3D Reconstruction Method, Device and Computer Equipment Based on Point Cloud

Abstract

A 3D reconstruction method based on point cloud, comprising: acquiring point cloud data and vehicle trajectory data; generating point cloud map data according to the point cloud data and the vehicle trajectory data; extracting geometric information and semantic information from the point cloud map data; determining a parameterization strategy corresponding to the corresponding point cloud map data according to the semantic information; performing parameterization processing on the corresponding point cloud map data according to the determined parameterization strategy and the geometric information; and performing 3D reconstruction according to the parameterized point cloud map data.

Description

Technical Field

[0001] The present application relates to a three-dimensional reconstruction method, device, computer device and storage medium based on point cloud. Background Art

[0002] Three-dimensional reconstruction refers to reconstructing the surface of an object based on the geometric information extracted from the object. During the process of autonomous driving, the traditional method is to obtain the point cloud data collected by in-vehicle sensors and extract the geometric information in the point cloud data for three-dimensional reconstruction, which can fully reproduce the detailed information of the vehicle's surrounding environment.

[0003] However, without affecting the driving safety and normal driving of the vehicle, it is not necessary to fully reproduce the detailed information of the vehicle's surrounding environment. If the traditional method is used for three-dimensional reconstruction, it will lead to a large amount of data for three-dimensional reconstruction, resulting in low efficiency of three-dimensional reconstruction. Summary of the Invention

[0004] According to various embodiments disclosed in the present application, a three-dimensional reconstruction method, device, computer device and storage medium based on point cloud are provided.

[0005] A three-dimensional reconstruction method based on point cloud includes:

[0006] Obtaining point cloud data and vehicle trajectory data;

[0007] Generating point cloud map data according to the point cloud data and the vehicle trajectory data;

[0008] Extracting geometric information and semantic information from the point cloud map data;

[0009] Determining a parameterization strategy corresponding to the corresponding point cloud map data according to the semantic information;

[0010] Performing parameterization processing on the corresponding point cloud map data according to the determined parameterization strategy and the geometric information; and

[0011] Performing three-dimensional reconstruction on the parameterized point cloud map data.

[0012] A three-dimensional reconstruction device based on point cloud includes:

[0013] An obtaining module, configured to obtain point cloud data and vehicle trajectory data;

[0014] A generating module, configured to generate point cloud map data according to the point cloud data and the vehicle trajectory data;

[0015] An extracting module, configured to extract geometric information and semantic information from the point cloud map data;

[0016] A determination module, configured to determine a parameterization strategy corresponding to the corresponding point cloud map data according to the semantic information;

[0017] A parameterization module, configured to perform parameterization processing on the corresponding point cloud map data according to the determined parameterization strategy and the geometric information; and

[0018] A three-dimensional reconstruction module, configured to perform three-dimensional reconstruction on the point cloud map data after parameterization processing.

[0019] A computer device, including a memory and one or more processors, wherein computer-readable instructions are stored in the memory, and when the computer-readable instructions are executed by the processors, the one or more processors perform the following steps:

[0020] A computer device, including a memory and one or more processors, wherein computer-readable instructions are stored in the memory, and when the computer-readable instructions are executed by the one or more processors, the one or more processors perform the steps in the above-mentioned method embodiments.

[0021] One or more non-volatile computer-readable storage media storing computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the one or more processors perform the steps in the above-mentioned method embodiments.

[0022] Details of one or more embodiments of the present application are set forth in the following drawings and description. Other features and advantages of the present application will become apparent from the specification, drawings, and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0024] Figure 1 It is an application scenario diagram of a three-dimensional reconstruction method based on point cloud in one or more embodiments.

[0025] Figure 2 It is a schematic flowchart of three-dimensional reconstruction based on point cloud in one or more embodiments.

[0026] Figure 3 It is a schematic flowchart of the step of performing parameterization processing on the corresponding point cloud map data according to the determined parameterization strategy and geometric information in one or more embodiments.

[0027] Figure 4Schematic flowchart of the step of generating point cloud map data according to vehicle trajectory data and point cloud data in one or more embodiments.

[0028] Figure 5 Schematic diagram of calculating trajectory data corresponding to point cloud data of a corresponding frame in one or more embodiments.

[0029] Figure 6 Schematic flowchart of 3D reconstruction based on point cloud in another embodiment.

[0030] Figure 7 Block diagram of a 3D reconstruction device based on point cloud in one or more embodiments.

[0031] Figure 8 Block diagram of a computer device in one or more embodiments. Detailed implementation manners

[0032] In order to make the technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0033] The 3D reconstruction method based on point cloud provided by the present application can be applied to an application scenario of Figure 1 as shown in autonomous driving. In the autonomous driving vehicle, a first vehicle-mounted sensor 102, a second vehicle-mounted sensor 104, and a vehicle-mounted computer device 106 are pre-installed. The first vehicle-mounted sensor can be simply referred to as the first sensor, the second vehicle-mounted sensor can be simply referred to as the second sensor, and the vehicle-mounted computer device can be simply referred to as the computer device. The first sensor 102 communicates with the computer device 106, and the second sensor 104 communicates with the computer device 106. During the autonomous driving process, the first sensor 104 transmits the collected point cloud data to the computer device 104. The second sensor 102 transmits the collected vehicle trajectory data to the computer device 104. The computer device 104 generates point cloud map data according to the vehicle trajectory data and the point cloud data, and extracts geometric information and semantic information from the point cloud map data. Thus, the computer device 104 determines a parameterization strategy corresponding to the corresponding point cloud map data according to the semantic information. The computer device 104 performs parameterization processing on the corresponding point cloud map data according to the determined parameterization strategy and geometric information. The computer device 104 performs 3D reconstruction on the data after parameterization processing. The first sensor 104 can be but is not limited to a lidar, a laser scanner, etc. The second sensor 102 can be but is not limited to an RTK (Real-time kinematic) sensor, an IMU (Inertial measurement unit) sensor, a wheel speed meter, etc.

[0034] In one embodiment, as Figure 2 shown, a three-dimensional reconstruction method based on point cloud is provided. Taking the computer device in Figure 1 as an example for illustration, the method includes the following steps:

[0035] Step 202, obtain point cloud data and vehicle trajectory data.

[0036] During the autonomous driving process of the vehicle, the surrounding environment can be scanned by the first sensor installed on the vehicle to obtain corresponding point cloud data. The first sensor transmits the collected point cloud data to the computer device. The second sensor transmits the collected vehicle trajectory data to the computer device.

[0037] The first sensor can be a lidar, a laser scanner, etc. The point cloud data is the point cloud data collected by the first sensor within the visible range. The point cloud data records the objects within the visible range in the form of points, which is a set of point data corresponding to multiple points on the object surface. Multiple can refer to two or more than two. The point cloud data can be three-dimensional point cloud data, and each frame of point cloud data can include point data corresponding to multiple points. The point data can specifically include at least one of the three-dimensional coordinates corresponding to the point, the laser reflection intensity, and the color information, etc. The three-dimensional coordinates can be the coordinates of the point in the Cartesian coordinate system, specifically including the horizontal axis coordinate, the vertical axis coordinate, and the vertical axis coordinate of the point in the Cartesian coordinate system. The Cartesian coordinate system is a three-dimensional space coordinate system established with the first sensor as the origin, and the three-dimensional space coordinate system includes a horizontal axis (x-axis), a vertical axis (y-axis), and a vertical axis (z-axis). The three-dimensional space coordinate system established with the first sensor as the origin satisfies the right-hand rule.

[0038] The second sensor can be an RTK (Real-time kinematic) sensor, an IMU (Inertial measurement unit) sensor, a wheel speed meter, etc. The vehicle trajectory data can include the trajectory points of the vehicle within a preset time period and the position information of the trajectory points. The trajectory points can be arranged in chronological order. The position information of the trajectory points can specifically include the longitude of the vehicle at each trajectory point, the latitude of the vehicle at each trajectory point, the time when the vehicle arrives at each trajectory point, the speed of the vehicle at each trajectory point, the position coordinates of the vehicle at each trajectory point, etc.

[0039] In one embodiment, during the autonomous driving process, in areas with good satellite signals, the vehicle trajectory data transmitted by the second sensor can be directly used. However, in areas with poor satellite signals, the second sensor cannot collect the corresponding vehicle trajectory data. Therefore, the computer device can optimize the vehicle trajectory data transmitted by the second sensor to predict the vehicle trajectory data corresponding to areas with poor satellite signals, such as under overpasses, in underground garages, and under high-rise buildings, so as to obtain more accurate vehicle trajectory data. For example, the optimization methods can include Kalman filtering, factor graph optimization, etc.

[0040] Step 204: Generate point cloud map data based on the point cloud data and the vehicle trajectory data.

[0041] Since there may be objects in the point cloud data that affect the generation of the point cloud map data, an object refers to a living or non-living entity in the vehicle's surrounding environment. The object can be a dynamic object or a static object. For example, the objects that affect the generation of the point cloud map data can be dynamic objects such as vehicles or pedestrians driving on the road and static objects such as temporarily parked vehicles and trash cans. Therefore, the computer device can remove the objects in the point cloud data that affect the generation of the point cloud map data. Specifically, the computer device can perform object detection on the point cloud data to determine the positions of the objects to be removed in the point cloud data, and then remove the corresponding objects according to the detected object positions. The computer device can use deep learning object detection methods for object removal. For example, Spatial Pyramid Pooling Net (SPPNet for short), Feature Pyramid Networks (FPN for short), etc.

[0042] The computer device matches the point cloud data after object removal processing with the vehicle trajectory data. The point cloud data can include multiple frames of point cloud data. Each frame of point cloud data has a corresponding timestamp, from which the time sequence between multiple frames of point cloud data can be determined. Specifically, the computer device can determine the trajectory data corresponding to each frame of point cloud data in the point cloud data based on the timestamp of the point cloud data in the vehicle trajectory data. The computer device accumulates the trajectory data corresponding to multiple frames of point cloud data in chronological order to generate point cloud map data.

[0043] Step 206: Extract geometric information and semantic information from the point cloud map data.

[0044] Geometric information is information used to represent the surface characteristics of each object in the point cloud map data. For example, geometric information can specifically include the position coordinates of an object in three-dimensional space, the attitude information of the object in three-dimensional space, and the bounding box represented by the size of the object. The position coordinates can be represented by (horizontal axis coordinate, vertical axis coordinate, vertical axis coordinate), that is, (x, y, z). The attitude information can be represented by (roll angle, pitch angle, yaw angle), that is, (roll, pitch, yaw). The size can be represented by (length, width, height). The semantic information can include the category information corresponding to each object in the point cloud map data. For example, vehicles, people, trees, etc.

[0045] After the computing device generates the point cloud map data, it extracts the geometric information and semantic information of each object in the point cloud map data. There can be multiple information extraction methods. It can directly calculate the geometric information and semantic information of each object in the point cloud map data, or it can convert the point cloud map data into raster map data, so as to extract the geometric information and semantic information of each object in the raster map data.

[0046] When the computer device directly calculates the geometric information and semantic information of each object in the point cloud map data, the computing device can first perform voxelization processing on the point cloud map data to obtain a feature matrix. Voxelization processing refers to converting the point cloud map data into a volume grid. The volume grid can be represented by a feature matrix. The feature matrix can include multiple matrix units, and one matrix unit represents a volume grid. The size of the matrix unit can be length * width * height. The length, width, and height of each matrix unit can be the same. The computer device calls a pre-trained deep learning model. For example, the deep learning model can be a three-dimensional convolutional neural network model. The computer device inputs the feature matrix into the deep learning model, and through the deep learning model, performs prediction operations on the feature matrix, and outputs the geometric information and semantic information corresponding to each object in the point cloud map data.

[0047] The point cloud map data generated by the computer device is three-dimensional data. When the computer device uses the method of converting the point cloud map data into raster map data to extract the geometric information and semantic information of each object, the computer device can project the point cloud map data to obtain a two-dimensional raster map. The two-dimensional raster map can include multiple grids. The size of the grid can be length * width. The length and width of each grid can be different. Furthermore, the computer device extracts the geometric information and semantic information corresponding to each object in the two-dimensional raster map.

[0048] Step 208, determine the corresponding parameterization strategy according to the semantic information for the corresponding point cloud map data.

[0049] The semantic information may include the category information corresponding to each object in the point cloud map data. The computer device determines the corresponding parameterization strategy for the object according to the category information corresponding to each object. The parameterization strategy may parameterize the object using parameters such as points, lines, planes, and solids. For example, the computer device may represent a road that is uneven but does not affect driving as a plane, may regard fences and walls as cuboids, and may represent railings as long cylindrical bodies. The computer device may determine the parameters corresponding to the object according to the category information of the object. The parameters corresponding to the same object may be the same type of parameter or a combination of multiple parameters. The parameterization strategy can be used to determine the parameters that replace the corresponding objects in the point cloud map data. By replacing the corresponding objects with parameters, the point cloud map data corresponding to the object can be obtained, and unnecessary point cloud map data can be hidden, thus reducing the data volume for subsequent 3D reconstruction.

[0050] The parameterization strategy can be selected according to the accuracy requirements corresponding to the vehicle's environment, and specifically may include a low-precision parameterization strategy and a high-precision parameterization strategy. The low-precision parameterization strategy parameterizes the point cloud map data with fewer parameters, while the high-precision parameterization strategy parameterizes the point cloud map data with more parameters. When the accuracy requirements corresponding to the vehicle's environment are low without affecting vehicle driving safety and normal driving, the low-precision parameterization strategy can be adopted. For example, the computer device may regard a traffic light as composed of a light box and a pole. The light box can be represented by a cuboid, and the pole can be represented by a cylinder. When the accuracy requirements corresponding to the vehicle's environment are high, the high-precision parameterization strategy can be adopted. For example, the computer device uses a CAD (Computer Aided Design) model to parameterize the traffic light body.

[0051] Step 210, parameterize the corresponding point cloud map data according to the determined parameterization strategy and geometric information.

[0052] Step 212, perform 3D reconstruction on the parameterized point cloud map data.

[0053] After the computer device determines the parameterization strategy corresponding to each object, it can obtain the parameters corresponding to each object. Thus, the computer device determines information such as the size and position of the object according to the geometric information of the object, and determines parameter information such as the size and position of each parameter according to the shape and size of the object. The computer device performs parameterization processing on the corresponding object according to the parameters, that is, performs modeling processing on the corresponding object according to the parameters, and then obtains the parameter model corresponding to the point cloud map data.

[0054] When an object corresponds to only one parameter, the computer device performs parametric processing on the corresponding object according to the parameter. When an object corresponds to multiple parameters, the computer device performs parametric processing on the corresponding components of the object for each parameter, so as to realize the combined modeling of multiple parameters for the corresponding object and obtain the parameter model corresponding to the object. After all the objects in the point cloud map data are completely parametrically processed, three-dimensional reconstruction is performed on the parametrically processed point cloud map data, and the parametrically processed data is displayed in the point cloud map corresponding to the point cloud map data to obtain a three-dimensional environmental map.

[0055] In one embodiment, during the process of three-dimensional construction based on point cloud data, the computer device can also obtain partial environmental data in the vehicle's surrounding environment from the database. For example, obtain road information, building data, etc. around the vehicle from the database. For this part of the data, the computer device does not need to collect the corresponding point cloud data through the first sensor. It can further reduce the amount of data for three-dimensional reconstruction.

[0056] In this embodiment, the computer device generates point cloud map data by obtaining point cloud data and vehicle trajectory data. Geometric information and semantic information are extracted from the point cloud map data. Compared with the method of three-dimensional reconstruction that completely relies on geometric information, semantic information can more accurately express the meaning of objects in the map. At the same time, it effectively avoids the problem of inaccurate three-dimensional reconstruction caused by noise or data loss. Without affecting the safety of vehicle driving and normal driving, the computer device can determine the parametric strategy corresponding to the corresponding point cloud map data according to the semantic information, and can select the corresponding parametric strategy according to different categories corresponding to the point cloud map data. Thus, parametric processing is performed on the corresponding point cloud map data according to the determined parametric strategy and geometric information, realizing replacing the corresponding objects in the point cloud map data with parameters, and being able to hide the unnecessary point cloud map data, reducing the amount of data of the point cloud map data. Furthermore, the computer device performs three-dimensional reconstruction on the parametrically processed point cloud map data. Since the parametrically processed point cloud map data is the data with reduced amount of data, the amount of data for three-dimensional reconstruction is reduced, and thus the efficiency of three-dimensional reconstruction is effectively improved.

[0057] In one of the embodiments, as Figure 3 shown, the above method further includes the step of performing parametric processing on the corresponding point cloud map data according to the determined parametric strategy and geometric information, specifically including:

[0058] Step 302, obtaining the model parameters corresponding to the corresponding point cloud map data from the historical parameters according to the parametric strategy.

[0059] Step 304, performing parametric processing on the corresponding point cloud map data according to the geometric information and the determined model parameters.

[0060] The computer device determines the parameterization strategy corresponding to the object according to the category information corresponding to each object. The parameterization strategy can parameterize the object using parameters such as points, lines, surfaces, and volumes. The parameters corresponding to the same object can be the same type of parameter or a combination of multiple parameters. The parameterization strategy can be used to determine the parameters that replace the corresponding objects in the point cloud map data, and the computer device can obtain the model parameters corresponding to the corresponding point cloud map data according to the parameters corresponding to the object in the historical parameters. The historical parameters can include the parameter information corresponding to the objects reconstructed in three dimensions. The model parameters can include the shape, structure information, etc. of the object.

[0061] Geometric information is information used to represent the surface characteristics of each object in the point cloud map data. For example, the geometric information can specifically include the position coordinates of the object in three-dimensional space, the attitude information of the object in three-dimensional space, and the bounding box represented by the size of the object. The computer device determines information such as the position and size of the object from the geometric information, and then the computer device parameterizes the corresponding point cloud map data according to the model parameters at the position where the object is located.

[0062] In this embodiment, since the model parameters are obtained from the historical parameters, the obtained model parameters better meet the requirements of the current autonomous driving environment. At the same time, for repeatedly occurring object components, the existing parameters can be directly used, further improving the accuracy of three-dimensional reconstruction.

[0063] In one of the embodiments, as Figure 4 shown, the above method further includes the step of generating point cloud map data according to the point cloud data and the vehicle trajectory data, specifically including:

[0064] Step 402, identify the point cloud data corresponding to the preset object in the point cloud data, and delete the point cloud data corresponding to the preset object.

[0065] Step 404, match the point cloud data after the deletion process with the vehicle trajectory data to obtain the trajectory data corresponding to the point cloud data after deletion.

[0066] Step 406, generate point cloud map data according to the trajectory data corresponding to the point cloud data after deletion.

[0067] The computer device identifies the point cloud data in the point cloud data, determines the bounding box corresponding to the preset object, that is, the point cloud data corresponding to the preset object is the point cloud data in the bounding box. The computer device deletes the point cloud data in the bounding box to obtain the point cloud data after the deletion process. Since the point cloud data has no trajectory data and can only calculate the positions corresponding to each frame of point cloud data through timestamps, the computer device matches the point cloud data after the deletion process with the vehicle trajectory data.

[0068] In one embodiment, matching the processed point cloud data after deletion with the vehicle trajectory data to obtain the trajectory data corresponding to the point cloud data after deletion includes: determining target trajectory points in the vehicle trajectory data according to the timestamps of each frame of point cloud data; calculating the trajectory data corresponding to the corresponding frame of point cloud data according to the position coordinates of the target trajectory points, the timestamps of the target trajectory points, and a preset relationship; and obtaining the trajectory data corresponding to the point cloud data after deletion according to the trajectory data corresponding to each frame of point cloud data.

[0069] The point cloud data after deletion includes multiple frames of point cloud data. The computer device can determine target trajectory points in the vehicle trajectory data according to the timestamps of each frame of point cloud data. The target trajectory points can be two trajectory points adjacent to the timestamps of each frame of point cloud data, and each frame of point cloud data can be on the linear function formed by the two trajectory points, that is, the trajectory data corresponding to each frame of point cloud data can be calculated according to the linear relationship among the three. The computer device calculates the trajectory data corresponding to the corresponding frame of point cloud data according to the position coordinates of the target trajectory points, the timestamps of the target trajectory points, and a preset relationship. Specifically, the computer device calculates the abscissa of the trajectory data corresponding to the frame of point cloud data according to the abscissa of the target trajectory point, the timestamp of the target trajectory point, the timestamp of the corresponding frame of point cloud data, and a preset relationship. For example, the preset relationship can be a linear function relationship. Similarly, the computer device calculates the ordinate of the trajectory data corresponding to the frame of point cloud data according to the ordinate of the target trajectory point, the timestamp of the target trajectory point, the timestamp of the corresponding frame of point cloud data, and a preset relationship, so as to obtain the trajectory data corresponding to the frame of point cloud data. The computer device obtains the trajectory data corresponding to multiple frames of point cloud data according to the above calculation method, and further obtains the trajectory data corresponding to the point cloud data after deletion.

[0070] The schematic diagram for calculating the trajectory data corresponding to the corresponding frame of point cloud data can be as Figure 5 shown. In a two-dimensional coordinate system, the circles represent the positions of the target trajectory points, and the triangles represent the positions corresponding to the point clouds. The position coordinates of the target trajectory point with timestamp t1 are (x1, y1), the position coordinates of the target trajectory point with timestamp t3 are (x3, y3), and the position coordinates of the specified frame of point cloud data with timestamp t2 can be represented by (x2, y1). (x1, y1) and (x3, y3) are known, and (x2, y1) is unknown. x2 = x1+(x3 - x1)*t2 / (t3 - t1), y2 = y1+(y3 - y1)*t2 / (t3 - t1). Thus, the position coordinates of the specified frame of point cloud data with timestamp t2 can be obtained.

[0071] In this embodiment, the computer device determines target trajectory points in the vehicle trajectory data according to the timestamps of each frame of point cloud data, calculates the trajectory data corresponding to the corresponding frame of point cloud data according to the position coordinates of the target trajectory points, the timestamps of the target trajectory points, and a preset relationship, so as to obtain the trajectory data corresponding to the point cloud data after deletion based on the trajectory data corresponding to each frame of point cloud data. This can ensure that each frame of point cloud data is matched with the corresponding trajectory data, which is beneficial to improving the accuracy of the point cloud map data.

[0072] After obtaining the trajectory data corresponding to the point cloud data after deletion, the computer device can generate point cloud map data according to the trajectory data corresponding to the point cloud data after deletion. The point cloud map data includes the trajectory data corresponding to each frame of point cloud data. By deleting the point cloud data corresponding to the preset object, redundant point cloud data can be removed, avoiding the influence of redundant data on the point cloud map data, and improving the accuracy and effectiveness of the point cloud map data. Matching the point cloud data after the deletion process with the vehicle trajectory data to obtain the trajectory data corresponding to the point cloud data after deletion, and generating point cloud map data according to the trajectory data corresponding to the point cloud data after deletion. This further improves the accuracy of the point cloud map data, which is beneficial to improving the accuracy of the subsequent 3D model data.

[0073] In another embodiment, as Figure 6 shown, a 3D reconstruction method based on point cloud is provided, including the following steps:

[0074] Step 602, obtain point cloud data and vehicle trajectory data.

[0075] Step 604, generate point cloud map data according to the point cloud data and the vehicle trajectory data.

[0076] Step 606, extract geometric information and semantic information from the point cloud map data.

[0077] Step 608, divide the point cloud map data into regions according to the geometric information and the semantic information to obtain multiple map regions.

[0078] Step 610, determine the parameterization strategy corresponding to each map region according to the semantic information.

[0079] Step 612, perform parameterization processing on the point cloud map data of each map region according to the determined parameterization strategy and the geometric information to obtain the region model data corresponding to each map region.

[0080] Step 614, perform 3D reconstruction on the region model data corresponding to multiple map regions.

[0081] After a computer device extracts geometric information and semantic information from point cloud map data, it divides the point cloud map data into multiple map regions according to the geometric information and semantic information. The geometric information may specifically include the position coordinates of an object in three-dimensional space, the attitude information of the object in three-dimensional space, and the bounding box represented by the size of the object. The semantic information may include the category information corresponding to each object in the point cloud map data. For example, vehicles, people, trees, crosswalks, lane lines, etc.

[0082] Specifically, the computer device determines the region category corresponding to each object according to the category information of each object, and divides the point cloud map data according to the geometric information of each object according to the region category corresponding to each object, obtaining multiple regions. The region categories may be divided according to the impact on driving behavior or according to traffic rules. The region categories divided according to the impact on driving behavior may include a prohibited line-crossing region, a deceleration region, etc. For example, the prohibited line-crossing region may include white solid lines and yellow solid lines, etc., and the deceleration region may include the road region before a crosswalk. The region categories divided according to traffic rules may include a vehicle driving region, a traffic signal sign region, a non-driving region, a crosswalk region, etc. Multiple different types of objects may be included in the same region category. For example, the vehicle drivable region may include lane lines, vehicles, etc. To facilitate the distinction of different region categories, the same region category may be represented by the same color. For example, the crosswalk region is represented by white, the vehicle driving region is represented by green, and the non-driving region is represented by yellow.

[0083] The computer device determines the parameterization strategy corresponding to each region category according to the semantic information. The parameterization strategy corresponding to each region category may be the parameterization method corresponding to the region category. The parameterization method may include a high-precision parameterization method and a low-precision parameterization method. Since the impact degrees of different region categories on driving behavior may be different, and different impact degrees on driving behavior may result in different corresponding parameterization accuracy requirements, the parameterization methods corresponding to region categories may also be different. For example, the vehicle driving region and the traffic signal sign region have a greater impact on driving behavior, and a high-precision parameterization method may be selected. The high-precision parameterization method may be to parameterize a traffic light body through a CAD (Computer Aided Design) model. The non-driving region has a smaller impact on driving behavior, and a low-precision parameterization method may be selected. The low-precision parameterization method may be to parameterize an object with simple parameters such as points, lines, surfaces, and solids. For example, a traffic light may be regarded as composed of a light box and a pole. The light box may be represented as a cuboid, and the pole may be represented as a cylinder.

[0084] The computer device obtains the parameters and parameterization methods corresponding to each object according to the determined parameterization strategy. Thus, the computer device determines information such as the size and position of the object based on the geometric information of each object, and determines parameter information such as the parameter size and position according to information such as the shape and size of the object. The computer device performs parameterization processing on the point cloud map data corresponding to the objects in the corresponding map area according to the parameterization method, that is, performs modeling processing on the objects in the corresponding map area, and then obtains the area model data corresponding to each map area. The computer device can perform parameterization processing on each map area one by one, combine the area model data corresponding to the multiple processed map areas, and then perform three-dimensional reconstruction on the combined area model data.

[0085] In this embodiment, the computer device divides the point cloud map data into multiple map areas according to geometric information and semantic information, and thus determines the parameterization strategy corresponding to each area category according to the semantic information. Then, according to the determined parameterization strategy and geometric information, parameterization processing is performed on the point cloud map data of each map area, the area model data corresponding to the multiple processed map areas is combined, and then three-dimensional reconstruction is performed on the combined area model data. Since the semantic information of each area will affect the driving behavior during the automatic driving process, dividing the point cloud map data into areas can avoid affecting the driving behavior and achieve safe driving. By determining the parameterization strategy corresponding to each map area and performing parameterization processing on the point cloud map data of the same map area using the same parameterization strategy, the parameterization processing efficiency can be improved. At the same time, dividing the point cloud map data into areas can quickly locate the corresponding map area according to the area identifier during the subsequent update of the three-dimensional model data, so as to perform area data update.

[0086] In one of the embodiments, the above method further includes: obtaining updated point cloud data, and determining the area identifier corresponding to the object identifier according to the object identifier corresponding to the updated point cloud data; extracting the original point cloud data corresponding to the object identifier according to the map area corresponding to the area identifier; and replacing the original point cloud data with the updated point cloud data.

[0087] After the computer device divides the point cloud map data into multiple map regions according to geometric information and semantic information, each map region has a corresponding region identifier. The region identifier is the unique identifier used to mark the map region. Since the road environment data around the vehicle has a certain change frequency, the point cloud data of each object corresponding to the map region will also be updated accordingly. When the computer device obtains the update data, it can obtain the corresponding object identifier according to the update data. The update data can be the point cloud data of the object that changes during the autonomous driving process. The update data can be uploaded by the user, or can be obtained by uploading abnormal information when the computer device detects that the road environment data around the vehicle has changed and the vehicle cannot drive normally. Since the point cloud map data is divided into regions, each map region can include the point cloud data corresponding to multiple objects, that is, the point cloud data in the bounding box corresponding to each object. Therefore, the computer device can determine the region identifier of the map region corresponding to the object identifier, extract the point cloud data corresponding to the object identifier according to the map region corresponding to the region identifier, and then replace the point cloud data corresponding to the object identifier according to the update data.

[0088] In this embodiment, since there is a corresponding object identifier for the update data, the point cloud data that needs to be replaced can be quickly located according to the object identifier. The computer device determines the region identifier corresponding to the object identifier, which can further narrow down the search scope of the data. Then, the point cloud data corresponding to the object identifier is extracted according to the map region corresponding to the region identifier, and the point cloud data corresponding to the object identifier is replaced according to the update data, further improving the data update efficiency.

[0089] In one of the embodiments, extracting geometric information and semantic information from the point cloud map data includes: performing voxelization processing on the point cloud map data to obtain a feature matrix; calling a pre-trained deep learning model, inputting the feature matrix into the deep learning model, and performing predictive operations on the feature matrix through the deep learning model to output the geometric information and semantic information corresponding to the feature matrix.

[0090] When the computing resources of a computer device are greater than or equal to a preset threshold, the computer device can perform voxelization processing on the point cloud map data to obtain a feature matrix. Specifically, the computer device calculates the differences between the maximum and minimum values of the point cloud data coordinates in the X, Y, and Z directions. The computer device determines the length, width, and height of the data area based on these three differences. The data area contains all the point cloud data. The computer device can perform voxelization processing on the point cloud data according to a preset size. The preset size can be length * width * height. The computer device divides the data area along the X direction according to the length in the preset size, divides the data area along the Y direction according to the width in the preset size, and divides the data area along the Z direction according to the height in the preset size, thereby obtaining a feature matrix. The length, width, and height of the preset size can be the same. The order of multi-directional division of the data area is not limited.

[0091] For example, the computer device can first divide the data area along the X direction according to the length in the preset size, then divide the data area in the Y direction according to the width in the preset size, and finally divide the data area in the Z direction according to the height in the preset size, thereby obtaining a feature matrix. The computer device can also first divide the data area along the X direction according to the length in the preset size, then divide the data area in the Z direction according to the height in the preset size, and finally divide the data area in the Y direction according to the width in the preset size to obtain a feature matrix.

[0092] The computer device then inputs the feature matrix into a pre-trained deep learning model, performs prediction operations on the feature matrix through the deep learning model, and outputs the geometric information and semantic information corresponding to the feature matrix. For example, the deep learning model can be a three-dimensional convolutional neural network model. The deep learning model can specifically include an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer, etc. The computer device can perform operations corresponding to the network structure on the feature matrix in sequence according to the network structure of the deep learning model, and then obtain the geometric information and semantic information corresponding to the feature matrix output by the deep learning model. The geometric information can specifically include the position coordinates of an object in three-dimensional space, the pose information of the object in three-dimensional space, and the bounding box represented by the size of the object. The semantic information can include the category information corresponding to each object in the point cloud map data. For example, vehicles, people, trees, crosswalks, lane lines, etc.

[0093] In this embodiment, when the computing resources of the computer device are greater than or equal to a preset threshold, the computer device performs voxelization processing on the data area corresponding to the point cloud data, and can classify the point cloud data corresponding to the object when there are situations such as occlusion of the object. The feature matrix obtained after voxelization processing is input into the deep learning model, and the deep learning model performs prediction operations on the feature matrix to output the geometric information and semantic information corresponding to the feature matrix. Since the deep learning model is pre-trained, the information extraction efficiency is improved.

[0094] In one embodiment, extracting geometric information and semantic information from the point cloud map data includes: projecting the point cloud map data onto a preset perspective to obtain raster map data; and extracting features from the raster map data to obtain the geometric information and semantic information in the raster map data.

[0095] The point cloud map data is three-dimensional point cloud data. The computer device projects the acquired point cloud map data onto a preset perspective, thereby obtaining the raster map data corresponding to the preset perspective and realizing the conversion of the three-dimensional point cloud data into two-dimensional data. For example, the preset perspective can be a bird's-eye view or a front view. When the computer device projects the point cloud map data at a bird's-eye view, the raster map data corresponding to the bird's-eye view can be obtained. When the computer device projects the point cloud map data at a front view, the raster map data corresponding to the front view can be obtained.

[0096] The computer device can perform image recognition on the raster map data to obtain the geometric information and semantic information corresponding to each object in the raster map data. The geometric information can specifically include the position coordinates of the object in three-dimensional space, the attitude information of the object in three-dimensional space, and the bounding box represented by the size of the object. The semantic information can include the category information corresponding to each object in the point cloud map data. For example, vehicles, people, trees, crosswalks, lane lines, etc.

[0097] Further, the computer device can also perform multi-perspective projection on the point cloud map data, thereby obtaining the raster map data corresponding to each perspective. The multi-perspectives can include a bird's-eye view and a front view. The computer device performs image recognition on the raster map data corresponding to each perspective to obtain the geometric information and semantic information of the raster map data corresponding to each perspective. Furthermore, the computer device performs fusion processing on the geometric information and semantic information of the raster map data corresponding to the multi-perspectives, realizes the combination of the geometric information and semantic information of the same object under different perspectives, and can obtain more accurate geometric information and semantic information.

[0098] In one embodiment, the image recognition method may also be to call a pre-established neural network model, input the raster map data into the neural network model for calculation, and output the geometric information and semantic information corresponding to the raster map data. For example, the neural network model may be a two-dimensional convolutional neural network model. Since the deep learning model is pre-trained, the information extraction efficiency is improved.

[0099] In this embodiment, the computer device projects the point cloud map data onto a preset perspective to obtain raster map data, extracts features from the raster map data to obtain the geometric information and semantic information in the raster map data. Since the raster map data is two-dimensional data and the amount of data to be processed is small, accurate geometric information and semantic information can be obtained in the autonomous driving mode when the computing resources are limited and the real-time requirement is high.

[0100] In one embodiment, as Figure 7 shown, a three-dimensional reconstruction device based on point cloud is provided, including: an acquisition module 702, a generation module 604, an extraction module 706, a determination module 708, a parameterization module 710, and a three-dimensional reconstruction module 712, where:

[0101] The acquisition module 702 is configured to acquire point cloud data and vehicle trajectory data.

[0102] The generation module 704 is configured to generate point cloud map data according to the point cloud data and the vehicle trajectory data.

[0103] The extraction module 706 is configured to extract geometric information and semantic information from the point cloud map data.

[0104] The determination module 708 is configured to determine a parameterization strategy corresponding to the corresponding point cloud map data according to the semantic information.

[0105] The parameterization module 710 is configured to perform parameterization processing on the corresponding point cloud map data according to the determined parameterization strategy and the geometric information.

[0106] The three-dimensional reconstruction module 712 is configured to perform three-dimensional reconstruction on the parameterized point cloud map data.

[0107] In one embodiment, the parameterization module 710 is further configured to obtain model parameters corresponding to the corresponding point cloud map data from historical parameters according to the parameterization strategy; perform parameterization processing on the corresponding point cloud map data according to the geometric information and the determined model parameters.

[0108] In one embodiment, the generation module 704 is configured to identify the point cloud data corresponding to a preset object in the point cloud data, and delete the point cloud data corresponding to the preset object; match the point cloud data after the deletion process with the vehicle trajectory data to obtain the trajectory data corresponding to the point cloud data after the deletion; generate point cloud map data according to the trajectory data corresponding to the point cloud data after the deletion.

[0109] In one embodiment, the generation module 704 is further configured to determine target trajectory points in the vehicle trajectory data according to the timestamps of each frame of point cloud data; calculate the trajectory data corresponding to the corresponding frame of point cloud data according to the position coordinates of the target trajectory points, the timestamps of the target trajectory points, and a preset relationship; obtain the trajectory data corresponding to the point cloud data after the deletion according to the trajectory data corresponding to each frame of point cloud data.

[0110] In one embodiment, the above device further includes: a partitioning module, configured to partition the point cloud map data according to geometric information and semantic information to obtain a plurality of map regions; determine a parameterization strategy corresponding to each map region according to the semantic information; perform parameterization processing on the point cloud map data of each map region according to the determined parameterization strategy and geometric information to obtain region model data corresponding to each map region; perform three-dimensional reconstruction on the region model data corresponding to the plurality of map regions.

[0111] In one embodiment, the above device further includes: an update module, configured to obtain update data, determine a region identifier corresponding to the object identifier according to the object identifier corresponding to the update data; extract the point cloud data corresponding to the object identifier according to the map region corresponding to the region identifier; replace the point cloud data corresponding to the object identifier according to the update data.

[0112] In one embodiment, the extraction module 706 is further configured to perform voxelization processing on the point cloud map data to obtain a feature matrix; call a pre-trained deep learning model, input the feature matrix into the deep learning model, and perform prediction operations on the feature matrix through the deep learning model to output the geometric information and semantic information corresponding to the feature matrix.

[0113] In one embodiment, the extraction module 706 is further configured to project the point cloud map data onto a preset perspective to obtain raster map data; perform feature extraction on the raster map data to obtain the geometric information and semantic information in the raster map data.

[0114] For the specific limitations of the 3D reconstruction device based on point cloud, reference can be made to the limitations of the 3D reconstruction method based on point cloud in the above text, which will not be elaborated here. Each module in the above 3D reconstruction device based on point cloud can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory of the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to each of the above modules.

[0115] In one embodiment, a computer device is provided, and its internal structure diagram can be as Figure 8 shown. The computer device includes a processor, a memory, a communication interface, and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store point cloud data. The communication interface of the computer device is used to connect and communicate with the first sensor and the second sensor. When the computer program is executed by the processor, it implements a 3D reconstruction method based on point cloud.

[0116] Those skilled in the art can understand that Figure 8 the structure shown in

[0117] is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0118] A non-volatile computer-readable storage medium storing one or more computer-readable instructions, when the computer-readable instructions are executed by one or more processors, cause the one or more processors to execute the steps in each of the above method embodiments.

[0119] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through computer-readable instructions. The computer-readable instructions can be stored in a non-volatile computer-readable storage medium. When the computer-readable instructions are executed, they can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in this application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0120] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0121] The above-described embodiments merely represent several implementation manners of this application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several modifications and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of the patent of this application should be subject to the appended claims.

Claims

1. A 3D reconstruction method based on point cloud, comprising: Obtaining point cloud data and vehicle trajectory data; Optimizing the vehicle trajectory data to predict the vehicle trajectory data corresponding to areas with poor satellite signals; The point cloud data is 3D point cloud data, and each frame of the point cloud data includes point data corresponding to multiple points, and the point data includes at least one of the 3D coordinates, laser reflection intensity, and color information corresponding to the point; Generating point cloud map data according to the point cloud data and the vehicle trajectory data; Extracting geometric information and semantic information from the point cloud map data, including: performing voxelization processing on the point cloud map data to obtain a feature matrix; and calling a pre-trained deep learning model, inputting the feature matrix into the deep learning model, and performing prediction operations on the feature matrix through the deep learning model to output the geometric information and semantic information corresponding to the feature matrix; Determining a parameterization strategy corresponding to the corresponding point cloud map data according to the semantic information; the parameterization strategy includes a low-precision parameterization strategy and a high-precision parameterization strategy; selecting a low-precision parameterization strategy or a high-precision parameterization strategy according to the accuracy requirement corresponding to the vehicle's environment; Performing parameterization processing on the corresponding point cloud map data according to the determined parameterization strategy and the geometric information; and Performing 3D reconstruction on the parameterized point cloud map data; The method further includes: dividing the point cloud map data into multiple map regions according to the geometric information and the semantic information; determining a parameterization strategy corresponding to each map region according to the semantic information; performing parameterization processing on the point cloud map data of each map region according to the determined parameterization strategy and the geometric information to obtain region model data corresponding to each map region; and performing 3D reconstruction on the region model data corresponding to multiple map regions; wherein, obtaining updated data, determining a region identifier corresponding to the object identifier according to the object identifier corresponding to the updated data; extracting the point cloud data corresponding to the object identifier according to the map region corresponding to the region identifier; and replacing the point cloud data corresponding to the object identifier according to the updated data.

2. The method according to claim 1, wherein The performing parameterization processing on the corresponding point cloud map data according to the determined parameterization strategy and the geometric information includes: Obtaining the model parameters corresponding to the corresponding point cloud map data from historical parameters according to the parameterization strategy; and Performing parameterization processing on the corresponding point cloud map data according to the geometric information and the determined model parameters.

3. The method according to claim 1, wherein The generating point cloud map data according to the point cloud data and the vehicle trajectory data includes: Identifying the point cloud data corresponding to a preset object in the point cloud data and deleting the point cloud data corresponding to the preset object; Matching the point cloud data after the deletion process with the vehicle trajectory data to obtain the trajectory data corresponding to the point cloud data after the deletion; and Generating point cloud map data according to the trajectory data corresponding to the point cloud data after the deletion.

4. The method according to claim 3, wherein The point cloud data after deletion includes multiple frames of point cloud data. Matching the point cloud data after deletion processing with the vehicle trajectory data to obtain the trajectory data corresponding to the point cloud data after deletion includes: Determining target trajectory points in the vehicle trajectory data according to the timestamps of each frame of point cloud data; Calculating the trajectory data corresponding to the corresponding frame of point cloud data according to the position coordinates of the target trajectory points, the timestamps of the target trajectory points, and a preset relationship; and Obtaining the trajectory data corresponding to the point cloud data after deletion according to the trajectory data corresponding to each frame of point cloud data.

5. The method according to any one of claims 1 to 4, characterized in that, Extracting geometric information and semantic information from the point cloud map data includes: Projecting the point cloud map data to a preset perspective to obtain raster map data; and Performing feature extraction on the raster map data to obtain geometric information and semantic information in the raster map data.

6. A three-dimensional reconstruction device based on point cloud, comprising: An acquisition module for acquiring point cloud data and vehicle trajectory data; Optimizing the vehicle trajectory data to predict the vehicle trajectory data corresponding to areas with poor satellite signals; the point cloud data is three-dimensional point cloud data, and each frame of the point cloud data includes point data corresponding to multiple points, and the point data includes at least one of the three-dimensional coordinates corresponding to the point, the laser reflection intensity, and color information; A generation module for generating point cloud map data according to the point cloud data and the vehicle trajectory data; An extraction module for extracting geometric information and semantic information from the point cloud map data, including: performing voxelization processing on the point cloud map data to obtain a feature matrix; and calling a pre-trained deep learning model, inputting the feature matrix into the deep learning model, and performing prediction operations on the feature matrix through the deep learning model to output the geometric information and semantic information corresponding to the feature matrix; A determination module for determining a parameterization strategy corresponding to the corresponding point cloud map data according to the semantic information; the parameterization strategy includes a low-precision parameterization strategy and a high-precision parameterization strategy; selecting a low-precision parameterization strategy or a high-precision parameterization strategy according to the accuracy requirements corresponding to the vehicle's environment; A parameterization module for parameterizing the corresponding point cloud map data according to the determined parameterization strategy and the geometric information; and A three-dimensional reconstruction module for performing three-dimensional reconstruction on the point cloud map data after parameterization processing; The device further includes: A division module for dividing the point cloud map data into multiple map regions according to the geometric information and the semantic information; determining the parameterization strategy corresponding to each map region according to the semantic information; parameterizing the point cloud map data of each map region according to the determined parameterization strategy and the geometric information to obtain region model data corresponding to each map region; and performing three-dimensional reconstruction on the region model data corresponding to multiple map regions; An update module, which acquires update data, determines a region identifier corresponding to the object identifier according to the object identifier corresponding to the update data; extracts point cloud data corresponding to the object identifier according to the map region corresponding to the region identifier; and replaces the point cloud data corresponding to the object identifier according to the update data.

7. The device according to claim 6, characterized in that, The parameterization module is further configured to obtain model parameters corresponding to the corresponding point cloud map data from historical parameters according to the parameterization strategy; and perform parameterization processing on the corresponding point cloud map data according to the geometric information and the determined model parameters.

8. A computer device, including a memory and one or more processors, wherein computer-readable instructions are stored in the memory, and when the computer-readable instructions are executed by the one or more processors, the one or more processors perform the following steps: Acquire point cloud data and vehicle trajectory data; optimize the vehicle trajectory data to predict vehicle trajectory data corresponding to regions with poor satellite signals; the point cloud data is three-dimensional point cloud data, and each frame of the point cloud data includes point data corresponding to a plurality of points, and the point data includes at least one of three-dimensional coordinates corresponding to the point, laser reflection intensity, and color information. Generate point cloud map data according to the point cloud data and the vehicle trajectory data. Extract geometric information and semantic information from the point cloud map data, including: performing voxelization processing on the point cloud map data to obtain a feature matrix; and invoking a pre-trained deep learning model, inputting the feature matrix into the deep learning model, and performing prediction operations on the feature matrix through the deep learning model to output geometric information and semantic information corresponding to the feature matrix. Determine a parameterization strategy corresponding to the corresponding point cloud map data according to the semantic information; the parameterization strategy includes a low-precision parameterization strategy and a high-precision parameterization strategy; select a low-precision parameterization strategy or a high-precision parameterization strategy according to the precision requirement corresponding to the environment where the vehicle is located. Perform parameterization processing on the corresponding point cloud map data according to the determined parameterization strategy and the geometric information; and Perform three-dimensional reconstruction on the parameterized point cloud map data. When the computer-readable instructions are executed, the following steps are further performed: divide the point cloud map data into multiple map regions according to the geometric information and the semantic information; determine a parameterization strategy corresponding to each map region according to the semantic information; perform parameterization processing on the point cloud map data of each map region according to the determined parameterization strategy and the geometric information to obtain region model data corresponding to each map region; and perform three-dimensional reconstruction on the region model data corresponding to multiple map regions; wherein, acquire update data, determine a region identifier corresponding to the object identifier according to the object identifier corresponding to the update data; extract point cloud data corresponding to the object identifier according to the map region corresponding to the region identifier; and replace the point cloud data corresponding to the object identifier according to the update data.

9. The computer device according to claim 8, wherein When the processor executes the computer-readable instructions, the following steps are further performed: Obtain the model parameters corresponding to the corresponding point cloud map data from the historical parameters according to the parameterization strategy; and Perform parameterization processing on the corresponding point cloud map data according to the geometric information and the determined model parameters.

10. The computer device according to claim 8, characterized in that, When the processor executes the computer-readable instructions, the following steps are also performed: Identify the point cloud data corresponding to the preset object in the point cloud data, and delete the point cloud data corresponding to the preset object; Match the point cloud data after the deletion process with the vehicle trajectory data to obtain the trajectory data corresponding to the point cloud data after the deletion; and Generate point cloud map data according to the trajectory data corresponding to the point cloud data after the deletion.

11. The computer device according to claim 10, wherein When the processor executes the computer-readable instructions, the following steps are also performed: Determine the target trajectory points in the vehicle trajectory data according to the timestamps of each frame of point cloud data; Calculate the trajectory data corresponding to the corresponding frame of point cloud data according to the position coordinates of the target trajectory points, the timestamps of the target trajectory points, and the preset relationship; and Obtain the trajectory data corresponding to the point cloud data after the deletion according to the trajectory data corresponding to each frame of point cloud data.

12. One or more non-volatile computer-readable storage media storing computer-readable instructions, when the computer-readable instructions are executed by one or more processors, cause the one or more processors to perform the following steps: Obtain point cloud data and vehicle trajectory data; optimize the vehicle trajectory data to predict the vehicle trajectory data corresponding to the area with poor satellite signals; the point cloud data is three-dimensional point cloud data, and each frame of the point cloud data includes point data corresponding to multiple points, and the point data includes at least one of the three-dimensional coordinates corresponding to the point, the laser reflection intensity, and the color information; Generate point cloud map data according to the point cloud data and the vehicle trajectory data; Extract geometric information and semantic information from the point cloud map data, including: performing voxelization processing on the point cloud map data to obtain a feature matrix; and calling a pre-trained deep learning model, inputting the feature matrix into the deep learning model, and performing predictive operations on the feature matrix through the deep learning model to output the geometric information and semantic information corresponding to the feature matrix; Determine the parameterization strategy corresponding to the corresponding point cloud map data according to the semantic information; the parameterization strategy includes a low-precision parameterization strategy and a high-precision parameterization strategy; select a low-precision parameterization strategy or a high-precision parameterization strategy according to the accuracy requirements corresponding to the vehicle's environment; Perform parameterization processing on the corresponding point cloud map data according to the determined parameterization strategy and the geometric information; and Perform three-dimensional reconstruction on the point cloud map data after the parameterization processing. When the computer-readable instructions are executed by the processor, the following steps are further performed: dividing the point cloud map data into multiple map regions according to the geometric information and the semantic information; determining a parameterization strategy corresponding to each map region according to the semantic information; performing parameterization processing on the point cloud map data of each map region according to the determined parameterization strategy and the geometric information to obtain region model data corresponding to each map region; and performing three-dimensional reconstruction on the region model data corresponding to the multiple map regions; wherein, obtaining updated data, determining a region identifier corresponding to the object identifier according to the object identifier corresponding to the updated data; extracting the point cloud data corresponding to the object identifier according to the map region corresponding to the region identifier; and replacing the point cloud data corresponding to the object identifier according to the updated data.

13. The storage medium according to claim 12, wherein When the computer-readable instructions are executed by the processor, the following steps are further performed: obtaining model parameters corresponding to the corresponding point cloud map data from historical parameters according to the parameterization strategy; and performing parameterization processing on the corresponding point cloud map data according to the geometric information and the determined model parameters.

14. The storage medium according to claim 12, characterized in that, When the computer-readable instructions are executed by the processor, the following steps are further performed: identifying the point cloud data corresponding to a preset object in the point cloud data and deleting the point cloud data corresponding to the preset object; matching the point cloud data after the deletion process with the vehicle trajectory data to obtain the trajectory data corresponding to the point cloud data after the deletion; and and generating point cloud map data according to the trajectory data corresponding to the point cloud data after the deletion.

15. The storage medium according to claim 14, characterized in that, When the computer-readable instructions are executed by the processor, the following steps are further performed: determining a target trajectory point in the vehicle trajectory data according to the time stamp of each frame of point cloud data; calculating the trajectory data corresponding to the corresponding frame of point cloud data according to the position coordinates of the target trajectory point, the time stamp of the target trajectory point, and a preset relationship; and and obtaining the trajectory data corresponding to the point cloud data after the deletion according to the trajectory data corresponding to each frame of point cloud data.

Images