A ship multi-source perception data fusion in-port target clustering method

Abstract

The application provides a ship multi-source perception data fusion in-port target clustering method, and belongs to the technical field of ship berthing perception and ship autonomous positioning. The method first acquires multi-source perception data of a ship-borne camera, a thermal imager and a laser radar and completes time synchronization, then realizes target detection and instance segmentation of a visible light video and target detection of an infrared video through YOLO11n-seg and YOLO11n algorithms respectively, projects laser radar point cloud data to a visible light image coordinate system and extracts effective point cloud indexes after verifying the effectiveness of cross-modal target detection results based on a calculation of an intersection over union, adopts a lightweight clustering algorithm based on local geometric features to construct a target point cloud set and generate a space bounding box, and finally calculates the distance and azimuth angle of the target relative to the ship, and fuses RTK positioning information of the ship to obtain the absolute position of the port target. The application solves the problems of under-segmentation of in-port dense targets, low multi-sensor data fusion efficiency and poor detection stability in poor visibility in the prior art.

Description

Technical Field

[0001] This invention relates to the field of ship berthing sensing and ship autonomous positioning technology, and more particularly to a method for clustering targets in port using multi-source sensing data fusion. Background Technology

[0002] The development of intelligent navigation decision-making systems for ships heavily relies on comprehensive perception of the navigation environment. Single-sensor systems, due to their physical limitations, struggle to provide complete environmental information under various weather conditions and complex scenarios. Multi-source sensor fusion technology, by integrating data from heterogeneous sensors such as lidar, cameras, and millimeter-wave radar, enables all-weather, multi-angle environmental perception, providing reliable data support for autonomous ship decision-making. Especially in restricted waterways such as ports, high-precision 3D environmental reconstruction and dynamic target recognition play a decisive role in ensuring navigation safety.

[0003] In existing technologies, Euclidean distance-based clustering methods have significant limitations when handling densely packed targets in ports. When the distance between ships and dock facilities or adjacent ships is small, the spatial adjacency characteristics of point cloud data lead to severe undersegmentation in traditional clustering algorithms. Although some studies have attempted to improve target segmentation accuracy using deep learning methods, the unique characteristics of maritime scenarios mean that purely data-driven solutions suffer from insufficient generalization. Furthermore, insufficient spatiotemporal synchronization accuracy of multi-sensor data and low efficiency in extracting features from heterogeneous data further constrain the overall performance of ship environmental perception systems. Especially under poor visibility conditions, existing fusion algorithms often struggle to maintain stable target detection capabilities, directly impacting the decision-making reliability of intelligent navigation systems. Summary of the Invention

[0004] To address the technical problems of existing technologies, such as the reliance on single sensing devices, high computational complexity, and undersegmentation of targets, this invention provides a port target clustering method based on the fusion of multi-source sensing data from ships. This invention primarily utilizes a multi-source sensing data fusion mechanism involving shipborne cameras, thermal imagers, and lidar, combined with a cross-modal target verification strategy based on intersection-union ratio (IU) and a lightweight point cloud clustering algorithm based on local geometric features (angular continuity and second-order differential curvature). This improves target detection accuracy in complex port scenes, solves the problem of undersegmentation of point clouds among dense ships, and achieves real-time and efficient multi-target spatial localization and absolute position calculation.

[0005] The technical means employed in this invention are as follows:

[0006] A method for clustering targets within a port based on the fusion of multi-source ship sensing data includes: S1. Acquire visible light video data collected by the shipborne camera, infrared video data collected by the shipborne thermal imager, and port environment point cloud feature data output by the shipborne lidar. S2. The visible light video data is subjected to visual target detection and instance segmentation using a YOLO11n-seg-based visual detection algorithm to obtain the visual target category, confidence level, and pixel-level segmentation mask; the infrared video data is subjected to infrared target detection using a YOLO11n-based infrared detection algorithm to obtain the infrared target category and location information. S3. Based on the spatial positional relationship between the visually detected target and the infrared detected target, the validity of the target detection result is determined by cross-union ratio calculation, and the pixel-level segmentation mask of the valid target is output. S4. Project the port environment point cloud feature data onto the image coordinate system of the visible light video data, and extract the effective point cloud index based on the pixel-level segmentation mask of the effective target; S5. Based on the effective point cloud index, a lightweight port target clustering algorithm based on local geometric features is used to construct a point cloud set for each target and generate a spatial bounding box for the effective port targets. S6. Traverse and find the minimum depth value and corresponding angular characteristic value of all valid targets, and calculate the relative distance and azimuth of the ship. S7. Integrate the ship's position information collected by the shipborne positioning equipment to calculate the absolute position of each port target.

[0007] Further, step S1 includes: S11. Initialize the multi-sensor time synchronization system, read the visible light video stream and its initial timestamp sequence, and read the infrared video stream and its initial timestamp sequence. S12. Read the point cloud depth feature matrix, angle feature matrix and corresponding initial timestamp output by the lidar, and load the extrinsic parameter transformation matrix between the camera and the lidar. S13. Based on the timestamp alignment strategy, output visible light video frames and infrared video frames that are synchronized with the current point cloud data time.

[0008] Further, step S2 includes: S21. Input the visible light video frame into the visual detection model based on a lightweight segmentation network, combine it with the ship visual detection dedicated pt weight file, and output the visual detection results and visual segmentation mask containing the categories of ship, dock, shoreline and obstruction. S22. Input the infrared video frame into the infrared detection model based on the lightweight detection network, combine it with the special pt weight file for ship infrared detection, and output the infrared detection results including the categories of ship, dock, shoreline and obstruction.

[0009] Further, step S3 includes: S31. Extract and cache the coordinates of the two-dimensional bounding boxes of each target in the visual detection results and infrared detection results; S32. Determine whether the bounding boxes of the visually detected target and the infrared detected target have spatial overlap. If there is overlap, calculate the intersection-union ratio; otherwise, determine it as an invalid target and remove it. S33. Set an intersection-union ratio (IU) threshold. When the IU value is greater than the threshold, it is determined to be a valid target, and the corresponding visual segmentation mask is retained. S34. Output the set of valid pixel coordinates of all valid targets in the image coordinate system.

[0010] Further, step S4 includes: S41. Project the point cloud depth feature matrix onto the visible light image plane using the extrinsic transformation matrix to establish the correspondence between the three-dimensional point cloud and the two-dimensional pixels; S42. Traverse the set of valid pixel coordinates and extract the depth feature value corresponding to each pixel coordinate; S43. Establish and output a set of three-dimensional valid point cloud indexes for valid targets.

[0011] Further, step S5 includes: S51. Initialize the target identifier array, breadth-first search queue, angle continuity threshold, and local smoothness threshold; S52. Traverse the set of valid 3D point cloud indexes, assign the current target identifier to the unmarked point cloud indexes and add them to the search queue; S53. Using the current point cloud index as the center point, use a breadth-first search strategy to find unlabeled related point cloud indexes in the neighborhood; S54. Determine geometric continuity based on the angular feature relationship between the neighboring point cloud and the central point cloud, and assign the same target label to the neighboring points that meet the angular continuity condition. S55. For neighboring points that do not meet the angular continuity condition, calculate the local geometric curvature based on the depth features of adjacent point clouds within the same bundle. When the local geometric curvature is less than the smoothness threshold, they are determined to be the same target and assigned the same target identifier. S56. When the search queue is empty, increment the target identifier and continue traversing until all valid point cloud indexes have been processed. S57. Construct the point cloud set corresponding to each target identifier and generate the minimum bounding box.

[0012] Further, in step S54, the angular continuity condition includes: A composite angle determination condition is constructed based on the included angle of the lidar scanning beam and the difference in horizontal azimuth angle between point clouds. When the composite angle is greater than the angle continuity threshold, it is determined that the two point clouds have spatial geometric continuity.

[0013] Furthermore, in step S55, the formula for calculating the local geometric curvature is as follows:

[0014] in, This represents the depth feature value of the current point. This represents the depth value of the preceding point. This represents the depth feature value of the post-position point.

[0015] Further, step S6 includes: S61. Traverse all point cloud depth feature values ​​within the bounding box of each target space, and extract the minimum depth value and its corresponding angle feature value. S62. Based on the minimum depth value and angle characteristic value, and combined with the lidar installation parameters, calculate the distance and azimuth of the target relative to the ship's coordinate system.

[0016] Further, step S7 includes: S71. Obtain the ship's geographic coordinates and heading angle information output by the shipborne RTK positioning equipment; S72. Based on the ship's geographical coordinates, heading angle, and the target's distance and azimuth relative to the ship, calculate the absolute position coordinates of each target in the geodetic coordinate system through coordinate transformation.

[0017] Compared with the prior art, the present invention has the following advantages: 1. The port target clustering method based on the fusion of multi-source sensing data of ships provided by this invention integrates multi-source sensing data of visible light, infrared and lidar, and combines cross-modal cross-interference-exchange-comparison verification strategy to solve the problems of the limitations of single sensor perception and poor detection stability in poor visibility, thereby improving the accuracy and robustness of port target detection.

[0018] 2. This invention employs a lightweight point cloud clustering algorithm based on local geometric features, combined with angular continuity determination and second-order difference curvature calculation, which effectively solves the problem of under-segmentation of point clouds of dense targets within the port. Moreover, the clustering algorithm has low computational complexity and fast segmentation speed, realizing efficient clustering and geometric spatial measurement of targets within the port.

[0019] 3. This invention achieves accurate spatiotemporal synchronization of multi-sensor data and efficient feature fusion of heterogeneous data. The sensor fusion method is simple to implement and has high fusion efficiency, which enhances the anti-interference ability of the perceived information. At the same time, it combines shipborne RTK positioning information to complete the absolute position calculation of the target, thereby improving the positioning accuracy and overall efficiency of the ship perception system in port navigation scenarios.

[0020] In summary, by applying the technical solution of this invention, the problems of incomplete information due to the single sensing device, under-segmentation of dense targets caused by traditional clustering algorithms, low efficiency of multi-sensor fusion, and poor detection stability in existing technologies are addressed. Through multi-source sensing data fusion, cross-modal target verification, lightweight local geometric feature clustering, and absolute position calculation, high-precision and high-efficiency detection and positioning of targets within the harbor are achieved, effectively overcoming the shortcomings of existing technologies. Therefore, the technical solution of this invention solves the problems of low detection accuracy, poor clustering effect, high difficulty in multi-sensor fusion, and unsatisfactory ranging and positioning effect in existing technologies.

[0021] Based on the above reasons, this invention can be widely applied in fields such as ship berthing perception, ship autonomous positioning, and ship intelligent navigation decision-making. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the method of the present invention.

[0024] Figure 2 This is a schematic diagram illustrating data acquisition as provided in an embodiment of the present invention.

[0025] Figure 3 This is a schematic diagram of visual inspection provided in an embodiment of the present invention.

[0026] Figure 4 This is a schematic diagram of decision-level fusion of visible light video and infrared video provided in an embodiment of the present invention.

[0027] Figure 5 This is a schematic diagram illustrating the effective point cloud coordinate acquisition provided in an embodiment of the present invention.

[0028] Figure 6 This is a schematic diagram of the point cloud clustering and segmentation algorithm provided in an embodiment of the present invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product or device.

[0031] like Figure 1 As shown, this invention provides a method for clustering targets within a port using multi-source sensing data fusion from ships, including: S1. Acquire visible light video data collected by the shipborne camera, infrared video data collected by the shipborne thermal imager, and port environment point cloud feature data output by the shipborne lidar. S2. The visible light video data is subjected to visual target detection and instance segmentation using a YOLO11n-seg-based visual detection algorithm to obtain the visual target category, confidence level, and pixel-level segmentation mask; the infrared video data is subjected to infrared target detection using a YOLO11n-based infrared detection algorithm to obtain the infrared target category and location information. S3. Based on the spatial positional relationship between the visually detected target and the infrared detected target, the validity of the target detection result is determined by cross-union ratio calculation, and the pixel-level segmentation mask of the valid target is output. S4. Project the port environment point cloud feature data onto the image coordinate system of the visible light video data, and extract the effective point cloud index based on the pixel-level segmentation mask of the effective target; S5. Based on the effective point cloud index, a lightweight port target clustering algorithm based on local geometric features is used to construct a point cloud set for each target and generate a spatial bounding box for the effective port targets. S6. Traverse and find the minimum depth value and corresponding angular characteristic value of all valid targets, and calculate the relative distance and azimuth of the ship. S7. Integrate the ship's position information collected by the shipborne positioning equipment to calculate the absolute position of each port target.

[0032] In specific implementation, as a preferred embodiment of the present invention, such as Figure 2As shown, step S1 includes: S11. Initialize the multi-sensor time synchronization system and read the visible light video stream. and its initial timestamp sequence Read infrared video stream and its initial timestamp sequence ; S12. Read the point cloud depth feature matrix output by the lidar. Angular feature matrix and corresponding initial timestamp Load the extrinsic transformation matrix between the camera and the lidar. ; S13. Based on a timestamp alignment strategy, output visible light video frames and infrared video frames synchronized with the current point cloud data time. In this embodiment, traverse... to All within the time period and Output the corresponding visible light video stream and infrared video stream .

[0033] In specific implementation, as a preferred embodiment of the present invention, such as Figure 3 As shown in step S21, the visible light video frame is input into a visual detection model based on a lightweight segmentation network. Combined with a dedicated pt weight file for ship visual detection, the visual detection results and visual segmentation mask containing categories of ships, docks, shorelines, and navigational obstructions are output. In this embodiment, the visual detection model adopts the YOLO11n-seg model.

[0034] S22. Input the infrared video frames into an infrared detection model based on a lightweight detection network, and combine it with a dedicated PT weight file for ship infrared detection to output infrared detection results including categories such as ships, docks, shorelines, and navigational obstructions. In this embodiment, the infrared detection model adopts the YOLO11n model.

[0035] In this embodiment, a visible light video stream is output. Port target classification results Visual segmentation mask Infrared video stream Test results .

[0036] In specific implementation, as a preferred embodiment of the present invention, such as Figure 4 As shown, step S3 includes: S31. Extract and cache the visual inspection results. Compared with infrared detection results The two-dimensional bounding box coordinates of each target in the data; S32. Determine the visual detection target. With infrared detection target If the bounding boxes have spatial overlap, the intersection-union ratio is calculated; otherwise, the target is deemed invalid and removed. S33. Set an intersection-over-union (IoU) ratio threshold. When the IoU value is greater than the threshold, it is determined to be a valid target, and the corresponding visual segmentation mask is retained. In this embodiment, the crossover-union ratio threshold is preferably 0.4. S34. Output the set of valid pixel coordinates of all valid targets in the image coordinate system. .

[0037] In specific implementation, as a preferred embodiment of the present invention, such as Figure 5 As shown, step S4 includes: S41. Using the aforementioned external parameter transformation matrix Point cloud depth feature matrix Project the image onto the visible light image plane to establish the correspondence between the three-dimensional point cloud and the two-dimensional pixels; S42. Traverse the set of valid pixel coordinates Extract the depth feature values ​​corresponding to each pixel coordinate. ; S43. Establish and output a set of 3D valid point cloud indexes for valid targets. .

[0038] In specific implementation, as a preferred embodiment of the present invention, such as Figure 6 As shown, step S5 includes: S51. Initialize the target identifier array, breadth-first search queue, angle continuity threshold, and local smoothness threshold; in this embodiment, initialize one array for storing different targets. array ,initialization Initialize angle threshold Smoothness threshold ; S52. Traverse the set of three-dimensional valid point cloud indexes. Indexing unlabeled point clouds Assign a current target identifier and add it to the search queue. ; S53, Indexed by the current point cloud Using the central point as the center, a breadth-first search strategy is used to find unlabeled associated point cloud indexes within the neighborhood; S54. Determine geometric continuity based on the angular feature relationship between the neighboring point cloud and the center point cloud, and assign the same target index to neighboring points that satisfy the angular continuity condition. ; S55. For neighboring points that do not meet the angular continuity condition, calculate the local geometric curvature based on the depth features of adjacent point clouds within the same line bundle. When the local geometric curvature is less than the smoothness threshold, they are determined to be the same target and assigned the same target identifier; that is, it is considered that the current two points are approximately locally collinear, and the current ID is assigned to the neighboring point. And continue to add to the search queue. ; S56, when the search queue If the target identifier is empty, increment the target identifier and continue traversing until all valid point cloud indexes have been processed. S57. Construct the point cloud set corresponding to each target identifier. Generate the minimum bounding box.

[0039] In a specific implementation, as a preferred embodiment of the present invention, the angular continuity condition in step S54 includes: A composite angle determination condition is constructed based on the included angle of the lidar scanning beams and the difference in horizontal azimuth angle between point clouds. When the composite angle is greater than the angle continuity threshold, the two point clouds are determined to have spatial geometric continuity. In this embodiment, based on the distance between the two points... Angle and Determine the geometric continuity of the angle. If If two points have high 3D spatial continuity, then the depth feature curvature based on second-order difference is defined for three adjacent points of the same line bundle. calculate.

[0040] In a specific implementation, as a preferred embodiment of the present invention, the formula for calculating the local geometric curvature in step S55 is as follows:

[0041] in, This represents the depth feature value of the current point. This represents the depth value of the preceding point. This represents the depth feature value of the post-position point.

[0042] In a specific implementation, as a preferred embodiment of the present invention, step S6 includes: S61. Traverse all point cloud depth feature values ​​within the bounding box of each target space, and extract the minimum depth value and its corresponding angle feature value. S62. Based on the minimum depth value and angle characteristic value, and combined with the lidar installation parameters, calculate the distance and azimuth of the target relative to the ship's coordinate system.

[0043] In a specific implementation, as a preferred embodiment of the present invention, step S7 includes: S71. Obtain the ship's geographic coordinates and heading angle information output by the shipborne RTK positioning equipment; S72. Based on the ship's geographical coordinates, heading angle, and the target's distance and azimuth relative to the ship, calculate the absolute position coordinates of each target in the geodetic coordinate system through coordinate transformation.

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for clustering targets within a port using multi-source ship sensing data fusion, characterized in that, include: S1. Acquire visible light video data collected by the shipborne camera, infrared video data collected by the shipborne thermal imager, and port environment point cloud feature data output by the shipborne lidar. S2. The visible light video data is subjected to visual target detection and instance segmentation using a YOLO11n-seg-based visual detection algorithm to obtain the visual target category, confidence level, and pixel-level segmentation mask; the infrared video data is subjected to infrared target detection using a YOLO11n-based infrared detection algorithm to obtain the infrared target category and location information. S3. Based on the spatial positional relationship between the visually detected target and the infrared detected target, the validity of the target detection result is determined by cross-union ratio calculation, and the pixel-level segmentation mask of the valid target is output. S4. Project the port environment point cloud feature data onto the image coordinate system of the visible light video data, and extract the effective point cloud index based on the pixel-level segmentation mask of the effective target; S5. Based on the effective point cloud index, a lightweight port target clustering algorithm based on local geometric features is used to construct a point cloud set for each target and generate a spatial bounding box for the effective port targets. S6. Traverse and find the minimum depth value and corresponding angular characteristic value of all valid targets, and calculate the relative distance and azimuth of the ship. S7. Integrate the ship's position information collected by the shipborne positioning equipment to calculate the absolute position of each port target.

2. The method for clustering targets within a port based on the fusion of multi-source sensor data of ships according to claim 1, characterized in that, Step S1 includes: S11. Initialize the multi-sensor time synchronization system, read the visible light video stream and its initial timestamp sequence, and read the infrared video stream and its initial timestamp sequence. S12. Read the point cloud depth feature matrix, angle feature matrix and corresponding initial timestamp output by the lidar, and load the extrinsic parameter transformation matrix between the camera and the lidar. S13. Based on the timestamp alignment strategy, output visible light video frames and infrared video frames that are synchronized with the current point cloud data time.

3. The method for clustering targets within a port based on multi-source sensor data fusion of ships according to claim 1, characterized in that, Step S2 includes: S21. Input the visible light video frame into the visual detection model based on a lightweight segmentation network, combine it with the ship visual detection dedicated pt weight file, and output the visual detection results and visual segmentation mask containing the categories of ship, dock, shoreline and obstruction. S22. Input the infrared video frame into the infrared detection model based on the lightweight detection network, combine it with the special pt weight file for ship infrared detection, and output the infrared detection results including the categories of ship, dock, shoreline and obstruction.

4. The method for clustering targets within a port based on the fusion of multi-source sensor data of ships according to claim 1, characterized in that, Step S3 includes: S31. Extract and cache the coordinates of the two-dimensional bounding boxes of each target in the visual detection results and infrared detection results; S32. Determine whether the bounding boxes of the visually detected target and the infrared detected target have spatial overlap. If there is overlap, calculate the intersection-union ratio; otherwise, determine it as an invalid target and remove it. S33. Set an intersection-union ratio (IU) threshold. When the IU value is greater than the threshold, it is determined to be a valid target, and the corresponding visual segmentation mask is retained. S34. Output the set of valid pixel coordinates of all valid targets in the image coordinate system.

5. The port target clustering method based on multi-source sensing data fusion of ships according to claim 1, characterized in that, Step S4 includes: S41. Project the point cloud depth feature matrix onto the visible light image plane using the extrinsic transformation matrix to establish the correspondence between the three-dimensional point cloud and the two-dimensional pixels; S42. Traverse the set of valid pixel coordinates and extract the depth feature value corresponding to each pixel coordinate; S43. Establish and output a set of three-dimensional valid point cloud indexes for valid targets.

6. The method for clustering targets within a port based on the fusion of multi-source sensor data of ships according to claim 1, characterized in that, Step S5 includes: S51. Initialize the target identifier array, breadth-first search queue, angle continuity threshold, and local smoothness threshold; S52. Traverse the set of valid point cloud indices, assign the current target identifier to the unmarked point cloud indexes and add them to the search queue; S53. Using the current point cloud index as the center point, use a breadth-first search strategy to find unlabeled related point cloud indexes in the neighborhood; S54. Determine geometric continuity based on the angular feature relationship between the neighboring point cloud and the central point cloud, and assign the same target label to the neighboring points that meet the angular continuity condition. S55. For neighboring points that do not meet the angular continuity condition, calculate the local geometric curvature based on the depth features of adjacent point clouds within the same bundle. When the local geometric curvature is less than the smoothness threshold, they are determined to be the same target and assigned the same target identifier. S56. When the search queue is empty, increment the target identifier and continue traversing until all valid point cloud indexes have been processed. S57. Construct the point cloud set corresponding to each target identifier and generate the minimum bounding box.

7. The method for clustering targets within a port based on multi-source sensor data fusion of ships according to claim 1, characterized in that, In step S54, the angular continuity condition includes: A composite angle determination condition is constructed based on the included angle of the lidar scanning beam and the difference in horizontal azimuth angle between point clouds. When the composite angle is greater than the angle continuity threshold, it is determined that the two point clouds have spatial geometric continuity.

8. The method for clustering targets within a port based on the fusion of multi-source sensor data of ships according to claim 1, characterized in that, In step S55, the formula for calculating the local geometric curvature is as follows: in, This represents the depth feature value of the current point. This represents the depth value of the preceding point. This represents the depth feature value of the post-position point.

9. The method for clustering targets within a port based on the fusion of multi-source sensor data of ships according to claim 1, characterized in that, Step S6 includes: S61. Traverse all point cloud depth feature values ​​within the bounding box of each target space, and extract the minimum depth value and its corresponding angle feature value. S62. Based on the minimum depth value and angle characteristic value, and combined with the lidar installation parameters, calculate the distance and azimuth of the target relative to the ship's coordinate system.

10. The method for clustering targets within a port based on the fusion of multi-source sensor data of ships according to claim 1, characterized in that, Step S7 includes: S71. Obtain the ship's geographic coordinates and heading angle information output by the shipborne RTK positioning equipment; S72. Based on the ship's geographical coordinates, heading angle, and the target's distance and azimuth relative to the ship, calculate the absolute position coordinates of each target in the geodetic coordinate system through coordinate transformation.

Images