Point cloud data processing method, electronic device, and vehicle

Abstract

The application belongs to the technical field of automatic driving and environment perception, and discloses a point cloud data processing method, an electronic device and a vehicle, wherein the point cloud data processing method applied to the vehicle comprises the following steps: obtaining to-be-processed point cloud data through a sensor on the vehicle; and deleting points in the to-be-processed point cloud data that match voxels in a preset occlusion object voxel mask; wherein the preset occlusion object voxel mask is determined in advance according to continuous multiple frames of point cloud data obtained by the sensor. Thus, fixed occlusion object filtering with low delay and low calculation cost can be realized without relying on a deep learning model, and the change of the sensor installation position or the cargo state can be automatically adapted, thereby avoiding the blind area problem caused by a static preset region.

Description

Technical Field

[0001] This invention relates to the fields of autonomous driving and environmental perception technology, and in particular to a point cloud data processing method, electronic device, and vehicle. Background Technology

[0002] With the rapid development of autonomous driving, robot navigation, and industrial automation technologies, 3D point cloud-based environmental perception systems are playing an increasingly important role in obstacle detection and obstacle avoidance decisions in complex scenarios. In practical engineering applications, vehicles (such as automated guided vehicles, AGVs) or robot platforms often have physical structures that remain relatively stationary compared to the sensors, such as load-bearing material pallets or sensor mounting brackets. These fixed obstructions inevitably enter the sensor's field of view, generating frequent point cloud data that can cause the perception system to misinterpret them as obstacles, leading to false alarms or incorrect obstacle avoidance behaviors. Summary of the Invention

[0003] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the first objective of this invention is to propose a point cloud data processing method that can achieve low-latency, low-computational-overhead fixed-obstacle filtering without relying on deep learning models, and automatically adapt to changes in sensor installation location or cargo status, avoiding blind spot problems caused by statically preset areas.

[0004] The second objective of this invention is to provide an electronic device.

[0005] The third objective of this invention is to provide a vehicle.

[0006] To achieve the above objectives, a first aspect of the present invention provides a point cloud data processing method, which is applied to a vehicle. The method includes: acquiring point cloud data to be processed through sensors on the vehicle; deleting points in the point cloud data to be processed that match voxels within a preset occlusion object pixel mask; wherein the preset occlusion object pixel mask is predetermined based on multiple consecutive frames of point cloud data acquired by the sensors.

[0007] According to the point cloud data processing method of the present invention, point cloud data to be processed is acquired by a sensor, and occlusion object pixel masks based on multiple consecutive frames are used for matching and deletion. Therefore, low-latency, low-computational-overhead fixed occlusion filtering can be achieved without relying on a deep learning model, and it automatically adapts to changes in sensor installation position or loading status, avoiding blind spot problems caused by static preset areas.

[0008] In addition, the point cloud data processing method according to the above embodiments of the present invention may further include the following additional technical features: According to an embodiment of the present invention, the preset occlusion pixel mask is determined by the following steps: dividing the space of each frame of point cloud data in the multi-frame point cloud data into multiple spatial units according to the same spatial range and the same size; for each spatial unit, counting the total number of times the spatial unit is occupied in the multi-frame point cloud data; and determining the preset occlusion pixel mask from the multiple spatial units based on the total number of times.

[0009] According to an embodiment of the present invention, determining the preset occlusion object pixel mask from the plurality of spatial units based on the total number of times includes: calculating the occupancy frequency based on the total number of times and the number of frames of the multi-frame point cloud data; and determining the spatial units with an occupancy frequency greater than a preset frequency threshold as the preset occupancy object pixel mask.

[0010] According to an embodiment of the present invention, before dividing the plurality of voxel units, the method further includes: spatially cropping the point cloud data, wherein the cropped area is the near-field space of the sensor, wherein the near-field space is a spatial region that is less than a preset distance threshold from the sensor.

[0011] According to one embodiment of the present invention, the step of counting the total number of times the spatial unit is occupied in the multi-frame point cloud data includes: downsampling each frame of the point cloud data to obtain the voxel centroid points in each frame of the point cloud data; and for each voxel centroid point, counting the number of times the voxel centroid point falls into the spatial unit in the multi-frame point cloud data.

[0012] According to one embodiment of the present invention, downsampling the point cloud data includes: dividing the point cloud data into multiple voxel regions according to a preset voxel side length, determining the centroid of each voxel region, and obtaining multiple voxel centroid points.

[0013] According to one embodiment of the present invention, the method further includes: updating the preset occlusion object pixel mask according to the number of times the voxels within the preset occlusion object pixel mask are matched by point cloud data.

[0014] According to an embodiment of the present invention, the method further includes: in response to detecting that the occluder corresponding to the preset occluder pixel mask has shifted or deformed, updating the preset occluder pixel mask based on the continuous multi-frame point cloud data reacquired by the sensor.

[0015] To achieve the above objectives, a second aspect of the present invention provides an electronic device comprising: a memory, a processor, and a computer program stored in the memory, wherein the computer program, when executed by the processor, implements the point cloud data processing method of the aforementioned embodiments of the present invention.

[0016] According to the embodiments of the present invention, the electronic device can achieve low-latency and low-computational-overhead fixed obstruction filtering without relying on deep learning models by executing computer programs through a processor, and automatically adapt to changes in sensor installation position or cargo status, avoiding blind spot problems caused by static preset areas.

[0017] To achieve the above objectives, a third aspect of the present invention provides a vehicle that includes the electronic equipment described in the foregoing embodiments of the present invention.

[0018] According to the vehicle of the present invention, by employing the electronic equipment of the above embodiments of the present invention, it is possible to achieve fixed obstruction filtering with low latency and low computational overhead without relying on deep learning models, and automatically adapt to changes in sensor installation position or cargo status, thereby avoiding blind spot problems caused by static preset areas.

[0019] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a point cloud data processing method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a point cloud data processing method according to another embodiment of the present invention; Figure 3 This is a flowchart illustrating a point cloud data processing method according to yet another embodiment of the present invention; Figure 4 This is a flowchart illustrating a point cloud data processing method according to another embodiment of the present invention; Figure 5 This is a block diagram of an electronic device according to an embodiment of the present invention; Figure 6 This is a block diagram of a vehicle according to an embodiment of the present invention. Detailed Implementation

[0021] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0022] The point cloud data processing method, electronic device, and vehicle according to embodiments of the present invention are described below with reference to the accompanying drawings.

[0023] Figure 1This is a flowchart illustrating a point cloud data processing method according to an embodiment of the present invention.

[0024] Specifically, in some embodiments of the present invention, such as Figure 1 As shown, the point cloud data processing method is applied to vehicles, and the point cloud data processing method includes: S101 acquires point cloud data to be processed through sensors on the vehicle.

[0025] Specifically, in this embodiment, the sensor is a depth camera or LiDAR, fixedly mounted on the vehicle, used to acquire real-time three-dimensional spatial information of the environment in front of or around the vehicle. The raw data output by the sensor can be a 3D point cloud or a depth image. If the sensor directly outputs a 3D point cloud, it is used as the point cloud data to be processed; if the sensor outputs a depth map, each pixel in the depth map needs to be converted into point coordinates in three-dimensional space according to the camera's intrinsic parameters, thereby generating 3D point cloud data. The acquired point cloud data to be processed contains discrete points on the surface of objects in the environment, and each point contains at least (x, y, z) three-dimensional coordinate information. Before proceeding to the subsequent filtering steps, necessary basic preprocessing can be performed on the point cloud, such as removing invalid points or outlier noise points, to improve data quality.

[0026] S102, delete the points in the point cloud data to be processed that match the voxels in the preset occlusion object pixel mask; wherein, the preset occlusion object pixel mask is predetermined based on multiple consecutive frames of point cloud data acquired by the sensor.

[0027] Specifically, in this embodiment, for each point in the point cloud data to be processed, the voxel index to which the point belongs is first calculated based on the preset voxel side length. This is done by dividing the point's x, y, and z coordinates by the voxel side length and rounding down to obtain three integer values ​​as the voxel index. The formula for the voxel index is as follows:

[0028] Where g(p) represents the voxel index, (x,y,z) represents the three-dimensional coordinates of point p, and d represents the voxel side length. This indicates rounding down to the nearest integer.

[0029] Then, the voxel index is used to query a pre-defined voxel mask hash table. The hash table pre-stores the index set of all fixed-occlusion objects, with a query time complexity of constant time. If the voxel index exists in the mask set, the current point is determined to be a point on a fixed occlusion object and is deleted from the point cloud; otherwise, it is retained. After traversing all points, the filtered point cloud is obtained. This step can complete the matching and deletion of all points in linear time complexity. The calculation formula is as follows:

[0030] in, This represents the filtered point cloud. This represents the original point cloud of the current frame, where p represents a point in the point cloud, g(p) represents the voxel index of point p, and M represents the preset occlusion object pixel mask.

[0031] It should be noted that during vehicle initialization or upon receiving a specific trigger signal (such as changing vehicles or completing unloading), the system continuously collects N frames of point cloud data via sensors, where N is a preset positive integer, preferably ranging from 10 to 50 frames. These N frames of point cloud data are then processed through voxelization and statistical analysis to identify voxel units whose spatial positions remain unchanged in the sensor coordinate system—that is, voxels occupied by fixed occupants. Specifically, the total number of times each voxel unit is occupied in the N frames is counted, and the occupancy frequency is calculated. Voxel units with occupancy frequencies exceeding a set threshold are filtered out to form a preset occupant pixel mask. This mask is stored in the form of a hash table, representing the spatial distribution of fixed occupants that are relatively stationary to the sensor in the current scene. Once the mask is established, it can be used for rapid filtering of subsequent real-time point cloud frames.

[0032] Furthermore, in some embodiments of the present invention, such as Figure 2 As shown, the preset occlusion object pixel mask is determined through the following steps: S201 divides the space of each frame of point cloud data in the multi-frame point cloud data into multiple spatial units according to the same spatial range and the same size.

[0033] Specifically, in this embodiment, a unified spatial clipping range is first set based on the effective detection range of the sensor and the possible locations of fixed obstructions. This clipping area is the near-field space of the sensor, such as a cuboid area with the sensor as the origin, 2m in front, 1m to the left and right, and 0.5m up and down, or more generally, a 2m×2m×1m cube area in front of the sensor. The specific range can be adjusted according to the actual installation location and application scenario, but it must ensure that all objects that may become fixed obstruction sources are covered. Then, the space within the clipping area is divided into cubic units of the same size according to a preset voxel side length d (e.g., 0.05m), and each unit is called a voxel. All frames participating in mask construction use the exact same spatial clipping range and voxel size to ensure that the same voxel index has a consistent physical spatial correspondence between different frames. In addition, the present invention does not specifically limit the value range of the near-field space and can dynamically adjust it according to different scenarios.

[0034] S202, for each spatial unit, count the total number of times the spatial unit is occupied in multi-frame point cloud data.

[0035] Specifically, in this embodiment, the system first initializes a counter for each voxel unit within the clipping area and sets it to zero. Then, it iterates through each frame of the N frames of point cloud data collected. For the current frame, it obtains the set of voxel centroids obtained after downsampling and determines the voxel unit to which each centroid belongs (based on the correspondence between its coordinates and the voxel side length). If a voxel unit contains at least one centroid, it is marked as "occupied" in the current frame, and the counter corresponding to that voxel is incremented. Each voxel unit is counted at most once in a frame to avoid duplicate counting. After performing the above operations on all N frames, each voxel unit corresponds to a total number of occupancy counts. ,in A three-dimensional index for voxels.

[0036] S203, based on the total number of iterations, determine the preset occlusion object pixel mask from multiple spatial units.

[0037] Specifically, in this embodiment, the occupancy frequency of each voxel unit is calculated based on the total number of occupancy times in the multi-frame point cloud and the total number of frames in the point cloud, i.e., the total number of occupancy times divided by the total number of frames. Since the fixed occupant remains absolutely stationary relative to the sensor, the voxel units it occupies are almost always occupied in each frame, so the occupancy frequency of these voxel units is close to 1. The system pre-sets a frequency threshold, such as 0.9, and identifies voxel units with an occupancy frequency greater than this threshold as voxels occupied by the fixed occupant, and filters them out from all spatial units to form a voxel mask set. Finally, this mask set is stored in a hash table so that it can be quickly queried whether a voxel index belongs to the fixed occupant in the subsequent real-time filtering stage.

[0038] Furthermore, in some embodiments of the present invention, such as Figure 3 As shown, based on the total number of iterations, a preset occlusion object pixel mask is determined from multiple spatial units, including: S301 calculates the occupied frequency based on the total number of times and the number of frames of multi-frame point cloud data.

[0039] Specifically, in this embodiment, for each voxel unit, the total number of times it is occupied in multiple frames of point cloud data is divided by the total number of point cloud frames, and the resulting ratio is the occupancy frequency of that voxel unit. This frequency is between 0 and 1, reflecting the stability of the voxel's occupancy in the point cloud in the temporal domain: the higher the frequency, the greater the probability that the voxel is occupied in consecutive frames, and the more stable its spatial position; conversely, the lower the frequency, the more random the voxel's occupancy or the presence of a dynamic object at that location. Voxels corresponding to fixed occupants typically exhibit a high occupancy frequency close to 1 because their positions are constant.

[0040] S302, the spatial unit whose occupied frequency is greater than the preset frequency threshold is determined as the preset occlusion object pixel mask.

[0041] Specifically, in this embodiment, the system pre-sets a frequency threshold, which can be selected between 0.7 and 0.95 based on the stability requirements of the actual scene, with a preferred value of 0.9. For each voxel unit, its calculated occupancy frequency is compared with this threshold. If the occupancy frequency is greater than the threshold, the voxel unit is determined to be a fixed occluder and included in the voxel mask set; if the occupancy frequency is less than or equal to the threshold, the voxel unit is determined not to be a fixed occluder and is not included in the mask set. Through this threshold filtering, fixed occluders can be effectively distinguished from accidentally passing dynamic objects or noise points, ensuring the accuracy and robustness of the mask.

[0042] It should be noted that the formula for calculating the voxel mask is as follows:

[0043] Where M represents the preset occlusion object pixel mask. Indicates spatial coordinate index as voxel units, This indicates the frequency occupied by the voxel in a multi-frame point cloud. This indicates the preset frequency threshold.

[0044] Furthermore, in some embodiments of the present invention, before dividing the data into multiple voxel units, the method further includes: spatially cropping the point cloud data, wherein the cropped area is the near-field space of the sensor, and the near-field space is a spatial region that is less than a preset distance threshold from the sensor.

[0045] Specifically, in this embodiment, to avoid interference from distant, irrelevant static objects such as walls and pillars in subsequent voxel statistics and to reduce overall computational overhead, the system first performs spatial clipping on each frame of point cloud data. The clipping area is set as the sensor's near-field space, i.e., the spatial region within a preset distance threshold from the sensor. This threshold can be set according to the sensor's detection range and the actual distribution of fixed obstructions, for example, 3 meters or 5 meters. In specific implementation, the near-field space can be a cuboid region centered on the sensor, such as 5 meters in front, 3 meters to the left and right, and 2 meters up and down, or it can be a spherical region, i.e., retaining all point clouds within a set value (such as 3 meters) of Euclidean distance from the sensor. After spatial clipping, only the point cloud data falling within the near-field space is retained, and the remaining point clouds are directly discarded, thereby effectively reducing the number of point clouds participating in subsequent voxel segmentation and statistics, and improving processing efficiency.

[0046] Furthermore, in some embodiments of the present invention, such as Figure 4 As shown, the total number of times a spatial unit is occupied in multi-frame point cloud data includes: S401, downsample each frame of point cloud data to obtain the voxel centroids in each frame of point cloud data.

[0047] Specifically, in this embodiment, for each frame of the original point cloud, a voxel grid filter is used for downsampling. First, according to a preset voxel side length (consistent with the side length used in subsequent spatial unit division, for example, 0.05 meters), the three-dimensional space of the current frame's point cloud is divided into several voxel regions. For each non-empty voxel region, the average coordinates of all points within the region are calculated to obtain the centroid of that voxel region. The set of all centroids is the downsampled point cloud for that frame. Through downsampling, the number of points can be significantly reduced while preserving spatial structure information, and small displacements caused by sensor noise can be eliminated.

[0048] S402, for each voxel centroid, count the number of times the voxel centroid falls into a spatial cell in multi-frame point cloud data.

[0049] Specifically, in this embodiment, for each voxel centroid in each frame of point cloud obtained after downsampling, its spatial unit (i.e., voxel unit) is determined based on its three-dimensional coordinates. All N frames of point cloud are traversed, and a counter is maintained for each spatial unit. If a voxel centroid falls into a spatial unit, the counter for that spatial unit is incremented. It is important to note that even if the same spatial unit contains multiple centroids in a single frame of point cloud, it is only counted once to avoid duplicate counting within the same frame. After statistics are collected for all frames, the total value for each spatial unit represents the total number of times that spatial unit has been occupied across multiple frames of point cloud data.

[0050] Furthermore, in some embodiments of the present invention, downsampling of point cloud data includes: dividing the point cloud data into multiple voxel regions according to a preset voxel side length, determining the centroid of each voxel region, and obtaining multiple voxel centroid points.

[0051] Specifically, in this embodiment, the system first sets a fixed voxel side length, for example, 0.05 meters. This side length determines the granularity of downsampling: the smaller the side length, the more points are retained after downsampling, and the higher the spatial resolution; the larger the side length, the higher the data compression and the higher the computational efficiency. In practical applications, the side length can be selected according to the point cloud density and accuracy requirements. Then, the three-dimensional space of the current frame point cloud is divided into a regular voxel grid according to the voxel side length. For each non-empty voxel region (i.e., a region containing at least one original point), the average coordinates of all points in the region are calculated, and the x, y, and z components are averaged respectively to obtain the centroid of the voxel region. This centroid serves as the unique representative point of the voxel region. After traversing all non-empty voxel regions, the set of all centroids constitutes the downsampled point cloud. Through this downsampling method, the amount of point cloud data can be significantly reduced while maintaining the geometric structure of the environment, and the influence of small displacements caused by sensor noise on subsequent statistics can be suppressed.

[0052] Furthermore, in some embodiments of the present invention, the method further includes: updating the preset occlusion object pixel mask according to the number of times the voxels within the preset occlusion object pixel mask are matched by point cloud data.

[0053] Specifically, in this embodiment, during the real-time filtering phase, the system maintains a recent hit timestamp or a hit counter for each voxel in the mask to record the number of times the voxel is matched by the point cloud data or the last matching time. Each time filtering is performed on the point cloud to be processed, if a point in the current frame falls within a mask voxel, the voxel's hit timestamp is updated to the current time, or its hit counter is incremented. For voxels in the mask, the system periodically checks their recent hit status: if a voxel is not hit within multiple consecutive frames, it is determined that the voxel no longer belongs to a fixed occluder (e.g., due to displacement or deformation of the occluder), and it is deleted from the mask hash table to free up storage space. On the other hand, the system can also monitor voxels that are not covered by the mask but are frequently occupied: if a voxel is hit more than a preset threshold within multiple consecutive frames and is not currently in the mask, it is added to the mask set. Through the aforementioned bidirectional update mechanism, the mask can be dynamically maintained as the environment changes (such as slow movement of obstructions, sensor vibration, etc.), which reduces memory usage and improves the long-term robustness of the system while ensuring filtering accuracy.

[0054] Furthermore, in some embodiments of the present invention, the method further includes: in response to detecting that the occluder corresponding to the preset occluder pixel mask has shifted or deformed, updating the preset occluder pixel mask based on the continuous multi-frame point cloud data reacquired by the sensor.

[0055] Specifically, in this embodiment, the system includes a mask reset trigger mechanism to detect potential displacement or deformation of fixed obstructions. Triggering conditions include, but are not limited to: vehicle unloading, vehicle type switching, forklift positioning signal, or a sharp drop in the hit rate of voxels in the mask across multiple consecutive frames (e.g., overall hit rate below a preset threshold). When any triggering condition is met, the system determines that the original preset voxel mask can no longer accurately reflect the current distribution of fixed obstructions. At this time, the system automatically clears the currently stored mask hash table and re-executes the aforementioned mask construction steps, i.e., re-collecting multiple consecutive frames of point cloud data, performing spatial clipping, voxel partitioning, occupancy statistics, and frequency filtering to generate an updated voxel mask. This trigger mechanism can be actively triggered by signals on the vehicle control bus (such as forklift forklift positioning signals) or automatically initiated by the sensing module based on real-time monitoring of the mask hit rate. Through this reset and update mechanism, the system can quickly and adaptively reconstruct the mask after changes in the obstruction's state, ensuring the accuracy and reliability of filtering.

[0056] In summary, the point cloud data processing method according to embodiments of the present invention acquires point cloud data to be processed through sensors and performs matching and deletion using pixel masks of occluded objects predetermined based on multiple consecutive frames. Therefore, it can achieve low-latency, low-computational-overhead fixed-occlusion filtering without relying on deep learning models, and automatically adapts to changes in sensor installation location or loading status, avoiding blind spot problems caused by static preset areas.

[0057] Based on the point cloud data processing method proposed in the foregoing embodiments of the present invention, the present invention also proposes a computer-readable storage medium storing a point cloud data processing program thereon, which, when executed by a processor, implements the point cloud data processing method of the present invention as described above.

[0058] According to the computer-readable storage medium of the present invention, by executing a point cloud data processing program through a processor, it is possible to achieve low-latency and low-computational-overhead fixed occlusion filtering without relying on a deep learning model, and automatically adapt to changes in sensor installation location or cargo status, avoiding blind spot problems caused by static preset areas.

[0059] Figure 5 This is a block diagram of an electronic device according to an embodiment of the present invention.

[0060] like Figure 5 As shown, the electronic device 100 includes: a memory 10, a processor 20, and a computer program stored on the memory 10. When the computer program is executed by the processor 20, it implements the point cloud data processing method of the above-described embodiments of the present invention.

[0061] According to the embodiments of the present invention, the electronic device can achieve low-latency and low-computational-overhead fixed obstruction filtering without relying on deep learning models by executing computer programs through a processor, and automatically adapt to changes in sensor installation position or cargo status, avoiding blind spot problems caused by static preset areas.

[0062] Figure 6 This is a block diagram of a vehicle according to an embodiment of the present invention.

[0063] like Figure 6 As shown, the vehicle 1000 includes an electronic device 100 as described in the above embodiments of the present invention.

[0064] According to the vehicle of the present invention, by employing the electronic equipment of the above embodiments of the present invention, it is possible to achieve fixed obstruction filtering with low latency and low computational overhead without relying on deep learning models, and automatically adapt to changes in sensor installation position or cargo status, thereby avoiding blind spot problems caused by static preset areas.

[0065] Furthermore, other components and functions of the vehicle in the embodiments of the present invention are known to those skilled in the art, and will not be described in detail here to reduce redundancy.

[0066] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0067] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0068] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0069] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0070] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0071] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0072] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0073] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A point cloud data processing method, characterized in that, Applied to vehicles, the method includes: The point cloud data to be processed is acquired through sensors on the vehicle; Delete the points in the point cloud data to be processed that match the voxels in the preset occlusion object pixel mask; The preset occlusion object pixel mask is predetermined based on multiple consecutive frames of point cloud data acquired by the sensor.

2. The method according to claim 1, characterized in that, The preset occlusion object pixel mask is determined through the following steps: The space of each frame of point cloud data in the multi-frame point cloud data is divided into multiple spatial units according to the same spatial range and the same size; For each spatial unit, the total number of times the spatial unit is occupied in the multi-frame point cloud data is counted. Based on the total number of times, the preset occlusion object pixel mask is determined from the plurality of spatial units.

3. The method according to claim 2, characterized in that, The step of determining the preset occlusion object pixel mask from the plurality of spatial units based on the total number of times includes: The occupancy frequency is calculated based on the total number of times and the number of frames in the multi-frame point cloud data. The spatial units whose occupancy frequency is greater than a preset frequency threshold are determined as the preset occlusion object pixel mask.

4. The method according to claim 2, characterized in that, Before dividing the plurality of voxel units, the method further includes: The point cloud data is spatially cropped, and the cropped area is the near-field space of the sensor, wherein the near-field space is a spatial region that is less than a preset distance threshold from the sensor.

5. The method according to claim 2, characterized in that, The total number of times the spatial unit is occupied in the multi-frame point cloud data includes: The point cloud data of each frame is downsampled to obtain the voxel centroids of each frame of point cloud data; For each voxel centroid, the number of times the voxel centroid falls into the spatial cell in the multi-frame point cloud data is counted.

6. The method according to claim 5, characterized in that, Downsampling the point cloud data includes: The point cloud data is divided into multiple voxel regions according to a preset voxel side length, and the centroid of each voxel region is determined to obtain multiple voxel centroid points.

7. The method according to claim 1, characterized in that, The method further includes: The preset occlusion object mask is updated based on the number of times the voxels within the preset occlusion object mask are matched by the point cloud data.

8. The method according to any one of claims 1-7, characterized in that, The method further includes: In response to the detection that the occluder corresponding to the preset occluder pixel mask has shifted or deformed, the preset occluder pixel mask is updated based on the continuous multi-frame point cloud data reacquired by the sensor.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory, wherein the computer program, when executed by the processor, implements the point cloud data processing method according to any one of claims 1-7.

10. A vehicle, characterized in that, Including the electronic device as described in claim 9.

Images