A method for correcting point cloud data of an unmanned surface vehicle for a sea surface scene

Abstract

The application provides a method for correcting point cloud data of an unmanned ship in a sea surface scene, comprising: obtaining original point cloud data and external parameter calibration data information of an unmanned ship in a motion state, wherein the original point cloud data is point cloud data collected in a time period of a frame of point cloud data of a laser radar, and auxiliary estimation correction information corresponding to the original point cloud data is obtained; wherein the auxiliary estimation correction information is IMU angular velocity information corresponding to a time period of a frame of point cloud data of the unmanned ship and GPS information corresponding to the time period of the frame of point cloud data of the unmanned ship; based on the external parameter calibration data information and the auxiliary estimation correction information, the original point cloud data is corrected to obtain corrected point cloud data; and the corrected point cloud data is subjected to noise reduction processing to obtain clean point cloud. Thus, accurate and stable data sources are provided for subsequent target detection, distance measurement and other perception tasks relying on point cloud information.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to the field of point cloud data processing. BACKGROUND

[0002] As an active ranging device, the laser radar measures the distance to the target object in real time by actively emitting a laser beam, and has been widely used in the field of automatic driving of vehicles. Based on the active ranging characteristics of the laser radar, its application in the field of unmanned surface vehicles can provide data support for subsequent route planning, ship motion control and other tasks. Generally, a scanning laser radar emits a laser beam and simultaneously rotates the laser emitter at a constant speed for 360 degrees to detect objects in the surrounding environment. When the laser radar carrier moves, the time at which the laser echoes reflected by objects at different angles are received is inconsistent, which can cause distortion of the point cloud. Therefore, in actual applications, the point cloud needs to be corrected. The motion correction method of the point cloud is mainly suitable for vehicle-mounted laser radars, and is mainly adapted to flat roads. However, due to road regulations and road planning, vehicles rarely make large-scale turns in a short time. However, the sea environment is accompanied by waves, and the ship body will move irregularly up and down with the waves during navigation. In the case of narrow channels such as entering and leaving the port, large-scale turning maneuvers may be performed to avoid other ships or buoys. The above situations result in poor adaptability of general point cloud correction algorithms in sea environments.

[0003] In particular, the climate conditions of the sea environment are variable, and rain, fog and the like can easily cause noise of unordered point clouds. At the same time, in the case of entering and leaving the port, sea clutter generated by other ships is often encountered, and the sea clutter in the point cloud is a cluster of dense and irregular points at the tail of the target ship, which can cause incorrect distance determination and misclassification of the target ship, and thus affect safe navigation. SUMMARY

[0004] In order to overcome the above technical defects, the present application provides a point cloud data correction method, a storage medium and an equipment.

[0005] To achieve the above purpose, the present application is realized by the following technical solutions:

[0006] In a first aspect, the present application provides a point cloud data correction method.

[0007] In a second aspect, the present application provides a computer device, which comprises a processor and a memory, and the memory stores at least one instruction, at least one program, a code set or an instruction set. The at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to realize the point cloud data correction method according to any one of the first aspect.

[0008] In a third aspect, the present application provides a computer readable storage medium, wherein the readable storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the point cloud data correction method according to any one of the first aspect.

[0009] The present application has the following beneficial effects:

[0010] In the present application, the original point cloud data scanned by the laser radar of the unmanned ship is corrected and denoised, so as to remove the point cloud translation, orientation motion distortion, pitch and roll motion distortion caused by the up and down motion of the ship body with waves, and the point cloud noise around the object caused by frequent climate changes on the sea surface and sea clutter caused by the ship sailing trail, so as to obtain clean point cloud, thereby providing accurate and stable data source for subsequent target detection, ranging and other perception tasks relying on point cloud information.

[0011] In addition to the purposes, features and advantages described above, the present application has other purposes, features and advantages. The present application will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0012] The accompanying drawings, which form a part of the present application, are intended to provide further understanding of the present application, and the illustrative embodiments of the present application and their description serve to explain the present application. The accompanying drawings do not constitute an undue limitation on the present application. In the drawings:

[0013] Figure 1 is a flowchart of the point cloud correction method of the present application.

[0014] Figure 2 is a flowchart of the point cloud correction method of the present application.

[0015] Figure 3 is a network structure diagram of the denoising module of the present application. DETAILED DESCRIPTION

[0016] The embodiments of the present application will be described in detail below with reference to the accompanying drawings, but the present application can be implemented in various different ways as limited and covered by the claims.

[0017] In the present application, the point cloud data correction method of the unmanned ship facing the sea surface scene is mainly realized by the motion correction module and the denoising module, as shown in Figure 1 The specific implementation is as follows:

[0018] The motion correction module obtains the original point cloud data obtained by laser radar scanning when the motion posture of the unmanned surface vehicle on the sea is complex and is accompanied by up-down pitching motion at all times, then removes the angle change motion distortion by using the IMU angular velocity information obtained by external parameter calibration, and further obtains the corrected point cloud data by removing the translation motion distortion in combination with GPS information. The correction process lays a solid foundation for subsequent point cloud denoising and subsequent perception application.

[0019] The denoising module includes a sea clutter removal submodule and a denoising submodule. Since the corrected point cloud data also includes two sea surface scene point cloud noise sources of weather change and ship wake, the corrected point cloud data is identified by using a point cloud attribute estimation network, sea clutter is removed, noise around an object is estimated and corrected immediately, and noise points are projected to a clean surface to obtain clean point cloud, thereby providing clean and accurate point cloud information for subsequent target detection, ranging and other perception tasks.

[0020] The above correction process is described in detail as follows:

[0021] To solve the above-mentioned technologies in the background art, as shown in Figure 2 The application provides a point cloud data correction method, which comprises the following steps:

[0022] Step S1: obtaining original point cloud data and external parameter calibration data information of an unmanned vehicle in a motion state, wherein the original point cloud data is point cloud data collected in a time period of one frame of point cloud data of a laser radar; and the external parameter calibration data information includes the position relationship between a calibrated IMU and the laser radar.

[0023] When the unmanned vehicle is in motion, the laser radar installed on the unmanned vehicle moves together with the unmanned vehicle. When the laser radar is working, it scans and collects data at a certain frequency. The time period of one frame of point cloud data mentioned here is the time taken by the laser radar to complete one complete scanning and collect a group of point cloud data.

[0024] During the period when the unmanned vehicle is in motion, the laser radar continuously emits laser beams. When the laser beams reflect back after irradiating the surface of surrounding objects (such as the water surface, other ships, and shore scenery), the three-dimensional coordinates of the object surface points can be calculated according to the laser round-trip time, emission angle and other information. Numerous such points constitute the original point cloud data of this frame. These original point cloud data can preliminarily reflect the general situation of the environment around the unmanned vehicle in this particular time period, and provide basic data support for subsequent further analysis of the environment where the unmanned vehicle is located and path planning.

[0025] In addition to the laser radar, an inertial measurement unit (IMU) and other sensors are installed on the unmanned ship for external parameter calibration. Various types of information should be time-aligned. Therefore, the external parameter calibration data information includes the position relationship matrix of the calibrated IMU to the laser radar.

[0026] It should be noted that in the scenario where the unmanned ship needs to recalibrate the sensors, in order to avoid errors in subsequent correction, the external parameter calibration data information needs to be corrected in time after each calibration of the sensors of the unmanned ship.

[0027] Step S2: obtaining auxiliary estimation correction information corresponding to the original point cloud data; wherein the auxiliary estimation correction information includes IMU angular velocity information corresponding to a frame of point cloud data time period of the unmanned ship, and GPS information corresponding to a frame of point cloud data time period of the unmanned ship;

[0028] In this application, the auxiliary estimation correction information obtained by other sensors is used to correct the original point cloud data. The auxiliary estimation correction information generally includes IMU angular velocity information corresponding to a frame of point cloud data time period of the unmanned ship, and GPS information corresponding to a frame of point cloud data time period of the unmanned ship.

[0029] When the laser radar obtains the original point cloud data in a frame of point cloud data time period, the IMU obtains the IMU angular velocity information corresponding to a frame of point cloud data time period of the unmanned ship at the same time. The GPS obtains the GPS information corresponding to a frame of point cloud data time period of the unmanned ship at the same time.

[0030] The IMU angular velocity information corresponding to a frame of point cloud data time period of the unmanned ship is the angular velocity information measured by the IMU in a frame of point cloud data time period of the unmanned ship. This angular velocity information reflects the rotation speed of the unmanned ship around each axis in this specific time period. When the unmanned ship moves on the water surface, it will rotate due to its own turning, the influence of wind and waves, and other factors. The angular velocity information of the IMU is a quantitative description of these rotational movements.

[0031] The GPS information corresponding to a frame of point cloud data time period of the unmanned ship is the information obtained by the GPS on the change of the position of the unmanned ship in a frame of point cloud data time period. The GPS information reflects the geographical position and related motion state data of the unmanned ship in this specific time period. It determines the position of the unmanned ship in the earth coordinate system by receiving satellite signals, including longitude, latitude and altitude, and may also include the speed and heading of the unmanned ship.

[0032] Step S3: correcting the original point cloud data based on the external parameter calibration data information and the auxiliary estimation correction information to obtain corrected point cloud data;

[0033] After the external parameter calibration data information and the auxiliary estimation correction information are obtained, the original point cloud data can be corrected to obtain corrected point cloud data. The correction process is shown in Figure 3 The correction process is mainly completed in the motion correction module, and the correction process is described in detail below.

[0034] When the unmanned ship moves actively, the heading motion distortion and the pitch and roll motion distortion caused by the ship body moving up and down with the waves. When the scene changes constantly with the up-and-down pitch on the sea surface, the acceleration output by the IMU is very divergent and difficult to use. In this case, the IMU angular velocity information and the external parameter calibration data information are used to remove the angular change motion distortion of the original point cloud data, thereby obtaining the first corrected point cloud data. Therefore, it can be concluded that the first corrected point cloud data is the point cloud data after removing the angular change motion distortion of the original point cloud data. The point cloud data after removing the angular change distortion of the original point cloud data is described in detail below.

[0035] It should be noted that the frequency of the laser radar (10 Hz) is usually much lower than the frequency of the IMU (200 Hz), and the frequency of the laser radar and the frequency of the IMU are not the same in different cases. The application adopts the frequency under the usual condition.

[0036] The rotating scanning laser radar generates point clouds by uniformly rotating the laser beams around it. Specifically, for a point in the point cloud, the time from the current point time to the end state time of the laser radar can be estimated by its angle relative to the initial position of the laser radar. Specifically: for each point p i in the original point cloud data, determine the initial position angle θ i of each point p i in the laser radar; then according to the initial position angle θ i , determine the time difference value of each point at the current time relative to the end state time of a frame of point cloud data time period, that is: (unit: ms). Then:

[0037]

[0038] where is the angle change of t i relative to the previous time t0, respectively corresponding to the angular velocity of the previous time and the current time. The IMU data interval is , that is, 0.5 ms, then within the time Data α i ,have:

[0039]

[0040] The calculation method described by formula (1)-(2) can be used to calculate the time difference ( IMU data α i Perform m-1 iterations and sum up, and we can get The IMU rotation angle α within the time, that is, the IMU rotation angle α corresponding to the time difference, is transformed into the IMU rotation matrix corresponding to the time difference after matrix transformation.

[0041]

[0042] The previous article obtained the position relationship matrix from the calibrated IMU to the lidar After that, the inverse matrix of the position relationship can be obtained by inverse transformation.

[0043] Determine the IMU rotation matrix corresponding to the time difference And the position relationship matrix from the calibrated IMU to the lidar And the position relationship matrix from the calibrated IMU to the lidar The inverse matrix of Then, according to the above three matrices, the rotation matrix of each point in the original point cloud data relative to the final state of the lidar is estimated As shown in formula (3).

[0044]

[0045] After determining the rotation matrix of each point in the original data relative to the final state of the lidar, the rotation matrix is ​​estimated, and then the angle change motion distortion of each point in the original point cloud data is removed based on the rotation matrix estimate, thereby obtaining the first corrected point cloud data.

[0046] The first corrected point cloud data mentioned above only removes the angular motion distortion of the original point cloud data. However, when the UAV is in motion, the point cloud data in the translation direction will also be distorted. Therefore, it is necessary to introduce the GPS information corresponding to the time period of one frame of the UAV point cloud data to remove the translation motion distortion of the first corrected point cloud data. The specific process is as follows:

[0047] The GPS information frequency is generally 10Hz. In this application, the laser radar point cloud information is synchronized with the GPS information time, that is, the initial state and final state GPS coordinates can be obtained within a point cloud data time period. Specifically: Based on the GPS information, the GPS coordinate transformation information between the initial state and the final state of the laser radar within a frame of the point cloud time period can be determined, and then the GPS coordinate transformation information can be calculated to obtain the first translation matrix between the initial state and the final state of the laser radar in the rectangular coordinate system. Converting the GPS coordinates to a unified rectangular coordinate system yields the radar's initial rectangular coordinates (x0, y0, z0) and final rectangular coordinates (x1, y1, z1). The first translation matrix between the initial and final states is:

[0048]

[0049] The time interval of a point cloud cycle is short (100ms), so it can be approximately considered that the GPS coordinates change uniformly within a point cloud cycle. i The time difference between the current moment and the end state moment of the point cloud cycle is (Unit is ms).

[0050] At this time, the first translation matrix and the time difference can be used to determine each point p i , the radar rectangular coordinates corresponding to the current moment. That is, point p i The radar rectangular coordinates at this moment are

[0051] After obtaining the radar rectangular coordinates, the second translation matrix from the current moment to the final state of the laser radar can be determined for each point in the first corrected point cloud data based on the radar rectangular coordinates and the first translation matrix. That is, the radar rectangular coordinates are (x i ,y i ,z i ), substituted into formula (4), we can get point p i The second translation matrix from time to the final state of the lidar

[0052] After determining the second translation matrix After that, the translational motion distortion can be removed from each point in the first corrected point cloud data according to the second translation matrix to obtain second corrected point cloud data. The second corrected point cloud data is used as the corrected point cloud data, wherein the second corrected point cloud data is the point cloud data after removing the angular motion distortion and the translational motion distortion.

[0053] In summary, the current point coordinates can be converted to the coordinate system p in the final state of the laser radar by combining the rotation angle and translation vector of each point in the original point cloud data to the final state of the laser radar i-lidar , as shown in formula (5). This completes the correction of all points in the original point cloud data, and then obtains the motion-corrected point cloud data. The correction process using the above point cloud data significantly improves the adaptability of the motion correction algorithm under different sea conditions. This process lays a solid foundation for point cloud denoising and subsequent perception applications.

[0054]

[0055] Step S4: performing noise reduction processing on the corrected point cloud data to obtain clean point cloud data.

[0056] The corrected point cloud data has been corrected for various distortions introduced by the UAV's motion. However, since the UAV operates over the sea, the environmental and climatic conditions are highly variable. Rain, fog, and other factors can easily lead to disordered point cloud noise. Furthermore, entering and leaving ports often encounters sea clutter generated by the wakes of other ships. Therefore, noise reduction of these point cloud noises and sea clutter is necessary to obtain clean point cloud data.

[0057] like Figure 3 As shown in the figure, the processing of the corrected point cloud data is completed by using two submodules, namely the sea clutter removal submodule and the denoising submodule. Therefore, the specific composition of the above submodules is briefly described:

[0058] The two submodules have the same input data format (both are point clouds), so we divide the network into three processes for the two submodules. Figure 3 To ensure the pose and sequence stability of the point cloud and the efficiency of the network, the following are shown: a point-by-point feature extraction model, in order to solve the pose stability problem of the point cloud, a four-element space transfer network is added to migrate the input point cloud data to a new pose state that is conducive to network learning; a feature fusion model, in order to solve the sequential stability problem, after obtaining the processed point-by-point feature vector, the network sums it in the dimension of the point and merges it into a feature vector. Regardless of the order of the input points, the commutative law of addition ensures the consistency of the merged feature vector; a regression module, in order to ensure the efficiency of the network, the fused feature vector is processed using several one-dimensional convolutional or fully connected layers to obtain the final result of the network output.

[0059] The feature extraction module and feature fusion module are the feature processing stage. The two submodules in this stage have the same structure. For the regression module, the output of the regression module in the sea clutter removal submodule network is point P' iis the probability of sea clutter is a specific value, and the structure is a number of one-dimensional convolution layers; the output of the regression module in the denoising module network is a point Estimated offset d i , is a three-dimensional vector, and its structure consists of several fully connected layers.

[0060] Based on the above description of the component structure, the noise reduction processing of the corrected point cloud data is described in more detail:

[0061] The main sources of noise in the sea surface point cloud are sea clutter generated by the wakes of other ships and noise around objects caused by complex weather conditions such as rain and fog. Sea clutter points differ significantly from the other points and are generally located at the stern of a ship, so they can be represented and removed separately. Noise around objects is closer to the surface points and is caused by incorrect reflections of the laser beam due to weather changes. It can be removed by relocating it to a suitable location on the edge of the object, while ensuring the accuracy and uniform distribution of the surface point cloud. The noisy sea surface point cloud can be represented as:

[0062]

[0063] Where P' is the corrected point cloud data, each point is p' i ; O is the set of sea clutter points, each point is o j ;n i For each point p i Noise interference on the i are vectors of the same dimension.

[0064] That is to say, it is necessary to remove the sea clutter points from the corrected point cloud data to obtain the noise point cloud data after the sea clutter points are removed. Figure 3 , the process of removing sea clutter is described in detail. The specific process is as follows:

[0065] like Figure 3 As shown in the upper network, the input is the corrected point cloud data P', that is, each point P' in the corrected point cloud output by the motion correction module i , the output is each point P' in P' i is the probability of sea clutter Specifically, through the probability function g(Ρ' i ), to estimate each point p' in the rectified point cloud data i The probability of belonging to the sea clutter point is calculated, and then the probability of each point of the corrected point cloud data belonging to the sea clutter is compared with the probability threshold according to the set probability threshold. If it exceeds the threshold, the point that exceeds the probability threshold is regarded as a sea clutter point and filtered out. The point that does not exceed the probability threshold is regarded as a normal data point and retained, and finally the noise point cloud after the sea clutter points are removed is obtained. The specific value of the probability threshold can be set by yourself, and this article does not impose any specific restrictions on this.

[0066]

[0067] Among them, P' i for p' i A collection of points in a field. Indicates filtering out sea clutter points exceeding the threshold in P', such as Figure 3 Shown in the middle position.

[0068] After obtaining the noise point cloud data after removing the sea clutter points, denoising is performed on the noise point cloud data, such as Figure 3 This process is described in detail as shown below:

[0069] like Figure 3 The lower network shows that the input is the noisy point cloud after removing the sea clutter points Every point The output is Every point in Offset d i Specifically, the noise point cloud output by the sea clutter removal submodule Hit the mark Estimated offset Then according to the noise point cloud data and the estimated offset d i , get a clean point cloud That is, the position of each point estimated by the denoising submodule network is right The same operation is performed on each point in the image to obtain a clean point cloud after noise reduction. As shown in formula (9), each point is now placed at the correct position.

[0070]

[0071] in The clean point cloud estimated by the denoising module network, The set of sea clutter points estimated by the sea clutter removal submodule network, The noise point cloud after sea clutter points are removed can also be expressed as d i is the offset of each point learned by the denoising submodule network, d i Interference with noise i Related.

[0072] In summary, the application removes the point cloud translation, heading motion distortion, pitch and roll motion distortion caused by the up and down motion of the boat body with waves, and the point cloud noise around the object caused by frequent climate changes on the sea surface and sea clutter caused by the ship sailing wake by correcting and denoising the original point cloud data scanned by the laser radar of the unmanned ship during the motion of the unmanned ship, so as to obtain clean point cloud, thereby providing accurate and stable data source for subsequent target detection, ranging and other perception tasks relying on point cloud information.

[0073] In some embodiments, a computer readable storage medium storing a computer program is provided, and the computer program, when executed by a processor, causes the processor to implement the point cloud data correction method of the first aspect.

[0074] Those of ordinary skill in the art can understand that all or some of the steps in the method disclosed above and the functional modules / units in the system and device can be implemented as software, firmware, hardware and appropriate combinations thereof. In a hardware implementation, the division between the functional blocks / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, one physical component can have multiple functions, or one function or step can be performed by several physical components in cooperation. Some or all of the physical components can be implemented as software executed by a processor, such as a central processor, a digital signal processor or a microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software can be distributed on a computer readable storage medium, which can include computer readable storage media (or non-transitory media) and communication media (or transitory media).

[0075] As known to those of ordinary skill in the art, the term computer readable storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer readable storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. In addition, it is known to those of ordinary skill in the art that communication media generally includes computer readable instructions, data structures, program modules or other data in modulated data signals such as carrier waves or other transmission mechanisms, and can include any information delivery medium.

[0076] For example, the computer readable storage medium can be an internal storage unit of the network management device, such as a hard disk or a memory of the network management device. The computer readable storage medium can also be an external storage device of the network management device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc.

[0077] In some embodiments, a device is provided, comprising a processor and a memory storing a computer program; the processor is configured to execute the computer program and implement the point cloud data correction method provided in the first aspect of the present application when executing the computer program.

[0078] It should be understood that the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0079] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, 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 correction method for unmanned boat point cloud data for sea scenes, characterized by: include: Obtaining original point cloud data and external parameter calibration data information of the unmanned vehicle in motion, wherein the original point cloud data is point cloud data collected within a time period of one frame of laser radar point cloud data; the external parameter calibration data information includes a position relationship matrix from a calibrated IMU to the laser radar; Acquire auxiliary estimation and correction information corresponding to the original point cloud data; wherein the auxiliary estimation and correction information is IMU angular velocity information corresponding to a time period of one frame of point cloud data of the unmanned vehicle, and GPS information corresponding to a time period of one frame of point cloud data of the unmanned vehicle; Based on the external parameter calibration data information and the auxiliary estimation correction information, the original point cloud data is corrected to obtain corrected point cloud data; Performing noise reduction processing on the corrected point cloud data to obtain a clean point cloud; The correcting process of the original point cloud data based on the external parameter calibration data information and the auxiliary estimation correction information to obtain the corrected point cloud data includes: Using the IMU angular velocity information and the external parameter calibration data information, the original point cloud data is subjected to angular change motion distortion removal to obtain first corrected point cloud data, wherein the first corrected point cloud data is the point cloud data obtained by removing the angular change motion distortion from the original point cloud data; Using the GPS information, the first corrected point cloud data is subjected to translational motion distortion removal to obtain second corrected point cloud data, and the second corrected point cloud data is used as the corrected point cloud data, wherein the second corrected point cloud data is the point cloud data after removing the angular change motion distortion and the point cloud data after removing the translational motion distortion.

2. The method according to claim 1, characterized in that Using the IMU angular velocity information and the external parameter calibration data information, the original point cloud data is removed from the angle change motion distortion to obtain the point cloud data after the angle change motion distortion is removed, including: Based on each point in the original point cloud data, determining an initial position angle of each point at the laser radar; Based on the initial position angle, determining a time difference between the current moment of each point and the final state moment of the time period of the frame of point cloud data; Based on the time difference, determining IMU angular velocity information corresponding to the time difference; Integrate the IMU angular velocity information to determine the IMU rotation angle corresponding to the time difference; Perform matrix changes based on the IMU rotation angle to determine the rotation matrix corresponding to the time difference; Based on the calibrated IMU-to-lidar position relationship matrix, determining an inverse matrix corresponding to the position relationship; Determine, based on the rotation matrix corresponding to the time difference, the position relationship matrix, and the inverse matrix, an estimated rotation matrix of each point in the original point cloud data relative to the final state of the laser radar; Based on the rotation matrix estimation, the angle change motion distortion is removed for each point in the original point cloud data to obtain first corrected point cloud data.

3. The method according to claim 1, characterized in that The step of using the GPS information to remove translational motion distortion from the point cloud data after removing the angle change motion distortion to obtain second corrected point cloud data includes: Based on the GPS information, determining GPS coordinate transformation information between an initial state and a final state of the laser radar within a frame point cloud time period; Determine, based on the GPS coordinate transformation information, a first translation matrix between an initial state and a final state of the laser radar in a rectangular coordinate system; Determining, based on the time difference and the first translation matrix, the radar rectangular coordinates corresponding to each point in the first corrected point cloud data at a current moment; Determine, based on the radar rectangular coordinates and the first translation matrix, a second translation matrix from the current moment to the final state of the laser radar for each point in the first corrected point cloud data; Based on the second translation matrix, the translation motion distortion is removed from each point in the first corrected point cloud data to obtain second corrected point cloud data.

4. The method according to claim 1, wherein The denoising process is performed on the corrected point cloud data to obtain a clean point cloud, including: removing sea clutter points from the corrected point cloud data to obtain noisy point cloud data after sea clutter points are removed; De-noising is performed on each point in the noisy point cloud data to obtain a clean point cloud.

5. The method according to claim 4, characterized in that Removing sea clutter points from the corrected point cloud data includes: determining, based on each point in the corrected point cloud data, a first probability that each point in the point cloud data is sea clutter; Determining whether the first probability is greater than a probability threshold; If so, the point cloud corresponding to the first probability of sea clutter is regarded as a sea clutter point and filtered out.

6. The method according to claim 5, characterized in that The step of performing denoising on each point in the noisy point cloud data to obtain a clean point cloud comprises: determining, based on each point in the noisy point cloud data, an estimated offset from each point in the noisy point cloud data; A clean point cloud is obtained based on the estimated offset and the noisy point cloud data.

7. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the point cloud data correction method according to any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that The readable storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the point cloud data correction method according to any one of claims 1 to 6.

Images