Measurement method of ship waterline based on dynamic segmentation of point cloud freeboard of laser radar

Abstract

The present invention relates to the field of image processing technology, and specifically to a method for measuring the waterline of a ship by dynamically segmenting the freeboard of a point cloud based on a laser radar, comprising the following steps: acquiring point cloud data through a laser radar, selecting multiple scale levels from the point cloud data, and decomposing the data of each level respectively to generate a multi-level point cloud representation; calculating the curvature and normal vector of each level of the point cloud data, extracting morphological features, and obtaining key morphological features. The present invention adopts a laser radar to obtain multi-level point cloud data, thereby improving the accuracy and efficiency of the measurement of the waterline of the ship. The calculation of curvature and normal vector is utilized, combined with the extraction of morphological features, to enhance the data recognition capability in a complex marine environment. By performing multi-level classification on the point cloud data and dynamic adjustment of the time series, real-time monitoring of changes in ship load is achieved, ensuring accurate reflection of real-time data.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a method for measuring a ship's waterline by dynamically segmenting a point cloud freeboard based on a laser radar. Background Art

[0002] The measurement of a ship's waterline is crucial for safe ship operation and performance evaluation. Measuring the waterline can help determine the ship's load status and stability, and is a key parameter in maritime safety and ship design. By analyzing the ship's waterline image, it is possible to monitor overload or imbalance, so that timely adjustments can be made to ensure the ship's safe navigation.

[0003] However, existing technologies mainly rely on traditional visual recognition methods, which are easily disturbed in complex environments with changing lighting and water surface reflections, affecting the stability and accuracy of measurement results. They are particularly inadequate in dynamic environments and are difficult to respond to changes in the ship's load status in real time, limiting their application in maritime operations that require rapid response. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a measurement method for ship waterline by dynamically segmenting point cloud freeboard based on laser radar.

[0005] In order to achieve the above object, the present invention adopts the following technical scheme, a method for measuring the waterline of a ship based on the dynamic segmentation of the point cloud freeboard of a laser radar, comprising the following steps:

[0006] Acquire point cloud data through a laser radar, select multiple scale levels from the point cloud data, decompose the data of each level respectively, and generate a multi-level point cloud representation; calculate the curvature and normal vector of each level of point cloud data, extract morphological features, and obtain key morphological features;

[0007] Perform multi-level classification on the key morphological features, mark the results of each classification on the corresponding point cloud data, and obtain labeled data; sort the labeled data according to the time series, compare and adjust the sorted labeled data frame by frame, and generate dynamically adjusted point cloud data;

[0008] Based on the dynamically adjusted point cloud data, the wave point cloud probability distribution is calculated, and a model is established to generate a wave characteristic distribution according to the model; a threshold is adjusted according to the wave characteristic distribution, and real-time segmentation is performed to obtain segmented point cloud data;

[0009] The error points of the segmented point cloud data are eliminated, the segmented point cloud data are cleaned using geometric matching, and the boundary between the ship's side and the waterline is divided by boundary strength to generate a segmentation result of the ship's side and the waterline.

[0010] Preferably, the steps of obtaining the multi-level point cloud representation are:

[0011] Obtain point cloud data through laser radar, select and decompose each level in the point cloud data to form a multi-level point cloud structure;

[0012] Based on the multi-level point cloud structure, the point cloud data decomposition coefficient of each level is calculated, and the calculation formula is:

[0013] ;

[0014] in, For the The point cloud data decomposition coefficient of the layer, For the Tier The coordinate values ​​of the points, is the total number of points in each layer of point cloud;

[0015] The point cloud data decomposition coefficients are combined to construct and improve a multi-level point cloud representation to obtain a multi-level point cloud representation.

[0016] Preferably, the steps of acquiring the key morphological features are:

[0017] Selecting a level of point cloud data from the multi-level point cloud representation, performing local processing on the selected point cloud data, and obtaining a local point cloud set for each point;

[0018] Based on the local point cloud set, the curvature and normal vector of each point are calculated using the following formula:

[0019] ;

[0020] in, For the The curvature value of a point, and are the normal vectors of the neighboring points and the current point respectively, For the The number of neighboring points of a point;

[0021] Based on the curvature value and the normal vector, the morphological features of the point cloud data are extracted to obtain key morphological features.

[0022] Preferably, the steps of acquiring the labeled data are:

[0023] Based on the key morphological features, multi-level classification is performed to calculate the probability of each data point belonging to each category. The calculation formula is:

[0024] ;

[0025] in, For the Points belong to the category The probability of For the Points for category The score, is the total number of categories, Representative Points for category score;

[0026] Based on the classification probability, a classification label is assigned to each point cloud data point to form labeled data.

[0027] Preferably, the steps of acquiring the dynamically adjusted point cloud data are:

[0028] Arrange and sort the labeled data in chronological order, check whether the time label of each frame of data is correct, and form a labeled data set arranged in time series;

[0029] Based on the annotated data set arranged in time series, data comparison is performed frame by frame, and for each frame of annotated data, the difference with the previous frame is checked, and the changed point cloud part is identified;

[0030] The point cloud portion with the change identified is reclassified and the label is updated to generate dynamically adjusted point cloud data matching the time change.

[0031] Preferably, the steps of acquiring the wave characteristic distribution are:

[0032] Based on the dynamically adjusted point cloud data, calculating the wave impact probability corresponding to each point cloud element;

[0033] According to the wave impact probability, the wave point cloud probability distribution of each point cloud element is calculated, and the calculation formula is:

[0034] ;

[0035] in, Indicate point The probability distribution of wave point cloud, is the attenuation parameter, is the wave center position, is the number of basic blocks affected by the wave;

[0036] Based on the wave point cloud probability distribution, the wave equation model is applied to fit the data, the deviation between the wave point cloud data and the model prediction is improved by adjusting the model parameters, a wave characteristic distribution model describing the wave motion and characteristics is established, and the wave characteristic distribution is obtained.

[0037] Preferably, the steps of obtaining the segmented point cloud data are:

[0038] According to the wave characteristic distribution model, the wave impact intensity of each point cloud element is judged to determine the dynamic threshold of the current wave condition;

[0039] Using the dynamic threshold, calculate whether each point cloud element exceeds the dynamic threshold, the expression is:

[0040] ;

[0041] in, For point A binary indicator of whether it is split. For point The probability distribution of wave point cloud, is the dynamic threshold;

[0042] According to the binary indicator of whether it is segmented, the point cloud data is divided into two parts: affected by the wave and unaffected, and the segmented point cloud data is obtained.

[0043] Preferably, the steps for obtaining the segmentation result of the ship's side and the waterline are:

[0044] Based on the segmented point cloud data, identifying and eliminating error points caused by wave dynamic segmentation to obtain processed point cloud data;

[0045] Cleaning the processed point cloud data, checking whether the point cloud data correctly corresponds to the physical position, and obtaining cleaned point cloud data;

[0046] The boundary between the ship's side and the waterline is divided from the cleaned point cloud data, the geometric features and boundary strength of the point cloud are analyzed to determine the positions of the ship's side and the waterline, and a segmentation result of the ship's side and the waterline is generated.

[0047] Compared with the prior art, the advantages and positive effects of the present invention are:

[0048] The present invention uses laser radar to obtain multi-level point cloud data, which improves the accuracy and efficiency of ship waterline measurement. The calculation of curvature and normal vectors, combined with the extraction of morphological features, enhances the data recognition ability in complex marine environments. By performing multi-level classification and dynamic adjustment of time series on point cloud data, real-time monitoring of ship load changes is achieved, ensuring accurate reflection of real-time data. Dynamically adjusted point cloud data allows real-time correction of wave effects, and reduces interference from environmental factors by establishing a wave characteristic model. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram of the steps of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0051] See also Figure 1 The present invention provides a technical solution, a method for measuring the waterline of a ship based on the dynamic segmentation of the point cloud freeboard of a laser radar:

[0052] Obtain point cloud data through laser radar, select multiple scale levels from the point cloud data, decompose the data of each level respectively, and generate a multi-level point cloud representation; calculate the curvature and normal vector of each level of point cloud data, extract morphological features, and obtain key morphological features;

[0053] Perform multi-level classification on key morphological features, mark the results of each classification on the corresponding point cloud data, and obtain labeled data; sort the labeled data according to the time series, compare and adjust the sorted labeled data frame by frame, and generate dynamically adjusted point cloud data;

[0054] Based on the dynamically adjusted point cloud data, the wave point cloud probability distribution is calculated, and a model is established to generate the wave characteristic distribution according to the model; the threshold is adjusted according to the wave characteristic distribution, and real-time segmentation is performed to obtain the segmented point cloud data;

[0055] The error points of the segmented point cloud data are eliminated, the segmented point cloud data is cleaned using geometric matching, and the boundary between the ship's side and the waterline is divided by boundary strength to generate the segmentation result of the ship's side and the waterline.

[0056] The steps to obtain the multi-level point cloud representation are:

[0057] Obtain point cloud data through laser radar, select and decompose each level in the point cloud data to form a multi-level point cloud structure;

[0058] Based on the multi-level point cloud structure, the point cloud data decomposition coefficient of each level is calculated. The calculation formula is:

[0059] ;

[0060] in, For the The point cloud data decomposition coefficient of the layer, For the Tier The coordinate values ​​of the points, is the total number of points in each layer of point cloud;

[0061] The decomposition coefficients of the combined point cloud data are used to construct and improve the multi-level point cloud representation to obtain the multi-level point cloud representation.

[0062] Specifically, the point cloud data obtained by lidar is selected and decomposed into multiple levels. This process is based on the structural complexity and spatial distribution characteristics of the point cloud data. Selecting the appropriate level can help to more carefully analyze the geometric and spatial characteristics of the object. The selection and decomposition of each level are based on the density and distribution of the point cloud, which is determined by calculating the neighborhood density of each point and its distribution in space. This not only ensures the pertinence and efficiency of data processing, but also can adjust the level of detail of the level according to the specific application scenario. The data decomposition of each level aims to extract the most representative features of the level for subsequent processing such as feature extraction and classification, so as to obtain a preliminary multi-level point cloud structure.

[0063] formula The benefit is that by calculating the synthesis coefficient of each level of segmented data, the characteristics of point cloud data at different levels can be effectively measured and compared, making the processing of point cloud data more refined and adaptable.

[0064] Calculation process: A point cloud layer contains three points, whose coordinates are , , , then the modulus of each point is 1, substituting into the formula, we get:

[0065] ;

[0066] This result shows that the distribution in the three directions is balanced, which is very helpful for understanding the expansion degree and uniformity of point cloud data in each direction.

[0067] The calculated decomposition coefficients of each layer of point cloud data are used to construct and improve the multi-level point cloud representation. These decomposition coefficients are calculated by accumulating the standardized values ​​of each point relative to its own coordinates. This process makes full use of the spatial characteristics and hierarchical structure information of the original point cloud data. By considering the position information and relative importance of each point in the point cloud, the spatial distribution and structural differences of the point cloud data are effectively captured. This is especially important when processing large-scale or complex structured point cloud data. It can ensure the accurate representation of the data and the effectiveness of subsequent processing. The use of this method not only improves the flexibility and adaptability of point cloud processing, but also ensures the accuracy and efficiency of data processing, and obtains a multi-level point cloud representation.

[0068] The steps to obtain key morphological features are:

[0069] Selecting a level of point cloud data from the multi-level point cloud representation, performing local processing on the selected point cloud data, and obtaining a local point cloud set of each point;

[0070] Based on the local point cloud set, the curvature and normal vector of each point are calculated using the following formula:

[0071] ;

[0072] in, For the The curvature value of a point, and are the normal vectors of the neighboring points and the current point respectively, For the The number of neighboring points of a point;

[0073] Based on the curvature value and normal vector, the morphological features of the point cloud data are extracted to obtain the key morphological features.

[0074] Specifically, a level of point cloud data is selected from the multi-level point cloud representation. The data is collected by the lidar system, which involves scanning a large range of environments. The data of each point contains position and reflection intensity information. The purpose of local processing from these data is to refine the analysis. By focusing on the point cloud set in a small area, the slight changes and features of the point cloud data can be observed more accurately. Such processing can help improve the accuracy and efficiency of morphological feature extraction in the subsequent steps. This step obtains a local point cloud set for each point;

[0075] formula The benefit of is that by calculating the difference in normal vectors between each point in the point cloud data and its neighboring points, the change in the local geometric shape of the point cloud surface can be accurately evaluated, which helps to better identify and classify morphological features in complex environments; parameters To pass the statistical point The total number of neighboring points in the local point cloud is obtained; parameter For each neighboring point in the point cloud dataset The normal vector is obtained; parameters To pass through the point cloud data set The normal vector is obtained.

[0076] Calculation process: Set a point j in the point cloud data, and it has neighboring points. The normal vectors of these neighboring points are , and the normal vector of point j is By calculating the sum of the squares of the differences between the normal vector of each neighboring point and the normal vector of point j, and then dividing by the number of neighboring points , and we get the curvature of point j :

[0077] ;

[0078] For example, ;

[0079] ;

[0080] ;

[0081] This result shows that the calculated curvature value , indicating that the local surface around point j has certain geometric changes. Its higher value indicates that the surface changes in this area are more drastic, which is suitable for further morphological feature extraction;

[0082] Based on the calculated curvature value and normal vector information, the morphological feature extraction of the point cloud data is performed. This processing uses point cloud data with higher curvature to identify areas with significant geometric changes. By focusing on these areas, the key features in the point cloud data can be more accurately identified and classified. Through these morphological features, the key morphological features in the point cloud data are extracted and confirmed, which provides key input information for subsequent data analysis and application.

[0083] The steps to obtain labeled data are:

[0084] Based on the key morphological features, multi-level classification is performed to calculate the probability of each data point belonging to each category. The calculation formula is:

[0085] ;

[0086] in, For the Points belong to the category The probability of For the Points for category The score, is the total number of categories, Representative Points for category score;

[0087] Based on the classification probability, a classification label is assigned to each point cloud data point to form labeled data.

[0088] Specifically, a feature set is selected from the key morphological features obtained in the previous step to prepare data for the input of the multi-level classification algorithm, ensuring that the features of each data point are fully recorded. The selection of key morphological features is based on the curvature and normal vector information calculated in the previous step. These features include the local density of points, local shape index, and curvature change. These data are encoded and standardized through a specific data processing process to meet the input requirements of the classification algorithm, ensuring the integrity and consistency of the feature data, and the effective correspondence of the data in the multi-level classification algorithm.

[0089] formula The benefit of is that it can effectively process scores from different categories and convert them into a probability distribution through exponential and normalization processing, which helps the multi-class classification algorithm to deal with possible score deviations or outliers; parameter For this point, The score is calculated by the weighted sum of the feature weights in the machine learning model and the input features; the parameter is the total number of classifications, which is directly determined by the setting of the classification problem. For example, in a problem with three possible classifications, Set to 3; parameter for The exponent of is calculated by the mathematical exponential function; the parameter It is the sum of the exponentials of all classification scores, obtained by first calculating the exponential of each classification score and then summing them.

[0090] Calculation process: Set point The score is (category ), and the total score of this point is (where R is the total number of categories, for example 3), calculated according to the formula:

[0091] ;

[0092] Normalize this result to probability and get the point belongs to the category The probability is 0.14 (or 14%). This result shows that the classification algorithm has 14% confidence in the classification of this point, which means that this point belongs to the category The probability is 14%. If the result is greater than a certain threshold (for example, 50%), it can be considered that the point does belong to the category .

[0093] Based on the classification probability, the most likely classification label is assigned to each point cloud data point. In this way, a complete annotated data set is formed. This process involves traversing all point cloud data points and determining the final label of each point based on the classification probability. This classification operation not only considers the classification probability of a single point, but may also combine the label information of surrounding points for spatial consistency checks to ensure the spatial continuity of the classification results. Through such meticulous operations, the final generated annotated data set can be directly used for subsequent analysis tasks, such as model training or trend prediction.

[0094] The steps for obtaining dynamically adjusted point cloud data are:

[0095] Sort the labeled data in chronological order, check whether the time label of each frame of data is correct, and form a labeled data set arranged in time series;

[0096] Based on the annotated data set arranged in time series, data comparison is performed frame by frame. For each frame of annotated data, the difference with the previous frame is checked and the changed point cloud parts are identified.

[0097] The changed point cloud parts are reclassified and the labels are updated to generate dynamically adjusted point cloud data that matches the time changes.

[0098] Specifically, first, the annotated data generated in the previous step is sorted in chronological order, and the time label of each frame of data is checked and confirmed to be correct. Through this process, the time order of each data point is ensured to be correct. This process involves the parsing of time labels and the application of sorting algorithms to ensure that the data is arranged in the order of actual occurrence. This is the basic requirement for processing time series data and has a direct impact on the overall quality of the data and the accuracy of subsequent processing. The sorted data set provides a standardized time series framework for further inter-frame comparison and data analysis. This sorting process is a key step in generating annotated data sets arranged in time series.

[0099] Based on the obtained annotated dataset arranged in time series, data comparison is performed frame by frame. This process involves detailed comparison of the point cloud data in each frame with the adjacent frames to identify the differences between frames, including position changes, density differences, and label changes of the point clouds. These changes partially reflect the dynamic changes of the object in continuous time frames. This process is a key link in providing a decisive basis for subsequent adjustments.

[0100] Based on the comparison, necessary adjustments are made to the marked changed point cloud parts. This process includes reclassifying or updating the labels of the identified changed areas. According to the comparison results of each frame, the classification and labels of the point cloud data are updated in real time to reflect the dynamic changes between consecutive frames. During the adjustment process, each label update must be based on strict classification logic and update rules to ensure the continuity and consistency of the data. Through this frame-by-frame comparison and adjustment, dynamically adjusted point cloud data that adapts to time changes is eventually formed.

[0101] The steps to obtain the wave characteristics distribution are:

[0102] Based on the dynamically adjusted point cloud data, the wave impact probability corresponding to each point cloud element is calculated;

[0103] According to the wave impact probability, the wave point cloud probability distribution of each point cloud element is calculated. The calculation formula is:

[0104] ;

[0105] in, Indicate point The probability distribution of wave point cloud, is the attenuation parameter, is the wave center position, is the number of basic blocks affected by the wave;

[0106] Based on the probability distribution of wave point cloud, the wave equation model is applied to fit the data. The deviation between the wave point cloud data and the model prediction is improved by adjusting the model parameters. A wave characteristic distribution model describing the wave motion and characteristics is established to obtain the wave characteristic distribution.

[0107] Specifically, to calculate the probability of wave impact based on dynamically adjusted point cloud data, it is first necessary to identify and classify each point cloud element, determine its environmental conditions and its spatial relationship with other point cloud elements. In this process, the coordinate data and physical properties of each point cloud element are recorded and analyzed in detail, and the time tags of the point cloud data also need to be accurately extracted to ensure the accuracy of subsequent probability calculations. By collecting this data, subsequent probability calculations can be performed more accurately. The environmental data and physical properties of each point cloud element will directly affect the calculation results of its wave impact probability. Through this detailed data processing and analysis, it is ensured that the calculation of the wave impact probability is both accurate and has practical application value.

[0108] formula The benefit of this method is that by considering the locations and attenuation parameters of multiple wave centers, the probability distribution of wave point clouds can be described. This multi-center and multi-attenuation parameter setting enables the model to flexibly adapt to different ocean environments and wave conditions, improving the applicability and prediction accuracy of the model. is the attenuation parameter, which is usually determined by experiments or field data analysis to adapt to the characteristics of different wave dynamics; is the wave center position, which is obtained based on oceanographic surveying data or real-time monitoring system to ensure the spatial accuracy of the model; parameter is the number of basic blocks affected by waves, which is preset according to the design requirements of the wave model and the actual application scenario, and can be adjusted through actual observation data to optimize the model performance; parameter Represents a specific point cloud location, obtained directly from the point cloud dataset, and the coordinates of each point reflect its precise position in space.

[0109] Calculation process: Wave center position is (100,200), the attenuation parameter is 0.1, the number of basic blocks is 10, point The position is (102,205), and the formula becomes:

[0110] ;

[0111] Calculate each Distance , substitute into the formula to calculate the value of each item:

[0112] ;

[0113] all The mean value of is 0.4947, so:

[0114] ;

[0115] The result shows that the wave point cloud probability distribution value of point (102,205) is 0.4947, which means that this location has a higher possibility of wave activity under the current model. This probability value can be used to further analyze wave behavior or provide safety warnings.

[0116] The wave equation model is applied for fitting and the wave point cloud data is analyzed. By adjusting the model parameters, the model can be better adapted to the actually observed wave point cloud data. This process includes multiple iterative calculations to find the best parameter settings. Each iteration adjusts the parameters based on the previous fitting results to reduce the error between the prediction and the actual data. In this way, the established model can not only describe the current state of the waves, but also predict the future direction of the waves, thereby providing more reliable data support for navigation or marine structure design.

[0117] The steps to obtain the segmented point cloud data are:

[0118] According to the wave characteristic distribution model, the wave impact intensity of each point cloud element is judged to determine the dynamic threshold of the current wave condition;

[0119] Using the dynamic threshold, calculate whether each point cloud element exceeds the dynamic threshold. The expression is:

[0120] ;

[0121] in, For point A binary indicator of whether it is split. For point The probability distribution of wave point cloud, is the dynamic threshold;

[0122] According to the binary indicator of whether it is segmented, the point cloud data is divided into two parts: affected by the wave and unaffected, and the segmented point cloud data is obtained.

[0123] Specifically, first, based on the established wave characteristic distribution model, the specific wave impact intensity of each point cloud data is analyzed to calculate the dynamic threshold that adapts to different wave conditions, ensuring that the threshold can not only reflect the dynamic changes of the current wave, but also be accurately applicable to the segmentation requirements under various wave intensity conditions. By dynamically adjusting the threshold, it can respond to environmental changes in real time, thereby improving the accuracy and efficiency of point cloud data processing.

[0124] Calculation process: The probability distribution of the wave point cloud at a certain point is 0.75, and the threshold is set is 0.5, according to the expression:

[0125] ;

[0126] The calculation result is 1, which means that the point will be segmented. This process shows that this point in the point cloud is considered to be segmented because the wave impact intensity is greater than the set threshold. The result shows that if the value of T(x) is 1, it means that the point is selected as a segmentation point due to the significant wave impact, so that those areas that are more affected by waves can be processed in a targeted manner.

[0127] The steps to obtain the segmentation results of the ship's side and waterline are as follows:

[0128] Based on the segmented point cloud data, the error points caused by wave dynamic segmentation are identified and eliminated to obtain the processed point cloud data;

[0129] Clean the processed point cloud data, check whether the point cloud data correctly corresponds to the physical position, and obtain the cleaned point cloud data;

[0130] The boundary between the ship's side and the waterline is divided from the cleaned point cloud data, the geometric characteristics and boundary strength of the point cloud are analyzed to determine the position of the ship's side and the waterline, and the segmentation result of the ship's side and the waterline is generated.

[0131] Specifically, based on the segmented point cloud data, a detailed data inspection is first performed to determine the types of error points that may be introduced by wave dynamic segmentation, which include isolated points or point group deviations. Detection and elimination operations are performed on these data points, and the geometric features and relative position information of the point cloud are used for evaluation. The relative distance and angular deviation between each point and its neighboring points are used to determine whether it is an error point. Through this series of operations, misleading data points are accurately removed, thereby improving the overall quality of the data set and the accuracy of subsequent processing, and obtaining processed point cloud data.

[0132] The processed point cloud data is cleaned. First, the position information of each point cloud data element is corrected to ensure that it reflects its true physical position. In this process, the coordinate data of each point is recalibrated, and the actual coordinates of the point cloud data are compared with the coordinates of the expected model. By adjusting the data points to the best fitting position, it is ensured that each point is not only numerically correct but also spatially accurate, which further enhances the availability and accuracy of the point cloud data and obtains the cleaned point cloud data.

[0133] The boundary between the ship's side and the waterline is divided from the cleaned point cloud data, and a detailed geometric feature analysis is performed on each segment of the point cloud to determine the boundary strength in the point cloud. This includes analyzing the density, layout and changes of the point cloud. The exact position of the ship's side and the waterline is determined based on these features. In this step, the natural dividing lines are identified by comparing the structural differences of adjacent point cloud layers. These dividing lines clarify the structural boundaries of the hull, thereby generating the segmentation results of the ship's side and the waterline.

[0134] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for measuring the waterline of a ship based on the dynamic segmentation of point cloud freeboard of a laser radar, characterized in that: The following steps are involved: Acquire point cloud data through a laser radar, select multiple scale levels from the point cloud data, decompose the data of each level respectively, and generate a multi-level point cloud representation; calculate the curvature and normal vector of each level of point cloud data, extract morphological features, and obtain key morphological features; Performing multi-level classification on the key morphological features, marking the result of each classification on the corresponding point cloud data, and obtaining labeled data; Sorting the annotated data according to a time series, comparing and adjusting the sorted annotated data frame by frame, and generating dynamically adjusted point cloud data; Based on the dynamically adjusted point cloud data, the wave point cloud probability distribution is calculated, and a model is established to generate a wave characteristic distribution according to the model; a threshold is adjusted according to the wave characteristic distribution, and real-time segmentation is performed to obtain segmented point cloud data; Eliminating error points from the segmented point cloud data, cleaning the segmented point cloud data using geometric matching, dividing the boundary between the ship's side and the waterline by boundary strength, and generating a segmentation result of the ship's side and the waterline; The steps for obtaining the multi-level point cloud representation are: Obtain point cloud data through laser radar, select and decompose each level in the point cloud data to form a multi-level point cloud structure; Based on the multi-level point cloud structure, the point cloud data decomposition coefficient of each level is calculated, and the calculation formula is: ； in, For the The point cloud data decomposition coefficient of the layer, For the Tier The coordinates of the points, is the total number of points in each layer of point cloud; Combining the point cloud data decomposition coefficients to construct and improve a multi-level point cloud representation to obtain a multi-level point cloud representation; The steps for obtaining the wave characteristic distribution are: Based on the dynamically adjusted point cloud data, calculating the wave impact probability corresponding to each point cloud element; According to the wave impact probability, the wave point cloud probability distribution of each point cloud element is calculated, and the calculation formula is: ； in, Indicate point The probability distribution of wave point cloud, is the attenuation parameter, is the wave center position, is the number of basic blocks affected by the wave; Based on the wave point cloud probability distribution, a wave equation model is applied to fit the data, the deviation between the wave point cloud data and the model prediction is improved by adjusting the model parameters, a wave characteristic distribution model describing the wave motion and characteristics is established, and the wave characteristic distribution is obtained; The steps for obtaining the segmented point cloud data are as follows: According to the wave characteristic distribution model, the wave impact intensity of each point cloud element is judged to determine the dynamic threshold of the current wave condition; Using the dynamic threshold, calculate whether each point cloud element exceeds the dynamic threshold, the expression is: ； in, For point A binary indicator of whether it is split. For point The probability distribution of wave point cloud, is the dynamic threshold; According to the binary indicator of whether it is segmented, the point cloud data is divided into two parts: affected by the wave and unaffected, and the segmented point cloud data is obtained.

2. The method for measuring the waterline of a ship by dynamically segmenting the freeboard of a point cloud based on laser radar according to claim 1 is characterized in that: The steps for obtaining the key morphological features are: Selecting a level of point cloud data from the multi-level point cloud representation, performing local processing on the selected point cloud data, and obtaining a local point cloud set for each point; Based on the local point cloud set, the curvature and normal vector of each point are calculated using the following formula: ； in, For the The curvature value of a point, and are the normal vectors of the neighboring points and the current point respectively, For the The number of neighboring points of a point; Based on the curvature value and the normal vector, the morphological features of the point cloud data are extracted to obtain key morphological features.

3. The method for measuring the waterline of a ship by dynamically segmenting the freeboard of a point cloud based on laser radar according to claim 1 is characterized in that: The steps for obtaining the labeled data are: Based on the key morphological features, multi-level classification is performed to calculate the probability of each data point belonging to each category. The calculation formula is: ； in, For the Points belong to the category The probability of For the Points for category The score, is the total number of categories, Representative Points for category score; Based on the classification probability, a classification label is assigned to each point cloud data point to form labeled data.

4. The method for measuring the waterline of a ship based on point cloud freeboard dynamic segmentation of laser radar according to claim 1 is characterized in that: The steps for obtaining the dynamically adjusted point cloud data are as follows: Arrange and sort the labeled data in chronological order, check whether the time label of each frame of data is correct, and form a labeled data set arranged in time series; Based on the annotated data set arranged in time series, data comparison is performed frame by frame, and for each frame of annotated data, the difference with the previous frame is checked, and the changed point cloud part is identified; The point cloud portion with the change identified is reclassified and the label is updated to generate dynamically adjusted point cloud data matching the time change.

5. The method for measuring the waterline of a ship based on point cloud freeboard dynamic segmentation of laser radar according to claim 1 is characterized in that: The steps for obtaining the segmentation result of the ship's side and the waterline are as follows: Based on the segmented point cloud data, identifying and eliminating error points caused by wave dynamic segmentation to obtain processed point cloud data; Cleaning the processed point cloud data, checking whether the point cloud data correctly corresponds to the physical position, and obtaining cleaned point cloud data; The boundary between the ship's side and the waterline is divided from the cleaned point cloud data, the geometric features and boundary strength of the point cloud are analyzed to determine the positions of the ship's side and the waterline, and a segmentation result of the ship's side and the waterline is generated.

Images