Point cloud data processing method, device, non-volatile storage medium and processor

Abstract

The present invention discloses a point cloud data processing method, device, non-volatile storage medium and processor. The point cloud data processing method includes: collecting multiple frames of point cloud data reflected by a target object through a laser radar installed on an autonomous driving vehicle, obtaining any frame of point cloud data from the multiple frames of point cloud data, and based on the arbitrary frame of point cloud data, fusing a predetermined number of frames of point cloud data to obtain fused point cloud fusion data; converting the point cloud fusion data into a two-dimensional pseudo image; using a convolutional neural network detection head to detect the two-dimensional pseudo image to obtain physical feature information of the target object, and determining the tracking ID of the target object based on the two-dimensional pseudo image; and determining the spatial position of the target object based on the physical feature information and the tracking ID of the target object. The present invention solves the technical problem that the tracking method in the prior art depends on the rules designed for specific scenarios, which makes it easy to be restricted in use in various scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud data processing, and in particular to a point cloud data processing method, device, non-volatile storage medium and processor. Background Art

[0002] Existing 3D object tracking algorithms used in autonomous vehicles primarily utilize Hungarian matching and 3D Kalman filtering methods. These employ manually designed rules, such as target location, size, and type, and matching parameters adjusted for different scenarios to track detected objects across consecutive frames. These existing solutions suffer from the following drawbacks:

[0003] 1. This tracking method relies on rules designed for specific scenarios and does not utilize a large amount of annotation information from rich scenarios. Its use in various scenarios is easily restricted.

[0004] 2. The detection algorithm and tracking algorithm are separated and cannot jointly improve the algorithm accuracy of each module.

[0005] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0006] Embodiments of the present invention provide a point cloud data processing method, device, non-volatile storage medium, and processor to at least solve the technical problem in the prior art that tracking methods rely on rules designed for specific scenarios, resulting in their use being easily restricted in various scenarios.

[0007] According to one aspect of an embodiment of the present invention, a point cloud data processing method applied to a vehicle is provided, comprising: collecting, by means of a laser radar installed on an autonomous vehicle, multiple frames of point cloud data reflected by a target object, wherein the target object includes at least one vehicle located outside the autonomous vehicle; obtaining any one frame of point cloud data from the multiple frames of point cloud data, and based on the any one frame of point cloud data, fusing a predetermined number of frames of point cloud data to obtain fused point cloud fusion data; converting the point cloud fusion data into a two-dimensional pseudo image; detecting the two-dimensional pseudo image using a detection head of a convolutional neural network to obtain physical feature information of the target object, the physical feature information of the target object including at least one of the following: physical size information, spatial position information, initial position image, and motion offset image of the target object; determining the tracking ID of the target object based on the two-dimensional pseudo image; and determining the spatial position of the target object based on the physical feature information and the tracking ID of the target object.

[0008] Optionally, based on any frame of point cloud data, a predetermined number of frames of point cloud data are fused, and this step includes: obtaining at least one frame of point cloud data to be fused that is adjacent to any frame of point cloud data; according to the positioning information of each frame of point cloud data to be fused, compensating any frame of point cloud data and at least one adjacent frame of point cloud data to be fused to the vehicle coordinate system to generate fused point cloud fusion data.

[0009] Optionally, before determining the spatial position of the target object based on the physical feature information and the tracking ID of the target object, the method includes: comparing the two-dimensional pseudo image obtained within a first preset time period with the two-dimensional pseudo image within a second preset time period, wherein the time node of the first preset time period is earlier than the time node of the second preset time period; judging whether the target object in the two-dimensional pseudo image within the second preset time period disappears; when it is determined that there is a disappeared target object in the two-dimensional pseudo image, deleting the tracking ID of the disappeared target object; when it is determined that the target object in the two-dimensional pseudo image still exists, keeping the tracking ID of the target object that still exists.

[0010] Optionally, comparing the two-dimensional pseudo image obtained within the first preset time period with the two-dimensional pseudo image within the second preset time period includes: when determining that a new target object appears on the two-dimensional pseudo image within the second preset time period, assigning a new tracking ID to the newly added target object.

[0011] Optionally, it is determined whether the target object appears in the two-dimensional pseudo image for the first time within a first preset time period; when it is determined that the target object appears in the two-dimensional pseudo image for the first time within the first preset time period, it is determined whether the target object still appears in the two-dimensional pseudo image within a second preset time period; when it is determined that the target object still appears in the two-dimensional pseudo image within the second preset time period, the response peak of the target object in the two-dimensional pseudo image is enhanced, and the response peak includes the outline of the target object.

[0012] Optionally, determining the spatial position of the target object according to the physical feature information and the tracking ID of the target object includes: fitting the corresponding motion offset image of the target object with the initial position image according to the tracking ID of the target object to obtain the spatial position of the target object.

[0013] Optionally, the two-dimensional pseudo image is detected using a category attribute function to obtain category attributes of the target object in the two-dimensional pseudo image, wherein the category attributes at least include type information of the target object.

[0014] According to another aspect of an embodiment of the present invention, a point cloud data processing device is also provided, including: an acquisition module, used to acquire multiple frames of point cloud data reflected by a target object through a laser radar installed on an autonomous driving vehicle, wherein the target object includes at least one vehicle located outside the autonomous driving vehicle; a fusion module, used to obtain any frame of point cloud data from the multiple frames of point cloud data, and based on the any frame of point cloud data, fuse a predetermined number of frames of point cloud data to obtain fused point cloud fusion data; a conversion module, used to convert the point cloud fusion data into a two-dimensional pseudo image; a detection module, using a detection head of a convolutional neural network to detect the two-dimensional pseudo image to obtain physical feature information of the target object, the physical feature information of the target object including at least one of the following: physical size information, spatial position information, initial position image, and motion offset image of the target object; a first determination module, used to determine the tracking ID of the target object based on the two-dimensional pseudo image; and a second determination module, used to determine the spatial position of the target object based on the physical feature information and the tracking ID of the target object.

[0015] According to another aspect of an embodiment of the present invention, a non-volatile storage medium is further provided. The non-volatile storage medium includes a stored program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the above-mentioned point cloud data processing method.

[0016] According to another aspect of an embodiment of the present invention, a processor is further provided, and the processor is used to run a program, wherein the above-mentioned point cloud data processing method is executed when the program is run.

[0017] In an embodiment of the present invention, a laser radar installed on an autonomous vehicle collects multiple frames of point cloud data reflected by a target object. By extracting any one frame of point cloud data from the multiple frames, and based on this arbitrary frame of point cloud data, a predetermined number of frames of point cloud data are fused to obtain fused point cloud data. This fused point cloud data is then converted into a two-dimensional pseudo-image. The two-dimensional pseudo-image is then detected using a convolutional neural network detection head to obtain physical feature information of the target object. The target object's tracking ID is determined based on the two-dimensional pseudo-image, and the target object's spatial position is determined based on the physical feature information and the tracking ID of the target object. This method fully utilizes the tracking ID of the target object and, by monitoring the tracking IDs of target objects in a wide range of scenarios, ensures the tracking module's scene generalization capability. This method eliminates the need for manually designed matching rules and manual parameter adjustment, and the entire process is automated, improving deployment efficiency and scalability. This method addresses the technical problem of existing tracking methods relying on rules designed for specific scenarios, which limits their use in a variety of scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0019] Figure 1 is a flowchart of a first embodiment of an optional point cloud data processing method according to the present invention;

[0020] Figure 2 is a flowchart of an optional method for fusing point cloud data of a predetermined number of frames according to an embodiment of the present invention;

[0021] Figure 3 is a flow chart of a second embodiment of an optional point cloud data processing method according to the present invention;

[0022] Figure 4 4 is a structural block diagram of an optional point cloud data processing device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] According to an embodiment of the present invention, a method embodiment of a point cloud data processing method applied to a vehicle is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0026] Figure 1 is a point cloud data processing method according to an embodiment of the present invention, such as Figure 1 As shown, the method includes the following steps:

[0027] Step S101, collecting multiple frames of point cloud data reflected by a target object using a laser radar installed on the autonomous vehicle, wherein the target object includes at least one vehicle located outside the autonomous vehicle;

[0028] Step S102: obtaining any frame of point cloud data from multiple frames of point cloud data, and fusing a predetermined number of frames of point cloud data based on the any frame of point cloud data to obtain fused point cloud fusion data;

[0029] Step S103, converting the point cloud fusion data into a two-dimensional pseudo image;

[0030] Step S104: Detecting the two-dimensional pseudo image using a detection head of a convolutional neural network to obtain physical feature information of the target object, where the physical feature information of the target object includes at least one of the following: physical size information, spatial position information, an initial position image, and a motion offset image of the target object;

[0031] Step S105, determining the tracking ID of the target object based on the two-dimensional pseudo image;

[0032] Step S106: determining the spatial position of the target object according to the physical feature information and the tracking ID of the target object.

[0033] Through the above steps, a laser radar installed on an autonomous vehicle collects multiple frames of point cloud data reflected by a target object. By extracting any frame of point cloud data from the multiple frames, and based on the arbitrary frame of point cloud data, a predetermined number of frames of point cloud data are fused to obtain fused point cloud data. The fused point cloud data is then converted into a two-dimensional pseudo-image. The two-dimensional pseudo-image is detected using a convolutional neural network detection head to obtain physical feature information of the target object. The tracking ID of the target object is determined based on the two-dimensional pseudo-image, and the spatial position of the target object is determined based on the physical feature information and the tracking ID of the target object. This method can fully utilize the tracking ID of the target object and ensure the scene generalization capability of the tracking module by supervising the tracking ID of the target object in a large number of rich scenes. This method does not require manual design of matching rules or manual parameter adjustment. The entire process is automated, improving deployment efficiency and scalability. This method solves the technical problem that existing tracking methods rely on rules designed for specific scenarios, which easily limits their use in various scenarios.

[0034] Alternatively, as Figure 2As shown, based on any frame of point cloud data, a predetermined number of frames of point cloud data are fused. This step includes: step S201, obtaining at least one frame of point cloud data to be fused adjacent to the any frame of point cloud data; step S202, compensating the any frame of point cloud data and the at least one adjacent frame of point cloud data to be fused to the vehicle coordinate system according to the positioning information of each frame of point cloud data to be fused, thereby generating fused point cloud fusion data. Specifically, taking the fusion processing of n frames of point cloud data as an example, the n-1 (n>1) frames of point cloud data before the current frame are converted to the vehicle coordinate system of the current n-th frame according to the corresponding vehicle positioning information, thereby obtaining the current batch of n frames of 3D point cloud fusion data. The value of each point in the point cloud data is x, y, z, r, Δt, where x, y, z represent the coordinate position of the point cloud in space. Specifically, x, y, z represent the length, width, and height of the point cloud, respectively, r represents pixels, and Δt is the time difference relative to the current frame.

[0035] According to a specific embodiment of the present application, point cloud fusion data is converted into a two-dimensional pseudo-image (2D feature map) by means of voxels, etc., so as to facilitate deep feature extraction using a 2D convolutional neural network. The 2D convolutional neural network structure generally adopts a Feature Pyramid Network (FPN) network structure, which uses multi-level feature maps to predict objects of different sizes, where high-level features carry high-level semantic information and a larger receptive field, which is suitable for detecting large objects, and shallow features carry low-level detail semantic information and a smaller receptive field, which is suitable for detecting small objects. FPN gradually fuses deep and shallow features, and finally obtains a feature map with rich expressive capabilities that can extract both target spatial features and target semantic features.

[0036] Furthermore, a convolutional neural network detection head is used to detect two-dimensional artifacts and obtain physical feature information of the target object. Specifically, the two-dimensional artifacts are extracted through a detection head containing a few convolutional layers to obtain the physical dimensions of the target object, such as length, width, height, and orientation angle; the spatial location of the target center; the initial position response characteristics of the target object, namely the initial position image; and the motion offset characteristics of the target object relative to the first frame of the previous batch, namely the motion offset image.

[0037] Alternatively, as Figure 3As shown, before determining the spatial position of a target object based on physical feature information and the tracking ID of the target object, the method includes: step S301, comparing a two-dimensional pseudo image obtained within a first preset time period with a two-dimensional pseudo image within a second preset time period, wherein the time node of the first preset time period is earlier than the time node of the second preset time period; step S302, determining whether the target object in the two-dimensional pseudo image within the second preset time period has disappeared; step S303, if it is determined that the target object in the two-dimensional pseudo image has disappeared, deleting the tracking ID of the disappeared target object; if it is determined that the target object in the two-dimensional pseudo image still exists, retaining the tracking ID of the target object that still exists. This can improve the tracking accuracy of the target object and effectively prevent the disappeared target object from affecting the tracking effect.

[0038] Furthermore, in order to improve the tracking accuracy of the target object, the position image in the current batch of frames and the position image obtained in the previous batch can be averaged to optimize the lines of the target object. According to a certain threshold, the target tracking ID that disappears in the current batch of frames is deleted, and the tracking ID of the target object that still exists in the current batch of frames is retained.

[0039] In order to further improve the tracking accuracy of the target object, the two-dimensional pseudo image obtained within the first preset time period is compared with the two-dimensional pseudo image within the second preset time period, including: when it is determined that a new target object appears on the two-dimensional pseudo image within the second preset time period, a new tracking ID is assigned to the newly added target object.

[0040] Optionally, a determination is made as to whether the target object appears in the two-dimensional pseudo image for the first time within a first preset time period. If this is determined to be the first time the target object appears in the two-dimensional pseudo image within the first preset time period, a determination is made as to whether the target object still appears in the two-dimensional pseudo image within a second preset time period. If this is determined to be the case, a response peak of the target object in the two-dimensional pseudo image is enhanced, the response peak including the outline of the target object. Simultaneously, a category attribute of the target object is obtained using an activation function. This allows for timely detection and tracking of newly appearing target objects.

[0041] Optionally, determining the spatial position of the target object based on the physical feature information and the tracking ID of the target object includes fitting a motion-shifted image of the target object with the initial position image based on the tracking ID of the target object to obtain the spatial position of the target object. Fitting the motion-shifted image of the target object with the initial position image is performed by adding the obtained motion-shifted image to the initial position image to obtain a final spatial position of the target object in the current batch of frames.

[0042] Optionally, a category attribute function is used to detect the two-dimensional pseudo image and obtain the category attributes of the target object in the two-dimensional pseudo image. The category attributes include, but are not limited to, the target object's category, spatial position, length, width, height, size, orientation angle, and tracking ID number. This allows for accurate identification and classification of the target object, improving the reliability of target tracking.

[0043] According to another specific embodiment of the present application, a point cloud data processing device is also provided. Figure 4 This is a block diagram of an optional point cloud data processing device according to an embodiment of the present invention, comprising: an acquisition module 40, a fusion module 41, a conversion module 42, a detection module 43, a first determination module 44, and a second determination module 45. The acquisition module 40 is configured to use a laser radar installed on an autonomous vehicle to acquire multiple frames of point cloud data reflected by a target object, wherein the target object includes at least one vehicle located outside the autonomous vehicle. The fusion module 41 is configured to extract any one frame of point cloud data from the multiple frames of point cloud data and, based on the one frame of point cloud data, fuse a predetermined number of frames of point cloud data to obtain fused point cloud data. The conversion module 42 is configured to convert the fused point cloud data into a two-dimensional pseudo image. The detection module 43 uses a convolutional neural network detection head to detect the two-dimensional pseudo image and obtain physical feature information of the target object. The physical feature information of the target object includes at least one of the following: physical size information, spatial position information, an initial position image, and a motion offset image of the target object. The first determination module 44 is configured to determine the tracking ID of the target object based on the two-dimensional pseudo image. The second determination module 45 is configured to determine the spatial position of the target object based on the physical feature information and the tracking ID of the target object.

[0044] In an embodiment of the present invention, a point cloud data processing device collects multiple frames of point cloud data reflected by a target object using a laser radar installed on an autonomous vehicle. The device then extracts any one frame of point cloud data from the multiple frames and fuses a predetermined number of frames of point cloud data based on the single frame to obtain fused point cloud data. The fused point cloud data is then converted into a two-dimensional pseudo-image. A convolutional neural network detection head detects the two-dimensional pseudo-image to obtain physical feature information of the target object. The target object's tracking ID is determined based on the two-dimensional pseudo-image, and the target object's spatial position is determined based on the physical feature information and the tracking ID. This method fully utilizes the tracking ID of the target object and, by monitoring the tracking IDs of target objects across a wide range of scenarios, ensures the tracking module's scene generalization capabilities. This method eliminates the need for manually designed matching rules and manual parameter adjustment, automating the entire process and improving deployment efficiency and scalability. This method addresses the technical issue of existing tracking methods, which rely on rules designed for specific scenarios, limiting their use in a variety of scenarios.

[0045] According to another specific embodiment of the present application, the point cloud data processing device includes: an acquisition module 40, a fusion module 41, a conversion module 42, a detection module 43, a first determination module 44, and a second determination module 45. The conversion module 42 converts the 3D point cloud data fused by the fusion module 41 into 2D feature data (that is, the point cloud fusion data is converted into a two-dimensional pseudo image), and then uses a two-dimensional convolutional neural network (2D CNN) to obtain the target feature data in the scene. Among them, the detection module 43 includes a first module, a second module, and a third module, such as Figure 3 As shown. The first module is used to obtain the size and orientation attributes of the target object. Specifically, the first module processes the above-mentioned target object feature data through a convolutional neural network to obtain the physical size, length, width, height, orientation angle, etc. of the target object. The second module is used to obtain the position attribute of the target object. Specifically, the second module processes the above-mentioned target feature data through a convolutional neural network to obtain the spatial position of the target center of the target object. The third module is used to obtain the tracking identity attribute of the target object. Specifically, the third module processes the above-mentioned target feature data through a convolutional neural network to obtain the initial position image of the target object and the motion offset image of the target object. The first determination module 44 obtains the final spatial position and tracking ID number of the target object in the current frame through the third module in the detection module 43, and obtains the category of the target object based on the response value. The second determination module 45 sends the various attributes of the target object obtained by the above-mentioned modules to the terminal. The various attributes of the target object include but are not limited to: the category, spatial position, length, width, height, size, orientation angle and tracking ID number of the target object.

[0046] According to another specific embodiment of the application, a non-volatile storage medium is also provided, which includes a stored program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the steps of the point cloud data processing method in the above embodiment.

[0047] According to another specific embodiment of the present application, a processor is further provided, which is used to run a program, wherein the program executes the steps of the point cloud data processing method in the above embodiment when running.

[0048] Accurate environmental perception is key to enabling autonomous vehicles to navigate reliably, make informed decisions, and drive safely in complex and dynamic environments. Unmanned vehicles or mobile surveying vehicles are typically equipped with a variety of sensors, such as cameras, millimeter waves, and lidar. Traditionally, image data captured by cameras can provide two-dimensional semantic and texture information, and has the advantages of low cost and high efficiency, making it one of the most commonly used data in perception tasks. However, image data lacks three-dimensional geographic information. Therefore, dense, accurate point cloud data with three-dimensional geographic information collected by lidar is also used in perception tasks. In addition, lidar is insensitive to changes in lighting conditions and can operate day and night. Even in adverse weather and environments, it can stably and accurately detect and perceive the surrounding environment. With the continuous breakthroughs in deep learning technology, 3D point cloud target detection and multi-target tracking based on deep neural networks have gradually become key technologies for autonomous driving perception modules. However, the perception algorithms currently used in 3D point cloud data are often plagued by the following problems:

[0049] 1. The detection algorithm and tracking algorithm modules are separated. The tracking algorithm does not utilize the supervision information of the object target ID and cannot improve the accuracy of the detection algorithm.

[0050] 2. Heuristic tracking algorithms rely on manually designed rules, which are limited by the engineer's domain and prior knowledge. Matching parameters need to be adjusted for different scenarios, resulting in low tracking accuracy.

[0051] 3. Heuristic tracking algorithms require manual design of matching rules and involve a lot of manual parameter tuning, resulting in low deployment efficiency and scalability. For example, in autonomous driving scenarios, it is necessary to detect and track multiple categories of targets (cars, pedestrians, two-wheeled vehicles, etc.). When manually tuning parameters, specific debugging is required for each category.

[0052] The application of the point cloud data processing method of this application can improve the tracking algorithm's scene adaptability and the accuracy of the detection and tracking algorithm. This point cloud data processing method can be applied to vehicles with L2 or higher autonomous driving systems, which include: sensors such as lidar, an intelligent driving hardware platform, and an intelligent driving software platform. The application of the technical solution of this application has the following advantages:

[0053] 1. The detection and tracking algorithms are based on deep learning neural network training, making full use of the supervision information of the object target ID to ensure the scene generalization ability of the tracking module.

[0054] 2. There is no need to manually design matching rules. The entire process is automated and does not involve manual parameter adjustment, which improves deployment efficiency and scalability.

[0055] 3. Using the supervision information of the ID of the target object in the detection algorithm also improves the accuracy of the detection algorithm.

[0056] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0057] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0058] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0059] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0060] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0061] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc. Various media that can store program codes.

[0062] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A point cloud data processing method applied to a vehicle, characterized in that: include: Collecting multiple frames of point cloud data reflected by a target object using a laser radar installed on the autonomous vehicle, wherein the target object includes at least one vehicle located outside the autonomous vehicle; Acquire any one frame of point cloud data from the multiple frames of point cloud data, and fuse a predetermined number of frames of point cloud data based on the any one frame of point cloud data to obtain fused point cloud fusion data; Converting the point cloud fusion data into a two-dimensional pseudo image; Detecting the two-dimensional pseudo image using a detection head of a convolutional neural network to obtain physical feature information of the target object, wherein the physical feature information of the target object includes at least one of the following: physical size information, spatial position information, an initial position image, and a motion offset image of the target object; determining a tracking ID of the target object according to the two-dimensional pseudo image; determining a spatial position of the target object based on the physical feature information and the tracking ID of the target object; Before determining the spatial position of the target object according to the physical feature information and the tracking ID of the target object, the method includes: Comparing the two-dimensional pseudo image obtained within a first preset time period with the two-dimensional pseudo image obtained within a second preset time period, wherein a time node of the first preset time period is earlier than a time node of the second preset time period; Determining whether the target object in the two-dimensional pseudo image disappears within the second preset time period; When it is determined that the target object has disappeared in the two-dimensional pseudo image, deleting the tracking ID of the disappeared target object; when it is determined that the target object still exists in the two-dimensional pseudo image, keeping the tracking ID of the still existing target object; The position image in the current batch of frames and the position image obtained in the previous batch are averaged to optimize the lines of the target object. According to the threshold, the target tracking IDs that disappear in the current batch of frames are deleted, and the tracking IDs of the target objects that still exist in the current batch of frames are retained.

2. The point cloud data processing method according to claim 1, characterized in that: Based on any one frame of point cloud data, a predetermined number of frames of point cloud data are fused, and this step includes: Acquire at least one frame of point cloud data to be fused that is adjacent to any one frame of point cloud data; According to the positioning information of each frame of point cloud data to be fused, the arbitrary frame of point cloud data and at least one adjacent frame of point cloud data to be fused are compensated to the vehicle coordinate system to generate the fused point cloud fusion data.

3. The point cloud data processing method according to claim 1, characterized in that: Comparing the two-dimensional pseudo image obtained within a first preset time period with the two-dimensional pseudo image obtained within a second preset time period includes: When it is determined that a new target object appears on the two-dimensional pseudo image within the second preset time period, a new tracking ID is assigned to the newly added target object.

4. The point cloud data processing method according to claim 1, characterized in that: Determining whether the target object appears in the two-dimensional pseudo image for the first time within the first preset time period; When it is determined that the target object appears in the two-dimensional pseudo image for the first time within the first preset time period, determining whether the target object still appears in the two-dimensional pseudo image within the second preset time period; When it is determined that the target object still appears in the two-dimensional pseudo image within the second preset time period, a response peak of the target object in the two-dimensional pseudo image is enhanced, where the response peak includes a contour of the target object.

5. The point cloud data processing method according to claim 1, characterized in that: Determining the spatial position of the target object according to the physical feature information and the tracking ID of the target object includes: According to the tracking ID of the target object, the corresponding motion offset image of the target object is fitted with the initial position image to obtain the spatial position of the target object.

6. The point cloud data processing method according to claim 3, characterized in that: The two-dimensional pseudo image is detected using a category attribute function to obtain a category attribute of the target object in the two-dimensional pseudo image, wherein the category attribute at least includes type information of the target object.

7. A point cloud data processing device, characterized in that: The point cloud data processing device is used to execute the method according to any one of claims 1 to 6, comprising: an acquisition module, configured to acquire, by means of a laser radar installed on the autonomous vehicle, multiple frames of point cloud data reflected by a target object, wherein the target object includes at least one vehicle located outside the autonomous vehicle; a fusion module, configured to obtain any one frame of point cloud data from the multiple frames of point cloud data, and fuse a predetermined number of frames of point cloud data based on the any one frame of point cloud data to obtain fused point cloud fusion data; A conversion module, used for converting the point cloud fusion data into a two-dimensional pseudo image; a detection module, which detects the two-dimensional pseudo image using a detection head of a convolutional neural network to obtain physical feature information of the target object, wherein the physical feature information of the target object includes at least one of the following: physical size information, spatial position information, initial position image, and motion offset image of the target object; a first determining module, configured to determine a tracking ID of the target object according to the two-dimensional pseudo image; The second determining module is configured to determine the spatial position of the target object according to the physical feature information and the tracking ID of the target object.

8. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is run by the processor, the device where the non-volatile storage medium is located is controlled to execute the point cloud data processing method according to any one of claims 1 to 6.

9. An electronic device, characterized in that: include: a memory storing an executable program; A processor, wherein the processor is used to run a program, wherein the point cloud data processing method according to any one of claims 1 to 6 is executed when the program is run.

Images