Point cloud data processing method and device

Abstract

The embodiment of the invention discloses a point cloud data processing method and device. The method comprises the steps of obtaining full-amount point cloud data needing to be displayed, performing downsampling on the full-amount point cloud data to obtain sparse point cloud data, detecting an operation event of camera visual angle change, rendering a display page through the sparse point cloud data when the operation event is executed, and displaying the display page through the sparse point cloud data when the operation event is not executed. And generating a cache picture corresponding to a display page according to the full-amount point cloud data, and rendering the cache picture. Therefore, the point cloud display fluency can be improved, the requirement for computer configuration is reduced, meanwhile, the details of the point cloud can be reserved, and the marking integrity and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and device for processing point cloud data. Background Art

[0002] With the development of intelligent driving technology, the high-precision point cloud data provided by LiDAR is essential to improve the accuracy of target detection, tracking and scene understanding. Point cloud annotation, as an important link between raw data and high-level perception, involves detailed labeling of the three-dimensional spatial information generated by LiDAR to optimize machine learning algorithms and ensure that autonomous vehicles can safely cope with complex road conditions.

[0003] In order to improve the efficiency and quality of point cloud annotation, a multi-frame fusion method is usually used to process lidar data, integrating data from multiple time points to increase the amount of information, reduce noise and enhance the capture of object morphology. Currently, two main point cloud display solutions have been developed: full point cloud display and downsampled point cloud display, aiming to balance the relationship between computing resource consumption and annotation quality, support efficient point cloud annotation, and promote the advancement of intelligent driving systems.

[0004] However, existing point cloud display and annotation technologies face challenges. Full point cloud display requires processing all point cloud data, which not only requires extremely high computer configuration and easily causes lag, but also greatly increases the annotation cost. Although downsampling point cloud display reduces hardware pressure, it loses details due to point cloud sparseness, affecting the integrity and accuracy of annotation. Summary of the invention

[0005] In view of this, an embodiment of the present invention aims to provide a method and device for processing point cloud data, which can improve the fluency of point cloud display and reduce the requirements for computer configuration. At the same time, it can retain the details of the point cloud and improve the completeness and accuracy of annotation.

[0006] In a first aspect, an embodiment of the present invention provides a lane point cloud data processing method, the method comprising:

[0007] Acquire multiple frames of initial point cloud data collected by the acquisition device on the lane through the laser radar during driving;

[0008] Fusing the multiple frames of initial point cloud data to obtain full point cloud data to be displayed;

[0009] The full point cloud data is divided into a plurality of sub-areas, and the full point cloud data is downsampled according to the sub-areas to obtain sparse point cloud data, wherein the sampling parameters of the sub-areas are determined according to the distance between the sub-areas and the driving track of the acquisition device;

[0010] Detect whether a perspective change operation event is being performed;

[0011] When the operation event is executed, a display page is rendered in real time using the sparse point cloud data;

[0012] When the operation event is not executed, a cache image having a size consistent with the current display page is generated according to the full point cloud data, and the cache image is rendered in real time;

[0013] Render the lane marking data input by the user.

[0014] In some embodiments, the sampling parameters include a downsampling ratio and a parameter value of a Bloom filter.

[0015] In some embodiments, the distance between the sub-area and the driving trajectory of the collection device is the shortest distance between each point in each sub-area and each point on the driving trajectory of the collection device.

[0016] In a second aspect, an embodiment of the present invention provides a point cloud data processing method, the method comprising:

[0017] Get the full amount of point cloud data that needs to be displayed;

[0018] Downsampling the full point cloud data to obtain sparse point cloud data;

[0019] Detecting the operation event of view change;

[0020] When executing the operation event, rendering a display page through the sparse point cloud data;

[0021] When the operation event is not executed, a cache image corresponding to the display page is generated according to the full point cloud data, and the cache image is rendered.

[0022] In some embodiments, obtaining the full amount of point cloud data to be displayed includes:

[0023] Obtain initial point cloud data and posture information of multiple frames;

[0024] The initial point cloud data of the multiple frames are superimposed according to the posture information to obtain the full amount of point cloud data.

[0025] In some embodiments, downsampling the full point cloud data to obtain sparse point cloud data includes:

[0026] Segmenting the full point cloud data to obtain multiple sub-areas;

[0027] Determining the distance between each of the sub-areas and the driving trajectory of the collection device;

[0028] Determine a sampling parameter corresponding to each of the sub-areas according to the distance;

[0029] The full point cloud data is downsampled based on the sampling parameters to obtain sparse point cloud data.

[0030] In some embodiments, segmenting the full point cloud data to obtain multiple sub-areas specifically includes:

[0031] The full amount of point cloud data is segmented according to a predetermined area size to obtain a plurality of sub-areas.

[0032] In some embodiments, when executing the operation event, rendering a display page using the sparse point cloud data includes:

[0033] Deleting the cached image;

[0034] The sparse point cloud data is rendered in real time.

[0035] In some embodiments, when the operation event is not executed, generating a cache image corresponding to the display page according to the full point cloud data, and rendering the cache image includes:

[0036] When the operation event is not executed, detecting the cached image;

[0037] In response to the absence of a cached image, creating a cached image;

[0038] In response to the existence of a cached image or the completion of creation of the cached image, the cached image is rendered in real time according to the full point cloud data.

[0039] In a third aspect, an embodiment of the present invention provides a device for processing lane point cloud data, the device comprising:

[0040] The first acquisition unit is used to acquire multiple frames of initial point cloud data collected by the acquisition device on the lane through the laser radar during driving;

[0041] A second acquisition unit is used to fuse the multiple frames of initial point cloud data to obtain a full amount of point cloud data to be displayed;

[0042] a third acquisition unit, configured to divide the full point cloud data into a plurality of sub-regions, and downsample the full point cloud data according to the sub-regions to obtain sparse point cloud data, wherein a sampling parameter of the sub-region is determined according to a distance between the sub-region and a driving track of the acquisition device;

[0043] A first detection unit, used to detect whether an operation event of changing a viewing angle is being performed;

[0044] A first rendering unit, configured to render a display page in real time using the sparse point cloud data when executing the operation event;

[0045] A second rendering unit is used to generate a cache image with a size consistent with the current display page according to the full point cloud data when the operation event is not executed, and to render the cache image in real time;

[0046] The third rendering unit is used to render the lane line marking data input by the user.

[0047] In a fourth aspect, an embodiment of the present invention provides a point cloud data processing device, the device comprising:

[0048] The fourth acquisition unit is used to acquire the full amount of point cloud data to be displayed;

[0049] A fifth acquisition unit, configured to downsample the full point cloud data to acquire sparse point cloud data;

[0050] A second detection unit, used to detect an operation event of a viewing angle change;

[0051] A fourth rendering unit, configured to render a display page using the sparse point cloud data when executing the operation event;

[0052] The fifth rendering unit is used to generate a cache image corresponding to the display page according to the full point cloud data when the operation event is not executed, and render the cache image.

[0053] In a fifth aspect, an embodiment of the present invention provides an electronic device, comprising a memory and a processor, wherein the memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method described in the first aspect and the second aspect.

[0054] In a sixth aspect, an embodiment of the present invention provides a computer program product, wherein the computer program product comprises a computer program. When the computer program runs on a computer, the computer executes the methods described in the first and second aspects above.

[0055] In a seventh aspect, an embodiment of the present invention provides a computer-readable storage medium storing computer program instructions, wherein the computer program instructions, when executed by a processor, implement the method described in the first aspect and the second aspect.

[0056] The technical solution of the embodiment of the present invention obtains the full point cloud data to be displayed, downsamples the full point cloud data to obtain sparse point cloud data, detects the operation event of the camera view angle change, and renders the display page through the sparse point cloud data when the operation event is executed. When the operation event is not executed, a cache image corresponding to the display page is generated according to the full point cloud data, and the cache image is rendered. In this way, the fluency of the point cloud display can be improved, the requirements for computer configuration can be reduced, and at the same time, the details of the point cloud can be retained, and the annotation integrity and accuracy can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The above and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0058] Figure 1 is a flow chart of a method for processing lane point cloud data according to an embodiment of the present invention;

[0059] Figure 2 is a flow chart of obtaining full point cloud data according to an embodiment of the present invention;

[0060] Figure 3 is a schematic diagram of full point cloud data according to an embodiment of the present invention;

[0061] Figure 4 is a flow chart of obtaining sparse point cloud data according to an embodiment of the present invention;

[0062] Figure 5 is a schematic diagram of an operation event of an embodiment of the present invention;

[0063] Figure 6 is a schematic diagram of sparse point cloud data rendered according to an embodiment of the present invention;

[0064] Figure 7 is a schematic diagram of a cached image rendered in an embodiment of the present invention;

[0065] Figure 8 is a flow chart of a method for processing lane point cloud data according to another embodiment of the present invention;

[0066] Fig. 9 is a flow chart of a method for processing point cloud data according to an embodiment of the present invention;

[0067] Fig.10 is a schematic diagram of a device for processing lane point cloud data according to an embodiment of the present invention;

[0068] Fig.11 is a schematic diagram of a point cloud data processing device according to an embodiment of the present invention;

[0069] Fig.12is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0070] The present application is described below based on embodiments, but the present application is not limited to these embodiments. In the detailed description of the present application below, some specific details are described in detail. It is possible for those skilled in the art to fully understand the present application without the description of these details. In order to avoid confusing the essence of the present application, known methods, processes, flows, components and circuits are not described in detail.

[0071] In addition, persons of ordinary skill in the art will appreciate that the drawings provided herein are for illustration purposes and are not necessarily drawn to scale.

[0072] Unless the context clearly requires otherwise, the words "include", "comprising" and similar words throughout the application should be interpreted as including rather than exclusive or exhaustive; that is, the meaning is "including but not limited to".

[0073] In the description of this application, it should be understood that the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of this application, unless otherwise specified, the meaning of "plurality" is two or more.

[0074] The solutions described in this specification and in the examples, if they involve the processing of personal information, will be processed on the premise of having a legal basis (such as obtaining the consent of the subject of personal information, or being necessary for the performance of a contract, etc.), and will only be processed within the scope of regulations or agreements. If a user refuses to process personal information other than the necessary information for basic functions, it will not affect the user's use of basic functions.

[0075] Point cloud is a collection of a large number of three-dimensional coordinate points that represent the geometric shape of an object or environmental surface. LiDAR is one of the main sensors that generate point cloud data. It obtains the distance information of the target by emitting laser pulses and measuring the reflection time. Point cloud data not only contains spatial position, but also can include attributes such as color and reflection intensity. It is widely used in intelligent driving, robot navigation, virtual reality and other fields.

[0076] In the field of intelligent driving, point cloud annotation is an important bridge between raw perception data and high-level algorithm understanding. By marking point cloud data in detail, such as identifying vehicles, pedestrians, obstacles, etc., it can provide training samples for machine learning and deep learning algorithms, and improve the recognition accuracy and decision-making ability of autonomous driving systems. High-quality point cloud annotation is crucial to ensure that the system can safely and effectively cope with various complex road conditions.

[0077] Taking the point cloud annotation in the 4D lane line annotation scenario as an example, 4D lane line annotation is a key task in autonomous driving technology, which involves accurately annotating lane lines in four-dimensional space (3D space + time dimension). Among them, 4D lane line annotation refers to the process of annotating autonomous driving data, not only annotating the position of lane lines in three-dimensional space, but also annotating their changes over time, thus forming a four-dimensional annotation system. With the development of autonomous driving technology, the accurate perception and understanding of the surrounding environment by the perception system has become crucial. As an important part of the road, the accurate annotation of lane lines is of great significance for the path planning and driving safety of autonomous driving vehicles. Traditional 2D or 3D annotation methods cannot fully reflect the dynamic changes of lane lines during actual driving, so 4D annotation methods came into being. 4D annotation includes the position of lane lines in three-dimensional space and their changes over time, which can more comprehensively describe the motion trajectory and morphological changes of lane lines. Through advanced annotation tools and algorithms, 4D annotation can achieve millimeter-level accuracy to ensure the accuracy of data. The changes of lane lines over time are taken into account in the annotation process, and the status of lane lines at different time points can be captured, providing key information for the decision-making of autonomous driving vehicles. Therefore, by accurately marking lane lines and their changes, autonomous vehicles can better identify the road environment and improve driving safety. 4D annotation provides autonomous vehicles with richer road information, helps optimize path planning algorithms, and improves driving efficiency and comfort. In the face of complex scenarios such as occlusion and lane changes, 4D annotation can provide more accurate lane line information to help autonomous vehicles make correct decisions.

[0078] In order to improve the efficiency and quality of point cloud annotation, the industry usually uses multi-frame fusion to process lidar data. Multi-frame fusion integrates data from multiple time points to increase the amount of available information and reduce the noise and missing problems that may exist in a single-frame point cloud, thereby enhancing the capture of object morphology. In addition, the industry has developed two main point cloud display solutions:

[0079] Full point cloud display: Displays all point cloud data, ensuring the completeness of details and the accuracy of annotation. However, this method requires extremely high computer configuration to avoid lag and greatly increases the annotation cost.

[0080] Downsampled point cloud display: Reduce the pressure on computer hardware by reducing the density of point clouds. Although the system requirements are reduced, the sparseness of point clouds leads to loss of details, affecting the integrity and accuracy of annotations, which in turn has an adverse impact on subsequent applications.

[0081] In view of the contradiction between the high computing resource requirements and the inefficient annotation process faced by the prior art in point cloud display and annotation, as well as the loss of details and annotation quality due to downsampling, an embodiment of the present invention proposes a method for processing point cloud data, aiming to optimize the processing flow of point cloud data, reduce system requirements while maintaining high-quality annotation results, thereby meeting the needs of the rapid development of intelligent driving, solving the current problems in point cloud annotation, providing more efficient and accurate annotation tools, and promoting the further development of intelligent driving technology.

[0082] The point cloud data processing method of the embodiment of the present invention is executed by various electronic devices, including memory, processor (CPU, Central Processing Unit), image processor (GPU, Graphics Processing Unit), etc. These devices can run specially designed point cloud processing and annotation software, which are suitable for the needs of different application scenarios. Specifically, the point cloud data processing method can be executed by the following electronic devices:

[0083] Laptops and desktop computers: Equipped with high-performance CPUs, large RAM, and one or more dedicated GPUs, these devices are suitable for local deployment and are particularly suitable for tasks that require a lot of computing resources, such as multi-frame fusion, noise reduction, feature extraction, and object recognition. They can provide sufficient performance to support complex point cloud processing algorithms and can be installed with professional-level point cloud processing software.

[0084] Tablets and high-performance mobile phones: With the advancement of mobile device hardware technology, some high-end tablets and smartphones also have considerable computing and graphics processing capabilities. These devices have built-in efficient CPUs and integrated or independent GPUs, which can perform preliminary point cloud data processing and lightweight annotation tasks in a mobile environment. Although its performance may not be as powerful as that of a desktop system, it is sufficient for quick on-site preview and simple editing.

[0085] Edge computing devices: Small, low-power but high-performance computing units designed for smart driving and other IoT applications. These devices typically integrate the latest AI acceleration chips and can process point cloud data in real time close to the data source, reducing latency and optimizing bandwidth usage.

[0086] Furthermore, point cloud annotation software is installed on the electronic device. Point cloud annotation software is a tool specially designed for processing and annotating three-dimensional point cloud data. It is widely used in many fields such as autonomous driving, robot navigation, geographic information system, building information modeling, archeology, industrial inspection, etc. Through this software, users can accurately annotate a large amount of three-dimensional point cloud data collected from sensors such as lidar and depth cameras, providing basic support for subsequent data analysis, model training or application development.

[0087] Figure 1 It is a flowchart of a method for processing lane point cloud data according to an embodiment of the present invention. Figure 1 The lane point cloud data processing method shown is executed by a processor (CPU and / or GPU), and specifically includes the following steps:

[0088] Step S110: Obtain full point cloud data.

[0089] In this embodiment, the full amount of point cloud data is point cloud data obtained by fusion of multiple frames of point cloud data.

[0090] Specifically, Figure 2 FIG. 1 is a flowchart of obtaining full point cloud data according to an embodiment of the present invention. Figure 2 As shown, obtaining full point cloud data includes the following steps:

[0091] Step S111, obtaining multiple frames of initial point cloud data and posture information.

[0092] In this embodiment, multiple frames of initial point cloud data collected by the acquisition device through the laser radar on the lane during driving, as well as the posture information corresponding to each frame of initial point cloud data, are obtained. Point cloud annotation software is installed on the electronic device, and the user selects the point cloud data to be annotated by operating the point cloud annotation software. The point cloud annotation software can obtain the initial point cloud data and posture information based on the point cloud data selected by the user. Among them, the point cloud annotation software can obtain the initial point cloud data and posture information from the local memory, or receive the initial point cloud data and posture information sent by other devices (such as cloud servers).

[0093] Among them, the initial point cloud data and attitude information are data acquired by the acquisition device. Specifically, the embodiment of the present invention is described by taking the acquisition device as a vehicle as an example. A laser radar is installed on the vehicle, and the laser radar is usually installed at a predetermined position of the vehicle to ensure the maximum scanning range and field of view coverage. The laser radar quickly emits short-pulse laser beams in a rotating or fixed direction, and measures the time difference of these beams reflected from the surface of the object (time of flight method), thereby calculating the distance of the target and forming the initial point cloud data. The attitude information includes acceleration, angular velocity, position information, etc. Specifically, the accelerometer and gyroscope in the inertial measurement unit (IMU) measure the acceleration and angular velocity changes of the vehicle in real time to provide high-frequency attitude updates. The geographic location of the vehicle is determined by satellite signals through a receiver of the global positioning system (GPS, Global Positioning System). Further, the point cloud data of the laser radar is combined with the data of sensors such as IMU and GPS to update the position and attitude information of the vehicle in real time. The posture information generally includes position (x, y, z) and rotation (roll, pitch, yaw or quaternion), which are used to describe the position and direction of the sensor relative to the global coordinate system. The embodiment of the present invention mainly marks the lane line, so the setting direction of the laser radar is towards the lane.

[0094] In some embodiments, the collected original point cloud data can be cleaned to remove noise points or outliers to ensure data quality. The data of all sensors are unified into the same coordinate system to ensure the consistency of the initial point cloud data and posture information.

[0095] Thus, the initial point cloud data and posture information can be obtained through the acquisition device, and the initial point cloud data and posture information can be sent to the cloud server. The electronic device installed with the point cloud annotation software can obtain the initial point cloud data and posture information from the cloud server in advance and store it in the local memory. At this time, the point cloud annotation software can obtain the initial point cloud data and posture information from the local memory. Alternatively, the electronic device installed with the point cloud annotation software can directly obtain the initial point cloud data and posture information from the cloud server. At this time, the point cloud annotation software receives the initial point cloud data and posture information sent by the cloud server through the network.

[0096] Step S112: superimpose the multiple frames of initial point cloud data according to the posture information to obtain the full amount of point cloud data.

[0097] In this embodiment, the multiple frames of initial point cloud data are fused to obtain the full amount of point cloud data to be displayed. Specifically, the initial point cloud data of each frame is converted from the sensor coordinate system to the global coordinate system using the posture information of each frame. The multiple frames of point cloud data converted to the global coordinate system are merged. In some cases, there may be repeated areas in the point clouds between adjacent frames. Certain algorithms can be used to remove duplicate points, such as a spatial index structure based on a KD tree, or downsampling the point cloud using methods such as voxel filtering. Finally, the merged super-large point cloud data (that is, the full amount of point cloud data) is stored in a predetermined format.

[0098] in, Figure 3 Schematic diagram of the full point cloud data of an embodiment of the present invention. Figure 3 As shown, the full point cloud data of the embodiment of the present invention is mainly collected for lane lines. It should be noted that since the collection range of the full point cloud data is relatively wide, in order to facilitate viewing, Figure 3 Only a portion of the full point cloud data is shown as an example.

[0099] Step S120: Acquire sparse point cloud data.

[0100] In this embodiment, the full point cloud data is downsampled to obtain sparse point cloud data. Specifically, the full point cloud data is divided into multiple sub-areas, and the full point cloud data is downsampled according to the sub-areas to obtain sparse point cloud data, and the sampling parameters of the sub-areas are determined according to the distance between the sub-areas and the driving track of the acquisition device.

[0101] Figure 4 FIG. 1 is a flowchart of obtaining sparse point cloud data according to an embodiment of the present invention. Figure 4 As shown, downsampling the full point cloud data to obtain sparse point cloud data includes the following steps:

[0102] Step S121: segment the full point cloud data to obtain multiple sub-areas.

[0103] In this embodiment, the full point cloud data is segmented according to a predetermined region size to obtain a plurality of sub-regions. The predetermined region size may be in various forms, for example, the predetermined region size is a predetermined volume, such as one region for every 10 cubic meters.

[0104] Therefore, by breaking down the full point cloud dataset into smaller, manageable sub-regions, computing tasks can be processed in parallel on distributed systems, or processed incrementally on devices with limited memory. At the same time, smaller data blocks are easier to load into memory quickly for processing, reducing the time required to read large files, and only data from specific areas can be loaded as needed. Moreover, different downsampling strategies can be applied based on the characteristics of each sub-region, thereby reducing the amount of data without affecting key information.

[0105] Step S122: determining the distance between each of the sub-areas and the driving track of the collection device.

[0106] In this embodiment, the distance between each of the sub-areas and the driving track of the collection device is: the shortest distance between each point in each of the sub-areas and each point on the driving track of the collection device.

[0107] Specifically, the points in the sub-area are taken as the first candidate points, the points on the driving track of the collection device are determined as the second candidate points, the distances between the first candidate points and the second candidate points are determined, and the shortest distance is determined as the distance between the sub-area and the driving track of the collection device. The above steps are performed for each sub-area respectively to obtain the distance corresponding to each sub-area.

[0108] Step S123: Determine a sampling parameter corresponding to each of the sub-areas according to the distance.

[0109] In this embodiment, the sampling parameters corresponding to each sub-area are determined according to the distance between each sub-area and the driving track of the collection device. The sampling parameters include the downsampling ratio and the parameter value of the Bloom filter. The downsampling ratio is used to characterize the proportion of the original data retained. The higher the downsampling ratio, the more data points are retained from the original data.

[0110] Specifically, different downsampling ratios are set based on the distance between the sub-area and the driving trajectory of the collection device. For example, the closer the distance, the more points are retained (the larger the downsampling ratio); the farther the distance, the fewer points are retained (the smaller the downsampling ratio). At the same time, a Bloom filter with different parameter values ​​is created to estimate the existence of points, thereby reducing storage requirements. Among them, the Bloom filter can be configured according to the expected false alarm rate and number of elements.

[0111] Among them, the parameter value of the Bloom filter is the size of the Bloom filter, that is, the length of the internal bit array of the Bloom filter. The length of the internal bit array is usually expressed in bits. The size of the Bloom filter directly affects the false positive rate, storage efficiency, number of hash functions and other performance of the Bloom filter. One feature of the Bloom filter is that there may be false positives (i.e., it is believed that an element exists but it actually does not exist), but there will be no false negatives (i.e., it is believed that an element does not exist but it actually exists). The size of the Bloom filter is inversely proportional to the false positive rate. The larger the bit array, the lower the false positive rate. Larger Bloom filters require more memory space, but can accommodate more elements or reduce the false positive rate. Smaller Bloom filters save more memory, but have a higher false positive rate for the same number of elements. The Bloom filter uses multiple independent hash functions to map elements to different positions in the bit array. Generally speaking, the size of the Bloom filter determines the maximum number of hash functions that can be used, because each hash function occupies a part of the bit array space.

[0112] To determine the size m of the Bloom filter (the length of the bit array) and the number k of hash functions, calculations can be performed based on the expected number of elements to be inserted n and the acceptable false positive rate p.

[0113] Among them, the calculation formula of the size m of the Bloom filter is as follows:

[0114]

[0115] Wherein, n is the number of points in the sub-region, p is the preset false alarm rate, m is the size of the Bloom filter, and 0<p<1.

[0116] If m is not an integer, m needs to be rounded up to obtain the size of the Bloom filter.

[0117] Step S124: downsample the full point cloud data based on the sampling parameters to obtain sparse point cloud data.

[0118] In this embodiment, for each sub-area, the corresponding downsampling ratio is determined according to its distance from the driving trajectory of the acquisition device, and then according to the downsampling ratio, random sampling, uniform distribution or other suitable algorithms are used to select which points should be retained in the final sparse point cloud. At the same time, during the downsampling process, Bloom filters are used to quickly determine whether a point has been selected to avoid repeated selection. At the same time, Bloom filters can help save memory when processing a large number of points because they do not need to store the actual point positions, but only store the probability information of whether these points exist. Afterwards, the points in all the downsampled sub-areas are merged into a new point cloud data set, namely the sparse point cloud data. This process should keep the relative positions between the sub-areas unchanged to ensure the consistency of the spatial relationship.

[0119] By setting an appropriate downsampling ratio, the number of data points in the point cloud data can be selectively reduced, thereby reducing the size of the data for subsequent processing. For each downsampled sub-region, create an independent Bloom filter. Add the data points retained after downsampling to the corresponding Bloom filter. When it is necessary to query whether a specific location point exists, first check the corresponding Bloom filter. If the Bloom filter returns "may exist in the set", further verify the actual existence of the point; if it returns "definitely does not exist in the set", the point can be directly excluded.

[0120] In summary, before applying downsampling to the original point cloud data, the distance between each sub-region and the driving trajectory is calculated, and an appropriate downsampling ratio is selected based on the distance. Then, the point cloud data in each sub-region is downsampled according to the selected ratio. During or after the downsampling process, the downsampled points are added to the corresponding Bloom filter, which enables the Bloom filter to be used for subsequent quick queries on whether a specific point exists. The closer the sub-region is, the more data points need to be saved, so its Bloom filter may require a larger bit array and more hash functions to maintain a low false alarm rate. For sub-regions that are farther away, due to the higher downsampling ratio and fewer data points, a smaller Bloom filter can be used, and a slightly higher false alarm rate can be tolerated.

[0121] Step S130: the operation event is being executed.

[0122] In this embodiment, in the point cloud annotation software, the user interface usually includes an area dedicated to displaying point cloud data. In this display area, the user can perform a variety of interactive operations, such as zooming, rotating, and dragging, to adjust the viewing angle to better observe and annotate the point cloud data. In the embodiment of the present invention, these operations that cause changes in viewing angle are defined as "operation events". Continuously detect whether there are such operation events. Once it is detected that the user has performed any operation that affects the change in viewing angle, the event will be responded to and recorded immediately to ensure that all user viewing angle adjustments can be processed in a timely manner.

[0123] During the point cloud display process, there are two time intervals according to user interaction:

[0124] First time interval (dynamic display): When the user is performing an operation event, real-time rendering is performed through sparse point cloud data to ensure a high frame rate and smooth user experience.

[0125] Second time interval (static display): When the user does not perform any operation events, the display is achieved by rendering the cached images, which reduces computing resource consumption and improves rendering efficiency.

[0126] When the user starts a new operation event, the system will switch from the second time interval back to the first time interval. When entering the first time interval, in order to reduce memory usage, the system will delete the previously created cache images. This ensures that when the system needs to process a large amount of point cloud data, the performance will not be affected by the cache images occupying too much memory. When the user stops the operation event, the system will switch to the second time interval and start the process of creating cache images. The cache image will calculate information such as lighting, shadows, and texture mapping based on the full point cloud data during the first rendering, and store this information in the cache image. Subsequent rendering can directly use this pre-calculated information to further optimize performance.

[0127] Figure 5 FIG. 1 is a schematic diagram of an operation event of an embodiment of the present invention. Figure 5 As shown, the horizontal axis represents time t, and the vertical axis represents whether an operation event is being executed. The event of executing an operation is marked as 1, and the event of not executing an operation is marked as 0. Figure 5 In the embodiment shown, t 0 Start displaying point cloud data at time t 0 -t 1 No operation event is executed within the time interval; t 1 The time to start executing the operation event, t 2 The operation ends at time t 2 -t 3 No operation event is executed within the time interval; t 3 The time to start executing the operation event, t 4The operation ends at time t 4 -t 5 No operation event is executed within the time interval; t 5 The time to start executing the operation event, t 6 The moment operation ends.

[0128] The first time interval includes t 1 -t 2 ,t 3 -t 4 ,t 5 -t 6 The second time interval includes t 0 -t 1 ,t 2 -t 3 ,t 4 -t 5 .

[0129] In response to the operation event being executed, the process proceeds to steps S140 - S150 , where a display page is rendered using the sparse point cloud data.

[0130] In response to no operation event being executed, the process proceeds to steps S160 - S190 , generating a cached image corresponding to the display page according to the full point cloud data, and rendering the cached image.

[0131] Step S140: Delete cached images.

[0132] In this embodiment, the point cloud display process is divided into two time intervals according to user interaction conditions: an operation event period (a first time interval) and a no operation event period (a second time interval), and these two time intervals are executed alternately.

[0133] During the operation event (first time interval): When the user performs an operation event (such as perspective change operations such as zooming, rotating, or dragging), real-time rendering is performed through sparse point cloud data to ensure the immediacy and smoothness of the interaction.

[0134] No operation event period (second time interval): During the time period when the user does not perform any operation event, the cached image is rendered to achieve display, thereby improving rendering efficiency and reducing the consumption of computing resources.

[0135] Especially when switching from the second time interval to the first time interval, since cached images have been created and used in the second time interval, these cached images are deleted when entering the first time interval to reduce memory usage. This mechanism not only optimizes memory management, but also ensures high performance of the system when processing large amounts of point cloud data, thereby providing a smooth and efficient user experience.

[0136] by Figure 5 As an example, assume that the current time is t 3 time, because at t 2 -t 3 The page is rendered and displayed by using the cached image during the time interval. Therefore, there is a cached image at time t3 and it needs to be deleted. 3 -t 4 At other times within the time interval, there is no need to delete the cached images.

[0137] Step S150: Rendering sparse point cloud data in real time.

[0138] In this embodiment, when an operation event initiated by a user is being processed, a display page is rendered using the sparse point cloud data.

[0139] Specifically, when the user performs an operation event, it dynamically switches to pre-generated sparse point cloud data for rendering. Sparse point cloud data retains the key features of full point cloud data, but greatly reduces the number of points that need to be processed and displayed, thereby achieving a higher frame rate (FPS) and ensuring a smooth and non-stuttering user experience. In addition, this sparse point cloud-based rendering method can effectively reduce the consumption of computing resources without affecting the user experience, making it possible to respond to each user operation more efficiently and provide a smoother and more intuitive interactive experience.

[0140] Furthermore, real-time rendering is performed while processing user-initiated operation events, ensuring that each interaction is immediately reflected in the interface. In order to ensure a smooth user experience, the frames per second (FPS) of the picture are optimized to always maintain a high level. For example, FPS>30 improves the immediacy and responsiveness of the interaction, and ensures that users can enjoy a smooth, non-stuttering visual experience even in complex scenarios.

[0141] Figure 6 Schematic diagram of sparse point cloud data rendered in an embodiment of the present invention. Figure 3 Compared with the full point cloud data shown in FIG. 1 , the sparse point cloud data is a part of the full point cloud data. Meanwhile, the points in the sparse point cloud data are obviously less than those in the full point cloud data. Figure 6 The yellow lines in the figure are marked lane lines.

[0142] It should be noted that in actual display, the scene displayed by the sparse point cloud data is part of the full point cloud data. Figure 3 Only a part of the scene of the full point cloud data is shown, so Figure 6 In the scene with sparse point cloud data shown in Figure 1, part of it has been shown in Figure 3 The other part is not shown in Figure 3 middle.

[0143] Step S160: There is a cached image.

[0144] In this embodiment, when it is detected that no operation event initiated by the user is currently being executed, it is checked whether there is a cached image for use.

[0145] by Figure 5 As an example, assume that the current time is t 4 time, because at t 3 -t 4 During the time interval, the page is rendered in real time using sparse point cloud data. Therefore, due to the continuous real-time rendering during this period, no cached images are created, resulting in no available cached images at this time. It is necessary to create cached images. In this case, at t 4 At this moment, a process is started to create a cached image. This step is to prepare static images for the next time interval so that they can be quickly loaded and displayed when the user is not operating, thereby reducing unnecessary repeated rendering and saving computing resources. Once the cached image is created, at t 4 -t 5 At other times in the time interval, if it is detected that no new operation events have occurred, the cached images that have been created can be directly used for display. This not only improves the response speed of the electronic device, but also maintains the consistency of the visual effects, while reducing the load on the CPU and GPU.

[0146] Step S170: Create a cached image.

[0147] In this embodiment, in response to the absence of a cached picture, a cached picture is created.

[0148] Specifically, when it is detected that there is no cached image, the process of creating a cached image will be started. First, determine the area in the user interface of the point cloud annotation software that displays the point cloud data. This step ensures that the cached image can accurately reflect the view range currently viewed by the user. And measure the current scene size (including width and height) of the area. This step is the key to ensuring that the created cached image exactly matches what is displayed on the user interface. Then, based on the measured scene size above, create a blank cached image that matches it. At this point, the cached image is an unrendered image and only serves as a ready container, waiting for subsequent real-time rendering updates based on the full amount of point cloud data. In this way, it can be ensured that cached images that match the current view can be quickly generated when needed, thereby optimizing rendering performance and user experience.

[0149] Step S180: Render cached images in real time.

[0150] In this embodiment, in response to the existence of a cached image or the completion of creation of a cached image, the cached image is rendered in real time according to the full point cloud data.

[0151] Specifically, during the time interval when no operation events are performed, an efficient caching and rendering mechanism is started to optimize performance. Specifically, when rendering the first frame, information such as lighting, shadows, and texture mapping is calculated based on the full point cloud data, and this information is stored in the cached image. Then, the cached image is rendered for the first time using these calculation results to ensure the accuracy and details of the displayed content. When rendering the second frame and subsequent frames, since information such as lighting, shadows, and texture mapping has been stored in the cached image, there is no need to repeatedly calculate these complex information, but directly use the pre-calculated results in the cached image for fast rendering. This not only improves rendering efficiency and maintains a high frame rate (FPS), but also reduces CPU and GPU resource consumption, providing a smooth and consistent user experience. In this way, electronic devices can greatly improve rendering speed and performance without affecting visual effects, especially when there is no user interaction for a long time, effectively saving computing resources while maintaining high-quality visual output.

[0152] Figure 7 is a schematic diagram of a cached image rendered in an embodiment of the present invention. Figure 7 As shown in the figure, the rendered cached image and the full point cloud data are consistent in user perception and can maintain high-quality visual output, but rendering by cached images can save a lot of computing resources.

[0153] Step S190: Rendering the annotation data.

[0154] In this embodiment, the lane line annotation data input by the user is rendered. The lane line annotation data includes other 3D objects in the annotation interface, such as lines, polygons, point rectangles, etc. Specifically, the user will change the geometric properties of the object in real time during the annotation process, and the interface needs to give the user real-time page feedback, so the annotation data is rendered in the display page.

[0155] It should be noted that in the embodiment of the present invention, during the time interval when the operation event is being executed, the user will not make annotations. Therefore, by rendering the display page with sparse point cloud data, the operation jamming problem caused by the huge amount of point cloud data can be avoided. At the same time, during the time interval when no operation event is executed, the user may make annotations. By creating a cached image and rendering the cached image based on the full point cloud data, the obtained static image can truly reflect the overall picture of the annotation scene, so that the user can quickly and accurately identify the object to be annotated, and improve the efficiency and accuracy of annotation. Moreover, the full point cloud data is cached as a picture. When the full point cloud needs to be rendered, the picture is actually rendered. Compared with directly rendering the point cloud, a large amount of calculations can be completed in advance, and the effect of real-time rendering can be achieved, and the computer does not require a high configuration.

[0156] The embodiment of the present invention obtains the full amount of point cloud data to be displayed, downsamples the full amount of point cloud data to obtain sparse point cloud data, detects the operation event of the camera viewing angle change, and renders the display page through the sparse point cloud data when the operation event is executed. When the operation event is not executed, a cache image corresponding to the display page is generated according to the full amount of point cloud data, and the cache image is rendered. In this way, the smoothness of the point cloud display can be improved, the requirements for computer configuration can be reduced, and at the same time, the details of the point cloud can be retained, and the annotation integrity and accuracy can be improved.

[0157] Figure 8 FIG. 1 is a flow chart of a method for processing lane point cloud data according to another embodiment of the present invention. Figure 8 As shown, the method for processing lane point cloud data in an embodiment of the present invention includes the following steps:

[0158] Step S210: Acquire multiple frames of initial point cloud data of the lane collected by the acquisition device through the laser radar during driving.

[0159] Step S220: Fusing the multiple frames of initial point cloud data to obtain the full amount of point cloud data to be displayed.

[0160] Step S230: divide the full point cloud data into multiple sub-areas, and downsample the full point cloud data according to the sub-areas to obtain sparse point cloud data, wherein the sampling parameters of the sub-areas are determined according to the distance between the sub-areas and the driving trajectory of the acquisition device.

[0161] Step S240: Detect whether a viewing angle change operation event is being executed.

[0162] Step S250: When executing the operation event, a display page is rendered in real time using the sparse point cloud data.

[0163] Step S260: When the operation event is not executed, a cache image having a size consistent with the current display page is generated according to the full point cloud data, and the cache image is rendered in real time.

[0164] Step S270: Rendering the lane line marking data input by the user.

[0165] In some embodiments, the sampling parameters include a downsampling ratio and a parameter value of a Bloom filter.

[0166] In some embodiments, the distance between the sub-area and the driving trajectory of the collection device is the shortest distance between each point in each sub-area and each point on the driving trajectory of the collection device.

[0167] In some embodiments, when executing the operation event, rendering a display page in real time using the sparse point cloud data includes:

[0168] Deleting the cached image;

[0169] The sparse point cloud data is rendered in real time.

[0170] In some embodiments, when the operation event is not executed, generating a cache image having a size consistent with the current display page according to the full point cloud data, and rendering the cache image in real time includes:

[0171] When the operation event is not executed, detecting the cached image;

[0172] In response to the absence of a cached image, creating a cached image;

[0173] In response to the existence of a cached image or the completion of creation of the cached image, the cached image is rendered in real time according to the full point cloud data.

[0174] The embodiment of the present invention obtains the full amount of point cloud data to be displayed, downsamples the full amount of point cloud data to obtain sparse point cloud data, detects the operation event of the camera viewing angle change, and renders the display page through the sparse point cloud data when the operation event is executed. When the operation event is not executed, a cache image corresponding to the display page is generated according to the full amount of point cloud data, and the cache image is rendered. In this way, the smoothness of the point cloud display can be improved, the requirements for computer configuration can be reduced, and at the same time, the details of the point cloud can be retained, and the annotation integrity and accuracy can be improved.

[0175] Fig. 9 is a flow chart of a method for processing point cloud data according to an embodiment of the present invention. Fig. 9As shown, the point cloud data processing method can be applied to the scene of lane line point cloud annotation, and can also be applied to other scenes involving point cloud processing such as point cloud annotation or point cloud display, and specifically includes the following steps:

[0176] Step S310: Obtain the full amount of point cloud data to be displayed.

[0177] Step S320: downsample the full point cloud data to obtain sparse point cloud data.

[0178] Step S330: Detect an operation event of a viewing angle change.

[0179] Step S340: When executing the operation event, a display page is rendered using the sparse point cloud data.

[0180] Step S350: When the operation event is not executed, a cache image corresponding to the display page is generated according to the full point cloud data, and the cache image is rendered.

[0181] In some embodiments, obtaining the full amount of point cloud data to be displayed includes:

[0182] Obtain initial point cloud data and posture information of multiple frames;

[0183] The initial point cloud data of the multiple frames are superimposed according to the posture information to obtain the full amount of point cloud data.

[0184] In some embodiments, downsampling the full point cloud data to obtain sparse point cloud data includes:

[0185] Segmenting the full point cloud data to obtain multiple sub-areas;

[0186] Determining the distance between each of the sub-areas and the driving trajectory of the collection device;

[0187] Determine a sampling parameter corresponding to each of the sub-areas according to the distance;

[0188] The full point cloud data is downsampled based on the sampling parameters to obtain sparse point cloud data.

[0189] In some embodiments, segmenting the full point cloud data to obtain multiple sub-areas specifically includes:

[0190] The full amount of point cloud data is segmented according to a predetermined area size to obtain a plurality of sub-areas.

[0191] In some embodiments, the determining of the distance between each of the sub-areas and the driving track of the collection device is specifically:

[0192] The shortest distance between each point in each sub-area and each point on the driving trajectory of the collection device is determined.

[0193] In some embodiments, the sampling parameters include a downsampling ratio and a parameter value of a Bloom filter.

[0194] In some embodiments, when executing the operation event, rendering a display page using the sparse point cloud data includes:

[0195] Deleting the cached image;

[0196] The sparse point cloud data is rendered in real time.

[0197] In some embodiments, when the operation event is not executed, generating a cache image corresponding to the display page according to the full point cloud data, and rendering the cache image includes:

[0198] When the operation event is not executed, detecting the cached image;

[0199] In response to the absence of a cached image, creating a cached image;

[0200] In response to the existence of a cached image or the completion of creation of the cached image, the cached image is rendered in real time according to the full point cloud data.

[0201] In some embodiments, the method further comprises:

[0202] Render the lane marking data input by the user.

[0203] The embodiment of the present invention obtains the full amount of point cloud data to be displayed, downsamples the full amount of point cloud data to obtain sparse point cloud data, detects the operation event of the camera viewing angle change, and renders the display page through the sparse point cloud data when the operation event is executed. When the operation event is not executed, a cache image corresponding to the display page is generated according to the full amount of point cloud data, and the cache image is rendered. In this way, the smoothness of the point cloud display can be improved, the requirements for computer configuration can be reduced, and at the same time, the details of the point cloud can be retained, and the annotation integrity and accuracy can be improved.

[0204] Fig.10 Schematic diagram of a device for processing lane point cloud data according to an embodiment of the present invention. Fig.10As shown, the processing device for lane point cloud data includes a first acquisition unit 51, a second acquisition unit 52, a third acquisition unit 53, a first detection unit 54, a first rendering unit 55, a second rendering unit 56, and a third rendering unit 57. Among them, the first acquisition unit 51 is used to acquire multiple frames of initial point cloud data collected by the acquisition device on the lane through the laser radar during driving. The second acquisition unit 52 is used to fuse the multiple frames of initial point cloud data to obtain the full point cloud data to be displayed. The third acquisition unit 53 is used to divide the full point cloud data into multiple sub-areas, and downsample the full point cloud data according to the sub-areas to obtain sparse point cloud data, and the sampling parameters of the sub-areas are determined according to the distance between the sub-areas and the driving track of the acquisition device. The first detection unit 54 is used to detect whether an operation event of changing the viewing angle is being executed. The first rendering unit 55 is used to render the display page in real time through the sparse point cloud data when the operation event is executed. The second rendering unit 56 is used to generate a cache image with the same size as the current display page according to the full point cloud data when the operation event is not executed, and render the cache image in real time. The third rendering unit 57 is used to render the lane line marking data input by the user.

[0205] The embodiment of the present invention obtains the full amount of point cloud data to be displayed, downsamples the full amount of point cloud data to obtain sparse point cloud data, detects the operation event of the camera viewing angle change, and renders the display page through the sparse point cloud data when the operation event is executed. When the operation event is not executed, a cache image corresponding to the display page is generated according to the full amount of point cloud data, and the cache image is rendered. In this way, the smoothness of the point cloud display can be improved, the requirements for computer configuration can be reduced, and at the same time, the details of the point cloud can be retained, and the annotation integrity and accuracy can be improved.

[0206] Fig.11 Schematic diagram of a point cloud data processing device according to an embodiment of the present invention. Fig.11 As shown, the processing device of point cloud data includes a fourth acquisition unit 61, a fifth acquisition unit 62, a second detection unit 63, a fourth rendering unit 64 and a fifth rendering unit 65. The fourth acquisition unit 61 is used to acquire the full point cloud data to be displayed. The fifth acquisition unit 62 is used to downsample the full point cloud data to obtain sparse point cloud data. The second detection unit 63 is used to detect an operation event of a change in viewing angle. The fourth rendering unit 64 is used to render a display page using the sparse point cloud data when the operation event is executed. The fifth rendering unit 65 is used to generate a cached image corresponding to the display page according to the full point cloud data when the operation event is not executed, and render the cached image.

[0207] The embodiment of the present invention obtains the full amount of point cloud data to be displayed, downsamples the full amount of point cloud data to obtain sparse point cloud data, detects the operation event of the camera viewing angle change, and renders the display page through the sparse point cloud data when the operation event is executed. When the operation event is not executed, a cache image corresponding to the display page is generated according to the full amount of point cloud data, and the cache image is rendered. In this way, the smoothness of the point cloud display can be improved, the requirements for computer configuration can be reduced, and at the same time, the details of the point cloud can be retained, and the annotation integrity and accuracy can be improved.

[0208] Fig.12 Schematic diagram of an electronic device according to an embodiment of the present invention. In this embodiment, the electronic device 7 includes a server, a terminal, etc. Fig.12 As shown, the electronic device 7 includes: at least one processor 71; a memory 72 connected to the at least one processor 71 for communication; and a communication component 73 connected to the scanning device for communication, the communication component 73 receives and sends data under the control of the processor 71; wherein the memory 72 stores instructions executable by at least one processor 71, and the instructions are executed by at least one processor 71 to implement a method for processing point cloud data.

[0209] Specifically, the electronic device includes: one or more processors 71 and a memory 72, Fig.12 A processor 71 is taken as an example. The processor 71 and the memory 72 may be connected via a bus or other means. Fig.12 The example of the bus connection is taken as an example. The memory 72 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules. The processor 71 executes various functional applications and data processing of the device by running the non-volatile software programs, instructions and modules stored in the memory 72, that is, realizing the above-mentioned point cloud data processing method.

[0210] The memory 72 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and applications required for at least one function; the data storage area may store a list of options, etc. In addition, the memory 72 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 72 may optionally include a memory remotely arranged relative to the processor 71, and these remote memories may be connected to an external device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0211] One or more modules are stored in the memory 72, and when executed by one or more processors 71, the point cloud data processing method in any of the above method embodiments is executed.

[0212] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.

[0213] The embodiment of the present invention obtains the full amount of point cloud data to be displayed, downsamples the full amount of point cloud data to obtain sparse point cloud data, detects the operation event of the camera viewing angle change, and renders the display page through the sparse point cloud data when the operation event is executed. When the operation event is not executed, a cache image corresponding to the display page is generated according to the full amount of point cloud data, and the cache image is rendered. In this way, the smoothness of the point cloud display can be improved, the requirements for computer configuration can be reduced, and at the same time, the details of the point cloud can be retained, and the annotation integrity and accuracy can be improved.

[0214] Another embodiment of the present invention relates to a non-volatile storage medium for storing a computer-readable program, wherein the computer-readable program is used for a computer to execute part or all of the above method embodiments.

[0215] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0216] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for processing lane point cloud data, characterized in that: The method comprises: Acquire multiple frames of initial point cloud data collected by the acquisition device on the lane through the laser radar during driving; Fusing the multiple frames of initial point cloud data to obtain full point cloud data to be displayed; The full point cloud data is divided into a plurality of sub-areas, and the full point cloud data is downsampled according to the sub-areas to obtain sparse point cloud data, wherein the sampling parameters of the sub-areas are determined according to the distance between the sub-areas and the driving track of the acquisition device; Detect whether a perspective change operation event is being performed; When the operation event is executed, a display page is rendered in real time using the sparse point cloud data; When the operation event is not executed, a cache image having a size consistent with the current display page is generated according to the full point cloud data, and the cache image is rendered in real time; Render the lane marking data input by the user.

2. The method according to claim 1, characterized in that The sampling parameters include a downsampling ratio and a parameter value of a Bloom filter.

3. The method according to claim 1, characterized in that The distance between the sub-areas and the driving track of the collection device is the shortest distance between each point in each sub-area and each point on the driving track of the collection device.

4. A method for processing point cloud data, characterized in that: The method comprises: Get the full amount of point cloud data that needs to be displayed; Downsampling the full point cloud data to obtain sparse point cloud data; Detecting the operation event of view change; When executing the operation event, rendering a display page through the sparse point cloud data; When the operation event is not executed, a cache image corresponding to the display page is generated according to the full point cloud data, and the cache image is rendered.

5. The method according to claim 4, characterized in that The acquisition of the full amount of point cloud data to be displayed includes: Obtain initial point cloud data and posture information of multiple frames; The initial point cloud data of the multiple frames are superimposed according to the posture information to obtain the full amount of point cloud data.

6. The method according to claim 4, characterized in that The downsampling of the full point cloud data to obtain sparse point cloud data includes: Segmenting the full point cloud data to obtain multiple sub-areas; Determining the distance between each of the sub-areas and the driving trajectory of the collection device; Determine a sampling parameter corresponding to each of the sub-areas according to the distance; The full point cloud data is downsampled based on the sampling parameters to obtain sparse point cloud data.

7. The method according to claim 6, characterized in that The segmentation of the full point cloud data to obtain multiple sub-areas is specifically as follows: The full amount of point cloud data is segmented according to a predetermined area size to obtain a plurality of sub-areas.

8. The method according to claim 4, characterized in that When executing the operation event, rendering and displaying a page by using the sparse point cloud data includes: Deleting the cached image; The sparse point cloud data is rendered in real time.

9. The method according to claim 4, characterized in that When the operation event is not executed, generating a cache image corresponding to the display page according to the full point cloud data, and rendering the cache image includes: When the operation event is not executed, detecting the cached image; In response to the absence of a cached image, creating a cached image; In response to the existence of a cached image or the completion of creation of the cached image, the cached image is rendered in real time according to the full point cloud data.

10. A lane point cloud data processing device, characterized in that: The device comprises: The first acquisition unit is used to acquire multiple frames of initial point cloud data collected by the acquisition device on the lane through the laser radar during driving; A second acquisition unit is used to fuse the multiple frames of initial point cloud data to obtain a full amount of point cloud data to be displayed; a third acquisition unit, configured to divide the full point cloud data into a plurality of sub-regions, and downsample the full point cloud data according to the sub-regions to obtain sparse point cloud data, wherein a sampling parameter of the sub-region is determined according to a distance between the sub-region and a driving track of the acquisition device; A first detection unit, used to detect whether an operation event of changing a viewing angle is being performed; A first rendering unit, configured to render a display page in real time using the sparse point cloud data when executing the operation event; A second rendering unit is used to generate a cache image with a size consistent with the current display page according to the full point cloud data when the operation event is not executed, and to render the cache image in real time; The third rendering unit is used to render the lane line marking data input by the user.

11. A point cloud data processing device, characterized in that: The device comprises: The fourth acquisition unit is used to acquire the full amount of point cloud data to be displayed; A fifth acquisition unit, configured to downsample the full point cloud data to acquire sparse point cloud data; A second detection unit, used to detect an operation event of a viewing angle change; A fourth rendering unit, configured to render a display page using the sparse point cloud data when executing the operation event; The fifth rendering unit is used to generate a cache image corresponding to the display page according to the full point cloud data when the operation event is not executed, and render the cache image.

12. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method according to any one of claims 1 to 9.

13. A computer program product, comprising a computer program, characterized in that: When the computer program is executed on a computer, the computer executes the method according to any one of claims 1 to 9.

14. A computer-readable storage medium storing computer program instructions, characterized in that: The computer program instructions, when executed by a processor, implement the method according to any one of claims 1 to 9.

Images