Ground target voxel camera visibility optimization method and device and storage medium

Abstract

The invention discloses a ground target voxel camera visibility optimization method and device and a storage medium, and the optimization method comprises the steps: calculating the occupancy state of each voxel in a 3D space according to LiDAR visibility; according to a point cloud distribution condition, calculating a category label and a point cloud gravity center position of each voxel; generating a virtual point for each pixel in the image; coordinates of each virtual point and the original point of the camera are converted into a self-vehicle coordinate system; under a self-vehicle coordinate system, each virtual point is connected with an original point of a camera to form a ray, voxels through which the ray passes are traversed by using a Ray-Casting algorithm, and each voxel is marked as' observed ',' empty 'or' unobserved 'based on an occupation state, a category label and a point cloud gravity center position of each voxel. According to the method, the number of unobserved ground target voxels can be reduced, and the recognition effect of the 3DOccupancy network on the ground target is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and in particular relates to a method, a device and a storage medium for optimizing the visibility of a ground target voxel camera in a 3D occupancy truth value system. Background Art

[0002] 3D Occupancy technology is a key technology in autonomous driving perception systems. Different from traditional 3D object detection, the 3D Occupancy model is used to predict each voxel in the 3D space. In order to enable the training of the 3D Occupancy model, each voxel in the 3D space needs to be labeled through a truth system, and each voxel is divided into different categories such as pedestrians, motor vehicles, and ground. Camera visibility is a link in the 3D Occupancy truth system, which provides a reference for the final voxel label generation by calculating the camera visibility of each voxel in the ego-vehicle coordinate system.

[0003] In the prior art, the camera visibility calculation process of a voxel specifically includes the following steps:

[0004] (1) Based on LiDAR visibility, the occupancy state of each voxel in the 3D space is calculated. If the voxel contains a point cloud target, the voxel is considered occupied.

[0005] (2) Generate a virtual point for each pixel in the image;

[0006] (3) Transform the virtual point and the camera origin coordinates to the vehicle coordinate system;

[0007] (4) In the ego-vehicle coordinate system, for each virtual point, connect the camera origin and the virtual point to form a ray, and then traverse the voxels that the ray passes through according to the Ray-Casting algorithm:

[0008] (4.1) Along each ray, traverse the voxels until the first occupied voxel is encountered, then terminate the traversal and exit the Ray-Casting function. Mark the currently occupied voxel as "observed" and the voxel that the ray passes through before the occupied voxel as "empty";

[0009] (4.2) Voxels that are not traversed by any ray are marked as “unobserved”. “Unobserved” includes voxels that are not traversed by the ray after the function (4.1) identifies the “observed” voxel and exits early.

[0010] (5) The truth system processes the voxels marked as "observed" or "empty" in the next step and ignores the "unobserved" voxels.

[0011] However, due to the particularity of ground targets, point clouds are distributed in a flat shape within the corresponding voxels. The existing technology fails to fully consider this situation, so in the Ray-Casting algorithm, the occupied ground voxels that the ray passes through are inaccurately set as "observed", resulting in "unobserved" for all subsequent ground targets, affecting the generation of 3D Occupancy truth values, and further affecting the network's recognition effect on ground areas. Summary of the invention

[0012] In order to solve the above technical problems, the present invention proposes a method, a device and a storage medium for optimizing the visibility of a ground target voxel camera.

[0013] In order to achieve the above object, the technical solution of the present invention is as follows:

[0014] In a first aspect, the present invention discloses a method for optimizing the visibility of a ground target voxel camera, comprising:

[0015] Step S1: Calculate the occupancy state of each voxel in the 3D space according to the LiDAR visibility. If the voxel contains a point cloud target, the voxel is in an occupied state.

[0016] Step S2: Calculate the category label of each voxel and the point cloud center of gravity position according to the point cloud distribution. The category label is: ground target or non-ground target.

[0017] Step S3: Generate a virtual point for each pixel in the image;

[0018] Step S4: transform the coordinates of each virtual point and the camera origin into the vehicle coordinate system;

[0019] Step S5: In the ego-vehicle coordinate system, each virtual point is connected to the camera origin to form a ray, and the Ray-Casting algorithm is used to traverse the voxels that the ray passes through, and each voxel is marked as "observed", "empty" or "unobserved" based on its occupancy state, category label, and point cloud centroid position;

[0020] Step S6: Perform the next step of processing on the voxels marked as "observed" or "empty", and ignore the "unobserved" voxels.

[0021] Based on the above technical solution, the following improvements can be made:

[0022] As a preferred solution, step S2 includes:

[0023] Step S2.1: According to the coordinates of each point in the point cloud, the voxel corresponding to the point is calculated to obtain the corresponding relationship between each point in the point cloud and each voxel;

[0024] Step S2.2: Obtain the coordinates and categories of all points corresponding to each voxel, and calculate the point cloud centroid position and category label of each voxel.

[0025] As a preferred solution, step S2.2 includes:

[0026] Get the coordinates of all points corresponding to each voxel;

[0027] Calculate the average value of the coordinates of all points of each voxel, and use the average value as the centroid position of the point cloud of the voxel;

[0028] Count the categories of all points of each voxel, and take the category with the largest number of points as the category of the voxel.

[0029] As a preferred solution, step S5 includes:

[0030] Step S5.1: In the ego-vehicle coordinate system, connect each virtual point with the camera origin to form a number of rays;

[0031] Step S5.2: using the Ray-Casting algorithm, along each ray, traverse the voxels that the ray passes through in sequence until the first occupied voxel is encountered, and the occupied voxel is the voxel in the occupied state;

[0032] Step S5.3: Determine the category label of the occupied voxel;

[0033] If the category label of the occupied voxel is a ground target, the plane where the point cloud gravity center of the original occupied voxel is located is taken as the upper plane, and together with the bottom plane of the original occupied voxel, a new voxel is formed.

[0034] Further determine whether the ray passes through the new voxel;

[0035] If the ray passes through a new voxel, the original occupied voxel is marked as "observed", and the voxel that the ray passed before the original occupied voxel is marked as "empty", the traversal is terminated, and the Ray-Casting function is exited;

[0036] If the new voxel is not traversed, the original occupied voxel is marked as "empty", and the traversal continues to find the next occupied voxel, and step S5.3 is repeated;

[0037] If the category label of the occupied voxel is a non-ground target, the original occupied voxel is marked as "observed", and the voxel that the ray passes through before the original occupied voxel is marked as "empty", the traversal is terminated, and the Ray-Casting function is exited.

[0038] In a second aspect, the present invention discloses a device for optimizing the visibility of a ground target voxel camera, comprising:

[0039] The occupancy state calculation module is used to calculate the occupancy state of each voxel in the 3D space according to the LiDAR visibility. If the voxel contains a point cloud target, the voxel is occupied.

[0040] Point cloud computing module, used to calculate the category label of each voxel and the point cloud center of gravity position according to the point cloud distribution. The category label is: ground target or non-ground target;

[0041] A virtual point generation module, used to generate a virtual point for each pixel in the image;

[0042] A coordinate transformation module is used to transform the coordinates of each virtual point and the camera origin into the ego-vehicle coordinate system;

[0043] The marking module is used to connect each virtual point with the camera origin in the ego-vehicle coordinate system to form a ray. The Ray-Casting algorithm is used to traverse the voxels that the ray passes through, and mark each voxel as "observed", "empty" or "unobserved" based on its occupancy state, category label, and point cloud center of gravity position.

[0044] The processing module is used to perform the next step of processing on the voxels marked as "observed" and "empty", and ignore the "unobserved" voxels.

[0045] As a preferred solution, the point cloud computing module includes:

[0046] A first calculation unit, used to calculate the voxel corresponding to each point in the point cloud according to the coordinates of the point, and obtain the corresponding relationship between each point in the point cloud and each voxel;

[0047] The second calculation unit is used to obtain the coordinates and categories of all points corresponding to each voxel, and calculate the point cloud centroid position and category label of each voxel.

[0048] As a preferred solution, the second computing unit includes:

[0049] An obtaining unit, used to obtain the coordinates of all points corresponding to each voxel;

[0050] A point cloud centroid position calculation unit is used to calculate the average value of the coordinates of all points of each voxel, and use the average value as the point cloud centroid position of the voxel;

[0051] The voxel category calculation unit is used to count the categories of all points of each voxel, and the category with the largest number of points is used as the category of the voxel.

[0052] As a preferred solution, the marking module includes:

[0053] A ray forming unit, used to connect each virtual point with the camera origin in the ego-vehicle coordinate system to form a number of rays;

[0054] A traversal unit is used to use the Ray-Casting algorithm to sequentially traverse the voxels that the ray passes through along each ray until the first occupied voxel is encountered, and the occupied voxel is a voxel in an occupied state;

[0055] A judgment unit, used to judge the category label of the occupied voxel;

[0056] If the category label of the occupied voxel is a ground target, the method in the ground target processing unit is executed;

[0057] If the category label of the occupied voxel is a non-ground target, the method in the non-ground target processing unit is executed;

[0058] The ground target processing unit is used to take the plane where the point cloud gravity center of the original occupied voxel is located as the upper plane, and together with the bottom plane of the original occupied voxel, form a new voxel.

[0059] Further determine whether the ray passes through the new voxel;

[0060] If the ray passes through a new voxel, the original occupied voxel is marked as "observed", and the voxel that the ray passed before the original occupied voxel is marked as "empty", the traversal is terminated, and the Ray-Casting function is exited;

[0061] If the new voxel is not passed through, the original occupied voxel is marked as "empty", and the traversal continues to find the next occupied voxel, and the method in the judgment unit is repeatedly executed;

[0062] The non-ground target processing unit is used to mark the originally occupied voxel as "observed" and the voxel that the ray passes through before the originally occupied voxel as "empty", terminate the traversal, and exit the Ray-Casting function.

[0063] In a third aspect, the present invention further discloses a storage medium storing one or more computer-readable programs, wherein the one or more programs include instructions suitable for being loaded by a memory and executing any of the above-mentioned methods for optimizing the visibility of ground target voxel cameras.

[0064] The present invention discloses a method, device and storage medium for optimizing the visibility of ground target voxel cameras, which have the following beneficial effects:

[0065] The present invention fully considers the actual distribution characteristics of ground target voxels, constructs finer-grained ground target voxels, gives more accurate target voxel states, reduces the number of "unobserved" ground target voxels, and improves the recognition effect of the 3D Occupancy network on ground targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0067] Figure 1 A flowchart of an optimization method provided by an embodiment of the present invention.

[0068] Figure 2 A schematic diagram of rays and new voxels provided in an embodiment of the present invention.

[0069] Figure 3 (a) The ground target voxel truth label before optimization provided by an embodiment of the present invention;

[0070] Figure 3 (b) The ground target voxel truth label after optimization provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments 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.

[0073] The expression of “comprising” an element is an “open” expression, which merely means that a corresponding component or step exists, and should not be interpreted as excluding additional components or steps.

[0074] In order to achieve the purpose of the present invention, some embodiments of a method, device and storage medium for optimizing ground target voxel camera visibility are as follows: Figure 1 As shown, the optimization methods include:

[0075] Step S1: Calculate the occupancy state of each voxel in the 3D space according to the LiDAR visibility. If the voxel contains a point cloud target, the voxel is in an occupied state.

[0076] Step S2: Calculate the category label of each voxel and the point cloud center of gravity position according to the point cloud distribution. The category label is: ground target or non-ground target.

[0077] Step S3: Generate a virtual point for each pixel in the image;

[0078] Step S4: transform the coordinates of each virtual point and the camera origin into the vehicle coordinate system;

[0079] Step S5: In the ego-vehicle coordinate system, each virtual point is connected to the camera origin to form a ray, and the Ray-Casting algorithm is used to traverse the voxels that the ray passes through, and each voxel is marked as "observed", "empty" or "unobserved" based on its occupancy state, category label, and point cloud centroid position;

[0080] Step S6: Perform the next step of processing on the voxels marked as "observed" or "empty", and ignore the "unobserved" voxels.

[0081] In order to further optimize the implementation effect of the present invention, in some other embodiments, the remaining characteristic technologies are the same, except that step S2 includes:

[0082] Step S2.1: According to the coordinates of each point in the point cloud, the voxel corresponding to the point is calculated to obtain the corresponding relationship between each point in the point cloud and each voxel (i.e., which voxel each point corresponds to and which points each voxel contains);

[0083] Step S2.2: Obtain the coordinates and categories of all points corresponding to each voxel, and calculate the point cloud centroid position and category label of each voxel.

[0084] Further, step S2.2 includes:

[0085] Get the coordinates of all points corresponding to each voxel;

[0086] Calculate the average value of the coordinates of all points of each voxel, and use the average value as the centroid position of the point cloud of the voxel;

[0087] Count the categories of all points of each voxel, and take the category with the largest number of points as the category of the voxel.

[0088] In order to further optimize the implementation effect of the present invention, in some other embodiments, the remaining characteristic technologies are the same, except that step S5 includes:

[0089] Step S5.1: In the ego-vehicle coordinate system, connect each virtual point with the camera origin to form a number of rays;

[0090] Step S5.2: using the Ray-Casting algorithm, along each ray, traverse the voxels that the ray passes through in sequence until the first occupied voxel is encountered, and the occupied voxel is the voxel in the occupied state;

[0091] Step S5.3: Determine the category label of the occupied voxel;

[0092] If the category label of the occupied voxel is a ground target, the plane where the point cloud gravity center of the original occupied voxel is located is taken as the upper plane, and together with the bottom plane of the original occupied voxel, a new voxel is formed.

[0093] Further determine whether the ray passes through the new voxel;

[0094] If the ray passes through a new voxel, the original occupied voxel is marked as "observed", and the voxel that the ray passed before the original occupied voxel is marked as "empty", the traversal is terminated, and the Ray-Casting function is exited;

[0095] If the new voxel is not traversed, the original occupied voxel is marked as "empty", and the traversal continues to find the next occupied voxel, and step S5.3 is repeated;

[0096] If the category label of the occupied voxel is a non-ground target, the original occupied voxel is marked as "observed", and the voxel that the ray passes through before the original occupied voxel is marked as "empty", the traversal is terminated, and the Ray-Casting function is exited.

[0097] The present invention optimizes the camera visibility of ground target voxels by modifying the termination condition of the Ray-Casting function, thereby reducing the number of "unobserved" ground target voxels.

[0098] like Figure 2 As shown in the figure, offset is the point cloud centroid position of the ground target voxel, and the offset plane and the bottom plane of the original occupied voxel constitute a new fine-grained voxel. If the red ray passes through the new voxel, the Ray-Casting algorithm terminates early, and the subsequent voxels are "unobserved". If the blue ray does not pass through the new voxel, the Ray-Casting algorithm continues to traverse downward until the termination condition is met.

[0099] After experimental comparison, Figure 3 (a) Figure 3 As shown in (b), it can be seen that after optimizing the camera visibility of ground target voxels, the true value information of ground target voxels is more accurate and complete, which helps the model to recognize the ground area.

[0100] In addition, an embodiment of the present invention discloses a device for optimizing the visibility of a ground target voxel camera, comprising:

[0101] The occupancy state calculation module is used to calculate the occupancy state of each voxel in the 3D space according to the LiDAR visibility. If the voxel contains a point cloud target, the voxel is occupied.

[0102] Point cloud computing module, used to calculate the category label of each voxel and the point cloud center of gravity position according to the point cloud distribution. The category label is: ground target or non-ground target;

[0103] A virtual point generation module, used to generate a virtual point for each pixel in the image;

[0104] A coordinate transformation module is used to transform the coordinates of each virtual point and the camera origin into the ego-vehicle coordinate system;

[0105] The marking module is used to connect each virtual point with the camera origin in the ego-vehicle coordinate system to form a ray. The Ray-Casting algorithm is used to traverse the voxels that the ray passes through, and mark each voxel as "observed", "empty" or "unobserved" based on its occupancy state, category label, and point cloud center of gravity position.

[0106] The processing module is used to perform the next step of processing on the voxels marked as "observed" and "empty", and ignore the "unobserved" voxels.

[0107] In order to further optimize the implementation effect of the present invention, in some other implementations, the remaining feature technologies are the same, except that the point cloud computing module includes:

[0108] A first calculation unit, used to calculate the voxel corresponding to each point in the point cloud according to the coordinates of the point, and obtain the corresponding relationship between each point in the point cloud and each voxel;

[0109] The second calculation unit is used to obtain the coordinates and categories of all points corresponding to each voxel, and calculate the point cloud centroid position and category label of each voxel.

[0110] Further, the second computing unit includes:

[0111] An obtaining unit, used to obtain the coordinates of all points corresponding to each voxel;

[0112] A point cloud centroid position calculation unit is used to calculate the average value of the coordinates of all points of each voxel, and use the average value as the point cloud centroid position of the voxel;

[0113] The voxel category calculation unit is used to count the categories of all points of each voxel, and the category with the largest number of points is used as the category of the voxel.

[0114] In order to further optimize the implementation effect of the present invention, in some other embodiments, the remaining characteristic technologies are the same, except that the marking module includes:

[0115] A ray forming unit, used to connect each virtual point with the camera origin in the ego-vehicle coordinate system to form a number of rays;

[0116] A traversal unit is used to use the Ray-Casting algorithm to sequentially traverse the voxels that the ray passes through along each ray until the first occupied voxel is encountered, and the occupied voxel is a voxel in an occupied state;

[0117] A judgment unit, used to judge the category label of the occupied voxel;

[0118] If the category label of the occupied voxel is a ground target, the method in the ground target processing unit is executed;

[0119] If the category label of the occupied voxel is a non-ground target, the method in the non-ground target processing unit is executed;

[0120] The ground target processing unit is used to take the plane where the point cloud gravity center of the original occupied voxel is located as the upper plane, and together with the bottom plane of the original occupied voxel, form a new voxel.

[0121] Further determine whether the ray passes through the new voxel;

[0122] If the ray passes through a new voxel, the original occupied voxel is marked as "observed", and the voxel that the ray passed before the original occupied voxel is marked as "empty", the traversal is terminated, and the Ray-Casting function is exited;

[0123] If the new voxel is not passed through, the original occupied voxel is marked as "empty", and the traversal continues to find the next occupied voxel, and the method in the judgment unit is repeatedly executed;

[0124] The non-ground target processing unit is used to mark the originally occupied voxel as "observed" and the voxel that the ray passes through before the originally occupied voxel as "empty", terminate the traversal, and exit the Ray-Casting function.

[0125] In this embodiment, the specific content of the device for optimizing the visibility of a ground object voxel camera is similar to the content of the method for optimizing the visibility of a ground object voxel camera disclosed in the above embodiment, and will not be repeated here.

[0126] It is worth noting that the present invention can be applied in an OCC truth value system.

[0127] In addition, an embodiment of the present invention further discloses a storage medium storing one or more computer-readable programs, wherein the one or more programs include instructions suitable for being loaded by a memory and executing the method for optimizing the visibility of a ground target voxel camera disclosed in any of the above embodiments.

[0128] The present invention discloses a method, device and storage medium for optimizing the visibility of ground target voxel cameras, which have the following beneficial effects:

[0129] The present invention fully considers the actual distribution characteristics of ground target voxels, constructs finer-grained ground target voxels, gives more accurate target voxel states, reduces the number of "unobserved" ground target voxels, and improves the recognition effect of the 3D Occupancy network on ground targets.

[0130] It should be understood that the various techniques described herein can be implemented in combination with hardware or software, or a combination thereof. Thus, the method and apparatus of the present invention, or some aspects or portions of the method and apparatus of the present invention, can take the form of program codes (i.e., instructions) embedded in a tangible medium, such as a floppy disk, a CD-ROM, a hard disk drive, or any other machine-readable storage medium, wherein when the program is loaded into a machine such as a computer and executed by the machine, the machine becomes a device for practicing the present invention.

[0131] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for illustrating the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which shall fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for optimizing the visibility of ground object voxel cameras, characterized in that: include: Step S1: Calculate the occupancy state of each voxel in the 3D space according to the LiDAR visibility. If the voxel contains a point cloud target, the voxel is in an occupied state. Step S2: Calculate the category label of each voxel and the point cloud center of gravity position according to the point cloud distribution. The category label is: ground target or non-ground target. Step S3: Generate a virtual point for each pixel in the image; Step S4: transform the coordinates of each virtual point and the camera origin into the vehicle coordinate system; Step S5: In the ego-vehicle coordinate system, each virtual point is connected to the camera origin to form a ray, and the Ray-Casting algorithm is used to traverse the voxels that the ray passes through, and each voxel is marked as "observed", "empty" or "unobserved" based on its occupancy state, category label, and point cloud centroid position; Step S6: Perform the next step of processing on the voxels marked as "observed" and "empty", and ignore the "unobserved" voxels.

2. The optimization method according to claim 1, characterized in that: The step S2 comprises: Step S2.1: According to the coordinates of each point in the point cloud, the voxel corresponding to the point is calculated to obtain the corresponding relationship between each point in the point cloud and each voxel; Step S2.2: Obtain the coordinates and categories of all points corresponding to each voxel, and calculate the point cloud centroid position and category label of each voxel.

3. The optimization method according to claim 2, characterized in that: The step S2.2 comprises: Get the coordinates of all points corresponding to each voxel; Calculate the average value of the coordinates of all points of each voxel, and use the average value as the centroid position of the point cloud of the voxel; Count the categories of all points of each voxel, and take the category with the largest number of points as the category of the voxel.

4. The optimization method according to any one of claims 1 to 3, characterized in that: The step S5 comprises: Step S5.1: In the ego-vehicle coordinate system, connect each virtual point with the camera origin to form a number of rays; Step S5.2: using the Ray-Casting algorithm, along each ray, traverse the voxels that the ray passes through in sequence until the first occupied voxel is encountered, and the occupied voxel is the voxel in the occupied state; Step S5.3: Determine the category label of the occupied voxel; If the category label of the occupied voxel is a ground target, the plane where the point cloud gravity center of the original occupied voxel is located is taken as the upper plane, and together with the bottom plane of the original occupied voxel, a new voxel is formed. Further determine whether the ray passes through the new voxel; If the ray passes through a new voxel, the original occupied voxel is marked as "observed", and the voxel that the ray passes through before the original occupied voxel is marked as "empty", the traversal is terminated, and the Ray-Casting function is exited; If the new voxel is not passed through, the original occupied voxel is marked as "empty", and the traversal continues to find the next occupied voxel, and step S5.3 is repeated; If the category label of the occupied voxel is a non-ground target, the original occupied voxel is marked as "observed", and the voxel that the ray passes through before the original occupied voxel is marked as "empty", the traversal is terminated, and the Ray-Casting function is exited.

5. A device for optimizing the visibility of ground object voxel cameras, characterized in that: include: The occupancy state calculation module is used to calculate the occupancy state of each voxel in the 3D space according to the LiDAR visibility. If the voxel contains a point cloud target, the voxel is occupied. Point cloud computing module, used to calculate the category label of each voxel and the point cloud center of gravity position according to the point cloud distribution. The category label is: ground target or non-ground target; A virtual point generation module, used to generate a virtual point for each pixel in the image; A coordinate transformation module is used to transform the coordinates of each virtual point and the camera origin into the ego-vehicle coordinate system; The marking module is used to connect each virtual point with the camera origin in the ego-vehicle coordinate system to form a ray. The Ray-Casting algorithm is used to traverse the voxels that the ray passes through, and mark each voxel as "observed", "empty" or "unobserved" based on its occupancy state, category label, and point cloud center of gravity position. The processing module is used to perform the next step of processing on the voxels marked as "observed" and "empty", and ignore the "unobserved" voxels.

6. The optimization device according to claim 5, characterized in that The point cloud computing module includes: A first calculation unit, used to calculate the voxel corresponding to each point in the point cloud according to the coordinates of the point, and obtain the corresponding relationship between each point in the point cloud and each voxel; The second calculation unit is used to obtain the coordinates and categories of all points corresponding to each voxel, and calculate the point cloud centroid position and category label of each voxel.

7. The optimization device according to claim 6, characterized in that The second computing unit comprises: An obtaining unit, used to obtain the coordinates of all points corresponding to each voxel; A point cloud centroid position calculation unit is used to calculate the average value of the coordinates of all points of each voxel, and use the average value as the point cloud centroid position of the voxel; The voxel category calculation unit is used to count the categories of all points of each voxel, and the category with the largest number of points is used as the category of the voxel.

8. The optimization device according to any one of claims 5 to 7, characterized in that: The marking module comprises: A ray forming unit, used to connect each virtual point with the camera origin in the ego-vehicle coordinate system to form a number of rays; A traversal unit is used to use the Ray-Casting algorithm to sequentially traverse the voxels that the ray passes through along each ray until the first occupied voxel is encountered, and the occupied voxel is a voxel in an occupied state; A judgment unit, used to judge the category label of the occupied voxel; If the category label of the occupied voxel is a ground target, the method in the ground target processing unit is executed; If the category label of the occupied voxel is a non-ground target, the method in the non-ground target processing unit is executed; The ground target processing unit is used to take the plane where the point cloud gravity center of the original occupied voxel is located as the upper plane, and together with the bottom plane of the original occupied voxel, form a new voxel. Further determine whether the ray passes through the new voxel; If the ray passes through a new voxel, the original occupied voxel is marked as "observed", and the voxel that the ray passes through before the original occupied voxel is marked as "empty", the traversal is terminated, and the Ray-Casting function is exited; If the new voxel is not passed through, the original occupied voxel is marked as "empty", and the traversal continues to find the next occupied voxel, and the method in the judgment unit is repeated; The non-ground target processing unit is used to mark the original occupied voxel as "observed", and mark the voxel that the ray passes through before the original occupied voxel as "empty", terminate the traversal, and exit the Ray-Casting function.

9. A storage medium, characterized in that The storage medium stores one or more computer-readable programs, wherein the one or more programs include instructions, and the instructions are suitable for being loaded by a memory and executing the method for optimizing the visibility of a ground object voxel camera as described in any one of claims 1 to 4.

Images