Multi-source point cloud-driven dynamic spraying control method and system for explosion-proof robot clusters

Abstract

This application relates to a method and system for dynamic spraying control of an explosion-proof robot swarm driven by multiple source point clouds. The method includes acquiring multi-source point cloud data of a target surface through multiple sensors; performing preliminary processing on the multi-source point cloud data to remove noise points and redundant data, thereby improving data quality and processing efficiency; inputting the preliminary point cloud dataset into a point cloud fusion algorithm based on geometric constraints to generate fused point cloud data in a unified coordinate system; generating a high-precision geometric model of the target surface based on the fused point cloud data; and combining this with a spraying path scheme generated by the task allocation strategy of the explosion-proof robot swarm. This allows each robot to dynamically adjust its spraying operation according to the specific characteristics of the target surface and the spraying requirements, thus improving the spraying quality and explosion-proof effect of the explosion-proof robots.

Description

Technical Field

[0001] This application relates to the fields of intelligent control, robotics and information technology, and in particular to a method and system for dynamic spraying control of explosion-proof robot clusters driven by multi-source point cloud. Background Technology

[0002] In many sectors of industrial production, especially in locations involving flammable and explosive materials such as petrochemicals and coal mining, explosion protection is crucial. For equipment surface protection in these environments, spraying explosion-proof coatings is a common and effective method. Traditional spraying methods rely primarily on manual operation, requiring workers to wear protective equipment and enter hazardous areas to perform the work. However, manual spraying has several drawbacks. On the one hand, workers face significant safety risks operating in explosive environments, and even slight carelessness could trigger serious accidents such as explosions. On the other hand, manual spraying is inefficient and struggles to ensure uniformity and consistency, affecting the explosion-proof effect.

[0003] To address these issues, some companies and research institutions have begun experimenting with explosion-proof robots for painting operations. These robots can, to some extent, replace manual labor, reduce worker safety risks, and improve painting efficiency. They are typically equipped with sensors and control systems, allowing them to paint along pre-set paths.

[0004] However, existing spraying systems have a significant problem: the information acquired about the target surface is not comprehensive or accurate enough. Due to the lack of comprehensive consideration of the target surface's three-dimensional geometry, texture, and temperature distribution, the robot struggles to dynamically adjust the spraying plan according to the actual conditions of the target surface, resulting in poor spraying quality and failing to fully meet the explosion-proof requirements in complex environments. Summary of the Invention

[0005] The main purpose of this application is to provide a method and system for dynamic spraying control of explosion-proof robot clusters driven by multi-source point cloud, which can improve the spraying quality of explosion-proof robots.

[0006] To achieve the above objectives, embodiments of the present invention provide a method for dynamic spraying control of explosion-proof robot clusters driven by multi-source point cloud, the method comprising:

[0007] Multi-source point cloud data of the target surface is acquired through multiple sensors. The multi-source point cloud data includes three-dimensional geometric information collected by a laser scanner, texture information captured by a depth camera, and temperature distribution information recorded by an infrared sensor.

[0008] The multi-source point cloud data is subjected to preliminary processing to generate a preliminary point cloud dataset. The preliminary processing includes removing noise points and redundant data.

[0009] The initial point cloud dataset is input into a point cloud fusion algorithm based on geometric constraints, and fused point cloud data in a unified coordinate system is generated through point cloud registration and data alignment operations.

[0010] Based on the fused point cloud data, a high-precision geometric model of the target surface is generated using a surface reconstruction algorithm.

[0011] Based on the high-precision geometric model and combined with the task allocation strategy of the explosion-proof robot cluster, a spraying path scheme for the explosion-proof robot is generated. The spraying path scheme includes the spraying start position, the spraying end position, and the spraying trajectory point sequence.

[0012] The spraying task instruction is sent to the explosion-proof robot cluster according to the spraying path plan, so that each robot can perform the spraying operation according to the spraying path plan.

[0013] In summary, the technical solution of this application acquires multi-source point cloud data of the target surface through multiple sensors, including 3D geometric information collected by a laser scanner, texture information captured by a depth camera, and temperature distribution information recorded by an infrared sensor, which can comprehensively and accurately describe the characteristics of the target surface. Preliminary processing of the multi-source point cloud data, removing noise points and redundant data, improves data quality and processing efficiency. The preliminary point cloud dataset is input into a point cloud fusion algorithm based on geometric constraints to generate fused point cloud data in a unified coordinate system, laying the foundation for subsequent generation of a high-precision geometric model. The high-precision geometric model of the target surface generated based on the fused point cloud data can more accurately reflect the actual situation of the target surface. Combined with the spraying path scheme generated by the task allocation strategy of the explosion-proof robot cluster, each robot can dynamically adjust its spraying operation according to the specific characteristics of the target surface and spraying requirements, improving the uniformity and consistency of spraying, thereby enhancing the spraying quality and explosion-proof effect of the explosion-proof robots. Simultaneously, this method realizes automated and intelligent spraying operations for the explosion-proof robot cluster, reducing the safety risks of manual operation and improving spraying efficiency. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of a scenario for a multi-source point cloud-driven dynamic spraying control method for explosion-proof robot clusters in an embodiment of this application.

[0015] Figure 2 A flowchart of a multi-source point cloud-driven dynamic spraying control method for explosion-proof robot clusters is provided for embodiments of this application;

[0016] Figure 3 This is a schematic diagram illustrating the process of generating the preliminary point cloud dataset provided in the embodiments of this application;

[0017] Figure 4 A schematic diagram of the data denoising process provided in the embodiments of this application;

[0018] Figure 5 This is a schematic diagram of the process for generating fused point clouds provided in an embodiment of this application;

[0019] Figure 6 This is a schematic diagram of the registration matrix generation process provided in an embodiment of this application;

[0020] Figure 7 This application provides a schematic diagram of the process for generating high-precision geometric models in its embodiments.

[0021] Figure 8 A schematic diagram of the process generated for the spraying path scheme provided in the embodiments of this application;

[0022] Figure 9 Another schematic diagram of the process generated for the spraying path scheme provided in the embodiments of this application;

[0023] Figure 10 A schematic diagram of the structure of the multi-source point cloud-driven dynamic spraying control system for explosion-proof robot clusters provided in this embodiment of the application;

[0024] Figure 11 A schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] This application provides a method and system for dynamic spraying control of explosion-proof robot clusters driven by multi-source point cloud, which will be described in detail below.

[0027] In this embodiment, the multi-source point cloud-driven dynamic spraying control method for explosion-proof robot clusters is a comprehensive approach to controlling explosion-proof robot spraying operations. Specifically, it involves the acquisition, processing, and analysis of point cloud data from multiple sources, aiming to dynamically control the spraying operations of the explosion-proof robot cluster based on the actual characteristics of the target surface.

[0028] As shown in Figure 1, a scenario for dynamic spraying control of explosion-proof robot clusters driven by multi-source point cloud is provided. The scenario based on explosion-proof robot spraying operation mainly includes multiple explosion-proof robots, multiple sensors, and a control platform; wherein the explosion-proof robots, multiple sensors, and control platform are connected through a wireless network.

[0029] Taking a petrochemical production workshop as an example, this scenario involves a large amount of flammable and explosive chemicals. Explosion-proof coatings on equipment surfaces are used to prevent explosions. In this scenario, an explosion-proof robot is responsible for spraying the coating onto the equipment surfaces. This robot has multiple movable joints, allowing for flexible adjustment of the spraying angle and position.

[0030] Multiple sensors (such as sensor networks) are installed on explosion-proof robots or in suitable locations within the workshop to acquire multi-source point cloud data of the target surface. Laser scanners can acquire the three-dimensional geometric information of the target surface. By emitting a laser beam and measuring the time and angle of reflected light, they accurately determine the three-dimensional coordinates of each point on the target surface. For example, when spraying paint on the surface of a large oil storage tank, a laser scanner can quickly scan the entire tank surface to acquire its complex three-dimensional shape information. Depth cameras can capture the texture information of the target surface. By analyzing the reflection and refraction of light, they generate texture images of the target surface. For example, for equipment with uneven surfaces or special textures, depth cameras can clearly record these details. Infrared sensors are used to record the temperature distribution information of the target surface. They detect the infrared radiation emitted by the target surface and convert it into temperature data. In petrochemical workshops, the temperature of equipment surfaces may change due to the flow of internal media and chemical reactions; infrared sensors can monitor these temperature changes in real time.

[0031] After acquiring multi-source point cloud data from multiple sensors, the control platform processes and analyzes this data. First, preliminary processing is performed on the multi-source point cloud data to remove noise points and redundant data, improving data quality. Then, the preliminary point cloud dataset is input into a geometrically constrained point cloud fusion algorithm, generating fused point cloud data in a unified coordinate system through point cloud registration and data alignment operations. Next, based on the fused point cloud data, a surface reconstruction algorithm is used to generate a high-precision geometric model of the target surface. Finally, based on the high-precision geometric model and the task allocation strategy of the explosion-proof robot cluster, a spraying path scheme for the explosion-proof robots is generated, and spraying task instructions are sent to the explosion-proof robot cluster.

[0032] During actual operation, the explosion-proof robot performs spraying operations on the target surface according to the received spraying task instructions and the spraying path plan. During the spraying process, multiple sensors continuously monitor the target surface and feed real-time data back to the control platform. Based on the feedback data, the control platform dynamically adjusts the spraying path plan and spraying parameters to ensure spraying quality and explosion-proof effect. For example, if an abnormal temperature rise is detected in a certain area of ​​the target surface during spraying, the control platform can promptly adjust the spraying thickness and speed in that area to ensure coating quality.

[0033] refer to Figure 2 , Figure 2 This is a flowchart illustrating a multi-source point cloud-driven dynamic spraying control method for explosion-proof robot clusters, as provided in this application embodiment. The executing entity of this method can be a computer device (which can serve as a control platform). This computer device can be a single computer device or a cluster of multiple computer devices. The computer device can be a terminal device or a server, etc. The multi-source point cloud-driven dynamic spraying control method for explosion-proof robot clusters provided in this application embodiment specifically includes:

[0034] S10: Acquire multi-source point cloud data of the target surface through multiple sensors. The multi-source point cloud data includes three-dimensional geometric information collected by a laser scanner, texture information captured by a depth camera, and temperature distribution information recorded by an infrared sensor.

[0035] In this embodiment, the target surface refers to the surface of an object requiring spraying to achieve explosion-proof functionality. It can be the surface of various industrial equipment, containers, pipes, etc., used in flammable and explosive environments, or the walls, floors, etc., inside buildings where explosion risks may exist. The shape and structure of the target surface can be very complex, including different geometric forms such as planes, curved surfaces, and uneven surfaces, and may have different material properties, such as metals, plastics, and ceramics. Different target surfaces have different requirements for spraying path planning, spraying material selection, and spraying process parameters during spraying operations. For example, for target surfaces with large variations in surface curvature, more precise spraying path planning is needed to ensure coating uniformity; for target surfaces of different materials, matching spraying materials need to be selected to ensure coating adhesion and explosion-proof performance. Accurately identifying and understanding the characteristics and requirements of the target surface is a key foundation for realizing dynamic spraying control of explosion-proof robot swarms driven by multi-source point cloud.

[0036] In this embodiment, multi-source point cloud data refers to a collection of various information about a target surface obtained from different types of sensors. A laser scanner is a device that uses laser technology for three-dimensional measurement. By emitting a laser beam and measuring the time and angle of the reflected light, it can accurately acquire the three-dimensional coordinates of each point on the target surface, thereby obtaining the three-dimensional geometric information of the target surface. For example, in a large industrial plant, when measuring the surface of a complex piece of machinery, a laser scanner can quickly scan the entire surface of the equipment, generating high-precision three-dimensional point cloud data that clearly presents the shape and detailed features of the equipment.

[0037] Depth cameras capture texture information of a target surface by analyzing the reflection and refraction of light. They can generate texture images of the target surface, which are crucial for subsequent surface reconstruction and spraying path planning. For example, for objects with uneven surfaces or special textures, depth cameras can accurately record these details, providing richer information for subsequent processing.

[0038] Infrared sensors are devices that operate on the principle of infrared radiation. They detect the infrared radiation emitted by a target surface and convert it into temperature data, thus recording the temperature distribution information of the target surface. In industries such as petrochemicals, temperature changes on equipment surfaces can affect the adhesion and explosion-proof performance of coatings; therefore, accurately obtaining temperature distribution information is crucial. Acquiring multi-source point cloud data through multiple sensors can comprehensively and accurately describe the characteristics of the target surface, providing a rich data foundation for subsequent processing and analysis. From a technical perspective, acquiring multi-source point cloud data improves the understanding of target surface information, facilitating more precise surface reconstruction and spraying path planning, thereby improving spraying quality and explosion-proof performance.

[0039] In one embodiment, to acquire high-quality multi-source point cloud data, a high-precision laser scanner, depth camera, and infrared sensor can be used. For the laser scanner, a model with high resolution and fast scanning speed can be selected to ensure accurate 3D geometric information can be acquired in a short time. When installing the laser scanner, a suitable position and angle should be chosen to avoid occlusion and reflection interference. For the depth camera, regular calibration and adjustment can be performed to ensure the accuracy of the captured texture information. Simultaneously, multiple depth cameras can be used to capture images from different angles, and then the data can be fused to obtain more complete texture information. For the infrared sensor, a model with high sensitivity and a wide temperature range can be selected to adapt to different working environments. When installing the infrared sensor, it must be ensured that it is accurately aligned with the target surface to avoid interference from the external environment. Through these measures, the quality and reliability of the multi-source point cloud data can be improved.

[0040] S20: Perform preliminary processing on the multi-source point cloud data to generate a preliminary point cloud dataset. The preliminary processing includes removing noise points and redundant data.

[0041] In this embodiment of the application, the preliminary processing is a process of preprocessing the acquired multi-source point cloud data, with the aim of improving data quality and processing efficiency.

[0042] Noise points refer to erroneous data points generated during data acquisition due to various factors (such as sensor errors, environmental interference, etc.), which affect subsequent processing and analysis results. Redundant data refers to duplicate or unnecessary data points, which increase the complexity and computational load of data processing. Removing noise points and redundant data can make point cloud data more concise and accurate. For example, when acquiring data with a laser scanner, it may be affected by ambient light or object reflections, generating some isolated noise points; when fusing data from multiple sensors, some data may be repeatedly acquired, forming redundant data. By removing these noise points and redundant data, data interference can be reduced, improving the accuracy and efficiency of subsequent processing. From a technical perspective, the preliminary point cloud dataset after initial processing is of higher quality, providing a more reliable data foundation for subsequent point cloud fusion and surface reconstruction.

[0043] In one embodiment, a density-based clustering algorithm can be used to remove noise points. This algorithm calculates the density around each data point and identifies data points with densities below a certain threshold as noise points, removing them accordingly. For redundant data, data points with distances less than a certain threshold and similar features can be merged or deleted by comparing their distances and features. In practice, multi-source point cloud data can be classified first, and then different processing methods can be applied to different types of data. For example, for 3D geometric information acquired by a laser scanner, the spatial distribution density of data points can be the focus; for texture information captured by a depth camera, the similarity of texture features can be considered. These methods can effectively remove noise points and redundant data, generating a high-quality preliminary point cloud dataset.

[0044] S30: Input the preliminary point cloud dataset into the point cloud fusion algorithm based on geometric constraints, and generate fused point cloud data in a unified coordinate system through point cloud registration and data alignment operations.

[0045] In this embodiment, the geometrically constrained point cloud fusion algorithm is a method for fusing point cloud data from different sources. It utilizes the geometric features and constraints of the point cloud data to achieve point cloud registration and data alignment. Point cloud registration refers to the process of transforming point cloud data from different coordinate systems to the same coordinate system. It aligns point cloud data spatially by finding the relative pose relationships between them. Data alignment, based on point cloud registration, further adjusts the position and orientation of the data to accurately fuse point cloud data from different sources. For example, in multi-source point cloud data, the 3D geometric information acquired by a laser scanner and the texture information captured by a depth camera may be obtained in different coordinate systems. Through point cloud registration and data alignment operations, they can be unified into the same coordinate system, forming a complete fused point cloud data. Technically, the generated fused point cloud data in a unified coordinate system can more accurately reflect the true state of the target surface, providing more precise data support for subsequent surface reconstruction.

[0046] In one embodiment, firstly, a local coordinate system transformation is performed on each group of data points in the initial point cloud dataset to generate point cloud data in a local coordinate system. Then, based on geometric constraints, the relative pose relationships between the groups of point cloud data are calculated, and an initial registration matrix is ​​generated. Next, the initial registration matrix is ​​optimized using an iterative nearest-point algorithm, continuously adjusting the position and orientation of the point cloud data to minimize the errors between them. Finally, based on the optimized registration matrix, each group of point cloud data is mapped to a unified coordinate system to generate fused point cloud data. In practical applications, prior knowledge and constraints, such as the geometry of the target surface and the sensor installation location, can be combined to improve the accuracy and efficiency of point cloud registration and data alignment.

[0047] S40: Based on the fused point cloud data, a high-precision geometric model of the target surface is generated using a surface reconstruction algorithm.

[0048] In this embodiment, the surface reconstruction algorithm is a method for constructing a geometric model of a target surface based on fused point cloud data. Although the fused point cloud data contains a large amount of information about the target surface, it is only a series of discrete points and cannot directly represent the continuous geometric shape of the target surface. The surface reconstruction algorithm constructs a continuous geometric model of the target surface by processing and analyzing these discrete points.

[0049] In this embodiment, the high-precision geometric model is a three-dimensional model generated based on fused point cloud data and using a surface reconstruction algorithm, which can accurately reflect the actual shape and features of the target surface.

[0050] For example, when reconstructing the surface of a complex industrial part, the fused point cloud data may contain a large number of points on the part's surface. However, these points need to be connected using a surface reconstruction algorithm to form a complete, high-precision geometric model. A high-precision geometric model can more accurately reflect the actual shape and characteristics of the target surface, providing a more reliable basis for subsequent spraying path planning. Technically, the generated high-precision geometric model can improve the accuracy and rationality of spraying path planning, thereby improving spraying quality and efficiency.

[0051] In one embodiment, a surface reconstruction algorithm based on triangular meshes can be employed. First, the boundary point set of the target surface is extracted from the fused point cloud data, and an initial triangular mesh model is generated based on this set. Then, the initial triangular mesh model undergoes topology optimization to eliminate non-manifold structures, making the mesh more regular and stable. Next, based on the optimized triangular mesh model, an interpolation algorithm is used to supplement the geometric details of the target surface, generating a high-precision geometric model containing more details. Finally, the high-precision geometric model is smoothed to remove noise and irregularities from the model surface, resulting in the final high-precision geometric model. In practical applications, appropriate surface reconstruction algorithms and parameters can be selected based on the complexity and accuracy requirements of the target surface to generate a high-precision geometric model that meets the needs.

[0052] S50: Based on the high-precision geometric model and combined with the task allocation strategy of the explosion-proof robot cluster, a spraying path scheme for the explosion-proof robot is generated. The spraying path scheme includes the spraying start position, the spraying end position, and the spraying trajectory point sequence.

[0053] In this embodiment, the task allocation strategy refers to a method for rationally allocating the painting task of each robot based on the number, performance, and task requirements of the explosion-proof robot cluster. A high-precision geometric model provides detailed shape and feature information of the target surface. Combined with the task allocation strategy, a suitable painting path scheme can be generated for each explosion-proof robot. The starting and ending positions of the painting path scheme determine the start and end positions of the robot's painting operation, while the sequence of painting trajectory points describes the path the robot needs to traverse during the painting process. For example, when painting a large planar surface, the surface can be divided into multiple regions based on the high-precision geometric model. Then, one or more regions can be allocated to each explosion-proof robot according to the task allocation strategy, and the starting and ending positions of the painting, as well as the sequence of painting trajectory points for each robot, can be determined. Through reasonable task allocation and painting path planning, the painting efficiency and quality of the explosion-proof robot cluster can be improved. Technically, the generated painting path scheme enables the explosion-proof robots to complete painting tasks more efficiently and accurately, avoiding repeated painting and missed painting.

[0054] In one embodiment, a priority-based task allocation strategy can be employed. First, the number of spraying units required for the spraying task is calculated based on the target surface area of ​​the high-precision geometric model and the coverage capacity of the spraying material. Then, a spraying task allocation table for each explosion-proof robot is generated based on the number of spraying units and the number of robots in the cluster. Next, based on the spraying task allocation table and the geometric features of the high-precision geometric model, priority regions for the spraying tasks are divided, such as high-curvature regions, low-curvature regions, and edge regions. A corresponding priority region is assigned to each explosion-proof robot, and an initial spraying path scheme is generated by combining the assigned priority region with the coverage capacity of the spraying material. In practical applications, the spraying path scheme can be dynamically adjusted according to the robot's real-time status and task execution to improve the flexibility and adaptability of the spraying operation.

[0055] S60: Sends spraying task instructions to the explosion-proof robot cluster according to the spraying path plan, so that each robot can perform spraying operations according to the spraying path plan.

[0056] In this embodiment of the application, the spraying task instruction is a command to control the explosion-proof robot to perform spraying operations. It includes relevant information in the spraying path scheme, such as the starting position of the spraying, the ending position of the spraying, and the sequence of spraying trajectory points.

[0057] The control platform can send spraying task instructions to the explosion-proof robot cluster based on the generated spraying path plan. Upon receiving the instructions, each robot performs the spraying operation according to the plan. For example, after receiving the spraying task instruction, the explosion-proof robot will move to the designated position based on the starting point, and then spray sequentially according to the spraying trajectory points until it reaches the endpoint. By sending spraying task instructions to the explosion-proof robot cluster, remote control and automated operation of the robots can be achieved, improving the efficiency and accuracy of the spraying operation. Technically, this method reduces human intervention, lowers worker safety risks, and ensures the consistency and stability of the spraying operation.

[0058] In one embodiment, the control platform can send spraying task instructions to the explosion-proof robot cluster via a wireless network. Before sending the instructions, they need to be encrypted and verified to ensure their security and accuracy. Upon receiving the instructions, the explosion-proof robots parse and verify them, and then perform the spraying operation according to the information in the instructions. During the spraying process, the robots provide real-time feedback on their status information, such as position, speed, and spraying progress. The control platform can monitor and adjust the robots in real time based on this feedback. If any abnormality is detected, the control platform can promptly send new instructions to enable the robots to take corresponding measures, such as pausing spraying or adjusting their path. In this way, the explosion-proof robot cluster can be ensured to complete the spraying task safely and efficiently.

[0059] In one embodiment, reference Figure 3 Step S20 may specifically include the following steps:

[0060] Step S201: Obtain the spatial distribution density value of each group of data points in the multi-source point cloud data, and calculate the average spatial distribution density value of each group of data points.

[0061] In this embodiment, the spatial distribution density value is an indicator describing the density of data points in space; it represents the number of data points per unit volume. Calculating the spatial distribution density value for each group of data points helps us understand the distribution of data points, providing a basis for subsequent noise and redundant data removal. For example, in the 3D geometric information acquired by a laser scanner, data points may be denser in some areas and sparser in others. By calculating the spatial distribution density value, these differences can be clearly distinguished. The average spatial distribution density value is a comprehensive measure of the spatial distribution density of each group of data points, reflecting the overall distribution characteristics of the data points. Technically, obtaining the spatial distribution density value and the average spatial distribution density value helps to more accurately identify noise points and redundant data, improving the effectiveness of initial processing.

[0062] In one embodiment, a grid-based method can be used to calculate the spatial distribution density value. First, the 3D space containing the multi-source point cloud data is divided into several small grids. For each grid, the number of data points it contains is counted, and then the number of data points is divided by the grid volume to obtain the spatial distribution density value of the data points within that grid. For each group of data points, the spatial distribution density values ​​of the grids they belong to are summarized, and then the average value is calculated to obtain the average spatial distribution density value of that group of data points. This method allows for the rapid and accurate calculation of the spatial distribution density value and the average spatial distribution density value for each group of data points.

[0063] Step S202: For each group of data points in the multi-source point cloud data, if the spatial distribution density value of the group of data points is less than the first preset density threshold, then remove the group of data points to eliminate noise points.

[0064] In this embodiment, the first preset density threshold is a pre-set standard value used to determine whether a data point is a noise point. Typically, noise points are sparsely distributed in space, and their spatial density value is lower than that of normal data points. Therefore, by comparing the spatial density value of each group of data points with the first preset density threshold, noise points can be effectively identified and removed. For example, during laser scanning, some isolated data points may be generated due to external interference. The spatial density value of these data points is often very low. By setting an appropriate first preset density threshold, these noise points can be accurately removed. Technically, removing noise points can improve data quality, reduce the impact of noise on subsequent processing, and make the generated preliminary point cloud dataset more accurate and reliable.

[0065] In one embodiment, the first preset density threshold can be determined based on the material properties of the target surface and the level of environmental noise. For targets with smooth surfaces and high reflectivity, more noise points may be generated, in which case the first preset density threshold can be appropriately increased; while for targets with rough surfaces and low reflectivity, fewer noise points are generated, in which case the first preset density threshold can be appropriately decreased. In practice, a suitable first preset density threshold can be found through multiple experiments and tests. After calculating the spatial distribution density value of each group of data points, it is compared with the first preset density threshold. If it is less than the threshold, the group of data points is removed from the multi-source point cloud data.

[0066] In one embodiment, reference Figure 4 Step S202 can be implemented in the following way:

[0067] Step S2021: Based on the spatial distribution characteristics of multi-source point cloud data, calculate the spatial distribution density value of each group of data points, whereby the spatial distribution density value represents the number of data points per unit volume.

[0068] In this embodiment, the spatial distribution characteristics of multi-source point cloud data refer to the distribution patterns and features of data points in three-dimensional space. Data collected by different sensors may have different spatial distribution characteristics. For example, data collected by a laser scanner may be relatively uniformly distributed along the normal direction of the target surface, while data collected by a depth camera may be more densely distributed in areas with significant texture variations. Calculating the spatial distribution density value quantifies the density of data points in space, thus providing a basis for subsequent noise point identification. Taking multi-source point cloud data of a complex industrial part as an example, the data point distribution density in the raised and recessed areas of the part's surface may differ; calculating the spatial distribution density value clearly reflects this difference. Technically, accurately calculating the spatial distribution density value helps to more accurately identify noise points and improve the accuracy of data processing.

[0069] In one embodiment, kernel density estimation can be used to calculate the spatial distribution density value. Kernel density estimation is a nonparametric statistical method that estimates the spatial distribution density by placing a kernel function around each data point and then summing all kernel functions. A Gaussian kernel function can be chosen due to its good smoothness and computational advantages. For each data point, the effective range of the kernel function is determined based on a certain bandwidth, centered on that point. Then, the weights of other data points within that range are calculated, and finally, the weights of all data points are summed to obtain the spatial distribution density value of that point. This method can more accurately reflect the spatial distribution characteristics of multi-source point cloud data.

[0070] Step S2022: Set a first preset density threshold, which is determined based on the material properties of the target surface and the level of environmental noise.

[0071] In this embodiment, the material properties of the target surface include factors such as surface roughness and reflectivity. Different material properties affect the quality and noise level of the data acquired by the sensor. For example, a smooth surface with high reflectivity may generate more reflection noise, while a rough surface may lead to uneven distribution of data points. Environmental noise levels are affected by interference factors in the surrounding environment, such as light and electromagnetic interference. The setting of the first preset density threshold needs to comprehensively consider these factors to ensure accurate identification of noise points. Technically, a reasonable setting of the first preset density threshold can effectively remove noise points and improve data quality and reliability.

[0072] In one embodiment, the first preset density threshold can be determined through experiments and data analysis. First, data is collected multiple times from target surfaces of different materials, while simultaneously recording the ambient noise level. Then, the collected data is analyzed, and the spatial distribution density values ​​of normal data points and noise points are statistically analyzed. Based on these statistical results and the requirements of the actual application, a suitable first preset density threshold is determined. For example, for a smooth metal target, in an environment with good lighting and low electromagnetic interference, the first preset density threshold can be set relatively high; while for a rough plastic target, in an environment with low lighting and high electromagnetic interference, the first preset density threshold needs to be set relatively low.

[0073] Step S2023: For each group of data points, compare its spatial distribution density value with the first preset density threshold; if the spatial distribution density value of a certain group of data points is lower than the first preset density threshold, then determine that the group of data points is a noise point and remove it from the multi-source point cloud data.

[0074] In this embodiment, by comparing the spatial distribution density value with a first preset density threshold, it is possible to intuitively determine whether a data point is a noise point. When the spatial distribution density value of a group of data points is lower than the first preset density threshold, it indicates that the distribution of that group of data points in space is too sparse, and it is likely to be a noise point. Removing these noise points from the multi-source point cloud data can reduce the impact of noise on subsequent processing and improve data quality. For example, when processing multi-source point cloud data of a large tank, by comparing the spatial distribution density value with the first preset density threshold, some isolated data points are identified as noise points and removed, making the subsequently generated preliminary point cloud dataset more accurately reflect the true surface of the tank. From a technical perspective, this method is simple and effective, and can quickly and accurately remove noise points, improving the efficiency and accuracy of data processing.

[0075] In one embodiment, each group of data points can be compared using a loop-based approach. For each group of data points in the multi-source point cloud data, its spatial distribution density value is sequentially obtained and then compared with a first preset density threshold. If the spatial distribution density value is lower than the first preset density threshold, the group of data points is marked as noise points and removed in subsequent data processing. In practice, this process can be implemented using a computer program to improve processing efficiency.

[0076] Step S203: For the removed data points, calculate the distance deviation between each group of data points and its adjacent data points, and remove redundant data points based on the distance deviation to obtain a preliminary point cloud dataset.

[0077] In this embodiment, the distance deviation value is an indicator that measures the distance difference between a data point and its neighboring data points. Redundant data points are usually very close to their neighboring data points, and their distance deviation values ​​are small. By calculating the distance deviation value, these redundant data points can be identified and removed. For example, in the process of multi-source point cloud data fusion, some data may be collected repeatedly, resulting in some data points being very close in space; these data points are redundant. By calculating the distance deviation value, these redundant data points can be accurately identified. From a technical perspective, removing redundant data points can reduce data redundancy, improve data processing efficiency, and make the initial point cloud dataset more concise and effective.

[0078] In one embodiment, a neighborhood search method can be used to calculate the distance deviation between each set of data points and its neighboring data points. For each data point, neighboring data points are searched within a certain neighborhood, and then the distance between the data point and its neighboring data points is calculated. These distances are statistically analyzed to calculate the distance deviation value. A distance deviation threshold can be set; when the distance deviation value of a data point is less than the threshold, it is identified as a redundant data point and removed. In practice, the neighborhood range and distance deviation threshold can be reasonably adjusted according to the characteristics of the data and processing requirements to achieve the best redundant data removal effect.

[0079] In one embodiment, reference Figure 5 Step S30 can be implemented in the following way:

[0080] Step S301: Perform local coordinate system transformation on each group of data points in the preliminary point cloud dataset to generate point cloud data in the local coordinate system.

[0081] In this embodiment, local coordinate system transformation is the process of converting each group of data points in the initial point cloud dataset from its original coordinate system to a local coordinate system. Data collected by different sensors may have different coordinate systems. Through local coordinate system transformation, these data can be unified into a relatively consistent local coordinate system, facilitating subsequent point cloud registration and data alignment operations. For example, data collected by laser scanners and depth cameras may have different coordinate system origins and coordinate axis directions. Through local coordinate system transformation, they can be converted into a common local coordinate system, making the relationships between the data clearer. From a technical perspective, generating point cloud data in a local coordinate system helps improve the accuracy and efficiency of point cloud registration, laying the foundation for the subsequent generation of fused point cloud data in a unified coordinate system.

[0082] In one embodiment, Principal Component Analysis (PCA) can be used for local coordinate system transformation. PCA is a method for data dimensionality reduction and feature extraction that identifies the principal distribution directions of data points. For each set of data points, the principal component directions are obtained by calculating its covariance matrix and then solving for the eigenvalues ​​and eigenvectors of the covariance matrix. A local coordinate system is then established using the principal component directions as coordinate axes, and the original data points are projected into this local coordinate system to obtain point cloud data in the local coordinate system. This method effectively extracts the main features of the data points, making the local coordinate system more consistent with the actual distribution of the data.

[0083] Step S302: Calculate the relative pose relationship between each group of point cloud data according to the geometric constraints, and generate the initial registration matrix.

[0084] In this embodiment, geometric constraints refer to limitations set based on the geometry and features of the target surface, such as distances and angles between point cloud data. Calculating the relative pose relationships between each set of point cloud data determines their relative positions and orientations in space. The initial registration matrix is ​​a matrix used to describe these relative pose relationships; it can transform one set of point cloud data from its local coordinate system to the coordinate system of another set. For example, when registering point cloud data of two adjacent industrial parts, geometric constraints can be set based on the geometry and installation relationship of the parts, such as the surfaces of the two parts should be parallel and the positions of corresponding holes should be aligned. By calculating the relative pose relationships and generating the initial registration matrix, the two sets of point cloud data can be initially aligned. From a technical perspective, accurately calculating the relative pose relationships and generating the initial registration matrix provides a better initial value for subsequent point cloud registration optimization, improving the efficiency and accuracy of registration.

[0085] In one embodiment, a feature-matching-based method can be used to calculate the relative pose relationship and generate an initial registration matrix. First, some feature points, such as keypoints and edge points, are extracted from each set of point cloud data. Then, by comparing the feature descriptors between these feature points, corresponding feature points between the two sets of point cloud data are found. Based on the geometric relationships between the corresponding feature points, such as distance and angle, the relative pose relationship between the two sets of point cloud data is calculated. Finally, an initial registration matrix is ​​generated based on the relative pose relationship. In practical applications, feature extraction and matching algorithms, such as SIFT and SURF, can be used to improve the accuracy and efficiency of feature matching.

[0086] Step S303: Optimize the initial registration matrix using the iterative nearest point algorithm to generate the optimized registration matrix.

[0087] In this embodiment, the Iterative Closest Point (ICP) algorithm is a point cloud registration optimization algorithm. It iteratively searches for the optimal matching relationship between two sets of point cloud data, thereby optimizing the initial registration matrix. The initial registration matrix may only be a preliminary estimate and contains some error. Optimization through the ICP algorithm can further improve the registration accuracy, making the two sets of point cloud data more accurately aligned. For example, when registering point cloud data of two complex surfaces, the initial registration matrix may only roughly align them, but some deviations still exist. Through iterative optimization of the ICP algorithm, the registration matrix can be continuously adjusted to make the point cloud data of the two surfaces completely overlap. Technically, generating an optimized registration matrix improves the quality of the fused point cloud data, making the subsequently generated high-precision geometric model more accurate.

[0088] In one embodiment, the specific implementation process of the ICP algorithm is as follows: First, based on the initial registration matrix of each group of point cloud data in the preliminary point cloud dataset, the initial relative pose relationship between each group of point cloud data is calculated. Then, one group of point cloud data is selected as the reference point cloud, and the remaining point cloud data are used as the point clouds to be registered. For each point cloud to be registered and the reference point cloud, the nearest point pair between them is calculated using the iterative nearest point algorithm, and an error function is generated. According to the principle of minimizing the error function, the pose parameters of the point cloud to be registered are adjusted to gradually optimize the initial registration matrix. The above iterative process is repeated until the error function converges to a preset threshold range, generating the final optimized registration matrix. In practical applications, some acceleration strategies, such as KD-tree search, can be combined to improve the computational efficiency of the ICP algorithm.

[0089] In one embodiment, reference Figure 6 Step S303: Optimize the initial registration matrix using the iterative nearest point algorithm, which may specifically include:

[0090] Step S3031: Based on the initial registration matrix of each group of point cloud data in the preliminary point cloud dataset, calculate the initial relative pose relationship between each group of point cloud data.

[0091] In this embodiment, the initial relative pose relationship describes the relative position and orientation of each group of point cloud data under the action of the initial registration matrix. While the initial registration matrix provides a preliminary registration result, it may contain some errors. Therefore, further calculation of the initial relative pose relationship is needed to provide a basis for subsequent iterative optimization. For example, when registering point cloud data collected by multiple sensors, the initial registration matrix may only be a rough estimate. By calculating the initial relative pose relationship, the initial differences between each group of point cloud data can be understood more accurately. Technically, accurately calculating the initial relative pose relationship helps determine the starting point of the iteration, improving the convergence speed and optimization effect of the iterative closest point algorithm.

[0092] In one embodiment, the initial relative pose relationship can be calculated using matrix operations. For two sets of point cloud data, let their initial registration matrices be respectively... and Then the initial relative pose relationship matrix between them It can be done Calculated. Wherein yes The inverse matrix. In this way, the initial relative pose relationship between each set of point cloud data can be calculated quickly and accurately.

[0093] Step S3032: Select a set of point cloud data as the reference point cloud, and the remaining point cloud data as the point cloud to be registered.

[0094] In this embodiment, the reference point cloud is a set of point cloud data used as a registration benchmark, while the point cloud to be registered is the point cloud data that needs to be aligned with the reference point cloud. Selecting a suitable reference point cloud can simplify the registration process and improve its efficiency and accuracy. For example, when registering point cloud data collected by multiple sensors, a set of point cloud data with high data quality and wide coverage can be selected as the reference point cloud. From a technical perspective, rationally selecting the reference point cloud and the point cloud to be registered helps to clarify the optimization objective of the iterative nearest point algorithm, making the registration process more targeted.

[0095] In one embodiment, a reference point cloud can be selected based on the quality and coverage of the point cloud data. Point cloud data with high quality, high data density, and coverage of the main features of the target surface can be selected as the reference point cloud. The remaining point cloud data is used as the point cloud to be registered. In practical applications, the quality and coverage of the point cloud data can be evaluated through statistical analysis of each set of point cloud data, such as calculating the number of data points and their distribution density, thereby making a reasonable selection.

[0096] Step S3033: For each point cloud to be registered and a reference point cloud, calculate the nearest point pair between them using the iterative nearest point algorithm and generate an error function.

[0097] In this embodiment, the nearest point pair refers to the data point in the reference point cloud that is closest to each data point in the point cloud to be registered. By calculating the nearest point pair, the correspondence between the point cloud to be registered and the reference point cloud can be found. The error function is an indicator that measures the difference between the point cloud to be registered and the reference point cloud, and it is usually defined based on the distance between the nearest point pairs. For example, for each point cloud to be registered and the reference point cloud, the Euclidean distance between their nearest point pairs can be calculated, and then the sum of the squares of these distances can be used as the value of the error function. From a technical point of view, calculating the nearest point pair and generating the error function are the core steps of the iterative nearest point algorithm, and they provide a basis for subsequent pose parameter adjustments.

[0098] In one embodiment, a KD-tree search algorithm can be used to compute nearest point pairs. A KD-tree is an efficient spatial search data structure that can quickly find the nearest point in the reference point cloud for each data point in the point cloud to be registered. For each point cloud to be registered and a reference point cloud, first, a KD-tree for the reference point cloud is constructed. Then, for each data point in the point cloud to be registered, a search is performed in the KD-tree to find its nearest point. An error function is generated based on the distance between the nearest point pairs. The error function can be in the form of a sum of squares, i.e. Where n is the number of nearest point pairs, and di is the distance between the i-th nearest point pair. This method allows for efficient and accurate calculation of nearest point pairs and the error function.

[0099] Step S3034: Adjust the pose parameters of the point cloud to be registered according to the principle of minimizing the error function, so as to gradually optimize the initial registration matrix.

[0100] In this embodiment, minimizing the error function is the core optimization objective of the iterative nearest-point algorithm. The pose parameters of the point cloud to be registered include translation and rotation parameters. By adjusting these parameters, the position and orientation of the point cloud in space can be changed, thereby gradually reducing the value of the error function. For example, when the value of the error function is large, it indicates a significant difference between the point cloud to be registered and the reference point cloud, requiring adjustment of the pose parameters to reduce this difference. After each adjustment of the pose parameters, the nearest point pair and the error function are recalculated until the value of the error function reaches its minimum or meets the preset convergence condition. Technically, adjusting the pose parameters according to the principle of minimizing the error function can progressively optimize the initial registration matrix and improve the registration accuracy.

[0101] In one embodiment, the least squares method can be used to adjust the pose parameters of the point cloud to be registered. The least squares method is a method for finding optimal parameters by minimizing the sum of squared errors. For each iteration, a system of linear equations is constructed based on the current nearest point pairs and the error function, and then this system of equations is solved to obtain the adjustment amount of the pose parameters. The adjustment amount is applied to the pose parameters of the point cloud to be registered, updating the initial registration matrix. By continuously iterating this process, the initial registration matrix is ​​gradually optimized. In practical applications, some optimization algorithms, such as the Gauss-Newton method, can be combined to improve the efficiency and accuracy of the solution.

[0102] Step S3035: Repeat the above iterative process until the error function converges to the preset threshold range, and generate the final optimized registration matrix.

[0103] In this embodiment, the convergence of the error function to a preset threshold indicates that the difference between the point cloud to be registered and the reference point cloud has been reduced to an acceptable level. The preset threshold is a value set according to the actual application requirements and accuracy requirements. When the value of the error function is less than the preset threshold, it indicates that the registration process has achieved the expected accuracy, and the iteration can be stopped to generate the final optimized registration matrix. For example, when registering point cloud data of two high-precision parts, the preset threshold can be set relatively small to ensure the registration accuracy. From a technical perspective, repeating the iteration process until the error function converges ensures that the generated final optimized registration matrix has high accuracy, providing a reliable guarantee for the subsequent generation of fused point cloud data in a unified coordinate system.

[0104] In one embodiment, after each iteration, the value of the error function can be checked to see if it is less than a preset threshold. If it is less than the preset threshold, the iteration stops, and the final optimized registration matrix is ​​output; otherwise, the next iteration continues. To avoid the iteration process from getting stuck in an infinite loop, a maximum number of iterations can be set. When the maximum number of iterations is reached, even if the value of the error function has not yet converged to the preset threshold range, the iteration stops, and the current registration matrix is ​​output as the final result. In this way, the effectiveness and stability of the iteration process can be guaranteed.

[0105] Step S304: Based on the optimized registration matrix, map each group of point cloud data to a unified coordinate system to generate fused point cloud data.

[0106] In this embodiment, the optimized registration matrix accurately describes the relative pose relationships between the various groups of point cloud data. By mapping the various groups of point cloud data according to the optimized registration matrix, they can be unified into a common coordinate system, forming fused point cloud data. For example, when fusing point cloud data acquired by a laser scanner and a depth camera, the optimized registration matrix maps them to a unified coordinate system, allowing data acquired by different sensors to be accurately superimposed to form a complete and accurate fused point cloud data. Technically, generating fused point cloud data integrates information from multi-source point cloud data, providing a richer and more accurate data foundation for subsequently generating high-precision geometric models of the target surface.

[0107] In one embodiment, matrix multiplication can be used to map each set of point cloud data to a unified coordinate system. For each data point in each set of point cloud data, its coordinates in the local coordinate system are represented as a vector. Then, the optimized registration matrix is ​​multiplied by this vector to obtain the coordinates of the data point in the unified coordinate system. After transforming the coordinates of all data points, the fused point cloud data in the unified coordinate system is obtained. In practice, this process can be implemented using a computer program to improve computational efficiency.

[0108] In one embodiment, reference Figure 7 Step S40: Based on the fused point cloud data, a high-precision geometric model of the target surface is generated using a surface reconstruction algorithm, including:

[0109] Step S401: Extract the boundary point set of the target surface from the fused point cloud data, and generate an initial triangular mesh model based on the boundary point set.

[0110] In this embodiment, the boundary point set of the target surface refers to the set of points located at the edge of the target surface. These boundary points are crucial for determining the shape and extent of the target surface. Extracting the boundary point set from fused point cloud data can be achieved by analyzing the local geometric features of the data points. For example, for point cloud data of a 3D object, boundary points typically have different curvatures and neighborhood distributions than interior points. Generating an initial triangular mesh model from the boundary point set is the process of connecting discrete boundary points into a triangular mesh. A triangular mesh model is a 3D geometric representation method that can conveniently describe the shape of the target surface. Technically, generating an initial triangular mesh model provides a foundation for subsequent topology optimization and geometric detail enhancement, contributing to the construction of a more accurate, high-precision geometric model.

[0111] In one embodiment, a curvature-based method can be used to extract the boundary point set. First, the curvature value of each data point in the fused point cloud data is calculated. Points with larger curvature values ​​are usually located at boundaries or sharp features. Then, based on a preset curvature threshold, data points with curvature values ​​greater than the threshold are marked as boundary points.

[0112] Step S402: Perform topology optimization on the initial triangular mesh model to eliminate non-manifold structures in the mesh.

[0113] In this embodiment, non-manifold structures refer to parts of a triangular mesh model that do not conform to the definition of a manifold. A manifold is a topological space that locally possesses Euclidean space properties. In three-dimensional space, a triangular mesh model with a manifold structure should have good connectivity and directionality. Non-manifold structures may cause problems in subsequent processing and analysis, such as producing incorrect results when performing geometric detail addition and surface smoothing. The purpose of topology optimization is to eliminate non-manifold structures by adjusting the connectivity of the triangular mesh, making the mesh more regular and stable. For example, there may be some isolated triangles or discontinuous boundaries in the initial triangular mesh model; topology optimization can solve these problems. Technically, eliminating non-manifold structures can improve the quality of the triangular mesh model, providing a more reliable foundation for the subsequent generation of high-precision geometric models.

[0114] In one embodiment, topology optimization can be performed using mesh simplification and mesh repair methods. Mesh simplification reduces mesh complexity by merging some adjacent triangles while maintaining the basic shape of the mesh. Mesh repair addresses non-manifold structures, such as connecting isolated triangles to adjacent meshes or repairing discontinuous boundaries. Open-source mesh processing libraries, such as CGAL (Computational GeometryAlgorithms Library), can be used, which provides a rich set of mesh topology optimization algorithms and tools.

[0115] Step S403: Based on the optimized triangular mesh model, use an interpolation algorithm to supplement the geometric details of the target surface and generate a high-precision geometric model.

[0116] In this embodiment, although the optimized triangular mesh model already possesses a basic shape, it may lack some detailed information. Interpolation algorithms are methods for estimating unknown data points using known data points. Using interpolation algorithms, geometric details of the target surface can be supplemented based on the triangular mesh model. For example, for targets with minor protrusions or depressions on their surfaces, the optimized triangular mesh model may not accurately represent these details. Interpolation algorithms can generate new geometric detail points in these areas, making the model more realistic. Technically, supplementing geometric details can improve the quality of high-precision geometric models, making them more accurately reflect the actual shape of the target surface.

[0117] In one embodiment, key geometric feature points, such as vertices and edge points, of the target surface can be extracted from the optimized triangular mesh model, and the geometric relationships between these feature points can be calculated. Based on the geometric relationships of the key geometric feature points, a suitable interpolation algorithm is selected, such as linear interpolation, spline interpolation, or radial basis function interpolation. Linear interpolation is an interpolation method that estimates the value of intermediate points by connecting adjacent feature points; spline interpolation can generate smoother curves; and radial basis function interpolation can handle more complex geometries. Using the selected interpolation algorithm, new geometric detail points are generated between the key geometric feature points, and then these newly generated geometric detail points are merged with the original triangular mesh model to form a high-precision geometric model containing more geometric details.

[0118] Step S404: Smooth the high-precision geometric model to obtain the final high-precision geometric model.

[0119] In this embodiment, smoothing is used to remove noise and irregularities from the surface of the high-precision geometric model, making the model surface smoother. During the generation of the high-precision geometric model, some noise or irregular geometric shapes may be introduced; for example, new points generated by interpolation algorithms may cause tiny bumps or depressions on the surface. Smoothing can be achieved by averaging or filtering the vertices of the model surface. For example, for each vertex, the average value of its neighboring vertices is calculated, and then the vertex's position is updated to the average value. Technically, smoothing can improve the quality and aesthetics of the final high-precision geometric model, making it more suitable for practical applications.

[0120] In one embodiment, the Laplacian smoothing algorithm can be used to smooth high-precision geometric models. The Laplacian smoothing algorithm is a smoothing method based on vertex neighborhood information. It calculates the difference between each vertex and its neighboring vertices and adjusts the vertices according to certain weights. For each vertex, the centroid of its neighboring vertices is calculated, and then the vertex is moved a certain distance towards the centroid. Through multiple iterations of this process, the model surface can be gradually smoothed. In practical applications, the smoothing intensity and the number of iterations can be adjusted according to the specific characteristics of the model to achieve the best smoothing effect.

[0121] In one embodiment, based on the optimized triangular mesh model, the geometric details of the target surface are supplemented using an interpolation algorithm, which may include:

[0122] 1. Extract key geometric feature points of the target surface from the optimized triangular mesh model and calculate the geometric relationships between these feature points.

[0123] In this embodiment, key geometric feature points refer to representative and significant points on the target surface, such as vertices, edge points, and points with significant curvature changes. These feature points reflect the main shape and structural information of the target surface. Extracting key geometric feature points from the optimized triangular mesh model can be achieved by analyzing the topology and geometric properties of the triangular mesh. For example, for a complex triangular mesh model of an industrial part, vertices and edge points can be determined by examining the connection relationships of the triangles, while points with significant curvature changes can be filtered by calculating the curvature value of each vertex. Calculating the geometric relationships between these feature points, such as distance and angle, helps in selecting a suitable interpolation algorithm. Technically, accurately extracting key geometric feature points and calculating their geometric relationships provides a foundation for subsequent supplementation of geometric details, improving the accuracy and effectiveness of interpolation.

[0124] In one embodiment, curvature analysis can be used to extract key geometric feature points. For each vertex in the optimized triangular mesh model, its curvature value is calculated. Vertices with larger curvature values ​​typically indicate greater surface curvature variation and are candidate points for key geometric feature points. A curvature threshold can be set, and vertices with curvature values ​​greater than this threshold are marked as key geometric feature points. For the extracted key geometric feature points, the Euclidean distance and included angle between them are calculated, and this geometric relationship information is stored to provide a basis for subsequent interpolation algorithms.

[0125] 2. Based on the geometric relationships of key geometric feature points, select an appropriate interpolation algorithm, including linear interpolation, spline interpolation, or radial basis function interpolation.

[0126] In this application, different interpolation algorithms have different characteristics and applicable scenarios. Linear interpolation is a simple and direct interpolation method that generates a linear interpolation result between two points by connecting adjacent key geometric feature points. Linear interpolation is suitable for regions with relatively simple geometric relationships and gentle changes. Spline interpolation can generate smoother curves. It generates an interpolation curve by fitting a set of control points and is suitable for regions that require smooth transitions. Radial basis function interpolation is a distance-based interpolation method. It estimates the value of unknown points by placing a radial basis function around each key geometric feature point and then weighting and summing all radial basis functions. It is suitable for handling complex geometric shapes and irregular data distributions. Choosing a suitable interpolation algorithm based on the geometric relationships of key geometric feature points, such as distance, angle, and distribution density, can improve the accuracy and effectiveness of interpolation. From a technical perspective, choosing a suitable interpolation algorithm can more accurately supplement the geometric details of the target surface, making the generated high-precision geometric model more realistic.

[0127] In one embodiment, if the distance between key geometric feature points is small and their distribution is relatively uniform, and the geometric relationship changes smoothly, a linear interpolation algorithm can be selected. Linear interpolation algorithms are computationally simple, efficient, and can quickly generate interpolation results. If a smooth transition curve needs to be generated between key geometric feature points, such as when processing the rounded corners of a surface, a spline interpolation algorithm can be selected. Spline interpolation algorithms can generate smooth curves based on the position of key geometric feature points and the tangent direction, making the model surface more natural. For regions with complex geometries and irregular distribution of key geometric feature points, a radial basis function interpolation algorithm can be selected. Radial basis function interpolation algorithms have strong flexibility and adaptability, and can handle various complex geometric relationships.

[0128] 3. Using the selected interpolation algorithm, new geometric detail points are generated between key geometric feature points to fill the geometric blank areas on the target surface.

[0129] In this embodiment, geometric blank areas refer to regions on the target surface that lack sufficient detail information in the optimized triangular mesh model. These blank areas can be filled by generating new geometric detail points between key geometric feature points, making the model more complete. The selected interpolation algorithm estimates the positions of points within these blank areas based on the geometric relationships of the key geometric feature points and known information. For example, on the surface of an object with complex texture, the optimized triangular mesh model may only represent the approximate shape of the object, while blank areas exist in the texture details. Interpolation algorithms can generate new geometric detail points in these blank areas, simulating the texture effect. Technically, filling geometric blank areas can improve the completeness and accuracy of the high-precision geometric model, making it more consistent with the actual situation of the target surface.

[0130] 4. Merge the newly generated geometric detail points with the original triangular mesh model to form a high-precision geometric model containing more geometric details.

[0131] In this embodiment, merging the newly generated geometric detail points with the original triangular mesh model aims to integrate the supplementary geometric detail information into the model, forming a complete high-precision geometric model. The merging process needs to consider the topological relationship between the newly generated geometric detail points and the original triangular mesh model to ensure good connectivity and consistency in the merged model. For example, the newly generated geometric detail points need to be reasonably connected to the triangles in the original triangular mesh model to avoid isolated points or discontinuous boundaries. Technically, merging the newly generated geometric detail points can make the high-precision geometric model richer and more accurate, better reflecting the actual shape and details of the target surface.

[0132] In one embodiment, triangulation can be used to merge newly generated geometric detail points with the original triangular mesh model. First, the newly generated geometric detail points are added to the vertex set of the original triangular mesh model. Then, triangulation is performed on the point set containing all vertices to generate a new triangular mesh. During triangulation, the topology of the original triangular mesh model needs to be considered, preserving as many original triangle connections as possible. Triangulation algorithms, such as the Delaunay triangulation algorithm, can be used to ensure that the generated triangular mesh has good geometric properties. The newly generated triangular mesh serves as a high-precision geometric model containing more geometric details.

[0133] In one embodiment, reference Figure 8 Step S50: Based on the high-precision geometric model and combined with the task allocation strategy of the explosion-proof robot cluster, a spraying path scheme for the explosion-proof robot is generated, which may specifically include:

[0134] Step S501: Calculate the number of spraying units required for the spraying task based on the target surface area of ​​the high-precision geometric model and the coverage of the spraying material.

[0135] In this embodiment, the target surface area of ​​the high-precision geometric model refers to the total area of ​​the surface to be coated. The coverage capacity of the coating material refers to the area that a unit area of ​​coating material can cover, which is related to the properties of the coating material, the coating process, and other factors. Calculating the number of coating units required for the coating task is to rationally plan the task and ensure sufficient coverage of the target surface. For example, when coating the surface of a large industrial piece of equipment, given its target surface area and the coverage capacity of the coating material, calculations can determine how many coating units are needed to complete the coating task. Technically, accurately calculating the number of coating units can avoid waste or insufficiency of coating material, improving the efficiency and quality of the coating operation.

[0136] In one embodiment, the target surface area can be obtained by calculating the surface area of ​​a high-precision geometric model. For a triangular mesh model, the area of ​​each triangle can be added together to obtain the total area. The coverage capacity of the spraying material can be determined through experimental testing or by referring to the technical parameters of the spraying material. Dividing the target surface area by the coverage capacity of the spraying material yields the number of spraying units required for the spraying task. In actual calculations, considering factors such as losses and overlaps during the spraying process, a certain margin can be appropriately added.

[0137] Step S502: Generate a spraying task allocation table for each explosion-proof robot based on the number of spraying units and the number of explosion-proof robot clusters.

[0138] In this embodiment, the spraying task allocation table is used to clarify the spraying tasks that each explosion-proof robot needs to undertake. Based on the number of spraying units and the number of explosion-proof robot clusters, the spraying tasks of each robot can be reasonably allocated, making the workload of each robot relatively balanced. For example, if there are 10 spraying units and 2 explosion-proof robots, 5 spraying units can be allocated to each robot. Technically, generating a reasonable spraying task allocation table can improve the working efficiency of the explosion-proof robot cluster and avoid situations where some robots have excessively heavy or light workloads.

[0139] In one embodiment, an average allocation method can be used to generate the spraying task assignment table. The number of spraying units is divided by the number of explosion-proof robots in the cluster to obtain the average number of spraying units allocated to each robot. If the division is not even, the remainder can be distributed to some of the robots sequentially. For example, with 11 spraying units and 3 explosion-proof robots, each robot is initially assigned 3 spraying units, and the remaining 2 units are distributed to 2 of those robots. Furthermore, considering the differences in robot performance and work efficiency, the allocation results can be adjusted appropriately.

[0140] Step S503: Based on the spraying task allocation table and the geometric features of the high-precision geometric model, generate a corresponding spraying path scheme for each explosion-proof robot.

[0141] In this embodiment, the geometric features of the high-precision geometric model include information such as surface curvature, shape, and boundaries. These geometric features affect the planning of the spraying path. Based on the spraying task allocation table and the geometric features of the high-precision geometric model, generating a corresponding spraying path scheme for each explosion-proof robot can ensure that the robot can complete the spraying task efficiently and accurately. For example, for areas with large surface curvature, the spraying path may need to be denser and more detailed; for edge areas, special attention needs to be paid to the coverage area of ​​the spraying. From a technical perspective, generating a reasonable spraying path scheme can improve the uniformity and quality of spraying, and avoid missed spraying and repeated spraying.

[0142] In one embodiment, the priority areas for the spraying task are first divided based on the geometric characteristics of the high-precision geometric model, including high-curvature areas, low-curvature areas, and edge areas. High-curvature areas typically require more detailed spraying due to greater surface variations; low-curvature areas are relatively smooth and can use a more precise spraying path; edge areas require ensuring the integrity of the spraying. Then, according to the spraying task allocation table, a corresponding priority area is assigned to each explosion-proof robot. For each explosion-proof robot, an initial spraying path scheme is generated based on its assigned priority area and the coverage capacity of the spraying material. A grid-based path planning method can be used, dividing the priority area into several small grids, and then spraying the grids sequentially in a certain order. In practical applications, the initial spraying path scheme can also be optimized based on the robot's motion capabilities and the characteristics of the spraying equipment.

[0143] In one embodiment, reference Figure 9 Step S503 can be implemented in the following way:

[0144] Step S5031: Based on the geometric features of the target surface of the high-precision geometric model, divide the priority areas of the spraying task, including high curvature areas, low curvature areas and edge areas.

[0145] In this embodiment, the geometric features of the target surface in the high-precision geometric model reflect the shape and structural characteristics of the target surface. High-curvature regions refer to areas with significant changes in surface curvature. In these regions, the surface shape changes drastically, making spraying relatively difficult and requiring more precise spraying operations to ensure coating uniformity and quality. For example, on the surface of some complex mechanical parts, protruding or recessed areas typically have high curvature. Low-curvature regions, on the other hand, are relatively smooth surfaces with minimal curvature changes, making spraying operations relatively simple and allowing for more conventional spraying methods. Edge regions are the boundary parts of the target surface; these areas are prone to missed spraying and require special attention to ensure coating integrity. Prioritizing spraying tasks by dividing them into priority regions helps to rationally allocate spraying resources and formulate spraying strategies based on the characteristics of different regions. From a technical perspective, accurately dividing priority regions can improve the targeting and efficiency of spraying, and enhance the overall spraying quality.

[0146] In one embodiment, high-curvature regions and low-curvature regions can be divided by calculating the curvature value of each point in a high-precision geometric model. A curvature threshold is set, and regions containing points with curvature values ​​greater than the threshold are classified as high-curvature regions, while regions with curvature values ​​less than the threshold are classified as low-curvature regions. Edge regions can be determined by detecting triangles located at the boundaries in a triangular mesh model. In practice, various geometric analysis tools and algorithms can be used to achieve region division.

[0147] Step S5032: According to the spraying task allocation table, assign a corresponding priority area to each explosion-proof robot.

[0148] In this embodiment, the spraying task allocation table specifies the amount of spraying work each explosion-proof robot needs to undertake. Based on this allocation table, priority areas are rationally distributed among each explosion-proof robot, ensuring a relatively balanced workload for each robot and allowing them to fully utilize their performance advantages. For example, if an explosion-proof robot has high spraying precision, it can be assigned spraying tasks in high-curvature areas; while robots with faster spraying speeds can be assigned tasks in low-curvature areas. Technically, rationally allocating priority areas can improve the working efficiency of the explosion-proof robot cluster, avoiding situations where some robots are idle or overworked, while ensuring that each area receives appropriate spraying treatment.

[0149] In one embodiment, a greedy algorithm can be used to allocate priority areas based on the performance parameters of each explosion-proof robot and the workload in the spraying task allocation table. First, the spraying difficulty and workload of each priority area are evaluated. Then, areas are allocated sequentially from highest to lowest robot performance until all areas are allocated. During the allocation process, it must be ensured that the workload of each robot does not exceed its maximum carrying capacity.

[0150] Step S5033: For each explosion-proof robot, generate an initial spraying path scheme based on its assigned priority area and the coverage capacity of the spraying material.

[0151] In this embodiment, the coverage capacity of the sprayed material refers to the area that a unit area of ​​sprayed material can cover. It is related to the properties of the sprayed material, the spraying process, and the parameters of the spraying equipment. By combining the priority area assigned to each explosion-proof robot with the coverage capacity of the sprayed material, an initial spraying path scheme suitable for that robot can be generated. For high-curvature areas, due to the need for finer spraying, the path may need to be denser and more complex; for low-curvature areas, the path can be relatively simple and sparse. For example, in high-curvature areas, a spiral spraying path can be used to ensure the uniformity of the coating; in low-curvature areas, a straight path can be used to improve spraying efficiency. From a technical perspective, generating a reasonable initial spraying path scheme can improve the quality and efficiency of spraying and reduce waste of sprayed material.

[0152] In one embodiment, each priority region can be divided into several smaller sub-regions. The spraying sequence and parameters for each sub-region are then determined based on the coverage capacity of the spraying material. Taking a high-curvature region as an example, the spraying can begin from the center and expand outwards using a spiral path, adjusting the spraying speed and amount according to changes in curvature. For low-curvature regions, a parallel straight-line path can be used for spraying. After generating the initial spraying path scheme, it can be optimized and adjusted according to actual conditions to adapt to different working scenarios and requirements.

[0153] Accordingly, to better implement the above methods, this application also provides a multi-source point cloud-driven dynamic spraying control system for explosion-proof robot clusters. For example... Figure 10 As shown, the multi-source point cloud-driven dynamic spraying control system 80 for explosion-proof robot clusters includes:

[0154] The acquisition module 801 is used to acquire multi-source point cloud data of the target surface through multiple sensors. The multi-source point cloud data includes three-dimensional geometric information acquired by a laser scanner, texture information captured by a depth camera, and temperature distribution information recorded by an infrared sensor.

[0155] The preliminary processing module 802 is used to perform preliminary processing on the multi-source point cloud data to generate a preliminary point cloud dataset. The preliminary processing includes removing noise points and redundant data.

[0156] The point cloud processing module 803 is used to input the preliminary point cloud dataset into the point cloud fusion algorithm based on geometric constraints, and generate fused point cloud data in a unified coordinate system through point cloud registration and data alignment operations.

[0157] The reconstruction module 804 is used to generate a high-precision geometric model of the target surface based on the fused point cloud data and using a surface reconstruction algorithm.

[0158] The spraying scheme generation module 805 is used to generate a spraying path scheme for the explosion-proof robot based on the high-precision geometric model and the task allocation strategy of the explosion-proof robot cluster. The spraying path scheme includes the spraying start position, the spraying end position, and the spraying trajectory point sequence.

[0159] The instruction sending module 806 is used to send spraying task instructions to the explosion-proof robot cluster according to the spraying path plan, so that each robot can perform spraying operations according to the spraying path plan.

[0160] The implementation details of each module are provided in the preceding method embodiments and will not be repeated here. The technical effects achieved by each module and device are described in the foregoing method embodiments.

[0161] like Figure 11 As shown, this application embodiment also provides a computer device 90, which includes a processor 901 and a memory 902, wherein the memory 902 stores a computer program, and when the computer program is executed by the processor 901, the processor 901 performs the steps of any of the methods described above.

[0162] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A method for dynamic spraying control of explosion-proof robot clusters driven by multi-source point cloud, characterized in that, Includes the following steps: Multi-source point cloud data of the target surface is acquired through multiple sensors. The multi-source point cloud data includes three-dimensional geometric information collected by a laser scanner, texture information captured by a depth camera, and temperature distribution information recorded by an infrared sensor. The multi-source point cloud data is subjected to preliminary processing to generate a preliminary point cloud dataset. The preliminary processing includes removing noise points and redundant data. The initial point cloud dataset is input into a point cloud fusion algorithm based on geometric constraints, and fused point cloud data in a unified coordinate system is generated through point cloud registration and data alignment operations. Based on the fused point cloud data, a high-precision geometric model of the target surface is generated using a surface reconstruction algorithm. Based on the high-precision geometric model and combined with the task allocation strategy of the explosion-proof robot cluster, a spraying path scheme for the explosion-proof robot is generated. The spraying path scheme includes the spraying start position, the spraying end position, and the spraying trajectory point sequence. The spraying task instruction is sent to the explosion-proof robot cluster according to the spraying path plan, so that each robot can perform the spraying operation according to the spraying path plan. The process involves generating a spraying path scheme for the explosion-proof robots based on the high-precision geometric model and the task allocation strategy of the explosion-proof robot cluster. This includes: calculating the number of spraying units required for the spraying task based on the target surface area and the coverage capacity of the spraying material in the high-precision geometric model; generating a spraying task allocation table for each explosion-proof robot based on the number of spraying units and the number of explosion-proof robots in the cluster; dividing the target surface geometric features of the high-precision geometric model into priority regions for the spraying task, including high-curvature regions, low-curvature regions, and edge regions; allocating a corresponding priority region to each explosion-proof robot according to the spraying task allocation table; and generating an initial spraying path scheme for each explosion-proof robot, combining its allocated priority region and the coverage capacity of the spraying material, using a spiral spraying path for high-curvature regions and a straight path for low-curvature regions.

2. The method according to claim 1, characterized in that, The multi-source point cloud data is preliminarily processed to generate a preliminary point cloud dataset, including: Obtain the spatial distribution density value of each group of data points in the multi-source point cloud data, and calculate the average spatial distribution density value of each group of data points; For each group of data points in multi-source point cloud data, if the spatial distribution density value of the data points in the group is less than the first preset density threshold, the data points in the group are removed to eliminate noise points. For the removed data points, calculate the distance deviation between each group of data points and its adjacent data points, and remove redundant data points based on the distance deviation values ​​to obtain a preliminary point cloud dataset.

3. The method according to claim 2, characterized in that, The initial point cloud dataset is input into a geometrically constrained point cloud fusion algorithm. Point cloud registration and data alignment operations are then performed to generate fused point cloud data in a unified coordinate system, including: Perform local coordinate system transformation on each group of data points in the initial point cloud dataset to generate point cloud data in the local coordinate system; Based on geometric constraints, the relative pose relationships between each set of point cloud data are calculated, and an initial registration matrix is ​​generated. The initial registration matrix is ​​optimized using the iterative nearest point algorithm to generate the optimized registration matrix; Based on the optimized registration matrix, the point cloud data of each group are mapped to a unified coordinate system to generate fused point cloud data.

4. The method according to claim 3, characterized in that, The initial registration matrix is ​​optimized using the iterative nearest-point algorithm, including: Based on the initial registration matrix of each group of point cloud data in the preliminary point cloud dataset, the initial relative pose relationship between each group of point cloud data is calculated. Select a set of point cloud data as a reference point cloud, and use the remaining point cloud data as the point cloud to be registered; For each point cloud to be registered and a reference point cloud, the nearest point pair between them is calculated using the iterative nearest point algorithm, and an error function is generated. Based on the principle of minimizing the error function, the pose parameters of the point cloud to be registered are adjusted to gradually optimize the initial registration matrix; The process of optimizing the initial registration matrix using the iterative nearest point algorithm is repeated until the error function converges to a preset threshold range, generating the final optimized registration matrix.

5. The method according to claim 3, characterized in that, Based on the fused point cloud data, a high-precision geometric model of the target surface is generated using a surface reconstruction algorithm, including: Extract the boundary point set of the target surface from the fused point cloud data, and generate an initial triangular mesh model based on the boundary point set; The initial triangular mesh model is subjected to topology optimization to eliminate non-manifold structures in the mesh; Based on the optimized triangular mesh model, an interpolation algorithm is used to supplement the geometric details of the target surface, generating a high-precision geometric model. The high-precision geometric model is smoothed to obtain the final high-precision geometric model.

6. The method according to claim 5, characterized in that, Based on the optimized triangular mesh model, interpolation algorithms are used to supplement the geometric details of the target surface, including: Extract key geometric feature points from the target surface from the optimized triangular mesh model and calculate the geometric relationships between these feature points; Based on the geometric relationships of key geometric feature points, select an interpolation algorithm, including linear interpolation, spline interpolation, or radial basis function interpolation; Using the selected interpolation algorithm, new geometric detail points are generated between key geometric feature points to fill the geometric gaps on the target surface; The newly generated geometric detail points are merged with the original triangular mesh model to form a high-precision geometric model containing more geometric details.

7. The method according to claim 2, characterized in that, For each group of data points in multi-source point cloud data, if the spatial distribution density value of the data point group is less than a first preset density threshold, then the data point group is removed to eliminate noise points, including: Based on the spatial distribution characteristics of multi-source point cloud data, the spatial distribution density value of each group of data points is calculated, where the spatial distribution density value represents the number of data points per unit volume. A first preset density threshold is set, which is determined based on the material properties of the target surface and the level of environmental noise. For each set of data points, compare its spatial distribution density value with a first preset density threshold; If the spatial distribution density value of a certain group of data points is lower than the first preset density threshold, then the group of data points is determined to be noise points and removed from the multi-source point cloud data.

8. A multi-source point cloud-driven dynamic spraying control system for explosion-proof robot clusters, characterized in that, The system includes: The acquisition module is used to acquire multi-source point cloud data of the target surface through multiple sensors. The multi-source point cloud data includes three-dimensional geometric information collected by a laser scanner, texture information captured by a depth camera, and temperature distribution information recorded by an infrared sensor. The preliminary processing module is used to perform preliminary processing on the multi-source point cloud data to generate a preliminary point cloud dataset. The preliminary processing includes removing noise points and redundant data. The point cloud processing module is used to input the preliminary point cloud dataset into the point cloud fusion algorithm based on geometric constraints, and generate fused point cloud data in a unified coordinate system through point cloud registration and data alignment operations. The reconstruction module is used to generate a high-precision geometric model of the target surface based on the fused point cloud data and using a surface reconstruction algorithm. The spraying scheme generation module is used to generate a spraying path scheme for the explosion-proof robots based on the high-precision geometric model and the task allocation strategy of the explosion-proof robot cluster. The spraying path scheme includes the spraying start position, the spraying end position, and the sequence of spraying trajectory points. Specifically, it is used to calculate the number of spraying units required for the spraying task based on the target surface area and the coverage capacity of the spraying material in the high-precision geometric model; generate a spraying task allocation table for each explosion-proof robot based on the number of spraying units and the number of explosion-proof robots in the cluster; divide the priority areas of the spraying task based on the geometric features of the target surface in the high-precision geometric model, including high curvature areas, low curvature areas, and edge areas; allocate a corresponding priority area to each explosion-proof robot according to the spraying task allocation table; and generate an initial spraying path scheme for each explosion-proof robot based on its allocated priority area and the coverage capacity of the spraying material, using a spiral spraying path for high curvature areas and a straight path for low curvature areas. The instruction sending module is used to send spraying task instructions to the explosion-proof robot cluster according to the spraying path plan, so that each robot can perform spraying operations according to the spraying path plan.

Images