A method, device, equipment and medium for online detection of high-precision map changes

Through lidar, point cloud data is collected for distortion processing and matching high-precision maps, the problem of inability to detect changes in high-precision maps in real time in the existing technology is solved, and the safety of the autonomous driving system is improved.

CN114624672BActive Publication Date: 2024-11-05GUANGZHOU WERIDE TECH LTD CO
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202210326427.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-30
Publication Date
2024-11-05
Estimated Expiration
2042-03-30

AI Technical Summary

Technical Problem

Due to the content limitations of two-dimensional images and the delay in model processing, the existing high-precision map update solution cannot detect changes in the spatial structure of high-precision maps in real time, resulting in a reduction in the safety of autonomous driving.

Method used

Point cloud data is collected through lidar for the first distortion processing, converted into a point cloud depth map, and point sampling of the point cloud and the wire harness according to the predetermined spacing, obtaining a dimensionality reduction virtual point cloud, matching a high-precision map to mark the pixel position, and calculating Pixel ratio, generate map change results based on threshold.

Benefits of technology

Real-time detection of changes in spatial structure of high-precision maps is realized, and the safety of the autonomous driving system is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114624672B_ABST
    Figure CN114624672B_ABST
Patent Text Reader

Abstract

The present invention discloses an online detection method, device, equipment and medium for high-precision map changes, which are applied to an automatic driving system that caches a high-precision map corresponding to a dense point cloud and is provided with a laser radar. The method comprises: when receiving a laser point cloud collected by the laser radar in a current frame, performing a first distortion processing on the laser point cloud to obtain a reduced-dimensional point cloud and converting it into a point cloud depth map; performing point sampling on a line beam between the reduced-dimensional point cloud and the laser radar at a predetermined interval and performing a second distortion processing to obtain respectively corresponding reduced-dimensional virtual point clouds; using the reduced-dimensional point cloud and multiple reduced-dimensional virtual point clouds to respectively match the high-precision map, marking the pixel positions in the point cloud depth map according to the matching results, and obtaining multiple identification pixels; calculating the pixel ratio occupied by all identification pixels in the point cloud depth map, comparing the pixel ratio with a preset ratio threshold, and generating a map change result, so that the changes in the current environment can be known in real time, thereby improving the safety of automatic driving.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of map detection technology, and in particular to an online detection method, device, equipment and medium for high-precision map changes. Background Art

[0002] With the continuous development of science and technology, the relevant technologies of autonomous driving are becoming more and more mature. As the core technology of the autonomous driving system, HD Map (High Definition Map) technology not only has higher location accuracy than the traditional static electronic map for users, but also contains richer road and traffic elements, such as real-time dynamic traffic and road condition information.

[0003] In the case of poor GPS signals, high-precision maps can still provide reliable point cloud positioning observation information for the fusion of multiple sensors required for positioning, and can provide a data basis for lane semantic information annotation for semantic maps. It can be seen that high-precision maps are crucial for autonomous driving, and the update timeliness of their changes may have a great impact on autonomous driving safety.

[0004] Therefore, when the actual driving environment of the vehicle changes, the high-precision map should be updated in time. The existing high-precision map update scheme is usually implemented by using image and deep learning model detection methods, but due to the content limitations of two-dimensional images and the delay of model processing, the above scheme cannot detect the changes in the spatial structure of the high-precision map in real time, resulting in reduced safety of autonomous driving. Summary of the invention

[0005] The present invention provides a method, device, equipment and medium for online detection of changes in high-precision maps, which solves the technical problem that existing map update solutions cannot detect changes in the spatial structure of high-precision maps in real time due to content limitations of two-dimensional images and the delay of model processing, resulting in reduced safety of autonomous driving.

[0006] A first aspect of the present invention provides an online detection method for high-precision map changes, comprising:

[0007] When receiving a laser point cloud collected by the laser radar in a current frame, performing a first distortion process on the laser point cloud to obtain a reduced-dimensional point cloud and converting it into a point cloud depth map;

[0008] Performing point sampling on the line beams between the reduced-dimensional point cloud and the laser radar according to a predetermined interval and performing a second distortion process to obtain reduced-dimensional virtual point clouds corresponding to each group of the line beams;

[0009] The reduced-dimensional point cloud and the plurality of reduced-dimensional virtual point clouds are respectively matched with a preset high-precision map, and pixel positions in the point cloud depth map are marked according to the matching results to obtain a plurality of identification pixels;

[0010] The pixel ratio of all the identified pixels in the point cloud depth map is calculated, and a map change result is generated according to a comparison result between the pixel ratio and a preset ratio threshold.

[0011] Optionally, the laser point cloud is provided with a start timestamp and an end timestamp; when receiving the laser point cloud collected by the laser radar in the current frame, performing a first distortion process on the laser point cloud to obtain a reduced-dimensional point cloud and converting it into a point cloud depth map comprises:

[0012] When receiving the laser point cloud collected by the laser radar in the current frame, calculating the interpolation interval between the end timestamp and the start timestamp;

[0013] Acquire the first pose information of the laser point cloud corresponding to the starting time stamp, and perform linear interpolation on the first pose information in the interpolation interval to obtain the second pose information of the laser point cloud corresponding to the ending time stamp;

[0014] Calculating a posture change value between the second posture information and the first posture information;

[0015] According to the preset calibration external parameters and the pose change value, the laser point cloud corresponding to the termination time stamp is converted into a coordinate system to obtain a reduced-dimensional point cloud;

[0016] A point cloud depth map is generated according to the specification parameters corresponding to the laser radar and the three-dimensional coordinates corresponding to each of the reduced-dimensional point clouds.

[0017] Optionally, the specification parameters include the number of line beams and the scanning interval angle; the step of generating a point cloud depth map according to the specification parameters corresponding to the laser radar and the three-dimensional coordinates corresponding to each of the reduced-dimensional point clouds includes:

[0018] Calculating an angle ratio between a preset visual angle and the scanning interval angle;

[0019] Using the number of line bundles as the number of depth map rows and the angle ratio as the number of depth map columns, an initial depth map is created;

[0020] Calculating the point cloud distance between each of the reduced-dimensional point clouds and the laser radar respectively;

[0021] Each of the point cloud distances is used as a pixel value corresponding to each pixel position in the initial depth map to obtain a point cloud depth map.

[0022] Optionally, the step of sampling points on the line beams between the reduced-dimensional point cloud and the laser radar at a predetermined interval and performing a second distortion process to obtain reduced-dimensional virtual point clouds corresponding to each group of the line beams includes:

[0023] Performing point sampling on the line beams between the dimension reduction point cloud and the laser radar according to a predetermined spacing to obtain a virtual point cloud corresponding to each group of the line beams;

[0024] Obtaining the posture information corresponding to each of the virtual point clouds;

[0025] The pose information is used to perform a second distortion process on each of the virtual point clouds to obtain a reduced-dimensional virtual point cloud corresponding to each group of the wire harnesses.

[0026] Optionally, the identification pixel includes a first identification pixel and a second identification pixel; the step of using the reduced-dimensional point cloud and the plurality of reduced-dimensional virtual point clouds to match a preset high-precision map respectively, marking the pixel positions in the point cloud depth map according to the matching results, and obtaining a plurality of identification pixels includes:

[0027] Using each of the dimensionally reduced virtual point clouds to match a preset high-precision map, respectively, to determine a first point cloud grid;

[0028] Determine the reduced-dimensional point cloud to which the first point cloud grid having the Gaussian distribution information cached therein belongs as a first target point cloud, and determine the reduced-dimensional point clouds other than the first target point cloud as remaining point clouds;

[0029] Acquire a first pixel position corresponding to the first target point cloud from the point cloud depth map, and mark a first identification pixel at the first pixel position;

[0030] Matching the high-precision map with the remaining point cloud to determine a second point cloud grid;

[0031] Determining whether Gaussian distribution information is cached in each of the second point cloud grids;

[0032] If not, the pixel position of the remaining point cloud to which the second point cloud grid belongs is obtained from the point cloud depth map as the second pixel position, and a second identification pixel is marked at the second pixel position.

[0033] Optionally, the step of using each of the reduced-dimensional virtual point clouds to match a preset high-precision map to determine a first point cloud grid includes:

[0034] Obtaining first three-dimensional coordinates corresponding to each of the reduced-dimensional virtual point clouds;

[0035] Calculate first indexes respectively using each of the first three-dimensional coordinates;

[0036] The high-precision map is matched according to each of the first indexes, and the point cloud mesh corresponding to each of the first indexes is determined as the first point cloud mesh.

[0037] Optionally, the step of calculating the pixel ratio occupied by all the identified pixels in the point cloud depth map, and generating a map change result according to a comparison result of the pixel ratio with a preset ratio threshold, comprises:

[0038] Calculate the pixel ratio of the identified pixel in the point cloud depth map;

[0039] comparing the pixel ratio with a preset ratio threshold;

[0040] If the pixel ratio is greater than or equal to the preset ratio threshold, the reduced-dimensional point cloud or reduced-dimensional virtual point cloud corresponding to the identified pixel at the current moment is cached to a preset cloud server, a map change result is generated, and an alarm message is output.

[0041] Optionally, before the step of using the remaining point cloud to match the high-precision map to determine the second point cloud grid, the method further includes:

[0042] Convert the dimension-reduced point cloud to a map coordinate system corresponding to the high-precision map to obtain a map point cloud;

[0043] Using a preset three-dimensional perception frame to detect the map point cloud, determine the dynamic obstacles in the map point cloud, and mark the corresponding reduced-dimensional point cloud as the dynamic obstacle point cloud;

[0044] Selecting the remaining point clouds that overlap with the dynamic obstacle point cloud as the second target point cloud;

[0045] A third pixel position corresponding to the second target point cloud is obtained from the point cloud depth map, and a third identification pixel is marked at the third pixel position.

[0046] Optionally, the method further comprises:

[0047] Obtaining the pixel position of the noise point in the point cloud depth map;

[0048] Determine whether there are pixel values ​​at a plurality of undetermined pixel positions adjacent to the noise pixel position;

[0049] If the pixel value exists in all the pending pixel positions, retain the noise pixel position;

[0050] If the pixel value does not exist at any of the pending pixel positions, deleting the noise pixel position;

[0051] When all the noise pixel positions are determined, the point cloud depth map is updated using the point cloud depth map at the current moment to obtain updated identification pixels.

[0052] Optionally, it also includes:

[0053] Performing region search and clustering on the point cloud depth map to obtain a plurality of pixel regions consisting of search pixel positions;

[0054] Deleting pixel regions where the number of search pixel positions is less than a preset number threshold from the plurality of pixel regions, to obtain an updated point cloud depth map;

[0055] The point cloud depth map at the current moment is used to update the point cloud depth map to obtain an updated identification pixel.

[0056] A second aspect of the present invention provides an online detection device for high-precision map changes, the device comprising:

[0057] The point cloud acquisition and dedistortion module is used to perform a first distortion process on the laser point cloud when receiving the laser point cloud collected by the laser radar in the current frame, obtain a reduced-dimensional point cloud, and convert it into a point cloud depth map;

[0058] A point cloud virtualization module, used for sampling points on the line beams between the reduced-dimensional point cloud and the laser radar according to a predetermined interval and performing a second distortion process to obtain reduced-dimensional virtual point clouds corresponding to each group of the line beams;

[0059] A point cloud marking module, used to use the reduced-dimensional point cloud and the plurality of reduced-dimensional virtual point clouds to respectively match a preset high-precision map, and mark the pixel positions in the point cloud depth map according to the matching results to obtain a plurality of identification pixels;

[0060] The change alarm module is used to calculate the pixel ratio of all the identified pixels in the point cloud depth map, and generate a map change result based on the comparison result of the pixel ratio and a preset ratio threshold.

[0061] A third aspect of the present invention provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the online detection method for high-precision map changes as described in any one of the first aspect of the present invention.

[0062] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the online detection method for high-precision map changes as described in any one of the first aspects of the present invention.

[0063] It can be seen from the above technical solutions that the present invention has the following advantages:

[0064] The present invention collects point clouds of the surrounding environment through a laser radar set on an automatic driving system. When receiving the laser point cloud collected by the laser radar in the current frame, the laser point cloud is subjected to a first distortion process to obtain a reduced-dimensional point cloud, and the reduced-dimensional point cloud is converted into a point cloud depth map. After sampling the line beams between each point in the reduced-dimensional point cloud and the laser radar according to a predetermined spacing, the sampled point cloud is subjected to a second distortion process to obtain reduced-dimensional virtual point clouds corresponding to each group of line beams. The reduced-dimensional point cloud and multiple reduced-dimensional virtual point clouds are then used to match high-precision maps respectively, and the pixel positions in the point cloud depth map are marked according to the matching results to obtain multiple identification pixels, and the pixel ratios occupied by all identification pixels in the point cloud depth map are calculated. The map change results are generated according to the comparison results of the pixel ratios and the preset ratio thresholds. This solves the technical problem that the existing map update scheme cannot detect the changes in the spatial structure of the high-precision map in real time due to the content limitations of the two-dimensional image and the delay of the model processing. The laser point cloud collected by the lidar is divided into harness groups after dimensionality reduction, and the pixel ratio is determined according to the matching results of the point cloud corresponding to each group in the high-precision map. Finally, a map change warning is generated based on the comparison result of the pixel ratio with the preset ratio threshold. This can inform the changes in the current environment in real time and improve the safety of autonomous driving. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0066] Figure 1 A flowchart of the steps of an online detection method for high-precision map changes provided in Embodiment 1 of the present invention;

[0067] Figure 2 A flowchart of the steps of an online detection method for high-precision map changes provided in Embodiment 2 of the present invention;

[0068] Figure 3 A schematic diagram of a laser point cloud dedistortion process provided by an embodiment of the present invention;

[0069] Figure 4 A schematic diagram of a point cloud depth map provided by an embodiment of the present invention;

[0070] Figure 5 A schematic diagram of a virtual point cloud division process provided by an embodiment of the present invention;

[0071] Figure 6 A schematic diagram of a virtual point cloud provided by an embodiment of the present invention;

[0072] Figure 7 A schematic diagram of a first pixel position mark provided by an embodiment of the present invention;

[0073] Figure 8 A schematic diagram of a third pixel position mark provided by an embodiment of the present invention;

[0074] Fig. 9 A schematic diagram of a second pixel position mark provided by an embodiment of the present invention;

[0075] Fig.10 The high-precision ground provided by the embodiment of the present invention Figure 3 Dimensional schematic diagram;

[0076] Fig.11 A flowchart of the steps of an online detection method for high-precision map changes provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0077] Embodiments of the present invention provide a method, device, equipment and medium for online detection of changes in high-precision maps, which are used to solve the technical problem that existing map update solutions cannot detect changes in the spatial structure of high-precision maps in real time due to content limitations of two-dimensional images and the delay of model processing, resulting in reduced safety of autonomous driving.

[0078] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0079] See also Figure 1 , Figure 1 A flowchart of the steps of an online detection method for high-precision map changes provided in Example 1 of the present invention.

[0080] The present invention provides an online detection method for high-precision map changes, which is applied to an autonomous driving system. The autonomous driving system caches a high-precision map corresponding to a dense point cloud and is provided with a laser radar. The method comprises:

[0081] Step 101, when receiving the laser point cloud collected by the laser radar in the current frame, performing a first distortion process on the laser point cloud to obtain a reduced-dimensional point cloud and convert it into a point cloud depth map;

[0082] High Definition Map (HD Map) refers to a lightweight map that can provide lane-level road information and road prior information, and can quantitatively describe the environment in which the autonomous driving system is located. It can be done by dividing the dense map point cloud of the entire city into 1m*1m*1m cubic grids, and each grid only stores the mean and variance of the Gaussian distribution calculated after the three-dimensional coordinates of all points in the grid. Specific types may include but are not limited to grid distribution maps DistMap and octree maps OctoMap, etc. For the autonomous driving system, in order to ensure accuracy while reducing resource consumption, the distribution map within a radius of 200 meters with the location as the center can be loaded according to the location of the system.

[0083] Laser point cloud refers to point data generated by the laser radar (Light Detection and Ranging, LiDAR) installed by the autonomous driving system, such as the autonomous driving vehicle, which uses a multi-line laser to emit lasers to the surrounding environment of the vehicle, and based on the returned lasers. Each point has corresponding three-dimensional coordinates (x, y, z), distance, beam number, horizontal rotation angle and other parameters, such as a rotating 64-line laser radar. The specific model of the laser radar is not limited in this embodiment of the application.

[0084] In the embodiment of the present application, after the automatic driving system is started, the dense point cloud corresponding to its location is obtained from the cloud server or the network and the high-precision map is cached, and the laser radar is started to obtain the laser point cloud of the surrounding environment frame by frame. Since the acquisition of the laser point cloud is accompanied by motion distortion due to the movement of the vehicle to which the automatic driving system belongs, when the automatic driving system receives the laser point cloud collected by the laser radar in the current frame, the laser point cloud can be subjected to the first distortion processing according to the posture obtained by the inertial measurement unit IMU in the laser radar to obtain the corresponding reduced-dimensional point cloud.

[0085] At the same time, to facilitate the subsequent point cloud display, the reduced-dimensional point cloud can be converted into a point cloud depth map according to the number of laser radar beams and the scanning interval angle, combined with the distance between the reduced-dimensional point cloud and the laser radar.

[0086] Step 102, sampling the line beams between the dimension-reduced point cloud and the laser radar according to a predetermined spacing and performing a second distortion process to obtain dimension-reduced virtual point clouds corresponding to each group of line beams;

[0087] After the laser point cloud is subjected to a first distortion process to obtain a reduced-dimensional point cloud, point sampling can be performed on the line beam between each point in the reduced-dimensional point cloud and the laser radar at a predetermined interval to obtain a reduced-dimensional virtual point cloud corresponding to each group of line beams.

[0088] After obtaining multiple groups of reduced-dimensionality virtual point clouds, due to motion distortion of the point clouds, a second distortion process may be performed on each reduced-dimensionality virtual point cloud by a dedistortion operation similar to step 101, thereby obtaining reduced-dimensionality virtual point clouds corresponding to each group of line harnesses.

[0089] It should be noted that the predetermined spacing may be set to 1 meter. At the same time, in order to save consumption of computing resources, point sampling may not be performed on the wiring harness beyond the predetermined range, for example, 10 meters.

[0090] It should be noted that in the embodiment of the present application, the posture increment can be obtained by integrating the IMU data within the current frame time range, and the posture increment is used to transform the point cloud between coordinate systems to achieve dedistortion of the current frame point cloud. For the first distortion processing and the second distortion processing to achieve dedistortion, the same dedistortion scheme can be used, or different dedistortion schemes can be used.

[0091] Step 103, using the reduced-dimensional point cloud and multiple reduced-dimensional virtual point clouds to match the preset high-precision map respectively, marking the pixel positions in the point cloud depth map according to the matching results, and obtaining multiple identification pixels;

[0092] The identification pixel refers to the color information marked at the pixel position of the reduced dimensionality virtual point cloud or the reduced dimensionality point cloud in the point cloud depth map after matching the high-precision map according to the index corresponding to the reduced dimensionality virtual point cloud and the reduced dimensionality point cloud, and satisfying the existence condition of Gaussian distribution information in the matched grid.

[0093] After obtaining each reduced-dimensional virtual point cloud, the corresponding index can be calculated according to the three-dimensional coordinates of the points in each reduced-dimensional virtual point cloud, and then matched with the high-precision map to determine whether there is Gaussian distribution information in the grid where each reduced-dimensional virtual point cloud is located in the high-precision map. After that, the three-dimensional coordinates corresponding to the remaining reduced-dimensional point clouds other than the reduced-dimensional point clouds corresponding to the above-mentioned reduced-dimensional virtual point clouds are further matched with the high-precision map to determine whether there is Gaussian distribution information in the grid where the remaining reduced-dimensional point clouds are located. Finally, the corresponding pixel positions in the point cloud depth map are marked based on the above matching results to obtain multiple identification pixels.

[0094] Step 104, calculate the pixel ratio of all the marked pixels in the point cloud depth map, and generate a map change result based on the comparison result of the pixel ratio and a preset ratio threshold.

[0095] After marking multiple identified pixels, in order to determine the impact of the identified pixels on the HD map, the pixel ratio of all identified pixels in the point cloud depth map can be further calculated, and the pixel ratio can be compared with the preset ratio threshold to determine whether the pixel ratio exceeds the threshold for HD map changes. If the pixel ratio is greater than or equal to the preset ratio threshold, a map change result is generated, thereby achieving online change detection and early warning of HD maps.

[0096] In the embodiment of the present application, the laser radar set on the automatic driving system collects point clouds of the surrounding environment. When the laser point cloud collected by the laser radar in the current frame is received, the laser point cloud is subjected to a first distortion process to obtain a reduced-dimensional point cloud, and the reduced-dimensional point cloud is converted into a point cloud depth map. After sampling the line beams between each point in the reduced-dimensional point cloud and the laser radar according to a predetermined interval, the sampled point cloud is subjected to a second distortion process to obtain reduced-dimensional virtual point clouds corresponding to each group of line beams. The reduced-dimensional point cloud and multiple reduced-dimensional virtual point clouds are then matched with high-precision maps respectively, and the pixel positions in the point cloud depth map are marked according to the matching results to obtain multiple identification pixels. The pixel ratios occupied by all identification pixels in the point cloud depth map are calculated, and the map change results are generated according to the comparison results of the pixel ratios and the preset ratio thresholds. This solves the technical problem that the existing map update scheme cannot detect the changes in the spatial structure of the high-precision map in real time due to the content limitations of the two-dimensional image and the delay of the model processing. The laser point cloud collected by the lidar is divided into harness groups after dimensionality reduction, and the pixel ratio is determined according to the matching results of the point cloud corresponding to each group in the high-precision map. Finally, a map change warning is generated based on the comparison result of the pixel ratio with the preset ratio threshold. This can inform the changes in the current environment in real time and improve the safety of autonomous driving.

[0097] See also Figure 2 , Figure 2 A flowchart of the steps of an online detection method for high-precision map changes provided in Example 2 of the present invention.

[0098] The present invention provides an online detection method for high-precision map changes, which is applied to an autonomous driving system. The autonomous driving system caches a high-precision map corresponding to a dense point cloud and is provided with a laser radar. The method comprises:

[0099] Step 201, when receiving the laser point cloud collected by the laser radar in the current frame, performing a first distortion process on the laser point cloud to obtain a reduced-dimensional point cloud and converting it into a point cloud depth map;

[0100] Optionally, the laser point cloud is provided with a start time stamp and an end time stamp, and step 201 may include the following sub-steps S11-S15:

[0101] S11, when receiving the laser point cloud collected by the laser radar in the current frame, calculating the interpolation interval between the end timestamp and the start timestamp;

[0102] In an example of the present application, after the autonomous driving system is started, the laser radar collects laser point clouds from the external environment frame by frame. When the autonomous driving system receives the laser point cloud collected by the laser radar in the current frame, since the vehicle in which the autonomous driving system is located is in a moving state, each point of the laser point cloud is not generated at the same time. In order to dedistort the laser point cloud, the difference between the start timestamp and the end timestamp contained in the laser point cloud is calculated to obtain the interpolation interval.

[0103] S12, obtaining the first pose information corresponding to the laser point cloud at the starting time stamp, and performing linear interpolation on the first pose information in the interpolation interval to obtain the second pose information corresponding to the laser point cloud at the ending time stamp;

[0104] Linear interpolation refers to the interpolation method in which the interpolation function is a first-order polynomial, and the interpolation error at the interpolation node is zero. Compared with other interpolation methods, such as parabolic interpolation, linear interpolation is simple and convenient. The geometric meaning of linear interpolation is to use the straight line passing through points A and B in the overview diagram to approximate the original function. Linear interpolation can be used to approximate the original function, and can also be used to calculate the value that is not in the table during the table lookup process.

[0105] After obtaining the interpolation interval, since the laser point cloud needs to be reduced to the moment corresponding to the starting timestamp, the first pose information of the laser point cloud corresponding to the starting timestamp can be obtained from the inertial measurement unit imu first, and the first pose information can be linearly interpolated according to the interpolation interval to obtain the second pose information of the laser point cloud corresponding to the ending timestamp.

[0106] S13, calculating a posture change value between the second posture information and the first posture information;

[0107] After obtaining the second posture information, in order to achieve further coordinate transformation, the difference between the second posture information and the first posture information can be calculated to obtain the posture change value.

[0108] In one embodiment, the posture change value can also be calculated in the following manner: T_12=T_1.inverse()*T_2, where inverse() is an inverse matrix operation, which in this embodiment is to find the inverse matrix of the second posture information.

[0109] S14, according to the preset calibration external parameters and posture change values, the laser point cloud corresponding to the termination time stamp is converted into a coordinate system to obtain a reduced-dimensional point cloud;

[0110] Calibration extrinsic parameters refer to the parameters used to determine the relative position relationship between the camera coordinate system and the world coordinate system, where Pw is the world coordinate, Pc is the camera coordinate, and the relationship between them is Pc=RPw+T, where T=(Tx, Ty, Tz), is the translation vector, and R=R(α, β, γ) is the rotation matrix, which are the rotation angles γ around the z-axis of the camera coordinate system, β around the y-axis, and α around the x-axis. The six parameters (α, β, γ, Tx, Ty, Tz) are the camera extrinsic parameters. In the embodiment of the present application, the camera coordinate system is the coordinate system where the laser point cloud is located, and the world coordinate system is the coordinate system where the posture information is located, which can be calibrated and obtained by linear calibration or nonlinear calibration methods.

[0111] like Figure 3 As shown, a schematic diagram of a laser point cloud dedistortion process in an embodiment of the present application is shown.

[0112] In an example of the present application, after obtaining the posture change value T_12 and the calibration external parameter T_lidar2imu, the laser point cloud p2 corresponding to the termination timestamp t2 can be transformed to the imu coordinate system through P2'=T_lidar2imu*p2, and then P2' can be transformed to the imu coordinate system at the initial timestamp t1. The corresponding formula is P2"=T_12*P2', thereby achieving dimensionality reduction of the laser point cloud and obtaining a reduced dimensionality point cloud.

[0113] S15. Generate a point cloud depth map according to the specification parameters corresponding to the laser radar and the three-dimensional coordinates corresponding to each dimensionality reduction point cloud.

[0114] Furthermore, the specification parameters include the number of line beams and the scanning interval angle, and S15 may include the following sub-steps:

[0115] Calculating an angle ratio between a preset visual angle and a scanning interval angle;

[0116] The number of line bundles is used as the number of depth map rows, and the angle ratio is used as the number of depth map columns to create an initial depth map.

[0117] Calculate the point cloud distance between each dimension-reduced point cloud and the laser radar respectively;

[0118] Each point cloud distance is used as the pixel value corresponding to each pixel position in the initial depth map to obtain a point cloud depth map.

[0119] The number of beams refers to the number of beams used by the LiDAR. For example, for a 64-beam LiDAR, the number of beams is 64. The scanning interval angle refers to the angle between two adjacent points when the LiDAR uses the same beam for scanning. Taking the above 64-beam LiDAR as an example, the scanning interval angle is 0.2 degrees.

[0120] To determine the specific specifications of the current point cloud depth map, we can first calculate the angle ratio between the preset visual angle and the scanning interval angle, use the number of line bundles as the number of depth map rows, and use the angle ratio as the number of depth map columns to construct the initial depth map.

[0121] At the same time, based on the 3D coordinates contained in each reduced-dimensional point cloud and taking the laser radar as the origin, the distance between each reduced-dimensional point cloud and the laser radar can be calculated as the point cloud distance. After the above transformation, each reduced-dimensional point cloud can correspond to the pixel position in the initial depth map. At this time, the point cloud distance can be used as the pixel value corresponding to each pixel position in the initial depth map to obtain the point cloud depth map. See Figure 4 .

[0122] Among them, the setting of the preset visual angle can be set based on actual conditions. For example, if only detection is performed in front of the vehicle, it can be set to 180 degrees. If panoramic detection of the vehicle's surrounding environment is required, it can be set to 360 degrees. The embodiment of the present application does not limit the setting of specific angles.

[0123] Optionally, the imu data within the current frame time range can be integrated to obtain the pose increment, thereby performing a dedistortion operation on the current frame point cloud.

[0124] Step 202, sampling the line beams between the reduced-dimensional point cloud and the laser radar according to a predetermined spacing and performing a second distortion process to obtain reduced-dimensional virtual point clouds corresponding to each group of line beams;

[0125] Optionally, step 202 may include the following sub-steps:

[0126] Perform point sampling on the line bundles between the dimension reduction point cloud and the laser radar according to a predetermined spacing to obtain a virtual point cloud corresponding to each group of line bundles;

[0127] Obtain the posture information corresponding to each virtual point cloud;

[0128] The pose information is used to perform a second distortion process on each virtual point cloud to obtain a reduced-dimensional virtual point cloud corresponding to each group of wire harnesses.

[0129] In the embodiment of the present application, in order to further improve the accuracy of obstacle recognition, the line beams between the dimensionality reduction point cloud and the laser radar can be sampled at a predetermined interval to obtain a virtual point cloud corresponding to each group of line beams, such as Figure 5As shown, obstacle refers to the endpoint of the virtual point cloud, lidar is the origin corresponding to the laser radar, and all points corresponding to the group of beams are used as a virtual point cloud. Then, the pose information corresponding to each virtual point cloud is obtained, and each pose information is used to perform a second distortion process on each virtual point cloud, so as to obtain a reduced-dimensional virtual point cloud corresponding to each group of beams. The specific de-distortion process can refer to the implementation process of step 201.

[0130] In one example of the present application, the predetermined spacing may be an equal spacing, such as 1 meter, see Figure 6 As shown, Figure 6 The virtual point cloud diagram is limited to a range of 10 meters. Alternatively, the spacing may be increased according to a preset gradient, such as 0.5 meters.

[0131] Step 203, using each dimension-reduced virtual point cloud to match the high-precision map respectively, to determine a first point cloud grid;

[0132] In an example of the present application, step 203 may include the following sub-steps:

[0133] Obtain the first three-dimensional coordinates corresponding to each dimension-reduced virtual point cloud;

[0134] Calculate the first index respectively using each first three-dimensional coordinate;

[0135] The high-precision map is matched according to each first index, and the point cloud grid corresponding to each first index is determined as the first point cloud grid.

[0136] In an embodiment of the present application, after obtaining each reduced-dimensional virtual point cloud, the first three-dimensional coordinates corresponding to each reduced-dimensional virtual point cloud can be obtained, and the first index corresponding to each reduced-dimensional virtual point cloud can be calculated respectively according to the index calculation method pre-stored in the autonomous driving system and the first three-dimensional coordinates. The high-precision map is matched according to the first index, and the point cloud mesh corresponding to each first index is determined as the first point cloud mesh.

[0137] It should be noted that, taking the three-dimensional coordinates (x1, y1, z1) as an example, for a depth map with R rows and C columns, the index calculation method can be as follows:

[0138] Index S = (x1-1)*C+y1+(z1-1)*R*C

[0139] Step 204, determining the reduced-dimensional point cloud to which the first point cloud grid having the Gaussian distribution information cached belongs as the first target point cloud, and determining the reduced-dimensional point clouds other than the first target point cloud as the remaining point clouds;

[0140] In a specific implementation, if there is an obstacle in the point cloud grid in the high-precision map, the point cloud grid caches the corresponding Gaussian distribution information; if there is no obstacle, the point cloud grid does not store Gaussian distribution information.

[0141] After determining the first point cloud grid, all the first point cloud grids with cached Gaussian distribution information can be screened out, and the virtual dimensionality reduction point cloud should not have Gaussian distribution information in the first point cloud grid. If Gaussian distribution information is cached, it indicates that the obstacles in the high-precision map at this location have been removed. Therefore, the dimensionality reduction point cloud to which the first point cloud grid with cached Gaussian distribution information belongs can be determined as the first target point cloud.

[0142] At the same time, in order to facilitate the subsequent processing of the remaining dimensionality reduction point clouds, the dimensionality reduction point clouds except the first target point cloud can be determined as the remaining point clouds.

[0143] In another example of the present application, a line bundle can be used as a detection group, and each first point cloud grid can be detected in sequence from near to far according to the distance between each first point cloud grid and the laser radar to see whether Gaussian distribution information is cached; if it is detected that any first point cloud grid has Gaussian distribution information cached, the detection of the line bundle is stopped, and the reduced dimensionality point cloud to which it belongs is determined as the first target point cloud.

[0144] Step 205, obtaining a first pixel position corresponding to the first target point cloud from the point cloud depth map, and marking a first identification pixel at the first pixel position;

[0145] In an embodiment of the present application, the identification pixel includes a first identification pixel and a second identification pixel. After determining the first target point cloud, since the first target point cloud is part of the dimensionality reduction point cloud, its first pixel position in the point cloud depth map can be calculated according to the first target power source, and the first pixel position can be marked as the first identification pixel, indicating that there is a change of "obstacle removed" at this position.

[0146] It should be noted that the first pixel position may be marked by using a color pixel, for example, yellow is used to mark the first pixel position. Figure 7 shown.

[0147] Step 206, using the remaining point cloud to match the high-precision map to determine the second point cloud grid;

[0148] Optionally, step 206 may include the following sub-steps:

[0149] Get the second three-dimensional coordinates corresponding to the remaining point clouds;

[0150] Calculate the second index respectively using each second three-dimensional coordinate;

[0151] The high-precision map is matched according to each second index, and the point cloud grid corresponding to each second index is determined as the second point cloud grid.

[0152] After screening the remaining point clouds, in order to determine whether there are any additional obstacles in the high-precision map, the second three-dimensional coordinates corresponding to the remaining point clouds can be obtained, and the second indexes can be calculated respectively using the second three-dimensional coordinates. The high-precision map is further matched using each second index, and the point cloud mesh corresponding to each second index is determined as the second point cloud mesh.

[0153] The specific calculation method of the second index is similar to the calculation method of the first index, which will not be repeated here.

[0154] In an optional embodiment of the present application, before executing step 206, the method may further include the following steps S21-S24:

[0155] S21, converting the dimension-reduced point cloud into a map coordinate system corresponding to the high-precision map to obtain a map point cloud;

[0156] S22, using a preset three-dimensional perception frame to detect the map point cloud, determine the dynamic obstacles in the map point cloud, and mark the corresponding reduced-dimensional point cloud as the dynamic obstacle point cloud;

[0157] S23, selecting the remaining point clouds that overlap with the dynamic obstacle point cloud as the second target point cloud;

[0158] S24. Obtain a third pixel position corresponding to the second target point cloud from the point cloud depth map, and mark a third identification pixel at the third pixel position.

[0159] In the specific implementation, there may be dynamic obstacles such as vehicles and pedestrians in the HD map. In order to further screen the reduced-dimensional point cloud and reduce the resource consumption of the autonomous driving system for HD map search, all reduced-dimensional point clouds can be converted to the map coordinate system corresponding to the HD map to obtain the map point cloud, and then the map point cloud is detected using a preset three-dimensional perception box such as 3DBounding Box to determine the dynamic obstacles in the map point cloud, and the corresponding reduced-dimensional point cloud is marked as a dynamic obstacle point cloud. The remaining point cloud that overlaps with the dynamic obstacle point cloud is further selected as the second target point cloud, and the third pixel position corresponding to the second target point cloud is obtained from the point cloud depth map, and the third identification pixel is marked at the third pixel position.

[0160] Optionally, the detection of dynamic obstacles may also be achieved by using, for example, a deep learning model, which is not limited in the embodiments of the present application.

[0161] It should be noted that the third pixel position can be marked by using a color pixel, for example, using blue to mark the third pixel position, to indicate that the laser radar beam is hitting a dynamic obstacle, such as Figure 8 shown.

[0162] Step 207, determining whether Gaussian distribution information is cached in each second point cloud grid;

[0163] After determining the second point cloud grid, in order to determine whether there are new obstacles or map missing in the high-precision map, it is possible to further determine whether Gaussian distribution information is cached in the second point cloud grid.

[0164] Step 208, if not, obtaining the pixel position of the remaining point cloud to which the second point cloud grid belongs from the point cloud depth map as the second pixel position, and marking a second identification pixel at the second pixel position;

[0165] In this embodiment, if the first point cloud grid does not cache Gaussian distribution information, it indicates that there is no obstacle at this location in the high-precision map, but in fact the laser radar hits an obstacle at this location, that is, there is a new obstacle or map missing at this location in the high-precision map. At this time, the pixel position of the remaining point cloud to which the second point cloud grid belongs can be obtained from the point cloud depth map, which is used as the second pixel position, and the second identification pixel is marked at the second pixel position.

[0166] It should be noted that the second pixel position can be marked by using a color pixel, for example, red to mark the second pixel position, to indicate that there are new obstacles or map missing changes in the high-precision map at this position, such as Fig. 9 shown.

[0167] It should be noted that, in order to improve processing efficiency, in the process of marking the first pixel position, the second pixel position and the third pixel position, they can be marked separately in multiple point cloud depth maps of the same specification, or they can be marked separately in the same point cloud depth map.

[0168] Step 209, calculate the pixel ratio of all the marked pixels in the point cloud depth map, and generate a map change result based on the comparison result of the pixel ratio and the preset ratio threshold.

[0169] In an example of the present application, step 209 may include the following sub-steps:

[0170] Calculate the pixel ratio of the identified pixel in the point cloud depth map;

[0171] comparing the pixel ratio with a preset ratio threshold;

[0172] If the pixel ratio is greater than or equal to the preset ratio threshold, the reduced-dimensional point cloud or reduced-dimensional virtual point cloud corresponding to the identified pixel at the current moment is cached to a preset cloud server to generate a map change result.

[0173] In the specific implementation, since the change of dynamic obstacles is not part of the map change, the ratio of lidar rays (number of points) hitting the changed area to the total number of rays (number of points) is calculated, which can also be understood as the ratio of points in the current point cloud that do not match the distribution map on the vehicle to all points in the current point cloud. The pixel ratio of the first identification pixel and the second identification pixel in the point cloud depth map can be directly calculated. For example, the first pixel ratio of the first identification pixel to the point cloud depth map and the second pixel ratio of the second identification pixel to the point cloud depth map are calculated separately, and the sum of the first pixel ratio and the second pixel ratio is calculated to obtain the pixel ratio.

[0174] The pixel ratio is further compared with the preset ratio threshold. If the pixel ratio is greater than or equal to the preset ratio threshold, it indicates that the high-precision map has undergone a significant change at this time. The reduced-dimensional point cloud corresponding to the second identification pixel at the current moment and the reduced-dimensional virtual point cloud corresponding to the first identification pixel can be cached to the preset cloud server to facilitate the generation of the corresponding high-precision map in the cloud for offline modification. The high-precision map can be Fig.10 As shown, the map change results are generated and the warning information is output at the same time to inform the user that there is a big difference between the high-precision map and the current environment.

[0175] Optionally, if the pixel ratio is less than a preset ratio threshold, the reduced-dimensional point cloud or reduced-dimensional virtual point cloud corresponding to the identified pixel at the current moment is still cached to a preset cloud server to facilitate subsequent review and modification of the high-precision map.

[0176] Furthermore, in order to further improve the accuracy of map change detection, the method may further include the following steps S31-S35:

[0177] S31, obtaining the pixel position of the noise point in the point cloud depth map;

[0178] S32, determining whether there are pixel values ​​at multiple undetermined pixel positions adjacent to the noise pixel position;

[0179] S33, if all the undetermined pixel positions have pixel values, retain the noise pixel position;

[0180] S34, if there is no pixel value at any undetermined pixel position, delete the noise pixel position;

[0181] S35. After all noise pixel positions are determined, the point cloud depth map at the current moment is used to update the point cloud depth map to obtain updated identification pixels.

[0182] In the embodiment of the present application, due to the limited accuracy of the pose information used for laser point cloud dedistortion, there may be a deviation between the reduced dimensionality point cloud and the actual position. Although the noise does not affect the detection of larger map changes, it can be filtered to further improve the accuracy of map change detection, and some non-clustered changes can also be filtered out. First, the noise pixel position can be further obtained from the point cloud depth map to determine whether there are pixel values ​​at multiple pending pixel positions adjacent to the noise pixel position, such as determining whether there are pixel values ​​at the pending pixel positions of the nine-square grid around the noise pixel position; if all pending pixel positions have pixel values, the noise pixel position is retained, and if any pending pixel position does not have a pixel value, the noise pixel position is deleted; when all noise pixel positions are determined, the point cloud depth map at the current moment is used to update the point cloud depth map to obtain an updated identification pixel.

[0183] In another example of the present invention, after executing S35, the method may further include the following steps S36-S38:

[0184] S36, performing a region search and clustering on the point cloud depth map to obtain a plurality of pixel regions consisting of search pixel positions;

[0185] S37, deleting pixel regions where the number of search pixel positions is less than a preset number threshold from the multiple pixel regions, to obtain an updated point cloud depth map;

[0186] S38. Use the point cloud depth map at the current moment to update the point cloud depth map to obtain updated identification pixels.

[0187] Among them, the area search can adopt a variety of area search algorithms, such as the depth-first search algorithm (DFS), which is an algorithm for traversing or searching a tree or graph. Traverse the nodes of the tree along the depth of the tree and search the branches of the tree as deeply as possible. When all the edges where the node v is located have been explored, the search will trace back to the starting node of the edge where the node v was found. This process continues until all nodes reachable from the source node have been found. If there are still undiscovered nodes, one of them is selected as the source node and the above process is repeated. The whole process is repeated until all nodes have been visited.

[0188] In the embodiment of the present application, after the existing identification pixel is updated, a regional search can be performed on the point cloud depth map to determine the search pixel positions with large changes in the point cloud depth map, and each search pixel position is clustered to obtain a plurality of pixel regions composed of the search pixel positions. In order to filter the changes in small areas and improve the detection accuracy, the pixel regions with a number less than a preset number threshold can be deleted according to the number of search pixel positions in each pixel region to obtain the updated identification pixel.

[0189] In the embodiment of the present application, the laser radar set on the automatic driving system collects point clouds of the surrounding environment. When the laser point cloud collected by the laser radar in the current frame is received, the laser point cloud is subjected to a first distortion process to obtain a reduced-dimensional point cloud, and the reduced-dimensional point cloud is converted into a point cloud depth map. After sampling the line beams between each point in the reduced-dimensional point cloud and the laser radar according to a predetermined interval, the sampled point cloud is subjected to a second distortion process to obtain reduced-dimensional virtual point clouds corresponding to each group of line beams. The reduced-dimensional point cloud and multiple reduced-dimensional virtual point clouds are then matched with high-precision maps respectively, and the pixel positions in the point cloud depth map are marked according to the matching results to obtain multiple identification pixels. The pixel ratios occupied by all identification pixels in the point cloud depth map are calculated, and the map change results are generated according to the comparison results of the pixel ratios and the preset ratio thresholds. This solves the technical problem that the existing map update scheme cannot detect the changes in the spatial structure of the high-precision map in real time due to the content limitations of the two-dimensional image and the delay of the model processing. The laser point cloud collected by the lidar is divided into harness groups after dimensionality reduction, and the pixel ratio is determined according to the matching results of the point cloud corresponding to each group in the high-precision map. Finally, a map change warning is generated based on the comparison result of the pixel ratio with the preset ratio threshold. This can inform the changes in the current environment in real time and improve the safety of autonomous driving.

[0190] See also Fig.11 , Fig.11 This is a structural block diagram of an online detection device for high-precision map changes provided in Example 3 of the present invention.

[0191] The embodiment of the present invention provides an online detection device for high-precision map changes, which is applied to an autonomous driving system. The autonomous driving system caches a high-precision map corresponding to a dense point cloud and is provided with a laser radar. The device includes:

[0192] The point cloud acquisition and dedistortion module 1101 is used to perform a first distortion process on the laser point cloud when receiving the laser point cloud collected by the laser radar in the current frame, obtain a reduced-dimensional point cloud, and convert it into a point cloud depth map;

[0193] The point cloud virtualization module 1102 is used to perform point sampling on the line beams between the reduced-dimensional point cloud and the laser radar according to a predetermined interval and perform a second distortion process to obtain reduced-dimensional virtual point clouds corresponding to each group of line beams;

[0194] The point cloud marking module 1103 is used to match the preset high-precision map with the reduced-dimensional point cloud and multiple reduced-dimensional virtual point clouds respectively, and mark the pixel positions in the point cloud depth map according to the matching results to obtain multiple identification pixels;

[0195] The change alarm module 1104 is used to calculate the pixel ratio of all identified pixels in the point cloud depth map, and generate a map change result based on the comparison result of the pixel ratio and a preset ratio threshold.

[0196] Optionally, the laser point cloud is provided with a start time stamp and an end time stamp; the point cloud acquisition and dedistortion module 1101 includes:

[0197] The interpolation interval determination submodule is used to calculate the interpolation interval between the end timestamp and the start timestamp when receiving the laser point cloud collected by the laser radar in the current frame;

[0198] The posture information determination submodule is used to obtain the first posture information corresponding to the laser point cloud at the starting time stamp, and perform linear interpolation on the first posture information in the interpolation interval to obtain the second posture information corresponding to the laser point cloud at the ending time stamp;

[0199] A posture change calculation submodule, used to calculate the posture change value between the second posture information and the first posture information;

[0200] The point cloud dimensionality reduction submodule is used to convert the coordinate system of the laser point cloud corresponding to the termination time stamp according to the preset calibration external parameters and posture change values ​​to obtain the reduced-dimensional point cloud;

[0201] The depth map generation submodule is used to generate a point cloud depth map according to the specification parameters corresponding to the laser radar and the three-dimensional coordinates corresponding to each dimensionality reduction point cloud.

[0202] Optionally, the specification parameters include the number of line beams and the scanning interval angle; the depth map generation submodule is specifically used for:

[0203] Calculating an angle ratio between a preset visual angle and a scanning interval angle;

[0204] The number of line bundles is used as the number of depth map rows, and the angle ratio is used as the number of depth map columns to create an initial depth map.

[0205] Calculate the point cloud distance between each dimension-reduced point cloud and the laser radar respectively;

[0206] Each point cloud distance is used as the pixel value corresponding to each pixel position in the initial depth map to obtain a point cloud depth map.

[0207] Optionally, the point cloud virtualization module 1102 includes:

[0208] A virtual point cloud generation submodule is used to perform point sampling on the line beams between the dimension reduction point cloud and the laser radar according to a predetermined spacing to obtain a virtual point cloud corresponding to each group of line beams;

[0209] The posture information acquisition submodule is used to obtain the posture information corresponding to each virtual point cloud;

[0210] The virtual point cloud dedistortion submodule is used to perform a second distortion process on each virtual point cloud using the posture information to obtain a reduced-dimensional virtual point cloud corresponding to each group of wire harnesses.

[0211] Optionally, the identification pixel includes a first identification pixel and a second identification pixel; and the point cloud marking module 1103 includes:

[0212] A first point cloud grid determination submodule is used to use each dimension-reduced virtual point cloud to match a preset high-precision map to determine a first point cloud grid;

[0213] A point cloud screening submodule, used to determine the reduced-dimensional point cloud to which the first point cloud grid having the Gaussian distribution information cached is attached as the first target point cloud, and to determine the reduced-dimensional point clouds other than the first target point cloud as the remaining point clouds;

[0214] A first identification pixel marking submodule is used to obtain a first pixel position corresponding to a first target point cloud from a point cloud depth map, and mark a first identification pixel at the first pixel position;

[0215] A second point cloud mesh determination submodule is used to match the high-precision map with the remaining point cloud to determine the second point cloud mesh;

[0216] A Gaussian distribution information cache judgment submodule is used to judge whether Gaussian distribution information is cached in each second point cloud grid;

[0217] The second identification pixel marking submodule is used to obtain the pixel position of the remaining point cloud belonging to the second point cloud grid from the point cloud depth map as the second pixel position, and mark the second identification pixel at the second pixel position.

[0218] Optionally, the first point cloud mesh determination submodule is specifically used for:

[0219] Obtain the first three-dimensional coordinates corresponding to each dimension-reduced virtual point cloud;

[0220] Calculate the first index respectively using each first three-dimensional coordinate;

[0221] The high-precision map is matched according to each first index, and the point cloud grid corresponding to each first index is determined as the first point cloud grid.

[0222] Optionally, the change alarm module 1104 includes:

[0223] A pixel ratio calculation submodule, used to calculate the pixel ratio of the first identification pixel and the second identification pixel in the point cloud depth map;

[0224] A ratio comparison submodule, used to compare the pixel ratio with a preset ratio threshold;

[0225] The map change alarm submodule is used to cache the reduced-dimensional point cloud or reduced-dimensional virtual point cloud corresponding to the identified pixel at the current moment to a preset cloud server if the pixel ratio is greater than or equal to a preset ratio threshold, generate a map change result and output an alarm message.

[0226] Optionally, the point cloud marking module 1103 further includes:

[0227] The map transformation submodule is used to convert the dimension-reduced point cloud into the map coordinate system corresponding to the high-precision map to obtain the map point cloud;

[0228] The dynamic obstacle point cloud determination submodule is used to detect the map point cloud using a preset three-dimensional perception frame, determine the dynamic obstacles in the map point cloud, and mark the corresponding reduced-dimensional point cloud as the dynamic obstacle point cloud;

[0229] The second target point cloud determination submodule is used to select the remaining point clouds that overlap with the dynamic obstacle point cloud as the second target point cloud;

[0230] The third identification pixel marking submodule is used to obtain the third pixel position corresponding to the second target point cloud from the point cloud depth map, and mark the third identification pixel at the third pixel position.

[0231] Optionally, the device further comprises:

[0232] A noise pixel position determination module is used to obtain the noise pixel position in the point cloud depth map;

[0233] A pixel value determination module is used to determine whether there are pixel values ​​at a plurality of pending pixel positions adjacent to a noise pixel position;

[0234] A noise pixel position retaining module is used to retain the noise pixel position if all the pending pixel positions have pixel values;

[0235] A noise pixel position deletion module is used to delete the noise pixel position if there is no pixel value at any pending pixel position;

[0236] The first identification pixel updating module is used to update the point cloud depth map by using the point cloud depth map at the current moment after all noise pixel positions are determined, so as to obtain updated identification pixels.

[0237] Optionally, it further includes: a pixel region determination module, used to perform region search and clustering on the point cloud depth map to obtain a plurality of pixel regions consisting of search pixel positions;

[0238] A pixel region screening module is used to delete pixel regions whose number of search pixel positions is less than a preset number threshold from multiple pixel regions to obtain an updated point cloud depth map;

[0239] The second identification pixel updating module is used to update the point cloud depth map by using the point cloud depth map at the current moment to obtain an updated identification pixel.

[0240] An embodiment of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the online detection method for high-precision map changes as described in any embodiment of the present invention.

[0241] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed, the method for online detection of changes in a high-precision map as described in any embodiment of the present invention is implemented.

[0242] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, modules and sub-modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0243] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0244] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0245] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0246] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0247] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A high-precision map detection method, characterized in that: include: When receiving a laser point cloud collected by the laser radar in a current frame, performing a first distortion process on the laser point cloud to obtain a reduced-dimensional point cloud and converting it into a point cloud depth map; Performing point sampling on the line beams between the reduced-dimensional point cloud and the laser radar according to a predetermined interval and performing a second distortion process to obtain reduced-dimensional virtual point clouds corresponding to each group of the line beams; The reduced-dimensional point cloud and the plurality of reduced-dimensional virtual point clouds are respectively matched with a preset high-precision map, and pixel positions in the point cloud depth map are marked according to the matching results to obtain a plurality of identification pixels; Calculating the pixel ratio of all the identified pixels in the point cloud depth map, and generating a map change result according to a comparison result of the pixel ratio with a preset ratio threshold; The laser point cloud is provided with a start time stamp and an end time stamp; when receiving the laser point cloud collected by the laser radar in the current frame, the laser point cloud is subjected to a first distortion process to obtain a reduced-dimensional point cloud and convert it into a point cloud depth map, comprising: When receiving the laser point cloud collected by the laser radar in the current frame, calculating the interpolation interval between the end timestamp and the start timestamp; Acquire the first pose information of the laser point cloud corresponding to the starting time stamp, and perform linear interpolation on the first pose information in the interpolation interval to obtain the second pose information of the laser point cloud corresponding to the ending time stamp; Calculating a posture change value between the second posture information and the first posture information; According to the preset calibration external parameters and the pose change value, the laser point cloud corresponding to the termination time stamp is converted into a coordinate system to obtain a reduced-dimensional point cloud; A point cloud depth map is generated according to the specification parameters corresponding to the laser radar and the three-dimensional coordinates corresponding to each of the reduced-dimensional point clouds.

2. The method according to claim 1, characterized in that The specification parameters include the number of beams and the scanning interval angle; The step of generating a point cloud depth map according to the specification parameters corresponding to the laser radar and the three-dimensional coordinates corresponding to each of the reduced-dimensional point clouds comprises: Calculating an angle ratio between a preset visual angle and the scanning interval angle; Using the number of line bundles as the number of depth map rows and the angle ratio as the number of depth map columns, an initial depth map is created; Calculating the point cloud distance between each of the reduced-dimensional point clouds and the laser radar respectively; Each of the point cloud distances is used as a pixel value corresponding to each pixel position in the initial depth map to obtain a point cloud depth map.

3. The method according to claim 1, characterized in that The step of sampling the line beams between the reduced-dimensional point cloud and the laser radar at a predetermined interval and performing a second distortion process to obtain reduced-dimensional virtual point clouds corresponding to each group of the line beams comprises: Performing point sampling on the line beams between the dimension reduction point cloud and the laser radar according to a predetermined spacing to obtain a virtual point cloud corresponding to each group of the line beams; Obtaining the posture information corresponding to each of the virtual point clouds; The pose information is used to perform a second distortion process on each of the virtual point clouds to obtain a reduced-dimensional virtual point cloud corresponding to each group of the wire harnesses.

4. The method according to claim 1, characterized in that: The identification pixel includes a first identification pixel and a second identification pixel; the step of using the reduced-dimensional point cloud and the plurality of reduced-dimensional virtual point clouds to match a preset high-precision map respectively, marking the pixel positions in the point cloud depth map according to the matching results, and obtaining a plurality of identification pixels includes: Using each of the dimensionally reduced virtual point clouds to match a preset high-precision map, respectively, to determine a first point cloud grid; Determine the reduced-dimensional point cloud to which the first point cloud grid having the Gaussian distribution information cached therein belongs as a first target point cloud, and determine the reduced-dimensional point clouds other than the first target point cloud as remaining point clouds; Acquire a first pixel position corresponding to the first target point cloud from the point cloud depth map, and mark a first identification pixel at the first pixel position; Matching the high-precision map with the remaining point cloud to determine a second point cloud grid; Determining whether Gaussian distribution information is cached in each of the second point cloud grids; If not, the pixel position of the remaining point cloud to which the second point cloud grid belongs is obtained from the point cloud depth map as the second pixel position, and a second identification pixel is marked at the second pixel position.

5. The method according to claim 4, characterized in that The step of using each of the reduced-dimensional virtual point clouds to match a preset high-precision map to determine a first point cloud grid includes: Obtaining first three-dimensional coordinates corresponding to each of the reduced-dimensional virtual point clouds; Calculate first indexes respectively using each of the first three-dimensional coordinates; The high-precision map is matched according to each of the first indexes, and the point cloud mesh corresponding to each of the first indexes is determined as the first point cloud mesh.

6. The method according to claim 4, characterized in that Before the step of using the remaining point cloud to match the high-precision map to determine the second point cloud grid, the method further includes: Convert the dimension-reduced point cloud to a map coordinate system corresponding to the high-precision map to obtain a map point cloud; Using a preset three-dimensional perception frame to detect the map point cloud, determine the dynamic obstacles in the map point cloud, and mark the corresponding reduced-dimensional point cloud as the dynamic obstacle point cloud; Selecting the remaining point clouds that overlap with the dynamic obstacle point cloud as the second target point cloud; A third pixel position corresponding to the second target point cloud is obtained from the point cloud depth map, and a third identification pixel is marked at the third pixel position.

7. The method according to claim 1, characterized in that The step of calculating the pixel ratio occupied by all the identified pixels in the point cloud depth map, and generating a map change result according to a comparison result of the pixel ratio with a preset ratio threshold, comprises: Calculate the pixel ratio of all the identified pixels in the point cloud depth map; comparing the pixel ratio with a preset ratio threshold; If the pixel ratio is greater than or equal to the preset ratio threshold, the reduced-dimensional point cloud or reduced-dimensional virtual point cloud corresponding to the identified pixel at the current moment is cached to a preset cloud server, a map change result is generated, and an alarm message is output.

8. The method according to any one of claims 1 to 7, characterized in that: The method further comprises: Obtaining the pixel position of the noise point in the point cloud depth map; Determine whether there are pixel values ​​at a plurality of undetermined pixel positions adjacent to the noise pixel position; If the pixel value exists in all the pending pixel positions, retain the noise pixel position; If the pixel value does not exist at any of the pending pixel positions, deleting the noise pixel position; When all the noise pixel positions are determined, the point cloud depth map is updated using the point cloud depth map at the current moment to obtain updated identification pixels.

9. The method according to claim 8, characterized in that Also includes: Performing region search and clustering on the point cloud depth map to obtain a plurality of pixel regions consisting of search pixel positions; Deleting pixel regions where the number of search pixel positions is less than a preset number threshold from the plurality of pixel regions, to obtain an updated point cloud depth map; The point cloud depth map at the current moment is used to update the point cloud depth map to obtain an updated identification pixel.

10. An online detection device for high-precision map changes, characterized in that: include: The point cloud acquisition and dedistortion module is used to perform a first distortion process on the laser point cloud when receiving the laser point cloud collected by the laser radar in the current frame, obtain a reduced-dimensional point cloud, and convert it into a point cloud depth map; A point cloud virtualization module, used for sampling points on the line beams between the reduced-dimensional point cloud and the laser radar according to a predetermined interval and performing a second distortion process to obtain reduced-dimensional virtual point clouds corresponding to each group of the line beams; A point cloud marking module, used to use the reduced-dimensional point cloud and the plurality of reduced-dimensional virtual point clouds to respectively match a preset high-precision map, and mark the pixel positions in the point cloud depth map according to the matching results to obtain a plurality of identification pixels; A change alarm module, used to calculate the pixel ratio of all the identified pixels in the point cloud depth map, and generate a map change result according to the comparison result of the pixel ratio and a preset ratio threshold; The laser point cloud is provided with a start time stamp and an end time stamp; the point cloud acquisition and dedistortion module comprises: An interpolation interval determination submodule, for calculating an interpolation interval between the end timestamp and the start timestamp when receiving a laser point cloud collected by the laser radar in a current frame; A posture information determination submodule, used to obtain the first posture information of the laser point cloud corresponding to the starting time stamp, and perform linear interpolation on the first posture information in the interpolation interval to obtain the second posture information of the laser point cloud corresponding to the ending time stamp; A posture change calculation submodule, used to calculate a posture change value between the second posture information and the first posture information; A point cloud dimensionality reduction submodule is used to perform coordinate system conversion on the laser point cloud corresponding to the termination time stamp according to preset calibration external parameters and the posture change value to obtain a dimensionality-reduced point cloud; The depth map generation submodule is used to generate a point cloud depth map according to the specification parameters corresponding to the laser radar and the three-dimensional coordinates corresponding to each of the reduced-dimensional point clouds.

11. An electronic device, characterized in that: It includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the high-precision map detection method as described in any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the high-precision map detection method as described in any one of claims 1-9 is implemented.

Citation Information

Patent Citations

  • High-precision map generate method, device and storage medium

    CN109064506A

  • Map construction method and device, electronic device and computer readable storage medium

    CN113865580A

  • Three-dimensional target detection method based on point cloud and image data fusion

    CN114092780A

  • High definition map positioning method and system, platform and computer-readable storage medium

    WO2021016803A1