Point cloud data preprocessing method and device, and surface target lidar

Abstract

This invention relates to a point cloud data preprocessing method and apparatus, and a surface target lidar, within the field of intelligent unmanned surface vessels (USVs). Addressing the problem that existing simple Euclidean clustering methods are insufficient for handling ship wake waves, leading to misjudgments of target ship position and volume, and affecting the USV's judgment of the target's true attributes, the present invention provides a point cloud data preprocessing method for removing noise from surface point cloud data. The method includes: Step 1: Selecting a point in the target surface space and obtaining a point cloud from the point; Step 2: Obtaining a set of laser points whose distances meet a preset threshold from the point cloud; Step 3: Determining whether the number of laser points in the laser point set meets a preset requirement. If it does, the set is retained as the target point cloud cluster, and Step 4 is performed; otherwise, Steps 1 to 3 are repeated; Step 4: Obtaining target surface information based on the target point cloud cluster. This method is suitable for point cloud processing technology and for use in shipborne surface lidar sensors.

Description

Technical Field

[0001] It involves the field of intelligent unmanned smart ships, specifically shipborne lidar. Background Technology

[0002] Unmanned surface vessels (USVs) are unmanned surface vessels that navigate on water autonomously or remotely. They can be carried to a predetermined location by a large vessel and then released to perform missions, or they can autonomously navigate to near-shore targets. They are mainly used to perform dangerous missions that are not suitable for manned vessels. Once equipped with advanced control systems, sensor systems, communication systems, and weapon systems, they can perform a variety of wartime and non-wartime military missions.

[0003] When an unmanned surface vessel (USV) is traveling on water, it needs to avoid obstacles on the water surface to prevent collisions between the USV and obstacles that could damage the USV.

[0004] Domestic and international scholars have conducted some research on obstacle avoidance for unmanned surface vessels (USVs). Before obstacle avoidance, obstacle detection methods are crucial. Currently, the mainstream obstacle detection technologies can be divided into four categories: ultrasonic obstacle detection, stereo vision obstacle detection, multi-sensor fusion obstacle detection, and lidar obstacle detection. Ultrasonic obstacle detection mainly uses the echo characteristics of sound waves to calculate the location of obstacles. Due to the Doppler effect, data anomalies can occur if the vehicle speed reaches a certain level. Stereo vision obstacle detection mainly uses mathematical principles in image processing to extend two-dimensional visual technology to three-dimensional space, acquiring more information about target objects, but this increases cost and information processing complexity. Multi-sensor fusion obstacle detection mainly integrates, optimizes, and merges information from two or more similar or heterogeneous sensors to obtain more environmental feature representations. However, the difficulty lies in the fusion algorithm; it is actually difficult to choose a suitable fusion algorithm that can match multiple identical or different sensor devices. LiDAR obstacle detection technology mainly utilizes the time difference between the emitted and reflected laser lines to calculate the angle and distance of the obstacle point. This detection system has advantages such as high accuracy, strong resolution, and good real-time performance.

[0005] LiDAR sensors have been widely used in the field of unmanned surface vessels (USVs), and point cloud processing technology has become increasingly mature. However, in the actual working environment of USVs, there are various interference factors such as ship wakes, weeds, and fishing nets, as well as mutual interference between multiple sensors, including the interaction between marine radar and lidar. First, the aforementioned interference factors will introduce scattered noise into the raw lidar data. If the data is not preprocessed, this noise will significantly affect the fusion of multi-sensor information and subsequent target detection and tracking. Second, the simple Euclidean clustering method is insufficient for handling ship wakes, which can lead to misjudgments of the target ship's position and size, affecting the USV's judgment of the target's true attributes. Summary of the Invention

[0006] To address the problem that existing simple Euclidean clustering methods are insufficient in handling ship wakes, leading to misjudgments of target ship position and volume, and affecting the unmanned surface vessel's (USV) assessment of the target's true attributes, the technical solution provided by this invention is as follows:

[0007] A point cloud data preprocessing method for removing noise from water surface point cloud data, the method comprising:

[0008] Step 1: Select a point in the target water surface space and obtain a point cloud based on that point;

[0009] Step 2: Obtain a set of laser points whose distances satisfy a preset threshold based on the point cloud;

[0010] Step 3: Determine whether the number of laser points in the laser point set meets the preset requirements. If it does, retain the set as the target point cloud cluster and proceed to Step 4; otherwise, return to repeat Step 1 to Step 3.

[0011] Step 4: Obtain the target water surface information based on the target point cloud cluster.

[0012] Furthermore, in a preferred embodiment, the method further includes:

[0013] Step 5: Repeat steps 1 to 4 until the preset number of repetitions is reached.

[0014] Furthermore, a preferred embodiment is provided, wherein in step 1, the point cloud is obtained by using a KD-Tree data index structure to perform a nearest neighbor search of the points to obtain the point cloud.

[0015] Furthermore, a preferred embodiment is provided, wherein in step 2, the set is obtained by weighting the distance of the point cloud.

[0016] Based on the same inventive concept, the present invention also provides a point cloud data preprocessing apparatus for removing noise from water surface point cloud data, the apparatus comprising:

[0017] Module 1: Used to select a point in the target water surface space and obtain a point cloud based on the point;

[0018] Module 2: Used to obtain a set of laser points whose distances satisfy a preset threshold based on the point cloud;

[0019] Module 3: Used to determine whether the number of laser points in the laser point set meets the preset requirements. If it does, the set is retained as the target point cloud cluster, and the function of Module 4 is performed; if it does not meet the requirements, the function of Modules 1 to 3 is returned to be executed.

[0020] Module 4: Used to obtain the target water surface information based on the target point cloud cluster.

[0021] Furthermore, in a preferred embodiment, the apparatus further includes:

[0022] Module 5: Used to repeat the functions of modules 1 to 4 until a preset number of repetitions is reached.

[0023] Furthermore, a preferred embodiment is provided, wherein the point cloud is obtained in module 1 by using a KD-Tree data index structure to perform a nearest neighbor search of the points to obtain the point cloud.

[0024] Based on the same inventive concept, the present invention also provides a surface target lidar, wherein the central processing unit of the lidar is loaded with the point cloud data preprocessing method described above.

[0025] Based on the same inventive concept, the present invention also provides a computer storage medium for storing a computer program, wherein when the computer program stored in the storage medium is read by the processor of a computer, the computer executes the point cloud data preprocessing method.

[0026] Based on the same inventive concept, the present invention also provides a computer, including a processor and a storage medium, wherein the storage medium stores a computer program, and when the processor reads the computer program stored in the storage medium, the computer executes the point cloud data preprocessing method.

[0027] Compared with the prior art, the advantages of the technical solution provided by the present invention are as follows:

[0028] The point cloud data preprocessing method provided by this invention classifies point cloud data using weighted Euclidean distance and completes noise removal and point cloud clustering based on the KD-Tree structure. It has the advantages of being more suitable for specific scenarios of water surface target recognition tasks and providing faster and more accurate processing results.

[0029] The point cloud data preprocessing method provided by this invention utilizes the characteristics of scattered, sparse, and low reflectivity of ship wakes, calculates weighted Euclidean distance based on point cloud intensity, and accelerates noise removal and point cloud clustering of lidar point cloud data through KD-Tree data indexing structure, thereby improving the speed and accuracy of water surface target detection.

[0030] The point cloud data preprocessing method provided by this invention adopts an improved weighted Euclidean clustering method. The improved clustering method can effectively suppress scattered outliers in the original lidar data by further filtering the point cloud clusters obtained after intensity-weighted clustering.

[0031] The point cloud data preprocessing method provided by this invention can effectively suppress scattered outliers in the original lidar data by further filtering the point cloud clusters obtained after intensity-weighted clustering. The method yields accurate and real-time results, which can meet the requirements of point cloud data preprocessing in the process of multi-sensor data fusion.

[0032] The water surface target lidar provided by this invention classifies point cloud data using weighted Euclidean distance and performs noise removal and point cloud clustering based on a KD-Tree structure, resulting in faster and more accurate processing, thus improving the speed and accuracy of water surface target detection.

[0033] Suitable for point cloud processing technology and for use in shipborne surface lidar sensors. Attached Figure Description

[0034] Figure 1 This is a typical lidar point cloud image of a surface target mentioned in Implementation Method Eleven;

[0035] Where (a) represents the actual photograph taken, and (b) represents the lidar point cloud map;

[0036] Figure 2 This is a schematic diagram comparing the improvement effects of weighted clustering mentioned in Implementation Method Eleven;

[0037] Where (a) represents the target bounding box obtained by the classical Euclidean clustering method, and (b) represents the target bounding box obtained by the weighted clustering method;

[0038] Figure 3 This is a schematic diagram of the KD-Tree search process mentioned in Implementation Method Eleven;

[0039] Figure 4 This is a schematic diagram of the weighted Euclidean clustering flowchart mentioned in Implementation Method Eleven. Detailed Implementation

[0040] To make the advantages and benefits of the technical solution provided by the present invention clearer, the technical solution provided by the present invention will now be described in further detail with reference to the accompanying drawings, specifically:

[0041] Implementation Method 1: This implementation method provides a point cloud data preprocessing method for noise removal from water surface point cloud data. The method includes:

[0042] Step 1: Select a point in the target water surface space and obtain a point cloud based on that point;

[0043] Step 2: Obtain a set of laser points whose distances satisfy a preset threshold based on the point cloud;

[0044] Step 3: Determine whether the number of laser points in the laser point set meets the preset requirements. If it does, retain the set as the target point cloud cluster and proceed to Step 4; otherwise, return to repeat Step 1 to Step 3.

[0045] Step 4: Obtain the target water surface information based on the target point cloud cluster.

[0046] Implementation Method Two: This implementation method further defines the point cloud data preprocessing method provided in Implementation Method One. The method further includes:

[0047] Step 5: Repeat steps 1 to 4 until the preset number of repetitions is reached.

[0048] Implementation Method 3: This implementation method further defines the point cloud data preprocessing method provided in Implementation Method 1. In step 1, the point cloud is obtained by using a KD-Tree data index structure to perform a nearest neighbor search on the points to obtain the point cloud.

[0049] Implementation Method 4: This implementation method is a further limitation of the point cloud data preprocessing method provided in Implementation Method 1. In step 2, the set is obtained by obtaining it based on the weighted distance of the point cloud.

[0050] Implementation Method 5: This implementation method provides a point cloud data preprocessing device for removing noise from water surface point cloud data. The device includes:

[0051] Module 1: Used to select a point in the target water surface space and obtain a point cloud based on the point;

[0052] Module 2: Used to obtain a set of laser points whose distances satisfy a preset threshold based on the point cloud;

[0053] Module 3: Used to determine whether the number of laser points in the laser point set meets the preset requirements. If it does, the set is retained as the target point cloud cluster, and the function of Module 4 is performed; if it does not meet the requirements, the function of Modules 1 to 3 is returned to be executed.

[0054] Module 4: Used to obtain the target water surface information based on the target point cloud cluster.

[0055] Implementation Method Six: This implementation method further defines the point cloud data preprocessing apparatus provided in Implementation Method Five. The apparatus further includes:

[0056] Module 5: Used to repeat the functions of modules 1 to 4 until a preset number of repetitions is reached.

[0057] Implementation Method Seven: This implementation method further defines the point cloud data preprocessing device provided in Implementation Method Five. In module 1, the point cloud is obtained by using a KD-Tree data index structure to perform a nearest neighbor search of the points to obtain the point cloud.

[0058] Implementation Method 8: This implementation method provides a surface target lidar, wherein the central processing unit of the lidar is loaded with a point cloud data preprocessing method provided in any one of Implementation Methods 1 to 4.

[0059] Implementation Method Nine: This implementation method provides a computer storage medium for storing a computer program. When the computer program stored in the storage medium is read by the computer's processor, the computer executes the point cloud data preprocessing method provided in any one of Implementation Methods One to Four.

[0060] Implementation Method 10: This implementation method provides a computer, including a processor.

[0061] The processor and storage medium, wherein the storage medium stores a computer program, and when the processor reads the computer program stored in the storage medium, the computer executes the point cloud data preprocessing method provided in any one of embodiments one to four.

[0062] Implementation Method Eleven: Combination Figure 1 and 2 This embodiment describes a specific example of the point cloud data preprocessing method provided in any one of embodiments one to four. It is used to compare with existing technologies and demonstrate the advantages of the point cloud data preprocessing method provided in any one of embodiments one to four. It also serves to explain the point cloud data preprocessing method provided in any one of embodiments one to four. Specifically:

[0063] This implementation provides an improved lidar point cloud data preprocessing method based on the KD-Tree data indexing structure. In the working environment of unmanned surface vessels, interference from ship wakes is common. These wakes are identified by lidar as sparse, low-reflectivity scattered noise points, such as… Figure 1As shown in (b), the continuous point cloud at the stern of the unmanned surface vessel is the wake point cloud. Due to the low intensity of the wake point cloud, it is displayed as dark red during visualization. This invention utilizes the characteristics of ship wakes being scattered, sparse, and having low reflectivity. It calculates weighted Euclidean distance based on point cloud intensity and accelerates noise removal and point cloud clustering of lidar point cloud data through the KD-Tree data indexing structure, thereby improving the speed and accuracy of surface target detection. The specific implementation method is as follows.

[0064] First, select a point P in the space and use the KD-Tree data index structure to accelerate the nearest neighbor search of that point;

[0065] The weighted distance of the point cloud is calculated based on the point cloud intensity to obtain a set of laser points whose distances meet the set threshold. Euclidean clustering is a method of grouping point clouds into clusters based on the Euclidean distance between them. The Euclidean distance between point clouds is the most common distance metric, and its coordinates in the three-dimensional coordinate system of the lidar point cloud are (x... p ,y p ,z p The point P with coordinates (x) q ,y q ,z q The distance d between points Q) PQ The expression is as follows

[0066]

[0067] Based on this, the classic Euclidean clustering method is optimized by comprehensively considering both point cloud intensity and Euclidean distance. The weighted distance calculation formula for the point cloud is as follows:

[0068]

[0069] In the formula: d intensity d represents the weighted Euclidean distance (m), intensity represents the point cloud intensity;

[0070] Figure 2 A comparison chart showing the improvement effect of weighted clustering;

[0071] like Figure 2 As shown, due to the adoption of the improved weighted Euclidean clustering method, the improved clustering method can effectively suppress scattered outliers in the original lidar data by further filtering the point cloud clusters obtained after intensity-weighted clustering.

[0072] The number of laser point sets calculated is evaluated. If the required number is met, the set is retained as a target point cloud. If not, the set is removed, and a new initial point is selected.

[0073] Analyze the obtained detection results and process the point cloud data of the next frame. The Euclidean clustering method based on point cloud reflectivity can better suppress the interference of ship wakes on the clustering results. Further screening of the point cloud clusters obtained after intensity-weighted clustering can also effectively suppress the scattered outlier noise points in the original lidar data. The results obtained by this method are accurate and have strong real-time performance, and can meet the requirements of point cloud data preprocessing in the multi-sensor data fusion process.

[0074] Among them, Figure 3 It is a schematic diagram of the KD-Tree search flow chart; for a given query point P to be queried, it is necessary to start comparing from the root node of the KD-Tree, comparing the value P(k) corresponding to the query point P on the current node division dimension k and the threshold m. If P(k) < m, then access the left subtree, otherwise access the right subtree until reaching the leaf node Q, and Q is the nearest neighbor point of P, and the minimum distance D is the distance between Q and P. Then backtrack along the original search path to the root node. During this process, if there are child nodes that satisfy the distance less than D from P, they are included in the search range, and the nearest neighbor point is updated accordingly; if there are no other remaining points, that is, all search paths are empty, then the KT-Tree search process ends.

[0075] Figure 4 It is a weighted Euclidean clustering flow chart; for a certain point P in space, find the neighboring points based on the KD-Tree nearest neighbor search algorithm and calculate the weighted distance of the point cloud as the distance judgment criterion. The neighboring points less than the set threshold are divided into the output set Q. Judge the number of elements in Q. If the number of elements in Q continues to increase, then select other laser points except P in Q and repeat the above process; if the number of elements in Q no longer increases and the number of elements meets the threshold requirement, then retain this set as the output result, otherwise remove this set. Repeat the above process until there are no remaining laser points in space.

[0076] The above further describes the technical solutions provided by the present invention in several specific implementation manners to highlight the advantages and beneficial effects of the technical solutions provided by the present invention. However, the above several specific implementation manners are only used to describe the technical solutions provided by the present invention and are not used as a limitation to the present invention; any reasonable modifications and improvements to the present invention, reasonable combinations of implementation manners, equivalent replacements, etc. within the spirit and principle of the present invention should be included within the protection scope of the present invention.

Claims

1. A point cloud data preprocessing method for noise removal from water surface point cloud data, characterized in that, The method includes: Step 1: Select a point in the target water surface space and obtain a point cloud based on that point; Step 2: Obtain a set of laser points whose distances satisfy a preset threshold based on the point cloud; Step 3: Determine whether the number of laser points in the laser point set meets the preset requirements. If it does, retain the set as the target point cloud cluster and proceed to Step 4; otherwise, return to repeat Step 1 to Step 3. Step 4: Obtain the target water surface information based on the target point cloud cluster; Among them, a point in space is selected. The nearest neighbor search for this point is accelerated by using the KD-Tree data index structure. Based on point cloud intensity and Euclidean distance, the weighted distance calculation formula for point clouds is as follows: ； In the formula: This represents the weighted Euclidean distance (m). Represents the Euclidean distance (m). Indicates point cloud intensity; The nearest neighbor search of the points is performed using a KD-Tree data index structure to obtain the point cloud. Specifically: For a given query point P, compare the values ​​corresponding to query point P on the partition dimension k of the current node, starting from the root node of the KD-Tree. and threshold ,like If the left subtree is reached, the right subtree is visited; otherwise, the right subtree is visited until the leaf node Q is reached. Q is the nearest neighbor of P, and the minimum distance D is the distance between Q and P. Then, the search backtracks along the original search path to the root node. During this process, if there is a child node whose distance to P is less than D, it is included in the search range, and the nearest neighbor is updated accordingly. If there are no other remaining points, that is, all search paths are empty, the KT-Tree search process ends. Perform clustering: For a point P in space, the KD-Tree nearest neighbor search algorithm is used to find neighboring points and calculate the weighted distance of the point cloud as the distance judgment criterion. Neighboring points less than a set threshold are assigned to the output set Q. The number of elements in Q is judged. If the number of elements in Q continues to increase, other laser points other than P are selected in Q and the process is repeated. If the number of elements in Q no longer increases and the number of elements meets the threshold requirement, the set is retained as the output result. Otherwise, the set is removed and the process is repeated until there are no more laser points remaining in space.

2. The point cloud data preprocessing method according to claim 1, characterized in that, The method further includes: Step 5: Repeat steps 1 to 4 until the preset number of repetitions is reached.

3. A point cloud data preprocessing device for removing noise from water surface point cloud data, characterized in that: The device includes: Module 1: Used to select a point in the target water surface space and obtain a point cloud based on the point; Module 2: Used to obtain a set of laser points whose distances satisfy a preset threshold based on the point cloud; Module 3: Used to determine whether the number of laser points in the laser point set meets the preset requirements. If it does, the set is retained as the target point cloud cluster, and the function of Module 4 is performed; if it does not meet the requirements, the function of Modules 1 to 3 is returned to be executed. Module 4: Used to obtain the target water surface information based on the target point cloud cluster; Among them, a point in space is selected. The nearest neighbor search for this point is accelerated by using the KD-Tree data index structure. Based on point cloud intensity and Euclidean distance, the weighted distance calculation formula for point clouds is as follows: ； In the formula: This represents the weighted Euclidean distance (m). Represents the Euclidean distance (m). Indicates point cloud intensity; The nearest neighbor search of the points is performed using a KD-Tree data index structure to obtain the point cloud. Specifically: For a given query point P, compare the values ​​corresponding to query point P on the partition dimension k of the current node, starting from the root node of the KD-Tree. and threshold ,like If the left subtree is reached, the right subtree is visited; otherwise, the right subtree is visited until the leaf node Q is reached. Q is the nearest neighbor of P, and the minimum distance D is the distance between Q and P. Then, the search backtracks along the original search path to the root node. During this process, if there is a child node whose distance to P is less than D, it is included in the search range, and the nearest neighbor is updated accordingly. If there are no other remaining points, that is, all search paths are empty, the KT-Tree search process ends. Perform clustering: For a point P in space, the KD-Tree nearest neighbor search algorithm is used to find neighboring points and calculate the weighted distance of the point cloud as the distance judgment criterion. Neighboring points less than a set threshold are assigned to the output set Q. The number of elements in Q is judged. If the number of elements in Q continues to increase, other laser points other than P are selected in Q and the process is repeated. If the number of elements in Q no longer increases and the number of elements meets the threshold requirement, the set is retained as the output result. Otherwise, the set is removed and the process is repeated until there are no more laser points remaining in space.

4. The point cloud data preprocessing apparatus according to claim 3, characterized in that, The device further includes: Module 5: Used to repeat the functions of modules 1 to 4 until a preset number of repetitions is reached.

5. A surface target lidar, characterized in that, The central processing unit of the radar is loaded with the point cloud data preprocessing method according to any one of claims 1-2.

6. A computer storage medium for storing computer programs, characterized in that, When the computer program stored in the storage medium is read by the computer's processor, the computer executes the point cloud data preprocessing method according to any one of claims 1-2.

7. A computer, comprising a processor and a storage medium, wherein the storage medium stores a computer program, characterized in that, When the processor reads the computer program stored in the storage medium, the computer executes the point cloud data preprocessing method according to any one of claims 1-2.

Images