] This application realizes the accurate splicing of multi-frame ground point cloud data by combining inertial navigation positioning data and preset calibration parameters. Through the preprocessing step, the noise and interference in the point cloud data are removed to ensure the quality of the pure point cloud data. Subsequently, the pure point cloud data is converted to the inertial navigation coordinate system using the calibration parameters, and then converted to the world coordinate system according to the inertial navigation positioning data, realizing the unified coordinate representation of the point cloud data. This process ensures the consistency of the spliced ​​ground point cloud data in space and time, and provides a reliable basis for subsequent analysis and application. By matching the inertial navigation positioning data and the point cloud data with the timestamp, the rapid splicing of multi-frame world point cloud data is realized, avoiding the cumbersome process of manual alignment or relying on complex algorithms for splicing in the traditional method, and improving the splicing efficiency. Through the inverse coordinate conversion step, the spliced ​​multi-frame world point cloud data is converted back to the laser radar coordinate system at the time corresponding to the last frame, ensuring the integrity and consistency of the spliced ​​data, avoiding the data loss or deformation caused by the coordinate conversion, and ensuring the accuracy and reliability of the spliced ​​ground point cloud data in subsequent applications.

[0063] According to some embodiments, a sliding window method can be used for stitching, which can be specifically achieved through the following steps: based on the inertial navigation positioning data, multiple frames of inertial navigation point cloud data are converted to a world coordinate system to determine multiple frames of world point cloud data; and multiple frames of world point cloud data are stitched according to a preset sliding window method.

[0064] In this application, the inertial navigation positioning data usually includes three-dimensional position information such as longitude, latitude, and altitude, as well as attitude information such as pitch angle, yaw angle, roll angle, and timestamp. The origin and direction of the world coordinate system can be set in advance, and then the size of the sliding window can be set, which can determine the number of point cloud frames considered in the stitching process.

[0065] In some implementations, the timestamps of the inertial navigation positioning data and the inertial navigation point cloud data are first aligned. For each frame of the inertial navigation point cloud data, the inertial navigation point cloud data is converted to the world coordinate system according to the three-dimensional position information in the inertial navigation positioning data corresponding to its timestamp. According to the attitude information in the inertial navigation positioning data, the inertial navigation point cloud data is rotated to make it consistent with the direction of the world coordinate system. Repeat the above steps to convert each frame of the inertial navigation point cloud data to generate multiple frames of world point cloud data.

[0066] Then the size of the sliding window can be set to N frames, that is, the point cloud data of the current frame and the previous N-1 frames are considered. All frame point cloud data are unified into a unified coordinate system for splicing. When new point cloud data arrives, it is added to the sliding window and the old point cloud data in the window is removed. In some implementations, when the number of multi-frame splicing frames (N) is 4, the point cloud results of the last 3 frames in the world coordinate system are stored first. When the processing of these 4 frames of ground point cloud is completed, the new frame is spliced ​​with the ground point cloud in the last 3 frames of the world coordinate system that has been stored to form a new 4-frame ground point cloud, and then the ground point cloud in the last 3 frames of the world coordinate system is stored. When a new frame of ground point cloud arrives, it is processed according to the above process. This not only achieves the effect of multi-frame splicing but also improves the running time. As the sliding window slides, the spliced ​​point cloud data is continuously output until all frame point cloud data are processed.

[0067] This application can accurately convert multiple frames of inertial navigation point cloud data into a unified world coordinate system by utilizing inertial navigation positioning data. The high-precision position and orientation information provided by the inertial navigation system ensures that each frame of point cloud data can be accurately placed in the world coordinate system, thereby achieving high-precision stitching of multiple frames of world point cloud data. The use of a preset sliding window method to stitch multiple frames of world point cloud data not only improves the real-time performance of data processing, but also enhances the flexibility of the system. The sliding window method allows the system to adjust the window size according to actual needs, thereby optimizing computing resources and time costs while ensuring stitching accuracy.

[0068] According to some embodiments, the clustering results can be determined based on elevation expansion and Euclidean clustering, which can be specifically achieved through the following steps: obtaining a preset elevation expansion multiple of the stitched ground point cloud data; processing the stitched ground point cloud data according to the elevation expansion multiple to determine the elevation expanded ground point cloud data; performing Euclidean clustering on the elevation expanded ground point cloud data to determine the clustering results, and extracting the intermediate cliff data based on the clustering results.

[0069] The clustering method used in this application is Euclidean clustering. The spliced ​​ground point cloud data contains three-dimensional coordinates (X, Y, Z), where the Z coordinate can represent elevation data.

[0070] In some implementations, the Z value corresponding to each point can be extracted from the spliced ​​ground point cloud data as the corresponding elevation data, and then the elevation data is multiplied by a preset elevation expansion multiple, so that the Euclidean distance of the point cloud can be increased when the Euclidean clustering method is used to obtain a better distance result. In the present application, the preset elevation expansion digit can be 2 times. Using the Euclidean clustering method, clustering is performed based on the elevation-expanded ground point cloud data obtained after expanding the elevation, and the elevation-expanded ground point cloud data is divided into different groups, i.e., clustering results.

[0071] Then, the clustering results can be analyzed, which may specifically involve the analysis of features such as cluster center, cluster size, and shape. In some implementations, a threshold can be pre-set to filter clusters related to intermediate cliffs. For example, clustering results with elevation differences greater than a certain threshold can be considered to indicate the presence of intermediate cliffs, thereby obtaining intermediate cliff data. The intermediate cliff data can be used to characterize whether there is an intermediate cliff.

[0072] This application can fully cover the terrain information of the target road and its surroundings by splicing ground point cloud data. When processing elevation data, by expanding the elevation data based on the elevation expansion multiple, better Euclidean distance results can be obtained during Euclidean clustering. Subsequently, the elevation-expanded ground point cloud data after elevation expansion is processed using Euclidean clustering, which can more accurately identify and extract the intermediate cliff data of the target road. In practical applications, this can significantly reduce computing time and resource consumption, especially when processing large-scale point cloud data. By reasonably setting the clustering method and performing clustering based on the expanded elevation difference, the distribution characteristics of the ground point cloud data can be more accurately reflected, thereby obtaining more reliable clustering results.

[0073] According to some embodiments, the intermediate cliff data can be determined by obtaining the three-dimensional bounding box of the clustering result, which can be specifically achieved through the following steps: determining the corresponding three-dimensional bounding box according to the clustering result; determining the clustering information of the clustering result according to the three-dimensional bounding box; filtering the clustering result according to the clustering information to determine the target clustering result; when the number of target clustering results is less than two, determining that the intermediate cliff data of the target road is empty; when the number of target clustering results is greater than or equal to two, determining the two target clustering results with the largest number of ground points in the target clustering results as preferred clustering results; determining the average elevation data corresponding to the cliff high surface and the cliff low surface according to the preferred clustering result and the preset point cloud selection method; determining the average elevation difference according to the average elevation data; when the average elevation difference is less than the preset elevation difference threshold, determining that the intermediate cliff data is empty; when the average elevation difference is greater than or equal to the preset elevation difference threshold, determining that there is an intermediate cliff on the target road, and determining the size and position of the intermediate cliff according to the preferred clustering result as the intermediate cliff data.

[0074] In this application, a three-dimensional bounding box is a minimum cuboid that can completely enclose all point cloud data in a cluster. In some implementations, this application uses an OBB bounding box (Oriented Bounding Box). The clustering information may include the length and width of the clustering result, the position and size of the three-dimensional bounding box, and the amount of point cloud data contained in the clustering result.

[0075] According to some embodiments, since each point has three-dimensional coordinates, boundary information of a three-dimensional bounding box and the like can be determined, thereby obtaining clustering information.

[0076] Based on the length, width and preset length threshold and width threshold in the clustering information, some invalid clustering effects can be filtered out; based on the three-dimensional bounding box position in the clustering information and the preset position range limit, some clustering results far away from the center of the lidar can be filtered out; based on the size and position of the clustering results in the clustering information, as well as the preset position limit and size limit of the clustering results, some clustering results outside the preset enclosing range can be filtered out, and then the target clustering results can be obtained.

[0077] If the number of target clustering results is less than two, it can be directly determined that there is no intermediate cliff in the target road, that is, the intermediate cliff data is empty. If the number of target clustering results is greater than or equal to two, the number of ground points in the target clustering results can be detected and sorted, and the two target clustering results with the largest number can be determined as the preferred clustering results.

[0078] refer to Figure 5 , the preset point cloud selection method can be: extract the farthest edge point of the target clustering result that is closer to the center of the unmanned vehicle or the center of the lidar, that is Figure 5 The points in the border 1 shown are the points closest to the clustering results that are far from the center of the unmanned vehicle or the center of the lidar, that is, Figure 5 The points in the frame 2 shown in the figure are calculated. The average elevation data of the points in the frame 1 and the average elevation data of the points in the frame 2 are calculated, where the frame 1 and the frame 2 correspond to the high face of the cliff and the low face of the cliff respectively. The average elevation data of the two frames are subtracted to obtain the average elevation difference.

[0079] Figure 5 The extraction process of the points in the middle frame 1 and the frame 2 is: sort each target clustering result in the x direction. The selection method of the x direction can be specifically determined according to the laser radar coordinate system. In this application, Figure 5 The horizontal left direction is set as the x direction, the vertical upward direction is set as the z direction, and then the vertical right direction with the x direction as the front is set as the y direction.

[0080] According to the set size, the points are divided into different preset ranges in the x direction, and then the sorted points are processed from small to large in the x direction. The points in each preset range in the x direction are then sorted in the y direction to select the minimum or maximum y value point (determined according to the lidar coordinate system, in this application, the minimum y value point is selected in border 1, and the maximum y value point is selected in border 2).

[0081] The average elevation difference is compared with the preset elevation difference threshold. If the average elevation difference is less than the preset elevation difference threshold, it means that there is no intermediate cliff on the target road, that is, the intermediate cliff data is empty. If the average elevation difference is greater than or equal to the preset elevation difference threshold, it can be determined that there is an intermediate cliff on the target road, and then the size and position of the intermediate cliff can be calculated based on the preferred clustering result, and then the "existence of intermediate cliff" and the size and position of the intermediate cliff are determined as the intermediate cliff data.

[0082] The present application can accurately identify the distribution of clustering results by determining the three-dimensional bounding box and clustering information of the clustering results. Further filtering and optimization steps ensure that only clustering results directly related to the middle cliff are retained, thereby improving the accuracy of recognition. The rapid calculation of the three-dimensional bounding box and clustering information effectively reduces the complexity and time cost of data processing. The efficiency of data processing is further improved by preset filtering conditions and point cloud selection methods. When it is determined that there is an middle cliff, the size and position of the cliff can be accurately calculated (, which improves the accuracy of the detection of middle cliffs on the road.

[0083] According to some embodiments, the final clustering result can be further selected from the preferred clustering results, and the size and position of the intermediate cliff can be determined by combining the convex hull detection method and the three-dimensional bounding box. This can be specifically achieved through the following steps: when the average elevation difference is greater than or equal to the preset elevation difference threshold, it is determined that there is an intermediate cliff on the target road; according to the preset cliff data requirements, the final clustering result is determined in the preferred clustering results; based on the convex hull detection method, the optional edge points of the final clustering result are determined; according to the optional edge points and the preset edge point selection method, the target edge points are determined; according to the final clustering result, the three-dimensional bounding box of the target edge point is determined to determine the size and position of the intermediate cliff according to the three-dimensional bounding box; the size and position of the intermediate cliff are determined as the intermediate cliff data.

[0084] In this application, the preset cliff data requirement can be to select point cloud data with larger elevation data and closest to the center of the unmanned vehicle. The preset edge point selection method can be to fit the edge points to a straight line, and the edge points closer to the straight line will be retained, and then the points corresponding to the minimum elevation data and the points corresponding to the maximum elevation data are used as the final target edge points. The form after fitting can be referred to Figure 6 The border 3 in the figure is the target edge point.

[0085] The final clustering result is projected into a two-dimensional plane, and then the edge lines and edge points of the two-dimensional plane are extracted according to the convex hull detection method. The convex hull (convex hull) means that if the straight line segments connecting any two points in set A are inside A, then set A is called convex, that is, a polygon with no concave places. The convex hull (convex hull) can contain all the points in the point set. In this application, the pre-set convex hull detection function can be directly used.

[0086] In some implementations, when there is an intermediate cliff on the target road, the final clustering result is first determined from the preferred clustering results according to the preset cliff data requirements. The optional edge points of the final clustering result are obtained by directly calling the convex hull detection function. Then, according to the preset edge point selection method, the positions containing the optional edge points in the final clustering result are linearly fitted to obtain the target edge points.

[0087] Then, the final edge point in the direction far from the center of the unmanned vehicle can be extracted from the target edge point, and the three-dimensional bounding box of the final edge point can be calculated according to the above specific steps. Figure 7 Schematic diagram of the cliff bounding box shown in Figure 7 The cliff bounding box in is the 3D bounding box corresponding to the middle cliff. From the 3D bounding box, we can get the length of the middle cliff and the distance from the center of the unmanned vehicle, that is, the size and position. The size and position of the middle cliff are determined as the middle cliff data.

[0088] This application further screens the final clustering results from the preferred clustering results based on the preset cliff data requirements. This process ensures the accuracy and pertinence of the data and reduces unnecessary data processing. The convex hull detection method is used to determine the optional edge points of the final clustering results, which can effectively identify the outer contour of the clustering results and provide a reliable basis for the subsequent determination of the target edge points. According to the preset edge point selection method, the target edge points are selected from the optional edge points. This process allows flexible adjustment according to specific needs to adapt to different application scenarios and data characteristics. Based on the three-dimensional bounding box of the target edge point, the size and position of the intermediate cliff can be accurately calculated.

[0089] The following describes an apparatus embodiment of the present application, which can be used to execute the method embodiment of the present application. For details not disclosed in the apparatus embodiment of the present application, reference can be made to the method embodiment of the present application.

[0090] Figure 8 A block diagram of a device for detecting a road cliff based on a laser radar point cloud provided in an embodiment of the present application. Figure 8As shown, the device 800 for detecting a road mid-cliff based on a laser radar point cloud includes a preliminary point cloud data determination module 801 , a ground segmentation module 802 , a ground stitching module 803 and a mid-cliff data determination module 804 .

[0091] A preliminary point cloud data determination module 801 is used to extract preliminary point cloud data of a target road based on the received laser radar point cloud data;

[0092] The ground segmentation module 802 is used to extract ground point cloud data from the preliminary point cloud data according to a preset ground segmentation method;

[0093] The ground stitching module 803 is used to stitch the ground point cloud data based on the received inertial navigation positioning data and the preset calibration parameters to determine the stitched ground point cloud data;

[0094] The intermediate cliff data determination module 804 is used to determine the clustering result according to the spliced ​​ground point cloud data and the preset clustering method, and extract the intermediate cliff data of the target road according to the clustering result.

[0095] Optionally, the preliminary point cloud data determination module 801 is specifically used for:

[0096] Get LiDAR point cloud data;

[0097] For the LiDAR point cloud data, preset outlier points are filtered to determine preliminary point cloud data.

[0098] Optionally, the ground segmentation module 802 is specifically used for:

[0099] According to the ground segmentation method, determine the regional segmentation model, ground fitting method, and ground estimation method;

[0100] Based on the regional segmentation model, the preliminary point cloud data is divided into polar coordinate grids to determine the grid ground point cloud data;

[0101] According to the ground fitting method, plane fitting is performed on the grid ground point cloud data to determine the ground fitting plane;

[0102] According to the ground estimation method and the ground fitting plane, the ground point cloud data in the preliminary point cloud data is determined.

[0103] Optionally, the ground splicing module 803 is specifically used for:

[0104] According to the preset preprocessing requirements, the ground point cloud data is preprocessed to determine the pure point cloud data; wherein the pure point cloud data includes multiple frames;

[0105] Obtain inertial navigation positioning data and calibration parameters;

[0106] According to the calibration parameters, the multi-frame pure point cloud data is converted to the inertial navigation coordinate system to determine the multi-frame inertial navigation point cloud data;

[0107] According to the timestamps of the ground points corresponding to the multi-frame inertial navigation point cloud data, the inertial navigation positioning data corresponding to the timestamps are determined;

[0108] According to the inertial navigation positioning data, the multi-frame inertial navigation point cloud data is converted to the world coordinate system to determine the multi-frame world point cloud data, and the multi-frame world point cloud data is spliced;

[0109] According to the last frame of positioning data in the inertial navigation positioning data, the spliced ​​multi-frame world point cloud data is subjected to inverse coordinate transformation to determine the inverse inertial navigation point cloud data of the inertial navigation coordinate system at the time corresponding to the last frame;

[0110] According to the calibration parameters, the inverse coordinate transformation is performed on the inverse inertial navigation point cloud data to determine the spliced ​​ground point cloud data of the lidar coordinate system at the moment corresponding to the last frame.

[0111] Optionally, the ground stitching module 803 converts multiple frames of inertial navigation point cloud data into a world coordinate system according to the inertial navigation positioning data to determine multiple frames of world point cloud data, and stitches the multiple frames of world point cloud data, specifically for:

[0112] According to the inertial navigation positioning data, the multi-frame inertial navigation point cloud data is converted into the world coordinate system to determine the multi-frame world point cloud data;

[0113] According to the preset sliding window method, multiple frames of world point cloud data are stitched together.

[0114] Optionally, the clustering method includes Euclidean clustering; the intermediate cliff data determination module 804 is specifically used for:

[0115] Obtain the preset elevation magnification multiples of the spliced ​​ground point cloud data;

[0116] Process the spliced ​​ground point cloud data according to the elevation expansion multiple to determine the elevation expanded ground point cloud data;

[0117] Euclidean clustering is performed on the elevation-expanded ground point cloud data to determine the clustering results, and the intermediate cliff data is extracted based on the clustering results.

[0118] Optionally, when extracting the intermediate cliff data according to the clustering result, the intermediate cliff data determining module 804 is specifically used to:

[0119] According to the clustering results, determine the corresponding three-dimensional bounding box;

[0120] According to the three-dimensional bounding box, clustering information of the clustering result is determined;

[0121] Filter the clustering results according to the clustering information to determine the target clustering results;

[0122] When the number of target clustering results is less than two, it is determined that the middle cliff data of the target road is empty;

[0123] When the number of target clustering results is greater than or equal to two, the two target clustering results with the largest number of ground points in the target clustering results are determined as preferred clustering results;

[0124] According to the optimal clustering results and the preset point cloud selection method, the average elevation data corresponding to the high surface of the cliff and the low surface of the cliff are determined;

[0125] Based on the average elevation data, determine the average elevation difference;

[0126] When the average elevation difference is less than the preset elevation difference threshold, the intermediate cliff data is determined to be empty;

[0127] When the average elevation difference is greater than or equal to a preset elevation difference threshold, it is determined that there is an intermediate cliff on the target road, and the size and position of the intermediate cliff are determined according to the preferred clustering result as the intermediate cliff data.

[0128] Optionally, when the average elevation difference is greater than or equal to a preset elevation difference threshold, the intermediate cliff data determination module 804 determines that there is an intermediate cliff on the target road, and determines the size and position of the intermediate cliff according to the preferred clustering result as the intermediate cliff data, specifically for:

[0129] When the average elevation difference is greater than or equal to a preset elevation difference threshold, it is determined that there is an intermediate cliff on the target road;

[0130] According to the preset cliff data requirements, the final clustering results are determined from the preferred clustering results;

[0131] Based on the convex hull detection method, determine the optional edge points of the final clustering result;

[0132] Determine the target edge point according to the optional edge point and the preset edge point selection method;

[0133] According to the final clustering result, a three-dimensional bounding box of the target edge point is determined, so as to determine the size and position of the middle cliff according to the three-dimensional bounding box;

[0134] The size and position of the middle cliff are determined as the middle cliff data.

[0135] The device performs functions similar to the method provided above. For other functions, please refer to the previous description and will not be repeated here.

[0136] Fig. 9A schematic diagram of the structure of an electronic device provided in an embodiment of the present application, such as Fig. 9 As shown, the electronic device 900 of this embodiment may include: a memory 901 and a processor 902 .

[0137] The memory 901 stores a computer program. When the computer program is executed by the processor 902, the processor 902 executes the method in the above embodiment.

[0138] The processor 902 and the memory 901 are connected, for example, via a bus.

[0139] Optionally, the electronic device 900 may further include a transceiver. It should be noted that in actual applications, the number of transceivers is not limited to one, and the structure of the electronic device 900 does not constitute a limitation on the embodiments of the present application.

[0140] The processor 902 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor 902 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like.

[0141] The bus may include a path to transmit information between the above components. The bus may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0142] The memory 901 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0143] The memory 901 is used to store the application code for executing the solution of the present application, and the execution is controlled by the processor 902. The processor 902 is used to execute the application code stored in the memory 901 to implement the contents shown in the above method embodiment.

[0144] The electronic devices include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), and fixed terminals such as digital TVs, desktop computers, etc. It can also be a server, etc. Fig. 9 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0145] The electronic device of this embodiment can be used to execute the method of any of the above embodiments, and its implementation principle and technical effect are similar, which will not be described in detail here.

[0146] The present application also provides a non-transitory computer-readable storage medium having computer-readable instructions stored thereon. When the aforementioned instructions are executed by a processor, the processor executes the method in the above embodiment.

[0147] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a non-transient computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0148] The embodiments of the present application are described in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and its core idea of ​​the present application. At the same time, changes or deformations made by those skilled in the art based on the ideas of the present application, the specific implementation methods and the scope of application of the present application, all belong to the scope of protection of the present application. In summary, the content of this specification should not be construed as a limitation on the present application.

Claims

1. A method for detecting a cliff in the middle of a road based on a laser radar point cloud, characterized in that: include: Extracting preliminary point cloud data of the target road based on the received laser radar point cloud data; Extracting ground point cloud data from the preliminary point cloud data according to a preset ground segmentation method; Based on the received inertial navigation positioning data and preset calibration parameters, the ground point cloud data is stitched to determine stitched ground point cloud data; Determine a clustering result according to the spliced ​​ground point cloud data and a preset clustering method, and extract the middle cliff data of the target road according to the clustering result; The clustering method includes Euclidean clustering; The step of determining a clustering result based on the spliced ​​ground point cloud data and a preset clustering method, and extracting the intermediate cliff data of the target road based on the clustering result, includes: Obtaining a preset elevation magnification multiple of the spliced ​​ground point cloud data; Processing the spliced ​​ground point cloud data according to the elevation expansion multiple to determine elevation expanded ground point cloud data; Performing Euclidean clustering on the elevation-expanded ground point cloud data to determine the clustering result, and extracting the intermediate cliff data according to the clustering result; The extracting the intermediate cliff data according to the clustering result includes: According to the clustering result, determining a corresponding three-dimensional bounding box; Determining clustering information of the clustering result according to the three-dimensional bounding box; Filtering the clustering results according to the clustering information to determine a target clustering result; When the number of the target clustering results is less than two, determining that the intermediate cliff data of the target road is empty; When the number of the target clustering results is greater than or equal to two, two target clustering results with the largest number of ground points in the target clustering results are determined as preferred clustering results; According to the preferred clustering result and the preset point cloud selection method, the average elevation data corresponding to the high surface of the cliff and the low surface of the cliff are determined; Determining an average elevation difference based on the average elevation data; When the average elevation difference is less than a preset elevation difference threshold, determining that the intermediate cliff data is empty; When the average elevation difference is greater than or equal to the preset elevation difference threshold, it is determined that there is an intermediate cliff on the target road, and the size and position of the intermediate cliff are determined according to the preferred clustering result to serve as the intermediate cliff data.

2. The method according to claim 1, characterized in that The step of extracting preliminary point cloud data of the target road based on the received laser radar point cloud data comprises: Acquire the laser radar point cloud data; For the laser radar point cloud data, preset abnormal points are filtered to determine the preliminary point cloud data.

3. The method according to claim 1, characterized in that: The extracting ground point cloud data from the preliminary point cloud data according to a preset ground segmentation method includes: According to the ground segmentation method, determining a regional segmentation model, a ground fitting method, and a ground estimation method; Based on the region segmentation model, polar coordinate grid division is performed on the preliminary point cloud data to determine grid ground point cloud data; According to the ground fitting method, plane fitting is performed on the grid ground point cloud data to determine a ground fitting plane; According to the ground estimation method and the ground fitting plane, ground point cloud data in the preliminary point cloud data is determined.

4. The method according to claim 1, characterized in that The step of stitching the ground point cloud data based on the received inertial navigation positioning data and the preset calibration parameters to determine the stitched ground point cloud data includes: According to a preset preprocessing requirement, the ground point cloud data is preprocessed to determine clean point cloud data; wherein the clean point cloud data includes multiple frames; Acquiring the inertial navigation positioning data and the calibration parameters; According to the calibration parameters, the multiple frames of the clean point cloud data are converted into an inertial navigation coordinate system to determine the multiple frames of inertial navigation point cloud data; Determining inertial navigation positioning data corresponding to the timestamps of the ground points corresponding to the multiple frames of inertial navigation point cloud data; According to the inertial navigation positioning data, the multiple frames of inertial navigation point cloud data are converted into a world coordinate system to determine multiple frames of world point cloud data, and the multiple frames of world point cloud data are spliced; According to the last frame of positioning data in the inertial navigation positioning data, the spliced ​​multi-frame world point cloud data is subjected to inverse coordinate transformation to determine the inverse inertial navigation point cloud data of the inertial navigation coordinate system at the time corresponding to the last frame; According to the calibration parameters, the inverse coordinate transformation is performed on the inverse inertial navigation point cloud data to determine the spliced ​​ground point cloud data of the laser radar coordinate system at the time corresponding to the last frame.

5. The method according to claim 4, characterized in that The method of converting the multiple frames of inertial navigation point cloud data into a world coordinate system according to the inertial navigation positioning data to determine multiple frames of world point cloud data, and splicing the multiple frames of world point cloud data comprises: According to the inertial navigation positioning data, the multi-frame inertial navigation point cloud data is converted into a world coordinate system to determine the multi-frame world point cloud data; The multiple frames of world point cloud data are spliced ​​according to a preset sliding window method.

6. The method according to claim 1, characterized in that When the average elevation difference is greater than or equal to the preset elevation difference threshold, determining that there is an intermediate cliff on the target road, and determining the size and position of the intermediate cliff according to the preferred clustering result as the intermediate cliff data, includes: When the average elevation difference is greater than or equal to a preset elevation difference threshold, determining that there is an intermediate cliff on the target road; Determining a final clustering result from the preferred clustering results according to a preset cliff data requirement; Determining the optional edge points of the final clustering result based on the convex hull detection method; Determining a target edge point according to the optional edge point and the preset edge point selection method; Determine a three-dimensional bounding box of the target edge point according to the final clustering result, so as to determine the size and position of the middle cliff according to the three-dimensional bounding box; The size and position of the intermediate cliff are determined as the intermediate cliff data.

7. A device for detecting cliffs in the middle of a road based on laser radar point cloud, characterized in that: include: A preliminary point cloud data determination module, used to extract preliminary point cloud data of a target road based on the received laser radar point cloud data; A ground segmentation module, used to extract ground point cloud data from the preliminary point cloud data according to a preset ground segmentation method; A ground stitching module, used for stitching the ground point cloud data based on the received inertial navigation positioning data and preset calibration parameters to determine stitched ground point cloud data; An intermediate cliff data determination module, used to determine a clustering result according to the spliced ​​ground point cloud data and a preset clustering method, and extract the intermediate cliff data of the target road according to the clustering result; The clustering method includes Euclidean clustering, and the intermediate cliff data determination module is specifically used for: Obtaining a preset elevation magnification multiple of the spliced ​​ground point cloud data; Processing the spliced ​​ground point cloud data according to the elevation expansion multiple to determine elevation expanded ground point cloud data; Performing Euclidean clustering on the elevation-expanded ground point cloud data to determine the clustering result, and extracting the intermediate cliff data according to the clustering result; When the intermediate cliff data determination module extracts the intermediate cliff data according to the clustering result, it is specifically used to: According to the clustering result, determining a corresponding three-dimensional bounding box; Determining clustering information of the clustering result according to the three-dimensional bounding box; Filtering the clustering results according to the clustering information to determine a target clustering result; When the number of the target clustering results is less than two, determining that the intermediate cliff data of the target road is empty; When the number of the target clustering results is greater than or equal to two, two target clustering results with the largest number of ground points in the target clustering results are determined as preferred clustering results; According to the preferred clustering result and the preset point cloud selection method, the average elevation data corresponding to the high surface of the cliff and the low surface of the cliff are determined; Determining an average elevation difference based on the average elevation data; When the average elevation difference is less than a preset elevation difference threshold, determining that the intermediate cliff data is empty; When the average elevation difference is greater than or equal to the preset elevation difference threshold, it is determined that there is an intermediate cliff on the target road, and the size and position of the intermediate cliff are determined according to the preferred clustering result to serve as the intermediate cliff data.

8. An electronic device, characterized in that: include: processor; A memory storing a computer program, which, when executed by the processor, enables the processor to execute the method for detecting a cliff in the middle of a road based on a laser radar point cloud as described in any one of claims 1 to 6.

Images