Single multi-frame cooperative passable point cloud segmentation method and device based on laser radar

Abstract

This invention relates to the field of autonomous driving technology, and particularly to a method and apparatus for single-frame and multi-frame collaborative drivable point cloud segmentation based on LiDAR. The method includes: inputting raw LiDAR data into a single-frame point cloud segmentation algorithm to obtain a single-frame point cloud segmentation result; inputting the raw LiDAR data into a preset real-time localization and mapping algorithm to obtain a downsampled point cloud map; performing drivability segmentation on the downsampled point cloud map to obtain a grid-based drivability analysis result; and fusing the single-frame point cloud segmentation result and the grid-based drivability analysis result to obtain a point cloud map with semantic information. This solves the problems of existing point cloud segmentation and roadside detection methods in autonomous driving, such as lack of labeled point cloud data, difficulty in interpretability and deployment in real vehicles, and poor segmentation results.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a single-frame / multi-frame collaborative passable point cloud segmentation method and apparatus based on lidar. Background Technology

[0002] Semantic segmentation is a key task in achieving autonomous driving, helping the autonomous driving system better understand its surrounding environment by imbuing perceived data with human-like scene understanding. The essence of laser point cloud segmentation is to input laser point cloud data (a set of points) into an algorithm, which then assigns different definitions to each point (such as trees, obstacles, roads, etc.) to ultimately form a point cloud scene that conforms to human understanding. Semantic segmentation of laser point clouds has always been a major challenge due to the uneven distribution, discrete nature, and lack of semantic information (such as texture and color) of the point clouds. Furthermore, in addition to the traditional task of segmenting obstacle point clouds, those skilled in the art believe that extracting road edges as traversable boundaries is essential to better assist autonomous driving decision-making and control in urban scenarios. Existing point cloud semantic segmentation technologies can be broadly divided into two directions based on their application methods:

[0003] The first approach is based on deep learning (a data-driven method): this involves learning a semantic segmentation deep learning model using a large amount of labeled point cloud data. The differences between various semantic segmentation models primarily stem from different point cloud data processing methods to extract more point cloud features, such as PointNet++ and Cylindar3D. This method is currently popular due to its high segmentation accuracy, topping the leaderboard in point cloud segmentation accuracy. However, deep learning-based point cloud segmentation requires a large amount of accurately labeled point cloud data (datasets), which is a very time-consuming and difficult task. The task also requires extracting road edges, but unfortunately, there are currently no publicly available road edge datasets. Furthermore, the lack of interpretability of deep learning models makes it difficult to locate and resolve problems during runtime. Additionally, the massive number of parameters in semantic segmentation models makes real-world deployment, even with GPU acceleration, a significant challenge.

[0004] The second approach involves manually summarizing the geometric features of point clouds to assign corresponding semantic point cloud results. However, geometry-based segmentation methods are typically inaccurate and prone to false positives and false negatives, leading to uneven segmentation. This poses a significant challenge for subsequent autonomous driving systems using these uncertain segmentation results to understand the scene. To address these issues, many researchers have utilized machine learning methods to improve segmentation algorithms, such as clustering and fitting algorithms. However, clustering algorithms are prone to clustering the same object into different objects; fitting algorithms often require multiple trials to obtain the best fit. Both types of algorithms require computation on large amounts of point clouds, making real-time operation difficult. Furthermore, these methods exhibit incompatibility issues similar to deep learning models across different LiDAR platforms. Traditional algorithms based on the geometric features of point clouds also suffer from significant disadvantages in combating interference such as vehicle motion distortion, sensor noise, and scene changes.

[0005] In summary, point cloud segmentation and roadside detection, as important components of autonomous driving, still need to be addressed: lack of labeled point cloud data, poor interpretability (debuggability), difficulty in real-vehicle deployment, and unsatisfactory segmentation results. Summary of the Invention

[0006] This invention provides a single-frame and multi-frame collaborative passable point cloud segmentation method and device based on LiDAR. Since urban scenes are mostly structured buildings, algorithms based on the geometric features of point clouds are inherently universal, but they are still very challenging to implement. For example, the uneven density of the LiDAR point cloud in a single frame leads to the need to decouple the data from the distance to the vehicle when summarizing geometric features, as well as interference from the scene such as uphill and downhill slopes and occlusion. Therefore, existing point cloud segmentation and roadside detection methods in autonomous driving suffer from problems such as lack of labeled point cloud data, poor interpretability, difficulty in real-vehicle deployment, and poor segmentation results.

[0007] A first aspect of the present invention provides a single- and multi-frame collaborative traversable point cloud segmentation method based on lidar, comprising the following steps:

[0008] The raw LiDAR data is input into the single-frame point cloud segmentation algorithm to obtain the point cloud segmentation result of a single frame.

[0009] The original lidar data is input into a preset real-time positioning and mapping algorithm to obtain a point cloud map after reduced sampling, and the point cloud map after reduced sampling is segmented for accessibility to obtain grid-based accessibility analysis results.

[0010] By fusing the point cloud segmentation results of the single frame with the grid-based accessibility analysis results, a point cloud map with semantic information is obtained.

[0011] Optionally, the step of inputting the original lidar data into a single-frame point cloud segmentation algorithm to obtain the point cloud segmentation result of a single frame includes:

[0012] The original lidar data is converted and input into a single-frame point cloud segmentation algorithm for point cloud regularization to obtain near-ground point clouds.

[0013] The near-ground point cloud is divided into annular sector spatial segments to obtain an annular structure loaded point cloud;

[0014] Adaptive curb point detection is performed on the point cloud loaded by the ring structure to obtain the point cloud segmentation result of the single frame.

[0015] Optionally, the step of performing accessibility segmentation on the downsampled point cloud map to obtain accessibility analysis results includes:

[0016] The point cloud map after the reduction is rasterized to obtain a rasterized point cloud map;

[0017] A accessibility analysis is performed on each grid cell in the rasterized point cloud map to obtain the grid-based accessibility analysis results.

[0018] Optionally, the step of fusing the point cloud segmentation results of the single frame and the raster-based accessibility analysis results to obtain a point cloud map with semantic information includes:

[0019] The inconsistencies between the point cloud segmentation results of the single frame and the grid-based accessibility analysis results are analyzed.

[0020] Based on the inconsistencies, point cloud samples are added to the local grid, and the added point cloud samples are subjected to refined analysis to obtain the point cloud map with semantic information.

[0021] A second aspect of the present invention provides a single- and multi-frame collaborative traversable point cloud segmentation device based on lidar, comprising:

[0022] The first segmentation module is used to input the original LiDAR data into the single-frame point cloud segmentation algorithm to obtain the point cloud segmentation result of a single frame.

[0023] The second segmentation module is used to input the original lidar data into a preset real-time positioning and mapping algorithm to obtain a point cloud map after reduced sampling, and to perform accessibility segmentation on the point cloud map after reduced sampling to obtain grid-based accessibility analysis results.

[0024] The fusion module is used to fuse the point cloud segmentation results of the single frame and the raster-based accessibility analysis results to obtain a point cloud map with semantic information.

[0025] Optionally, the first segmentation module includes:

[0026] The filtering and extraction unit is used to convert the original lidar data into a single-frame point cloud segmentation algorithm for point cloud regularization to obtain near-ground point clouds.

[0027] A segmentation unit is used to perform annular sector spatial segmentation on the near-ground point cloud to obtain an annular structure loaded point cloud;

[0028] The curb point detection unit is used to perform adaptive curb point detection on the point cloud loaded by the ring structure to obtain the point cloud segmentation result of the single frame.

[0029] Optionally, the second segmentation module includes:

[0030] A rasterization unit is used to rasterize the downsampled point cloud map to obtain a rasterized point cloud map.

[0031] The mobility analysis unit is used to perform mobility analysis on each grid in the rasterized point cloud map to obtain the grid-based mobility analysis results.

[0032] Optionally, the fusion module includes:

[0033] The analysis unit is used to analyze the inconsistencies between the point cloud segmentation results of the single frame and the raster-based accessibility analysis results.

[0034] The fine-grained analysis unit is used to add point cloud samples to local grids based on the inconsistencies, and to perform fine-grained analysis on the added point cloud samples to obtain the point cloud map with semantic information.

[0035] A third aspect of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the single- and multi-frame collaborative drivable point cloud segmentation method based on lidar as described in the above embodiments.

[0036] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described single- and multi-frame collaborative drivable point cloud segmentation method based on lidar.

[0037] The single-frame and multi-frame collaborative passable point cloud segmentation method and apparatus based on LiDAR proposed in this invention can perform point cloud segmentation and passable boundary (roadside) segmentation in real time for autonomous driving in urban scenarios by fusing dual-stream analysis of single-frame point clouds and local maps. This results in a highly accurate and uniform semantic map of passable areas, helping vehicles better understand the scene and plan safe routes. It not only improves the accuracy of traditional geometric feature segmentation algorithms but also achieves good segmentation results for small objects such as roadside edges. Furthermore, it can greatly reduce the problem of uneven segmentation of point clouds and the difficulty of subsequent module processing caused by the uncertainty of the results.

[0038] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0039] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0040] Figure 1 A flowchart illustrating a single-frame / multi-frame collaborative passable point cloud segmentation method based on lidar provided in an embodiment of the present invention;

[0041] Figure 2 This is a visual representation of the input and output of a single- and multi-frame collaborative passable point cloud segmentation method based on lidar provided in an embodiment of the present invention.

[0042] Figure 3 A detailed flowchart of another single- and multi-frame collaborative passable point cloud segmentation method based on lidar provided in an embodiment of the present invention;

[0043] Figure 4 A modular flowchart of another single- and multi-frame collaborative passable point cloud segmentation method based on lidar provided in an embodiment of the present invention;

[0044] Figure 5 for Figure 4 A schematic diagram illustrating the principles of different modules;

[0045] Figure 6 This is a schematic diagram of the point cloud segmentation effect of a single frame provided in an embodiment of the present invention;

[0046] Figure 7 This is a single- or multi-frame semantic point cloud map provided in an embodiment of the present invention;

[0047] Figure 8 A block diagram illustrating a single- or multi-frame collaborative passable point cloud segmentation device based on lidar provided in an embodiment of the present invention.

[0048] Figure 9 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0049] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0050] The following description, with reference to the accompanying drawings, describes a single-frame / multi-frame collaborative passable point cloud segmentation method and apparatus based on lidar according to embodiments of the present invention.

[0051] Figure 1 This is a flowchart illustrating a single- and multi-frame collaborative passable point cloud segmentation method based on lidar, provided in an embodiment of the present invention.

[0052] like Figure 1 As shown, the single- and multi-frame collaborative drivable point cloud segmentation method based on lidar includes the following steps:

[0053] In step S101, the original lidar data is input into the single-frame point cloud segmentation algorithm to obtain the point cloud segmentation result of a single frame.

[0054] Furthermore, in one embodiment of the present invention, the original lidar data is input into a single-frame point cloud segmentation algorithm to obtain the point cloud segmentation result of a single frame, including:

[0055] The original lidar data is converted and input into a single-frame point cloud segmentation algorithm for point cloud regularization to obtain near-ground point clouds.

[0056] The near-ground point cloud is divided into annular fan-shaped spatial segments to obtain a point cloud with an annular structure.

[0057] Adaptive curb point detection is performed on the point cloud of the ring structure to obtain the point cloud segmentation result of a single frame.

[0058] Specifically, such as Figure 2 and 3 As shown, the original LiDAR data is transformed into the local coordinate system to obtain the target LiDAR point. The point cloud data of each frame in the target LiDAR point is input into the single-frame point cloud segmentation algorithm. The point cloud data is regularized to facilitate point cloud search and obtain the near-ground point cloud. Then, the neighborhood geometric feature data of the near-ground point cloud is adaptively extracted. Through special neighborhood general geometric features, coarse segmentation point clouds such as obstacles, curbs, and grass are extracted by analysis. That is, annular fan-shaped spatial segmentation and curb point detection are performed to obtain the single-frame point cloud segmentation result.

[0059] Among them, such as Figure 4 and 5 As shown, point cloud regularization includes circular regularization and sector regularization. The two regularization methods need to be performed simultaneously. Circular regularization comes from the point cloud coil of the scanning LiDAR mechanism itself, which is mainly responsible for detecting lawns, large obstacles, and curbs. Sector regularization packages point clouds with similar horizontal scanning angles into a bundle, which is mainly responsible for road surface protrusions. Performing the two regularization methods simultaneously can better find and analyze the relationship between a point and its surrounding points.

[0060] Adaptability is designed to address the issue of geometric features being affected by the distance from the target point to the vehicle, caused by the point cloud density being greater at nearby locations than at distant locations. Specifically, it is necessary to decouple geometric features from point cloud density by correlating the analysis region of the point cloud with the point cloud density (the distance between the point and the vehicle), thereby facilitating the extraction of geometric features of similar size by single-frame point cloud segmentation algorithms.

[0061] Unlike most methods that determine target points using a precise threshold, this invention uses a percentage change in a specific angle to extract point clouds. This specific angle reduces the instability of previous decoupling methods on the point cloud analysis area and geometric features. The percentage change ensures the universality of features across different LiDAR platforms, meaning that similar percentage changes can be obtained by changing certain parameters.

[0062] like Figure 6 As shown, both near and far point clouds are segmented, indicating that it greatly mitigates the side effect of denser points in the near and sparser points in the far, and can accurately distinguish obstacles and drivable boundaries (curb edges). However, geometry-based algorithms have certain drawbacks, namely, the segmented point clouds may be uneven, and some false detections may occur. In addition, due to the large number of point clouds in a single frame, increasing the complexity of the point cloud segmentation method to achieve higher accuracy would significantly increase the running time. Therefore, it is still necessary to use downsampled point cloud maps to help further analyze street scenes.

[0063] In step S102, the original lidar data is input into a preset real-time positioning and mapping algorithm to obtain a point cloud map after the reduction in data acquisition. The point cloud map after the reduction in data acquisition is then segmented for accessibility to obtain a grid-based accessibility analysis result.

[0064] Furthermore, in one embodiment of the present invention, the downsampled point cloud map is segmented for drivability to obtain drivability analysis results, including:

[0065] The point cloud map after the mining is reduced is rasterized to obtain a rasterized point cloud map;

[0066] A accessibility analysis is performed on each grid cell in the rasterized point cloud map to obtain the grid-based accessibility analysis results.

[0067] Specifically, such as Figure 3 and 5 As shown, raw LiDAR data can be input into algorithms capable of creating point cloud maps, such as SLAM (Simultaneous Localization and Mapping). First, the raw LiDAR data is converted to the local coordinate system to obtain the target LiDAR point. Then, feature points are extracted from the point cloud data of each frame of the target LiDAR point according to the curvature. Next, point cloud matching is performed based on the planar point cloud and corner points. Finally, downsampling is performed, and the downsampled point cloud is added to the point cloud of the world coordinate system or global coordinate system. The local map is then synthesized by accumulating the data. Finally, the downsampled global point cloud map is obtained through map leverage. Finally, the downsampled point cloud map is rasterized for accessibility segmentation to obtain raster-based accessibility analysis results, which can then be used to further analyze street scenes.

[0068] In step S103, the point cloud segmentation results of a single frame and the raster-based accessibility analysis results are fused to obtain a point cloud map with semantic information.

[0069] Furthermore, in one embodiment of the present invention, the point cloud segmentation results of a single frame and the raster-based accessibility analysis results are fused to obtain a point cloud map with semantic information, including:

[0070] The inconsistencies between point cloud segmentation results of a single frame and raster-based accessibility analysis results are analyzed.

[0071] Based on the inconsistencies, point cloud samples are added to local grids, and the added point cloud samples are subjected to refined analysis to obtain a point cloud map with semantic information. The added point cloud samples are derived from the dense single-frame segmented point cloud results.

[0072] Specifically, such as Figure 7 As shown, the inconsistencies between the coarse point cloud segmentation results obtained from single-frame point cloud segmentation and the raster analysis results are analyzed. Based on the inconsistencies, samples are added to the local raster and refined analysis is performed to obtain accurate and uniform results. Finally, the segmented point cloud is output to obtain a semantic map.

[0073] For example, after rasterizing and downsampling the point cloud map, a passability analysis is performed to segment large obstacles that are different from flat land and "flat land" to obtain the raster analysis results. Due to the severe downsampling of the point cloud map, the "flat land" here actually includes grass, flat land and roadside. Next, the point clouds of grass and roadside are extracted from the "flat land" point cloud.

[0074] By fusing the point cloud segmentation results of a single frame with the raster analysis results, we can understand the discrepancies in the analysis results caused by downsampling. Therefore, we use geometric features to help distinguish between grass and roadside point clouds, perform fine analysis on local point clouds, achieve the verification effect, complete uniform and accurate segmentation of passable areas, and finally output the labeled point cloud to form a point cloud map with semantic information.

[0075] It should be noted that the embodiments of the present invention do not require a GPU, only a CPU. It can obtain reliable passable areas in urban scenarios in real time without using a GPU, which helps intelligent vehicles achieve autonomous driving. This greatly reduces the computing power pressure on the GPU in autonomous driving tasks and helps the deployment of autonomous driving systems in real vehicles.

[0076] The single- and multi-frame collaborative drivable point cloud segmentation method based on lidar proposed in this embodiment of the invention has the following advantages:

[0077] Beneficial effects:

[0078] (1) High efficiency: It can accurately find the area that needs detailed analysis and obtain the results in real time to provide to the subsequent obstacle avoidance planning module;

[0079] (2) Feature universality: It can obtain universal features through geometric analysis of structured scenes, which are applicable to different urban scenes;

[0080] (3) Adjustability: After debugging, it can be used on different lidar platforms;

[0081] (4) Plug and play: It can work with different laser point cloud SLAM (simultaneous localization and mapping algorithm).

[0082] Next, referring to the accompanying drawings, a single- or multi-frame collaborative passable point cloud segmentation device based on lidar according to an embodiment of the present invention is described.

[0083] Figure 8 This is a block diagram of a single- and multi-frame collaborative passable point cloud segmentation device based on lidar according to an embodiment of the present invention.

[0084] like Figure 8 As shown, the single- and multi-frame collaborative passable point cloud segmentation device 80 based on lidar includes: a first segmentation module 801, a second segmentation module 802, and a fusion module 803.

[0085] The system comprises the following modules: a first segmentation module 801, which inputs raw LiDAR data into a single-frame point cloud segmentation algorithm to obtain a single-frame point cloud segmentation result; a second segmentation module 802, which inputs raw LiDAR data into a preset real-time positioning and mapping algorithm to obtain a downsampled point cloud map, and performs drivability segmentation on the downsampled point cloud map to obtain a raster-based drivability analysis result; and a fusion module 803, which fuses the single-frame point cloud segmentation result and the raster-based drivability analysis result to obtain a point cloud map with semantic information.

[0086] Furthermore, in one embodiment of the present invention, the first segmentation module 801 includes:

[0087] The filtering and extraction unit is used to convert the raw lidar data into a single-frame point cloud segmentation algorithm for point cloud regularization to obtain near-ground point clouds.

[0088] The segmentation unit is used to perform annular fan-shaped spatial segmentation on the near-ground point cloud to obtain a point cloud with an annular structure.

[0089] The curb point detection unit is used to perform adaptive curb point detection on the point cloud loaded in the ring structure to obtain the point cloud segmentation result of a single frame.

[0090] Furthermore, in one embodiment of the present invention, the second segmentation module 802 includes:

[0091] Rasterization unit is used to rasterize the point cloud map after mining downsampling to obtain a rasterized point cloud map;

[0092] The accessibility analysis unit is used to perform accessibility analysis on each grid cell in the rasterized point cloud map to obtain grid-based accessibility analysis results.

[0093] Furthermore, in one embodiment of the present invention, the fusion module 803 includes:

[0094] The analysis unit is used to analyze inconsistencies between the point cloud segmentation results of a single frame and the raster-based accessibility analysis results.

[0095] The fine-grained analysis unit is used to add point cloud samples to local grids based on inconsistencies, and to perform fine-grained analysis on the added point cloud samples to obtain a point cloud map with semantic information.

[0096] It should be noted that the foregoing explanation of the embodiment of the single- and multi-frame collaborative passable point cloud segmentation method based on lidar also applies to the single- and multi-frame collaborative passable point cloud segmentation device based on lidar in this embodiment, and will not be repeated here.

[0097] The single- and multi-frame cooperative traversable point cloud segmentation device based on lidar proposed in this embodiment of the invention has the following features:

[0098] Beneficial effects:

[0099] (1) High efficiency: It can accurately find the area that needs detailed analysis and obtain the results in real time to provide to the subsequent obstacle avoidance planning module;

[0100] (2) Feature universality: It can obtain universal features through geometric analysis of structured scenes, which are applicable to different urban scenes;

[0101] (3) Adjustability: After debugging, it can be used on different lidar platforms;

[0102] (4) Plug and play: It can work with different laser point cloud SLAM (simultaneous localization and mapping algorithm).

[0103] Figure 9 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. The electronic device may include:

[0104] The memory 901, the processor 902, and the computer program stored on the memory 901 and capable of running on the processor 902.

[0105] When the processor 902 executes the program, it implements the single- and multi-frame collaborative passable point cloud segmentation method based on lidar provided in the above embodiments.

[0106] Furthermore, electronic devices also include:

[0107] Communication interface 903 is used for communication between memory 901 and processor 902.

[0108] The memory 901 is used to store computer programs that can run on the processor 902.

[0109] The memory 901 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0110] If the memory 901, processor 902, and communication interface 903 are implemented independently, then the communication interface 903, memory 901, and processor 902 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0111] Optionally, in a specific implementation, if the memory 901, processor 902, and communication interface 903 are integrated on a single chip, then the memory 901, processor 902, and communication interface 903 can communicate with each other through an internal interface.

[0112] The processor 902 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0113] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described single- and multi-frame collaborative navigable point cloud segmentation method based on lidar.

[0114] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0115] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0116] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0117] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0118] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0119] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

[0120] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0121] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A single-frame / multi-frame collaborative drivable point cloud segmentation method based on lidar, characterized in that, Includes the following steps: The raw LiDAR data is input into a single-frame point cloud segmentation algorithm to obtain the point cloud segmentation result for a single frame, specifically including: The original lidar data is converted and input into a single-frame point cloud segmentation algorithm for point cloud regularization to obtain near-ground point clouds. The near-ground point cloud is divided into annular sector spatial segments to obtain an annular structure loaded point cloud; Adaptive roadside point detection is performed on the point cloud loaded by the ring structure to obtain the point cloud segmentation result of the single frame; The original lidar data is input into a preset real-time positioning and mapping algorithm to obtain a point cloud map after data reduction. The point cloud map after data reduction is then segmented for traversability to obtain grid-based traversability analysis results, specifically including: The original lidar data is input into the preset real-time positioning and mapping algorithm to convert the original lidar data to the local coordinate system and obtain the target lidar point. Feature points are extracted from the point cloud data of each frame in the target laser point, and the feature points are downsampled to obtain the downsampled feature points. The reduced-mining feature points are added to the point cloud in the world coordinate system or global coordinate system to obtain the reduced-mining point cloud map. The point cloud map after the reduction is rasterized to obtain a rasterized point cloud map; Perform a accessibility analysis on each grid cell in the rasterized point cloud map to obtain the grid-based accessibility analysis results; By fusing the point cloud segmentation results of the single frame with the grid-based accessibility analysis results, a point cloud map with semantic information is obtained.

2. The single- and multi-frame collaborative drivable point cloud segmentation method based on lidar according to claim 1, characterized in that, The point cloud segmentation results of the single frame and the raster-based accessibility analysis results are fused to obtain a point cloud map with semantic information, including: The inconsistencies between the point cloud segmentation results of the single frame and the grid-based accessibility analysis results are analyzed. Based on the inconsistencies, point cloud samples are added to the local grid, and the added point cloud samples are subjected to refined analysis to obtain the point cloud map with semantic information.

3. A single-frame / multi-frame collaborative passable point cloud segmentation device based on lidar, characterized in that, include: The first segmentation module is used to input the raw LiDAR data into the single-frame point cloud segmentation algorithm to obtain the point cloud segmentation result of a single frame, specifically including: The filtering and extraction unit is used to convert the original lidar data into a single-frame point cloud segmentation algorithm for point cloud regularization to obtain near-ground point clouds. A segmentation unit is used to perform annular sector spatial segmentation on the near-ground point cloud to obtain an annular structure loaded point cloud; The curb point detection unit is used to adaptively detect curb points on the point cloud loaded by the ring structure to obtain the point cloud segmentation result of the single frame; The second segmentation module is used to input the original LiDAR data into a preset real-time positioning and mapping algorithm to obtain a point cloud map after data reduction, and to perform accessibility segmentation on the point cloud map after data reduction to obtain grid-based accessibility analysis results, specifically including: The conversion unit is used to input the original lidar data into the preset real-time positioning and mapping algorithm to convert the original lidar data to the local coordinate system and obtain the target lidar point. The extraction and downsampling unit is used to extract feature points from the point cloud data of each frame in the target laser point and downsample the feature points to obtain downsampled feature points. The addition unit is used to add the reduced-mining feature points to the point cloud in the world coordinate system or the global coordinate system to obtain the reduced-mining point cloud map. A rasterization unit is used to rasterize the downsampled point cloud map to obtain a rasterized point cloud map. The mobility analysis unit is used to perform mobility analysis on each grid in the rasterized point cloud map to obtain the grid-based mobility analysis results. The fusion module is used to fuse the point cloud segmentation results of the single frame and the raster-based accessibility analysis results to obtain a point cloud map with semantic information.

4. The single- and multi-frame collaborative passable point cloud segmentation device based on lidar according to claim 3, characterized in that, The fusion module includes: The analysis unit is used to analyze the inconsistencies between the point cloud segmentation results of the single frame and the raster-based accessibility analysis results. The fine-grained analysis unit is used to add point cloud samples to local grids based on the inconsistencies, and to perform fine-grained analysis on the added point cloud samples to obtain the point cloud map with semantic information.

5. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the single- and multi-frame collaborative drivable point cloud segmentation method based on lidar as described in any one of claims 1-2.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the single- and multi-frame collaborative drivable point cloud segmentation method based on lidar as described in any one of claims 1-2.

Images