Target tracking and recognition system based on marine automatic charging robot

Abstract

The present invention provides a target tracking and identification system based on a ship-based automatic charging robot, which relates to the field of data processing technology. The system comprises: an image acquisition module for acquiring original image data in a port environment; a feature processing module for performing feature enhancement processing on the original image data; a target filtering module for identifying target candidate areas in the image, and performing matching screening based on hull structure shape parameters to eliminate background interference areas and non-target areas; a positioning guidance module for analyzing the position deviation of the target hull in the image, and generating guidance path data in combination with the current position parameters of the automatic charging robot. The present invention improves the accuracy of tracking and identification of the automatic charging robot.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, in particular to a target tracking and identification system based on a marine automatic charging robot. Background Art

[0002] In existing technology, marine automatic charging robots primarily rely on image recognition and wireless communication for target identification and positioning. Specifically, they place visual markers (such as QR codes or color labels) on the charging target, capture images with a camera, and extract feature information through image processing algorithms to achieve target identification and positioning. Furthermore, the robot typically uses wireless signal strength or ultrasonic ranging to correct its relative position to the target, thereby completing path planning and docking charging operations. This type of system has high recognition accuracy in static or regularly structured environments.

[0003] However, in actual port ship automatic charging application scenarios, the hull's exterior is subject to long-term erosion by the marine environment, and visual markers are easily corroded by seawater, covered with stains, or interfered with by sunlight reflections, resulting in the image recognition module being unable to accurately extract target features. In addition, there are background interference sources at the port operation site, such as crane arms, floating water surface reflections, and various types of docked ships, which make it easy for the robot to mistakenly identify non-target objects as targets during the recognition process, resulting in path deviation. Taking fishing boats as an example, their hulls often have no standardized markings and their paint is severely aged. If the image recognition system still relies on rule labels, the recognition success rate will drop significantly, seriously affecting the stability and safety of the charging operation. Summary of the Invention

[0004] The purpose of the present invention is to provide a target tracking and identification system based on a marine automatic charging robot, aiming to solve the problems mentioned in the background technology.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0006] A target tracking and recognition system based on a marine automatic charging robot, the system comprising:

[0007] An image acquisition module is used to obtain raw image data in a port environment, wherein the raw image data includes a hull appearance image and background scene information;

[0008] The feature processing module is used to perform feature enhancement processing on the original image data, including enhancing the contrast area in the image, removing the spot interference, and extracting the edge information of the hull structure to generate structural feature data;

[0009] The target filtering module is used to identify target candidate areas in the image based on structural feature data, and perform matching screening based on the hull structure shape parameters, eliminating background interference areas and non-target areas to generate target recognition data;

[0010] The positioning guidance module is used to analyze the position deviation of the target hull in the image based on the target recognition data, and generate guidance path data in combination with the current position parameters of the automatic charging robot. The guidance path data is used to control the charging robot to berth to the target hull to achieve accurate docking and automatic charging operations.

[0011] Preferably, the feature processing module includes:

[0012] A contrast enhancement unit is used to perform grayscale mapping adjustment on the brightness characteristics of the hull area and the background area in the image based on the original image data, perform local enhancement processing on the low-contrast area, and generate enhanced image data;

[0013] A light spot removal unit is used to identify and separate highlight areas in the image caused by sea surface reflection or metal reflection by analyzing the brightness distribution gradient and regional edge morphology based on the enhanced image data, correct the edges of the highlight areas, and perform image reconstruction on the interfered pixels to generate interference-corrected image data;

[0014] The structural edge extraction unit is used to extract continuous hull edge curves based on the interference correction image data and convert the extracted edge curve information into structural feature data.

[0015] Preferably, the target filtering module includes:

[0016] A candidate region recognition unit is used to identify and segment multiple target candidate regions in the image based on the structural feature data, the continuity of edges and the closedness of regions, extract the boundary coordinate information of each region, and generate candidate region data;

[0017] The structure matching unit is used to extract the hull structure shape parameters of each candidate area based on the candidate area data, compare the parameters with the preset standard hull structure features for similarity, and generate a structure matching result;

[0018] The interference elimination unit is used to eliminate candidate areas whose similarity is lower than a preset similarity threshold according to the structure matching results, and only retain the target area with the highest similarity to generate it as target recognition data.

[0019] Preferably, the positioning guidance module includes:

[0020] The deviation analysis unit is used to perform spatial transformation on the image position coordinates of the target area based on the target recognition data and the current position parameters of the automatic charging robot, analyze the relative position of the target hull in the actual environment, and generate offset data;

[0021] a path generation unit, configured to generate a guidance path from the current position to the target ship according to the offset data, the spatial offset relationship between the target ship and the automatic charging robot, and information on the obstacle layout in the port environment, divide the guidance path into continuous path nodes, and generate path data;

[0022] A path correction unit is used to adaptively correct path nodes based on the path data and the new environmental data perceived in real time during the movement of the charging robot, to ensure that the charging robot can avoid dynamic obstacles and unforeseen interference, and to generate corrected path data;

[0023] The path output unit is used to parse the corrected path data into motion control instructions and generate them to the charging robot control system, so that the charging robot can autonomously navigate to the target hull according to the corrected guidance path and complete the berthing operation.

[0024] Preferably, based on the original image data, grayscale mapping adjustment is performed on the brightness characteristics of the hull area and the background area in the image, and local enhancement processing is performed on the low-contrast area to generate enhanced image data, including:

[0025] According to the original image data, preliminary partitioning is performed to distinguish the hull area from the background area to obtain partition data;

[0026] According to the partition data, the brightness distribution of the pixels in each partition is counted, and the areas with brightness lower than the set brightness threshold are enhanced, and the pixel values ​​are adjusted to improve the local contrast to obtain the preliminary processed image data;

[0027] Based on the preliminary processed image data, a comprehensive analysis is performed on the brightness distribution of each partition of the entire image, and the grayscale mapping parameters between each partition are dynamically adjusted to improve the contrast between the hull area and the background area, thereby obtaining enhanced image data.

[0028] Preferably, based on the enhanced image data, by analyzing the brightness distribution gradient and the regional edge morphology, identifying and separating the highlight areas in the image caused by sea surface reflection or metal reflection, correcting the edges of the highlight areas, and performing image reconstruction on the interfered pixels to generate interference-corrected image data, including:

[0029] Based on the enhanced image data, the brightness gradient of each pixel in the image is calculated, and the highlight area is identified based on the set highlight threshold;

[0030] For the identified highlight area, edge detection and edge correction are performed based on its edge morphology and the spatial distribution characteristics of the surrounding pixels to reconstruct the boundary of the highlight area;

[0031] According to the reconstructed highlight area boundary, the disturbed pixels inside the highlight area are reconstructed in combination with the brightness and texture features of the surrounding normal area to generate interference correction image data.

[0032] Preferably, according to the structural feature data, based on the continuity of edges and the closedness of regions, multiple target candidate regions are identified and segmented in the image, the boundary coordinate information of each region is extracted, and candidate region data is generated, including:

[0033] Based on the structural feature data, the continuity of the edge information in the image is judged, the potential target area is preliminarily divided, and the partition data is generated;

[0034] Based on the partition data, for each divided area, the area with complete boundaries is screened out by calculating the closedness score, and its boundary coordinate information is calibrated to generate closed area data;

[0035] All closed area data are summarized to generate candidate area data.

[0036] Preferably, based on the candidate area data, the hull structure shape parameters are extracted for each candidate area, and the similarity is compared with the preset standard hull structure features to generate a structure matching result, including:

[0037] According to the candidate region data, the shape parameters of each candidate region are extracted to generate shape feature data;

[0038] Based on the shape feature data, the similarity between it and the preset standard hull structure features is calculated to obtain the similarity score of each candidate area;

[0039] According to the similarity scores of each candidate region, the target region with the highest similarity score is screened out to generate the structure matching result.

[0040] Preferably, based on the offset data, based on the spatial offset relationship between the target ship and the automatic charging robot, combined with the obstacle layout information in the port environment, a guidance path from the current position to the target ship is generated, the guidance path is divided into continuous path nodes, and path data is generated, including:

[0041] According to the offset data, the spatial relative coordinates of the current position of the automatic charging robot and the target hull are determined to generate spatial offset relationship data;

[0042] Collect obstacle layout information in the port environment, establish an environmental model, and generate obstacle distribution data;

[0043] Generate an obstacle-avoiding and smooth guidance path based on the spatial offset relationship data and obstacle distribution data, and decompose the entire guidance path into a series of continuous path nodes;

[0044] The generated path nodes are optimized to ensure the feasibility of the path and the smoothness of movement, and the path data is generated.

[0045] Preferably, according to the path data, and during the movement of the charging robot, the path nodes are adaptively corrected based on the new environmental data perceived in real time to ensure that the charging robot can avoid dynamic obstacles and unforeseen interference, and the corrected path data is generated, including:

[0046] During the movement of the charging robot, it continuously collects real-time environmental perception data, monitors dynamic obstacles and newly appearing interference objects at the port site, and generates new environmental data;

[0047] Based on the new environment data and path data, the spatial relationship between the generated path nodes and dynamic obstacles is analyzed to determine whether there is a collision or interference risk at the path nodes;

[0048] When there is a risk of collision or interference between path nodes, the affected path nodes are dynamically adjusted or reconstructed to generate corrected path data.

[0049] The above solution of the present invention includes at least the following beneficial effects:

[0050] In order to solve the problems that are difficult to overcome with existing technologies in actual port ship charging environments, such as the easy damage of visual markers, strong reflective interference, diverse target structures and complex backgrounds, the present invention proposes a system solution for target tracking and identification that does not rely on external labels or QR codes and is directly based on the image features of the hull structure. The image acquisition module collects the original image data in the port environment in real time, which is not affected by factors such as missing markers, stains and sunlight reflections, and can stably obtain complete information including the hull appearance and environmental background. The feature processing module uses multi-level image processing technologies such as contrast enhancement, spot removal and structural edge extraction to effectively highlight the hull structure features, significantly reduce the probability of interference with target identification caused by ambient light, seawater reflection and background debris, and improve the adaptability and robustness of the system in various extreme weather conditions and high dynamic backgrounds.

[0051] The target filtering module, based on structural feature data, accurately selects the target area that best matches the hull structure parameters from multiple candidate areas, avoiding misidentification and missed identification caused by labels or simple features alone. In particular, in actual scenarios where multiple ships are side by side and there is a lot of interference from dock facilities, it can still accurately lock the charging target. The positioning guidance module dynamically generates guidance path data by analyzing target recognition data and the current position parameters of the automatic charging robot, enabling the robot to adaptively avoid obstacles and accurately berth the target hull. This solution eliminates the problem of inaccurate target recognition and path planning in existing technologies due to label failure or environmental changes, effectively improving the stability, safety and practical application coverage of the automatic charging robot's operations, and significantly improving the automatic charging operation experience in unmanned ports and complex dock environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 1 is an architecture diagram of a target tracking and identification system based on a marine automatic charging robot provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0053] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0054] like Figure 1 As shown, an embodiment of the present invention proposes a target tracking and recognition system based on a marine automatic charging robot, the system comprising:

[0055] An image acquisition module is used to obtain raw image data in a port environment, wherein the raw image data includes a hull appearance image and background scene information;

[0056] The feature processing module is used to perform feature enhancement processing on the original image data, including enhancing the contrast area in the image, removing the spot interference, and extracting the edge information of the hull structure to generate structural feature data;

[0057] The target filtering module is used to identify target candidate areas in the image based on structural feature data, and perform matching screening based on the hull structure shape parameters, eliminating background interference areas and non-target areas to generate target recognition data;

[0058] The positioning guidance module is used to analyze the position deviation of the target hull in the image based on the target recognition data, and generate guidance path data in combination with the current position parameters of the automatic charging robot. The guidance path data is used to control the charging robot to berth to the target hull to achieve accurate docking and automatic charging operations.

[0059] In an embodiment of the present invention, the overall structure of the system includes an image acquisition module, a feature processing module, a target filtering module and a positioning guidance module. The modules cooperate with each other through the sequential data flow to achieve the accurate identification and autonomous berthing and charging of the target ship by the marine automatic charging robot in a port environment. First, the image acquisition module is deployed on the automatic charging robot body, and the original image data in the port environment is acquired in real time through high-definition imaging equipment. The original image data includes not only the appearance image of the target ship, but also the background scene information of the dock, water surface, other ships and surrounding structures. In order to adapt to the interference factors such as complex port lighting, reflections, stains, etc., the module supports stable operation under different lighting and climatic conditions, ensuring that the data source is extensive and has the robustness for practical application.

[0060] After acquiring the raw image data, the feature processing module systematically enhances its features. This module uses methods such as contrast adjustment, light reflection removal, and structural edge extraction to enhance the distinction between the target hull and the surrounding background. For example, when the hull surface is obscured by seawater reflections, direct sunlight, or dirt, the system intelligently processes areas of uneven brightness, reflective spots, or blurred outlines in the original image, making the hull structure more prominent in subsequent steps. After processing by this module, the output structural feature data includes enhanced key image features, significantly improving the recognition accuracy of the target hull by subsequent modules.

[0061] The target filtering module then analyzes the structural feature data to identify all candidate target regions in the image. Based on the continuity of the hull's edges, the closedness of the region, and structural shape parameters, the system screens all areas in the image that may represent the hull. For areas with distinct features that don't conform to the target hull's standard structure, such as dock edges and floating objects, the system effectively eliminates them based on the available structural features. This module ultimately generates target recognition data for the unique or optimal target region identified, laying the foundation for the robot's precise positioning and path planning.

[0062] The positioning guidance module further analyzes the relative spatial position between the target hull and the robot body in the current image coordinate system based on the target recognition data, and combines the real-time posture parameters collected by the automatic charging robot itself to generate guidance path data. This path data not only takes into account the actual berthing position of the target hull, but also integrates dynamic factors such as obstacles and moving ships in the port environment to ensure that the robot can approach the target along the optimal and safe path. Finally, the guidance path data is sent to the automatic charging robot motion control system in real time to realize the entire process of autonomous navigation, berthing and automatic charging. In the actual port operation environment, this embodiment can cope with various challenges such as complex hull appearance, cluttered background, and drastic environmental changes, ensuring that the charging robot accurately and efficiently completes the automatic docking operation.

[0063] In a preferred embodiment of the present invention, the feature processing module includes:

[0064] A contrast enhancement unit is used to perform grayscale mapping adjustment on the brightness characteristics of the hull area and the background area in the image based on the original image data, perform local enhancement processing on the low-contrast area, and generate enhanced image data;

[0065] A light spot removal unit is used to identify and separate highlight areas in the image caused by sea surface reflection or metal reflection by analyzing the brightness distribution gradient and regional edge morphology based on the enhanced image data, correct the edges of the highlight areas, and perform image reconstruction on the interfered pixels to generate interference-corrected image data;

[0066] The structural edge extraction unit is used to extract continuous hull edge curves based on the interference correction image data and convert the extracted edge curve information into structural feature data.

[0067] In an embodiment of the present invention, the feature processing module is subdivided into a contrast enhancement unit, a spot removal unit, and a structure edge extraction unit. The data flow between each unit is clear and mutually supportive. The system first processes the original image data through the contrast enhancement unit. The original image data is collected from the port environment and contains rich information about the hull area and the background area. The contrast enhancement unit automatically partitions the input original image to distinguish the brightness distribution of the hull and the background. For example, the water surface of some ports is highly reflective, and the contrast between the hull and the background is low under natural light. The system can perform pixel value stretching and local enhancement processing on the low-contrast area through adaptive grayscale mapping adjustment. This process significantly improves the distinguishability of the target hull area in the overall image, laying the foundation for subsequent highlight area recognition and structural contour extraction.

[0068] The enhanced image data is output to the spot removal unit, which specifically analyzes common highlight interference areas such as sea surface reflections and metal reflections. The system first accurately detects highlight areas in the image by calculating the brightness gradient of the pixels. It then performs edge correction based on the edge shape and the distribution of surrounding pixels. For example, in some cases, bolts or water stains on the hull surface can form irregular highlight spots. The system re-segments these boundaries and, combining the brightness and texture characteristics of adjacent pixels, reconstructs the affected pixel blocks, filling in the actual structural information and minimizing the impact of highlight areas on the judgment of the hull structure. The resulting interference-corrected image data is free of significant reflection interference and can be used for high-precision edge detection.

[0069] The interference-corrected image data is then input into the structural edge extraction unit, whose main task is to extract continuous and complete edge curves of the hull structure from images that eliminate highlight interference and enhance contrast. The system uses a multi-scale edge detection algorithm to adapt to changes in the curvature and length of the hull edge under different ship types and different viewing angles. For example, for long fishing boats or wide-body cargo ships, the algorithm can automatically adjust the detection parameters, capture continuous hull contour information, and uniformly convert these curve information into structural feature data. Ultimately, the structural feature data not only retains the most representative edge features of the target hull, but also effectively filters out background contours and noise data that are irrelevant to identification, making downstream target area screening and comparison highly reliable and stable.

[0070] In scenarios where the port ship docking environment is complex and the natural conditions are changeable, the feature processing module of this embodiment can continuously provide a high-quality image feature basis for the target recognition and tracking system, thereby ensuring that the automatic charging robot can achieve high robustness and high-precision target recognition capabilities.

[0071] The process of extracting continuous hull edge curves based on interference-corrected image data and converting the extracted edge curve information into structural feature data specifically includes:

[0072] In actual port image processing, the system first acquires interference-corrected image data after speckle removal. Most of the interference in this image due to sea surface reflections or metal reflections has been corrected, preserving the actual outline and structural details of the hull. Subsequently, the system uses a multi-scale edge detection method to scan the image line by line, initially detecting possible edge points by comparing the grayscale value changes of adjacent pixels. To enhance continuity recognition capabilities, the system further employs a pixel connectivity tracking algorithm, connecting all edge points belonging to the same hull outline in spatial order, ultimately forming one or more closed or nearly closed continuous edge curves.

[0073] In the process of generating structural feature data, the system will perform parameterized processing on each edge curve, such as measuring the length of the curve, calculating the main direction of the curve, analyzing the curvature change of the edge, and statistically analyzing the distribution of the turning angles of each edge segment. For example, the length can be determined by successively accumulating the distances between each adjacent edge point; for the curvature, it can be obtained by calculating the degree of change in the angle between the front and rear vectors in the neighborhood of each point. These parameterized feature information will be uniformly organized into structural feature data as the basic data source for subsequent target area identification and matching. This process does not require human intervention and is compatible with a variety of ship types and viewing angles. Even when the edge of the hull is partially blocked or partially blurred, sufficiently reliable edge structure information can still be extracted to ensure the robustness and adaptability of subsequent target recognition.

[0074] In a preferred embodiment of the present invention, the target filtering module includes:

[0075] A candidate region recognition unit is used to identify and segment multiple target candidate regions in the image based on the structural feature data, the continuity of edges and the closedness of regions, extract the boundary coordinate information of each region, and generate candidate region data;

[0076] The structure matching unit is used to extract the hull structure shape parameters of each candidate area based on the candidate area data, compare the parameters with the preset standard hull structure features for similarity, and generate a structure matching result;

[0077] The interference elimination unit is used to eliminate candidate areas whose similarity is lower than a preset similarity threshold according to the structure matching results, and only retain the target area with the highest similarity to generate it as target recognition data.

[0078] In an embodiment of the present invention, the target filtering module is specifically implemented by the candidate region identification unit, the structure matching unit and the interference elimination unit in sequence. The system first performs a comprehensive analysis of the input structural feature data through the candidate region identification unit. The structural feature data has effectively highlighted the contour edge and structural characteristics of the hull. The system uses edge continuity analysis technology to determine whether the edge lines of each area in the image form a closed and complete figure, and preliminarily partitions the image through pixel connectivity judgment, contour tracking and other methods. For each divided area, the system further calculates the closure score, and extracts the boundary coordinates of the area with a complete closed boundary to realize the identification and aggregation of multiple potential target candidate areas. The output of the candidate area data provides a structured information basis for subsequent fine feature comparison and interference elimination.

[0079] Next, the structural matching unit extracts the structural shape parameters of each candidate area data. This unit can extract multi-dimensional structural parameters including length ratio, curvature, and corner density for each candidate area, and compare these actually detected shape features with the standard hull structure features preset by the system in detail. The system adopts a multi-parameter fusion similarity calculation method, taking into account the global structure of the contour and the local detail features. For example, when there are only local differences between the shape of the candidate area and the preset standard, the system can determine its overall similarity based on the weight setting, and achieve high-precision recognition of hulls with complex structures and irregular shapes. All comparison results form a structural matching result, further clarifying the target area that is closest to the standard features.

[0080] The interference elimination unit adopts a threshold judgment mechanism based on the structural matching results to eliminate all candidate areas whose similarity is lower than the set threshold. For non-target areas introduced by background debris, dock facilities, temporary obstacles, etc., this unit can effectively remove interference, and ultimately only retain the target area with the highest similarity as the target recognition data output. In this way, even if there are a large number of non-target objects similar to the target hull in the actual port scene, the system can accurately eliminate irrelevant areas to ensure that subsequent positioning and path planning are based on unique and reliable target data. This embodiment effectively improves the anti-interference ability, adaptability and overall recognition accuracy of the target recognition system, and is particularly suitable for port operating environments with complex ship types and drastic background dynamic changes.

[0081] Among them, according to the structural matching results, candidate regions with similarity lower than the preset similarity threshold are eliminated, and only the target region with the highest similarity is retained and generated as target recognition data, which specifically includes:

[0082] During the target filtering phase, the system compares each candidate region against pre-set standard hull structural features across multiple parameters. This similarity is determined by calculating the differences between the candidate region's length ratio, curvature distribution, and corner characteristics, respectively, and the corresponding parameters of the standard features. These differences are then combined into a final similarity score using pre-set weights. For example, assuming the system assigns weights to length ratio, curvature, and corner density, the final score is the weighted sum of these individual scores.

[0083] Subsequently, the system will summarize the similarity scores of all candidate areas and compare each score with the preset similarity threshold. The setting of the threshold can be obtained by experimentally testing multiple sets of real-world scene samples, so that it can take into account both recognition accuracy and anti-interference ability. All candidate areas below the threshold will be directly eliminated, and only the target areas with the highest scores and exceeding the threshold will be retained. To ensure the uniqueness of identification, if there are multiple areas with very close scores, the system can further combine auxiliary information such as spatial position, boundary closure or historical trajectory for secondary screening, and finally output the optimal area data as target identification data. In actual applications, this screening and comparison process is fully automated, which is not only suitable for application scenarios with diverse hull shapes, but also facilitates the flexible adjustment of similarity threshold parameters for different port operations, achieving high configurability and high applicability of the system.

[0084] In a preferred embodiment of the present invention, the positioning guidance module includes:

[0085] The deviation analysis unit is used to perform spatial transformation on the image position coordinates of the target area based on the target recognition data and the current position parameters of the automatic charging robot, analyze the relative position of the target hull in the actual environment, and generate offset data;

[0086] a path generation unit, configured to generate a guidance path from the current position to the target ship according to the offset data, the spatial offset relationship between the target ship and the automatic charging robot, and information on the obstacle layout in the port environment, divide the guidance path into continuous path nodes, and generate path data;

[0087] A path correction unit is used to adaptively correct path nodes based on the path data and the new environmental data perceived in real time during the movement of the charging robot, to ensure that the charging robot can avoid dynamic obstacles and unforeseen interference, and to generate corrected path data;

[0088] The path output unit is used to parse the corrected path data into motion control instructions and generate them to the charging robot control system, so that the charging robot can autonomously navigate to the target hull according to the corrected guidance path and complete the berthing operation.

[0089] In this embodiment of the present invention, the positioning guidance module is composed of a deviation analysis unit, a path generation unit, a path correction unit, and a path output unit, which collaborate to complete the entire autonomous navigation and docking process. First, the deviation analysis unit receives target identification data from the target filtering module. Combined with the current position parameters acquired in real time by the automatic charging robot, it performs a spatial transformation on the target area within the image coordinate system, calculating the relative spatial position and offset between the target vessel and the robot. Based on the robot's own positioning on the port map, the system can quickly determine multi-dimensional deviations from the target vessel, including lateral, longitudinal, and distance deviations, and generates offset data, providing a quantitative basis for the dynamic generation of guidance paths.

[0090] The path generation unit then receives the offset data and, in combination with known information about the layout of obstacles in the port environment, such as docked vessels, dock facilities, and buoys, uses a path planning algorithm to generate an optimal guidance path from the robot's current position to the target vessel. This path not only considers the principle of shortest distance in space but also prioritizes obstacle avoidance and the dynamic nature of the environment. The system divides the complete guidance path into sequential path nodes, enabling the robot to gradually adjust its direction and speed during berthing, ensuring smooth and safe movement and outputting structured path data.

[0091] The path correction unit is responsible for adaptively correcting path nodes during the robot's actual movement, based on real-time environmental perception data collected by the charging robot. When new dynamic obstacles appear in the port environment, such as approaching robots, human activity, or temporarily moving vessels, the system detects them in real time and dynamically adjusts the established path to avoid collisions, ensuring real-time optimality of the path and the safe completion of the berthing mission. This corrected path data is continuously updated, providing the robot with the latest navigation instructions for subsequent actions.

[0092] Finally, the path output unit interprets the corrected path data into specific motion instructions recognizable by the robot motion control system, driving the automatic charging robot to autonomously complete the berthing operation of the target ship according to the corrected guidance path. In actual port operations, the collaborative work of these modules can significantly improve the navigation flexibility and docking accuracy of the charging robot, especially in dynamic, complex, obstacle-laden, and space-constrained ship docking environments. Through this series of processes, the embodiment can ensure that the charging robot can achieve efficient, intelligent, and autonomous charging operations in the changing and complex port environment.

[0093] Among them, according to the target recognition data and combined with the current position parameters of the automatic charging robot, the image position coordinates of the target area are spatially transformed, the relative position of the target hull in the actual environment is analyzed, and the offset data is generated, which specifically includes:

[0094] Target recognition data contains the position parameters of precisely selected target areas in the image coordinate system, such as the target area's center and boundary coordinates. Based on the charging robot's actual spatial location, the system uses the robot's onboard sensors (such as GPS, inertial measurement unit, and wheel encoders) to obtain its current position parameters, including the robot's 2D or 3D spatial coordinates and orientation angle within the port's operating area.

[0095] The system then uses a spatial mapping model to convert the target area's position from image coordinates to actual physical space coordinates, based on the calibration parameters of the current imaging device (such as focal length, imaging resolution, mounting height, and orientation). For example, if the camera uses a common pinhole model, the pixel coordinates of the target area can be mapped to the actual location on the port site by establishing a conversion relationship from image coordinates to ground coordinates. The spatial mapping process can pre-calibrate the port environment to determine the necessary parameters, and then adjust them in real time based on the current status of the robot and camera.

[0096] Finally, the system calculates the spatial offset between the actual spatial coordinates of the target hull and the current position parameters of the charging robot. This offset typically includes the lateral distance, longitudinal distance, and height difference (if any), and can be further decomposed into the direction and distance information required for the robot's movement. For example, if the target hull is 5 meters in front of the robot and 1 meter to the left, the offset data will clearly indicate that the robot needs to move 1 meter to the left and 5 meters forward to complete docking preparations. The offset data serves as an important input for subsequent path planning and dynamic navigation, ensuring that the robot can accurately adjust its motion path according to the actual on-site environment and achieve efficient and safe automatic berthing.

[0097] In a preferred embodiment of the present invention, based on the original image data and the brightness characteristics of the hull area and the background area in the image, grayscale mapping adjustment is performed, and local enhancement processing is performed on the low-contrast area to generate enhanced image data, including:

[0098] According to the original image data, preliminary partitioning is performed to distinguish the hull area from the background area to obtain partition data;

[0099] According to the partition data, the brightness distribution of the pixels in each partition is counted, and the areas with brightness lower than the set brightness threshold are enhanced, and the pixel values ​​are adjusted to improve the local contrast to obtain the preliminary processed image data;

[0100] Based on the preliminary processed image data, a comprehensive analysis is performed on the brightness distribution of each partition of the entire image, and the grayscale mapping parameters between each partition are dynamically adjusted to improve the contrast between the hull area and the background area, thereby obtaining enhanced image data.

[0101] In this embodiment of the present invention, the system uses a contrast enhancement unit to perform in-depth processing of the brightness characteristics of the ship hull and background areas in the raw image data. This unit first automatically partitions the raw image data and uses an intelligent algorithm to identify the boundary between the hull and the background. To address issues such as uneven lighting, sea surface reflections, and shadow interference that may arise in complex port environments, the partitioning process dynamically adjusts the threshold for the boundary, ensuring adaptability and accuracy. After partitioning, the system calculates the brightness distribution of pixels within each partition. For areas with brightness below the set threshold, an adaptive grayscale mapping algorithm is used to increase their brightness and contrast. This processing not only effectively enhances the visual prominence of the ship hull within the overall image but also compensates for the reduced contrast caused by inclement weather such as haze and rain, ensuring that details in the hull area are fully preserved. Based on the initially processed image, a comprehensive analysis of the brightness distribution of each partition is performed, and the grayscale mapping parameters between the partitions are dynamically adjusted to further enhance the contrast between the hull and the background, making the structural features of the target hull more prominent in the enhanced image data. This enhanced image data lays a solid foundation for subsequent highlight area identification and hull structural feature extraction. In actual applications, even in low-contrast scenes such as early morning, evening, and cloudy days, the system can still stably identify the hull outline, effectively improving the overall recognition robustness and adaptability.

[0102] In a preferred embodiment of the present invention, based on the enhanced image data, by analyzing the brightness distribution gradient and the regional edge morphology, the highlight area in the image caused by sea surface reflection or metal reflection is identified and separated, the edge of the highlight area is corrected, and image reconstruction is performed on the interfered pixels to generate interference-corrected image data, including:

[0103] Based on the enhanced image data, the brightness gradient of each pixel in the image is calculated, and the highlight area is identified based on the set highlight threshold;

[0104] For the identified highlight area, edge detection and edge correction are performed based on its edge morphology and the spatial distribution characteristics of the surrounding pixels to reconstruct the boundary of the highlight area;

[0105] According to the reconstructed highlight area boundary, the disturbed pixels inside the highlight area are reconstructed in combination with the brightness and texture features of the surrounding normal area to generate interference correction image data.

[0106] In this embodiment of the present invention, when processing enhanced image data, the speckle removal unit first calculates the brightness gradient of each pixel in the image to quickly identify highlight areas caused by sea surface reflections or metal reflections. Based on a set highlight threshold, the system clusters and labels all potential highlight pixels within the image, ensuring that highlight areas caused by different types of reflections can be separated. The speckle removal unit then uses edge detection and edge correction algorithms to analyze the spatial distribution characteristics of the edges of these areas, correcting irregular or damaged highlight boundaries into continuous, smooth curves that seamlessly connect with adjacent normal areas. For interfered pixels within the highlight areas, the system reconstructs them using methods such as pixel interpolation or texture compensation based on the brightness and texture characteristics of the surrounding normal areas, effectively restoring the original hull structural information. The resulting interference-corrected image data not only removes interference caused by strong reflections but also preserves the true features of the original hull to the greatest extent possible. This process is particularly effective in complex water environments, metal hulls, or situations with multiple light sources, significantly improving the accuracy of subsequent hull structure edge extraction and providing a reliable image foundation for downstream candidate area identification and hull target positioning.

[0107] Among them, according to the reconstructed highlight area boundary, the disturbed pixels inside the highlight area are reconstructed in combination with the brightness and texture features of the surrounding normal area to generate interference correction image data, specifically including:

[0108] After identifying and reconstructing the boundaries of the highlighted area, the system marks all pixels within the highlighted area as disturbed pixels. For these pixels, the system first selects several adjacent normal area pixels outside the highlighted area boundary. These pixels serve as reconstruction references, reflecting the brightness and texture distribution characteristics of that location in the image when it is not disturbed by the highlight. The reference area can be selected using methods such as an annular neighborhood, an extended window, or adaptive blocking to ensure that the reference data contains sufficient information and is close to the local image structure of the highlighted area.

[0109] During the specific reconstruction process, the system can use a variety of existing technologies to perform pixel compensation. One of the commonly used methods is interpolation, which sets the brightness value of the disturbed pixels in the highlight area to the average, weighted average, or median of the reference pixels. For example, spatial weighted interpolation can be used to give higher weights to the disturbed pixels near the boundary than to the normal pixels outside the boundary, so that the reconstructed brightness and texture transition more naturally toward the boundary. For image areas with texture, image restoration techniques (such as the "patch method" or "sample replication method" based on texture synthesis) can be combined to use texture blocks in the surrounding area to fill the pixels in the highlight area row by row or block by block, so that the brightness, contrast, and texture distribution remain continuous and consistent with the surrounding area.

[0110] For example, if the highlight area is located on the surface of the hull deck, and the normal pixels outside the boundary mostly show wood grain or metal texture, the system will prioritize collecting these wood grain or metal textures as references during reconstruction, and use spatial texture synthesis or block matching algorithms to expand similar textures into the highlight area, so that the reconstructed area has neither obvious brightness mutations nor texture breaks.

[0111] After the reconstruction process is completed, all interfered pixels are assigned brightness and texture characteristics consistent with the normal area. The final interference-corrected image data not only eliminates the information loss caused by highlights, but also restores the true structure of the hull to the greatest extent, providing a clear and stable visual basis for subsequent edge extraction and target recognition.

[0112] In a preferred embodiment of the present invention, based on the structural feature data, the image is identified and segmented into multiple candidate target regions based on edge continuity and region closure, the boundary coordinate information of each region is extracted, and candidate region data is generated, including:

[0113] Based on the structural feature data, the continuity of the edge information in the image is judged, the potential target area is preliminarily divided, and the partition data is generated;

[0114] Based on the partition data, for each divided area, the area with complete boundaries is screened out by calculating the closedness score, and its boundary coordinate information is calibrated to generate closed area data;

[0115] All closed area data are summarized to generate candidate area data.

[0116] In this embodiment of the present invention, a candidate region identification unit evaluates the continuity and closure of input structural feature data, gradually screening multiple candidate target regions. The system first uses edge information from the structural feature data to perform a connectivity analysis on all edge curves in the image, determining which edges can be closed to form complete regions. For each potential target region, the candidate region identification unit further calculates a closure score, filtering out regions with complete boundaries based on pixel distribution, connected region area, and contour closure. To address the potential for multiple target overlap or partial occlusion in port scenarios, the system uses partitioned data analysis to accurately distinguish overlapping or adjacent targets, reducing recognition errors. All rigorously screened and labeled closed region data is aggregated to generate candidate region data. This candidate region data not only contains the spatial boundary coordinates of each region but also provides a clear data foundation for subsequent structural shape parameter extraction and target screening. In actual port operations, this embodiment can effectively adapt to the diversity of vessels and complex environmental influences, achieving high-precision initial identification of target hulls, laying a solid recognition foundation for subsequent intelligent navigation and precise docking by the automatic charging robot.

[0117] Among them, the continuity of the edge information in the image is judged, the potential target area is preliminarily divided, and the partition data is generated, which specifically includes:

[0118] The system first receives structural feature data, typically the edge curve information of the ship hull extracted after enhancement and interference correction by the previous module. In this step, the system uses edge connectivity analysis techniques to traverse each edge curve in the image. The core of connectivity determination is to examine the adjacency of each edge pixel and determine whether it and its adjacent pixels belong to the same continuous edge. For example, using the eight-neighborhood pixel connectivity principle, if any of the eight surrounding pixels of a pixel belong to an edge, these points are considered connected.

[0119] Within the entire image, the system automatically groups all continuous pixel points into several groups through traversal and recursive search, each group corresponding to a continuous edge curve. Furthermore, the system divides the area where these continuous edge curves are close to or surrounded by each other into a potential target area. For edges with breakpoints or bifurcations, the system can fill or connect them through interpolation or smoothing algorithms to ensure the integrity and coherence of the area division. After completing the above operations, the system identifies a potential target partition for each continuous edge and records these partitions and their pixel index data as partition data to prepare for subsequent area closure analysis.

[0120] For example, if the input image is a ship edge map of a port scene, the system will automatically identify multiple connected edge lines with clear contours, and based on this, divide the image into multiple preliminary potential target areas. These areas may be the hull, or they may include non-target objects such as docks and pontoons.

[0121] Among them, according to the partition data, for each divided area, the area with complete boundaries is screened out by calculating the closedness score, and its boundary coordinate information is calibrated to generate closed area data, which specifically includes:

[0122] After initially obtaining all potential target areas (partition data), the system further performs a closure analysis on each partition. The calculation of the closure score is generally based on information such as the starting and ending points of the region edge curve, the boundary length, and the region area. Common closure judgment methods include:

[0123] First, the system determines whether the distance between the start and end points of each edge curve is less than a certain threshold. If the distance is close enough, the region is considered closed. Second, the ratio of the number of pixels along the region's boundary to the area of ​​pixels enclosed by the region is calculated. If the ratio is close to the theoretical value, the boundary is complete. For edge curves that are not completely closed but have small gaps, the system can automatically fill in the boundary using pixel interpolation or shortest path completion algorithms to ensure that the boundary meets the complete closure criteria.

[0124] After screening, the system classifies all areas determined to have complete boundaries as closed areas and extracts their boundary coordinate information. Specifically, boundary coordinate information includes the pixel sequence of each closed area's outer contour, the coordinates of its extreme points (such as the top left and bottom right corners), or the outline shape expressed as a polygonal point set.

[0125] Ultimately, all areas that pass the closure analysis and have their boundary coordinates calibrated are aggregated to form closed area data. This data not only significantly improves the accuracy of subsequent structural parameter extraction and target screening, but also automatically adapts to the complex port environment, where multiple targets are distributed, overlapped, and partially obscured. For example, in the presence of multiple ship hulls and dock edges, the system can accurately identify target areas with complete boundaries and suspected to be ship hulls, providing reliable input for the charging robot's subsequent automatic identification and path planning.

[0126] The calculation method of the closure score includes:

[0127] According to the structural feature data, the boundary pixels of each region are extracted to form a boundary pixel sequence and generate region boundary data;

[0128] For the region boundary data, the spatial distance between the starting point and the end point of the boundary is determined respectively, and the total length of the region boundary is counted to generate the boundary length data;

[0129] Based on the boundary length data, the ratio of the head-to-tail distance of the region boundary to the total length of the boundary is calculated to obtain the boundary closure parameter, where the head-to-tail distance is the Euclidean distance between the starting point and the end point of the boundary, and the total length of the boundary is the sum of the distances between all adjacent boundary pixels;

[0130] According to the region boundary data, the total number of pixels inside the region is calculated to obtain the region area parameter;

[0131] Compare the ratio of the area parameter to the square of the total length of the boundary, and use this ratio as an auxiliary parameter of the regional enclosure. The larger the area and the closer the ratio is to the theoretical circular area, the higher the enclosure.

[0132] The boundary closure parameter and the regional closure auxiliary parameter are weighted and summed according to the preset weights to obtain the closure score. The closure score ranges from 0 to 1. The higher the score, the more complete the regional boundary and the better the closure.

[0133] According to the closure score, the areas with scores higher than the preset closure threshold are screened out as closed areas with complete boundaries, and the closed area data are output.

[0134] In a preferred embodiment of the present invention, based on the candidate region data, the hull structure shape parameters are extracted for each candidate region, and the similarity is compared with the preset standard hull structure features to generate a structure matching result, including:

[0135] According to the candidate region data, the shape parameters of each candidate region are extracted to generate shape feature data;

[0136] Based on the shape feature data, the similarity between it and the preset standard hull structure features is calculated to obtain the similarity score of each candidate area;

[0137] According to the similarity scores of each candidate region, the target region with the highest similarity score is screened out to generate the structure matching result.

[0138] In this embodiment of the present invention, the structure matching unit systematically extracts structural shape parameters and performs similarity comparisons on candidate region data. Each candidate region is sequentially extracted for multi-dimensional shape features, including length ratio, curvature, and edge angle distribution. These features reflect the spatial form and local geometric characteristics of the actual hull structure. The system pre-defines standard hull structural features for each ship type, and the shape features of all candidate regions are comprehensively compared against these standard features. This similarity comparison utilizes a multi-feature fusion algorithm, weighting the similarities of different parameters to ensure that the evaluation results consider both the overall outline and detailed variations. During the scoring process, if a candidate region has a high degree of match with the standard features, it is considered the target hull. Finally, all candidate regions are sorted by similarity score, and the system automatically selects the region with the highest score as the target region and outputs it as the structure matching result. Through this refined comparison and automated screening, the system can accurately distinguish target hulls from non-target objects, even in complex scenarios such as multiple ships docked together or partially obscured, improving overall recognition accuracy and enhancing the positioning and docking capabilities of the automatic charging robot.

[0139] According to the candidate region data, shape parameters of each candidate region are extracted to generate shape feature data, which specifically includes:

[0140] After obtaining several candidate regions, the system treats each candidate region as an independent object for feature extraction. For each region, the system first traverses its boundary pixels or outer contour point set, and extracts key parameters reflecting the regional structure and morphology based on the spatial distribution, connectivity, and geometric relationships of the pixels. Common shape parameters include:

[0141] (1) Length ratio, i.e., the ratio of the length of the candidate region’s outer contour in the major axis direction to the length in the minor axis direction, used to distinguish between slender hulls and wide-body hulls;

[0142] (2) Curvature distribution refers to the degree of curvature calculated at various locations on the outer contour. It can be numerically quantified by measuring the angle formed between three consecutive contour points, reflecting the local turning or arc-shaped structural characteristics of the hull;

[0143] (3) Corner feature density: by counting the distribution and number of significant corners or acute angles on the regional contour, it is used to distinguish special structures such as the bow and stern of the hull, and the side.

[0144] In addition, supplementary parameters such as area, circumscribed rectangle, minimum enclosing circle, and circumference can also be extracted to enrich the expression of shape features.

[0145] For example, for a long fishing vessel, the major axis is much larger than the minor axis, so the system will extract a larger length ratio parameter. The curvature is more concentrated at the bow and stern, and the density of corner features is higher at both ends of the bow. All these parameters are integrated into a set of structured shape feature data, which serves as the basis for subsequent similarity analysis of candidate regions. This extraction process is automated and requires no human intervention, making it suitable for characterizing various types and sizes of hulls within the port.

[0146] Among them, based on the shape feature data, the similarity is calculated with the preset standard hull structure features to obtain the similarity score of each candidate area, including:

[0147] After extracting shape feature data for each candidate region, the system calls upon a built-in library of standard ship structure parameters. This library contains a set of standard shape feature parameters for common port vessel types (such as fishing vessels, cargo ships, and tugboats). These standard parameters, derived through prior sample analysis, expert definition, or historical data analysis, cover multiple dimensions such as principal axis length ratios, typical curvature distribution patterns, and corner characteristics.

[0148] During the similarity calculation process, the system compares the characteristic values ​​of the candidate region with those of the standard for each shape parameter. For example, if the length ratio of the candidate region is close to that of the standard ship model, it will receive a high score in this category. For curvature distribution and corner features, the system measures the overall degree of consistency between the two through traversal matching and difference statistics. The comparison results of all parameters are combined according to pre-set weights to form a final comprehensive similarity score.

[0149] For example, assuming a standard fishing boat has a length ratio of 4:1, the curvature distribution is mainly concentrated at the ends, and the corner features are distributed at one point each at the head and tail. When the shape parameters extracted from a candidate region are compared with the standard parameters, if the differences are small and the distribution of curvature and corner features is roughly consistent, the overall similarity score for that region will be higher; otherwise, the score will be lower.

[0150] If scores for multiple parameters differ, the system automatically adjusts the scores based on the weights of each parameter to prioritize the identification of the vessel's primary features. Ultimately, all candidate regions receive a comprehensive similarity score for subsequent selection of optimal targets. This process ensures accurate vessel identification while improving the system's adaptability to diverse vessel types and complex port environments. This step is fully automated, facilitating widespread application and eliminating the need for subjective human judgment.

[0151] More specifically, the similarity can be calculated using the following steps:

[0152] For each candidate region, five shape parameters are extracted, including length ratio, aspect ratio, average curvature, corner feature density, and region compactness, based on boundary coordinates and structural feature data.

[0153] Among them, the length ratio is the ratio of the length of the region's major axis to the length of the minor axis, the aspect ratio is the ratio of the width to the height of the region's circumscribed rectangle, the average curvature is the arithmetic mean of the curvatures of all points on the region's boundary curve, the corner feature density is the number of significant corner points per unit length, and the region compactness is the ratio of the region area to the area of ​​the circumscribed rectangle.

[0154] All parameters are normalized so that characteristic data of different physical dimensions can be uniformly compared;

[0155] Each normalized parameter is compared with the corresponding parameter of the preset standard hull structure feature, and the absolute difference between the two is calculated. For each parameter, the parameter difference is subtracted from 1 to obtain the single similarity score;

[0156] In order to reflect the importance of different parameters to the judgment of hull structure, weights are set for the five parameters according to actual application experience or prior knowledge;

[0157] All individual similarity scores are weighted and summed up according to their weights, and the weighted result is used as the similarity score of the candidate region;

[0158] Among them, the similarity score value range is between 0 and 1. The higher the score, the higher the overall similarity between the candidate area and the standard hull structure.

[0159] For example, the length ratio, aspect ratio, average curvature, corner feature density and regional compactness of a candidate region differ from the standard features by 0.1, 0.05, 0.15, 0.08 and 0.12 respectively. Assuming that the weights of each parameter are 0.3, 0.2, 0.2, 0.15 and 0.15 respectively, the final similarity score of the candidate region is: (1-0.1)×0.3+(1-0.05)×0.2+(1-0.15)×0.2+(1-0.08)×0.15+(1-0.12)×0.15.

[0160] The system sorts the similarity scores of all candidate areas or compares them with a preset threshold to obtain the target area that is closest to the standard hull structure for subsequent identification and operation.

[0161] In a preferred embodiment of the present invention, a guidance path from the current position to the target ship is generated based on the offset data, the spatial offset relationship between the target ship and the automatic charging robot, and the obstacle layout information in the port environment. The guidance path is divided into continuous path nodes to generate path data, including:

[0162] According to the offset data, the spatial relative coordinates of the current position of the automatic charging robot and the target hull are determined to generate spatial offset relationship data;

[0163] Collect obstacle layout information in the port environment, establish an environmental model, and generate obstacle distribution data;

[0164] Generate an obstacle-avoiding and smooth guidance path based on the spatial offset relationship data and obstacle distribution data, and decompose the entire guidance path into a series of continuous path nodes;

[0165] The generated path nodes are optimized to ensure the feasibility of the path and the smoothness of movement, and the path data is generated.

[0166] In an embodiment of the present invention, a path generation unit generates an optimal guidance path from the current position to the target vessel based on the spatial offset relationship between the target vessel and the automated charging robot, combined with the obstacle layout in the port environment. This process first accurately locates the relative spatial relationship between the robot and the target vessel using offset data. The system collects obstacle distribution data in real time within the port area, building a refined environmental model that incorporates all known static and dynamic obstacles into path planning. The path generation unit integrates the spatial offset relationship and obstacle distribution, and uses a dynamic path planning algorithm (e.g., one based on graph theory or heuristic search) to generate a smooth path that avoids obstacles while minimizing travel distance. The path is then refined into a series of continuous path nodes. Each path node corresponds to an intermediate navigation point, which the robot can follow sequentially during actual movement, achieving smooth transitions. To ensure path feasibility and smoothness, the system also optimizes path nodes, such as removing unnecessary turning points, smoothing curves, and adjusting turning radius. The resulting path data provides the charging robot with a clear motion reference, enabling it to autonomously, accurately, and safely navigate to the target vessel in a complex port environment.

[0167] Among them, according to the offset data, the spatial relative coordinates of the current position of the automatic charging robot and the target hull are determined, and the spatial offset relationship data is generated, which specifically includes:

[0168] The system first receives the physical coordinates of the target ship in the port environment, obtained through target recognition and spatial mapping, as well as the current actual coordinate information of the automatic charging robot body. The robot's actual coordinates can be obtained through a combination of multiple sensors such as differential GPS, lidar positioning, and inertial navigation units, and are updated in real time. The system then compares the target ship's position coordinates with the robot's current position parameters and calculates their lateral, longitudinal, and (if applicable) height distances in the same coordinate system. Specifically, the difference between the two points in the X, Y (and Z) axes is taken to determine the robot's spatial offset relative to the target ship.

[0169] To adapt to the coordinate system of the actual port, the system will pre-set a unified reference coordinate origin. The spatial offset relationship data is vector information containing lateral, longitudinal and height differences, which can be further interpreted as the direction and distance required for the robot to move. For example, if the robot is 3 meters west of the target ship and 2 meters south, the spatial offset relationship data will indicate that the robot needs to move a corresponding distance in the northeast direction to approach the target. This spatial offset provides an accurate basis for setting the start and end points of the subsequent path planning algorithm, and is compatible with the actual needs of ships of different types and sizes and different docking locations.

[0170] The obstacle layout information in the port environment is collected, the environment model is established, and the obstacle distribution data is generated, which specifically includes:

[0171] During the robot's autonomous navigation process, the system must continuously sense and record obstacles in the port's operating area. These obstacles may include other docked vessels, buoys, dock structures, operating vehicles, accumulated cargo, and even moving personnel. The system integrates multiple sensors, such as lidar, ultrasonic radar, and depth cameras, to perform a comprehensive scan of the operating area. Using the distance data transmitted by the sensors, the system automatically detects the spatial location and outline of obstacles.

[0172] The system aggregates the location information of all detected obstacles and stores it in a structured environment model. This environment model can be represented using a raster map, vector map, or point cloud data structure, and includes parameters such as the obstacle's center coordinates, outer contours, and occupied area. For dynamic obstacles, the system also periodically refreshes their spatial distribution to ensure the real-time and accuracy of the environment model during navigation. For example, if a large cargo ship arrives at a certain dock area, the system can immediately incorporate the ship's position on the map into the obstacle distribution data, effectively avoiding collisions during path planning.

[0173] Specifically, the method for establishing the environment model includes:

[0174] Based on the collected obstacle layout information in the port environment, the original obstacle data including the obstacle spatial coordinates, size characteristics and type attributes are obtained to describe the distribution status of all obstacles in the operation area of ​​the automatic charging robot;

[0175] Perform spatial consistency processing on the original obstacle data, use a unified coordinate system to transform and normalize the spatial position, boundary shape and size parameters of all obstacles, and generate standardized obstacle space data;

[0176] Based on standardized obstacle spatial data, obstacles are classified into different types, and static and dynamic obstacles are marked separately. In combination with the real-time motion status of the obstacles, the spatial position of dynamic obstacles is periodically updated to generate dynamic obstacle distribution information.

[0177] The spatial coordinates, boundary information and type parameters of all static and dynamic obstacles are aggregated according to the preset data structure to generate an environmental model that describes the global distribution of obstacles in the current port environment;

[0178] The environmental model is output as obstacle distribution data for path planning and path node optimization processing, thereby realizing dynamic obstacle avoidance and mobile path planning for the automatic charging robot's guided path.

[0179] Among them, based on the spatial offset relationship data and obstacle distribution data, an obstacle-avoiding and smooth guidance path is generated, and the entire guidance path is decomposed into a series of continuous path nodes, specifically including:

[0180] After obtaining the spatial offset relationship between the robot and the target vessel and complete obstacle distribution data, the system calls the path planning module to generate the optimal movement path. This module is typically based on A*, D*, RRT, or a modified version of the algorithm. It searches for a feasible path from the starting point to the end point within the environmental model, taking into account objectives such as path length, obstacle avoidance requirements, and smoothness of movement.

[0181] During the planning process, the system begins with the robot's current position as the starting point and the target vessel's position as the endpoint, treating obstacles as impassable areas. During the path search, all obstacles are automatically avoided, ensuring that each node on the path is within a safe zone. To enhance navigation stability, path generation considers parameters such as the number of turns and curvature, minimizing sharp turns, sharp angles, and frequent changes of direction, thereby improving the smoothness of the guidance path.

[0182] After planning, the system breaks down the complete path into a series of nodes, each containing spatial coordinates and heading information. These nodes are then sequentially transmitted to the robot as navigation instructions, guiding it toward the target vessel. For example, a path might start from the robot's current location, pass through an open area of ​​the harbor, avoid two anchored vessels, and a buoy, ultimately arriving at a berth next to the target vessel.

[0183] The generated path nodes are optimized to ensure path feasibility and movement smoothness, and path data is generated, specifically including:

[0184] After generating preliminary path nodes, the system enters the path optimization phase to further improve the feasibility of the navigation path and the smoothness of the robot's motion. This optimization process includes multiple processes, including node thinning, path smoothing, corner reduction, and local adjustments. First, the system analyzes the distances and corner angles between nodes. If any nodes are too dense or have excessive turning angles, they are automatically merged or smoothed to reduce unnecessary path jitter and frequent sharp turns.

[0185] Furthermore, the system can use curve-fitting algorithms (such as Bezier or spline curves) to smooth the overall path, making turns more natural and preventing sudden deceleration or deviation from the intended route. For areas of the environment that may change, the system can also dynamically fine-tune local nodes of the path to ensure that the path remains unobstructed regardless of static or dynamic obstacles.

[0186] The resulting output path data includes all optimized path nodes and their corresponding navigational attributes, enabling the robot to achieve smooth, safe, and efficient autonomous movement. For example, the optimized path features moderate distances between nodes and smooth turns, minimizing pauses and risks, even in obstructed harbor operations, enabling precise docking of the autonomous charging robot with the ship.

[0187] In a preferred embodiment of the present invention, based on the path data and during the movement of the charging robot, the path nodes are adaptively corrected based on the new environmental data perceived in real time to ensure that the charging robot can avoid dynamic obstacles and unforeseen interference, and the corrected path data is generated, including:

[0188] During the movement of the charging robot, it continuously collects real-time environmental perception data, monitors dynamic obstacles and newly appearing interference objects at the port site, and generates new environmental data;

[0189] Based on the new environment data and path data, the spatial relationship between the generated path nodes and dynamic obstacles is analyzed to determine whether there is a collision or interference risk at the path nodes;

[0190] When there is a risk of collision or interference between path nodes, the affected path nodes are dynamically adjusted or reconstructed to generate corrected path data.

[0191] In this embodiment of the present invention, a path correction unit adaptively corrects path nodes in real time to address environmental changes that may occur during the charging robot's actual movement. During the robot's movement, the system continuously collects real-time environmental perception data from the site, including information about port workers, other mobile equipment, unexpected obstacles, and dynamically docked vessels. This new environmental data is dynamically combined with the generated path data, enabling the system to analyze in real time the spatial relationship between each path node and dynamic obstacles. When a path node is identified as posing a risk of collision or interference with an obstacle, the system immediately initiates an adaptive path correction algorithm, dynamically adjusting the affected node locally or reconstructing the entire path. For example, if a new operating vessel crosses ahead, the system can predict the risk area in advance, adjust the path nodes to avoid the obstacle, and regenerate a safe and reliable new path. This corrected path data is output to the robot control system in real time, ensuring that the robot always follows the optimal and safe trajectory, avoiding all known and unforeseen interference. In the complex and ever-changing port environment, this embodiment significantly improves the automatic charging robot's emergency obstacle avoidance capabilities and environmental adaptability, effectively ensuring the continuity, safety, and efficiency of charging operations.

[0192] Among them, based on the new environment data and path data, the spatial relationship between the generated path nodes and dynamic obstacles is analyzed to determine whether there is a collision or interference risk among the path nodes, including:

[0193] During the robot's actual movement, the system continuously utilizes various environmental sensing devices (such as LiDAR, depth cameras, and ultrasonic radar) to collect new environmental data about the port's operating area in real time. This data includes the spatial coordinates, speed, and direction of movement of all dynamic obstacles within the area (such as moving vessels, vehicles, and personnel). The system compares and analyzes this real-time obstacle information with previously generated navigation path data, focusing on the spatial distance relationship between path nodes and dynamic obstacles.

[0194] Common methods for determining whether a path node presents a collision or interference risk include: for each path node, setting a safety buffer zone centered on the node and comparing this buffer zone with the spatial boundaries of all dynamic obstacles. If any obstacle is found to overlap with the buffer zone, or if the obstacle's predicted trajectory will enter the node's area within a short period of time, the system determines that the path node presents a potential collision or interference risk. This judgment not only considers the obstacle's current position but also makes predictions based on its movement trends. For example, by analyzing the obstacle's speed and direction, it can predict its position within a few seconds. This allows for early risk perception and proactive path protection.

[0195] For example, if a work vehicle is slowly crossing sideways 20 meters in front of a node on the robot's guidance path, the system will determine whether the two may intersect in the next few seconds based on the speed of the work vehicle and the distance between the path nodes. If there is a risk of intersection, the node will be marked as requiring adjustment.

[0196] When there is a risk of collision or interference between path nodes, the affected path nodes are dynamically adjusted or reconstructed to generate corrected path data, specifically including:

[0197] Once the system determines that a path node has a risk of collision or interference with a dynamic obstacle, it immediately enters the dynamic adjustment and path reconstruction phase. Dynamic adjustment uses the following steps:

[0198] First, the affected path node and its surrounding nodes within a certain range are identified as adjustment targets. The system then re-searches the feasible area around these nodes in the environmental model, attempting to make fine adjustments along the edge of the obstacle or against the direction of the obstacle's motion to move the node toward a safe area. The adjustment process ensures that the adjusted path maintains overall coherence and accessibility, without sudden turns or unusual curves.

[0199] If a dynamic obstacle significantly impacts the path or cannot be resolved with a single node adjustment, the system can trigger partial path reconstruction. This involves starting from the robot's current position, avoiding known obstacles, and regenerating a new path to the target in the remaining space. This reconstruction process can utilize the same search and optimization algorithms used for initial path planning, such as A*, D*, and RRT. After reconstruction, the system connects the newly generated path nodes with the unaffected portion of the original path to form a complete, corrected path.

[0200] The corrected path data is transmitted to the charging robot's control system in real time, allowing the robot to immediately continue navigating according to the updated path, effectively avoiding collisions or interference with dynamic obstacles. The entire path correction and reconstruction process is automatic and continuous, requiring no human intervention, and is adaptable to the port's high-frequency environmental changes and complex moving obstacles. For example, if a large number of people cross directly in front of the charging robot, the system can instantly adjust the path nodes and guide the robot to a safe area, ensuring safe, smooth, and efficient charging operations.

[0201] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. The target tracking and recognition system based on the marine automatic charging robot is characterized by: The system comprises: An image acquisition module is used to obtain raw image data in a port environment, wherein the raw image data includes a hull appearance image and background scene information; The feature processing module is used to perform feature enhancement processing on the original image data, including enhancing the contrast area in the image, removing the spot interference, and extracting the edge information of the hull structure to generate structural feature data; The target filtering module is used to identify target candidate areas in the image based on structural feature data, and perform matching screening based on the hull structure shape parameters, eliminating background interference areas and non-target areas to generate target recognition data; The positioning guidance module is used to analyze the position deviation of the target hull in the image based on the target recognition data, and generate guidance path data based on the current position parameters of the automatic charging robot. The guidance path data is used to control the charging robot to berth to the target hull to achieve accurate docking and automatic charging operations; The positioning guidance module includes: The deviation analysis unit is used to perform spatial transformation on the image position coordinates of the target area based on the target recognition data and the current position parameters of the automatic charging robot, analyze the relative position of the target hull in the actual environment, and generate offset data. Specifically, it includes: The target recognition data includes position parameters of the target candidate area in the image coordinate system; A spatial mapping model is used to convert the position parameters in the target recognition data into actual physical space coordinates; Calculate the spatial offset between the actual physical space coordinates and the current position parameters of the automatic charging robot; The path generation unit is used to generate a guidance path from the current position to the target ship based on the offset data, the spatial offset relationship between the target ship and the automatic charging robot, and the obstacle layout information in the port environment. The guidance path is divided into continuous path nodes and path data is generated, specifically including: According to the offset data, the spatial relative coordinates of the current position of the automatic charging robot and the target hull are determined to generate spatial offset relationship data; Collect obstacle layout information in the port environment, establish an environmental model, and generate obstacle distribution data; Generate an obstacle-avoiding and smooth guidance path based on the spatial offset relationship data and obstacle distribution data, and decompose the entire guidance path into a series of continuous path nodes; Optimize the generated path nodes to ensure path feasibility and movement smoothness, and generate path data; A path correction unit is used to adaptively correct path nodes based on the path data and the new environmental data perceived in real time during the movement of the charging robot, to ensure that the charging robot can avoid dynamic obstacles and unforeseen interference, and to generate corrected path data; The path output unit is used to parse the corrected path data into motion control instructions and generate them to the charging robot control system, so that the charging robot can autonomously navigate to the target hull according to the corrected guidance path and complete the berthing operation.

2. The target tracking and identification system based on the marine automatic charging robot according to claim 1 is characterized in that: The feature processing module includes: A contrast enhancement unit is used to perform grayscale mapping adjustment on the brightness characteristics of the hull area and the background area in the image based on the original image data, perform local enhancement processing on the low-contrast area, and generate enhanced image data; A light spot removal unit is used to identify and separate highlight areas in the image caused by sea surface reflection or metal reflection by analyzing the brightness distribution gradient and regional edge morphology based on the enhanced image data, correct the edges of the highlight areas, and perform image reconstruction on the interfered pixels to generate interference-corrected image data; The structural edge extraction unit is used to extract continuous hull edge curves based on the interference correction image data and convert the extracted edge curve information into structural feature data.

3. The target tracking and identification system based on the marine automatic charging robot according to claim 1 is characterized in that: The target filtering module includes: A candidate region recognition unit is used to identify and segment multiple target candidate regions in the image based on the structural feature data, the continuity of edges and the closedness of regions, extract the boundary coordinate information of each region, and generate candidate region data; The structure matching unit is used to extract the hull structure shape parameters of each candidate area based on the candidate area data, compare the parameters with the preset standard hull structure features for similarity, and generate a structure matching result; The interference elimination unit is used to eliminate candidate areas whose similarity is lower than a preset similarity threshold according to the structure matching results, and only retain the target area with the highest similarity to generate it as target recognition data.

4. The target tracking and identification system based on the marine automatic charging robot according to claim 2 is characterized in that: Based on the original image data, grayscale mapping adjustment is performed on the brightness characteristics of the hull area and the background area in the image, and local enhancement processing is performed on the low-contrast area to generate enhanced image data, including: According to the original image data, preliminary partitioning is performed to distinguish the hull area from the background area to obtain partition data; According to the partition data, the brightness distribution of the pixels in each partition is counted, and the areas with brightness lower than the set brightness threshold are enhanced, and the pixel values ​​are adjusted to improve the local contrast to obtain the preliminary processed image data; Based on the preliminary processed image data, a comprehensive analysis is performed on the brightness distribution of each partition of the entire image, and the grayscale mapping parameters between each partition are dynamically adjusted to improve the contrast between the hull area and the background area, thereby obtaining enhanced image data.

5. The target tracking and identification system based on the marine automatic charging robot according to claim 4 is characterized in that: Based on the enhanced image data, by analyzing the brightness distribution gradient and regional edge morphology, the highlight areas in the image caused by sea surface reflection or metal reflection are identified and separated, the edges of the highlight areas are corrected, and image reconstruction is performed on the interfered pixels to generate interference-corrected image data, including: Based on the enhanced image data, the brightness gradient of each pixel in the image is calculated, and the highlight area is identified based on the set highlight threshold; For the identified highlight area, edge detection and edge correction are performed based on its edge morphology and the spatial distribution characteristics of the surrounding pixels to reconstruct the boundary of the highlight area; According to the reconstructed highlight area boundary, the disturbed pixels inside the highlight area are reconstructed in combination with the brightness and texture features of the surrounding normal area to generate interference correction image data.

6. The target tracking and identification system based on the marine automatic charging robot according to claim 3 is characterized in that: Based on the structural feature data, edge continuity and region closure, multiple target candidate regions are identified and segmented in the image, the boundary coordinate information of each region is extracted, and candidate region data is generated, including: Based on the structural feature data, the continuity of the edge information in the image is judged, the potential target area is preliminarily divided, and the partition data is generated; Based on the partition data, for each divided area, the area with complete boundaries is screened out by calculating the closedness score, and its boundary coordinate information is calibrated to generate closed area data; All closed area data are summarized to generate candidate area data.

7. The target tracking and identification system based on the marine automatic charging robot according to claim 6 is characterized in that: Based on the candidate area data, the hull structure shape parameters are extracted for each candidate area, and the similarity is compared with the preset standard hull structure features to generate the structure matching results, including: According to the candidate region data, the shape parameters of each candidate region are extracted to generate shape feature data; Based on the shape feature data, the similarity between it and the preset standard hull structure features is calculated to obtain the similarity score of each candidate area; According to the similarity scores of each candidate region, the target region with the highest similarity score is screened out to generate the structure matching result.

8. The target tracking and identification system based on the marine automatic charging robot according to claim 1 is characterized in that: Based on the path data and the new environmental data perceived in real time during the movement of the charging robot, the path nodes are adaptively corrected to ensure that the charging robot can avoid dynamic obstacles and unforeseen interference, and the corrected path data is generated, including: During the movement of the charging robot, it continuously collects real-time environmental perception data, monitors dynamic obstacles and newly appearing interference objects at the port site, and generates new environmental data; Based on the new environment data and path data, the spatial relationship between the generated path nodes and dynamic obstacles is analyzed to determine whether there is a collision or interference risk at the path nodes; When there is a risk of collision or interference between path nodes, the affected path nodes are dynamically adjusted or reconstructed to generate corrected path data.

Images