Vision-based automatic box grabbing and releasing identification control system for open top box

Abstract

The invention discloses a vision-based automatic open top box catching and releasing recognition control system, which relates to the technical field of computer vision and comprises the steps of image acquisition, image preprocessing, open top box feature extraction, empty box and shielding judgment, box catching point generation, catching score sorting, control instruction generation, catching and releasing execution, exception handling and the like. By extracting the contour and edge features of the opening, the recognition of the open top box structure and the judgment of the empty box state are realized; preferably sorting the box grabbing points by combining a multi-dimensional scoring mechanism; failure retry and grabbing strategy self-adaptive updating are achieved through grabbing feedback. The method is suitable for automatic grabbing and releasing operation of the non-standard container of the port, and has the advantages of being high in recognition accuracy, high in grabbing stability, complex in adaptive scene and the like.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and more specifically to a vision-based automated open-top box grabbing and placing identification control system. Background Technology

[0002] With the rapid development of container terminals and intelligent logistics systems, robots are being widely used for cargo handling and placement. In ports, open-top containers are often used to store goods to meet special cargo requirements. However, the open tops of open-top containers lack standardized cover structures, resulting in irregular height differences when stacked. Furthermore, factors such as obstructions, reflections, and watermarks inside the containers can significantly impact the accuracy of vision-based automatic identification.

[0003] Existing automated container handling systems mostly rely on standard container structures, and key identification points such as the four corner lock holes and container dimensions have been accurately identified by trained models. However, open-top containers, due to their lack of fixed structural features, often require manual confirmation of container position and handling order during the handling process, which severely limits the level of automation.

[0004] Furthermore, open-top containers are often stacked together with other container types in practical applications, resulting in indistinct edges and backgrounds. This makes visual recognition models prone to misjudgments, leading to failed container grabbing, mis-grabbing, or damage to the container, especially at night or in rainy or foggy weather. Existing methods have failed to effectively solve the problem of automatic grabbing and placement recognition of open-top containers, lacking targeted control logic and processes, resulting in low operational efficiency. Summary of the Invention

[0005] The purpose of this invention is to provide a vision-based automated grabbing and placing control system for open-top containers, in order to address the shortcomings of the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a vision-based automated open-top container grabbing and placing control system, comprising:

[0007] Image acquisition module: Acquires image information of open-top containers in the port;

[0008] Image preprocessing module: performs preprocessing on the image information;

[0009] Feature extraction module: Uses the trained neural network model to extract open-top box features from the preprocessed image, obtaining the edge feature set and opening contour features of each visible box;

[0010] Target generation module: Based on the extracted opening contour features, determine whether the open-top box is empty or whether there are obstructions, and generate a set of candidate box grabbing points according to the grabbing priority rules;

[0011] Scoring and Ranking Module: Scores the credibility of each capture point in the candidate capture point set to determine the capture order;

[0012] Command generation module: Inputs the coordinates of the highest-scoring grabbing point and its corresponding box edge features into the control system to generate grabbing control commands;

[0013] Grab and release execution module: Executes grab and release actions according to the grab box control command, and monitors the grab status in real time;

[0014] Anomaly handling module: When a grabbing failure or pose deviation exceeding a set threshold is detected, the next grabbing target is selected from the candidate grabbing point set and the control process is repeated until the grabbing is successful or the open-top box is marked as ungrabable.

[0015] Preferably, the feature extraction module includes:

[0016] The preprocessed image is divided into multiple analysis regions, and the initial edge contour lines are determined based on the brightness gradient changes within the regions.

[0017] Based on the edge contour line, extract the boundary extension features in the vertical and horizontal directions to construct a candidate set of box edges;

[0018] Structural connectivity analysis is performed on the feature points in each candidate set to select regions that form closed rectangles or approximate rectangles as the target box contour;

[0019] Based on the continuity of the upper boundary of the target box profile and the internal brightness distribution characteristics, the opening area at the top of the box is identified, and a corresponding opening profile feature set is generated.

[0020] Preferably, the target generation module includes:

[0021] The mean and variance of the brightness of the internal area of ​​the box are calculated based on the opening contour features and compared with the set empty box threshold range to determine whether the open-top box is empty.

[0022] Occlusion analysis is performed on image regions that are detected as non-empty boxes, and regions whose brightness is less than 0.8 times the average of the background and whose area exceeds a set ratio are identified as occluded regions.

[0023] Within the unobstructed opening area, candidate areas for grabbing points that meet the grabbing conditions are selected based on edge smoothness and spatial symmetry.

[0024] A set of candidate capture points is generated by combining preset priority rules.

[0025] Preferably, the step of selecting candidate regions for grabbing points that meet the grabbing conditions includes: selecting the intersection of the upper edge of the opening and the left and right boundaries, and setting a fixed depth range vertically below it as the grabbing point evaluation zone, and analyzing the continuity and smoothness of the edge lines within the region; if the pixel continuity rate of the edge lines within the grabbing region is higher than 90%, and the difference in length between the left and right edges does not exceed 10 pixels, it is considered a candidate region with a complete structure and suitable for performing grabbing operations, and its center point coordinates are recorded as candidate grabbing points.

[0026] Preferably, the scoring and sorting module includes:

[0027] For each candidate grab point, a structural integrity score is calculated, based on the continuous pixel rate and boundary straightness of its edge region;

[0028] Extract the spatial position of each grab point in the image coordinate system, and calculate the symmetry score based on the offset distance from the geometric center of the box;

[0029] Priority levels are assigned based on the stacking height information of the containers at the grab point;

[0030] The three indicators of structural integrity score, symmetry score and stacking priority are weighted and fused to output the comprehensive credibility score of each grab point, and the grab order is arranged from high to low according to the score.

[0031] Preferably, the method for obtaining the comprehensive credibility score is as follows:

[0032] The overall credibility score is calculated as follows: (structural integrity score × 0.4) + (symmetry score × 0.3) + (stack priority score × 0.3). All grabbing points are sorted from highest to lowest credibility score to generate the final grabbing order list.

[0033] The method for obtaining the symmetry score is as follows: taking the geometric center of the box as the reference point, calculate the offset distance of the grab point relative to the center point in the image coordinate system, and determine its left and right symmetry. If the difference in offset distance between the grab point and the center is less than 10 pixels, the maximum allowable offset is set to 50 pixels when scoring. The symmetry score is linearly distributed according to the inverse proportional relationship between the offset distance from small to large.

[0034] The method for obtaining the stacking priority score is as follows: obtain the vertical stacking height of the boxes corresponding to the grab point, divide the height into several levels, set the highest priority to 100 points, and decrease the score by 20 points for each additional layer of stacking height, with a minimum score of 0 points.

[0035] Preferably, the capture and release execution module includes:

[0036] Collect grasping status data, including the closing angle of the grasping fixture, contact pressure value, vibration feedback, and weight change data after the object is grasped.

[0037] Preferably, the exception handling module includes:

[0038] During the gripping and releasing process, the clamp pressure, posture changes and gripping displacement data are collected in real time to determine whether the gripping is successful.

[0039] When the clamp pressure is detected to be lower than the preset gripping threshold or the pose deviation exceeds the set angle range, the gripping is deemed to have failed.

[0040] Based on the previous capture results, the failed capture points are removed from the candidate capture point set, and the next optimal capture point is selected again based on the original score;

[0041] Regenerate the corresponding grab box control command to execute a new round of grabbing. If the number of consecutive grabbing failures reaches the set limit, mark the current open-top box as ungrabable and terminate the task.

[0042] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0043] 1. This invention solves the problem that traditional standard container visual recognition models cannot adapt to complex working conditions such as the open-top container's uncovered structure, varied shapes, and severe obstruction by constructing a vision-based automatic grasping and placement recognition control method tailored to the structural characteristics of open-top containers. Through modular design, it achieves a closed-loop function encompassing image acquisition, preprocessing, feature extraction, grasping judgment, scoring and ranking, command control, and anomaly handling, enabling high-precision recognition and dynamic decision-making for grasping open-top containers in complex port environments.

[0044] 2. Compared to existing container grabbing methods that rely on manual assistance or rule-driven approaches, this invention significantly improves structural recognition accuracy, empty container identification reliability, grabbing success rate, and overall operational efficiency. Particularly in terms of grabbing failure feedback, optimal grabbing point selection, and adaptive instruction iteration, it achieves dual guarantees of stability and fault tolerance through a multi-dimensional scoring mechanism and retry logic. It is applicable to various practical application scenarios such as unmanned container yards, ports, and warehouses, providing technical support and an engineering foundation for the automated processing of non-standard containers. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0046] Figure 1 This is a flowchart of the system modules of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] For examples, please refer to Figure 1 As shown in this embodiment, a vision-based automated open-top container grabbing and placing control system includes:

[0049] Image acquisition module: Acquires image information of open-top containers in the port;

[0050] Image preprocessing module: performs preprocessing on the image information;

[0051] Feature extraction module: Uses the trained neural network model to extract open-top box features from the preprocessed image, obtaining the edge feature set and opening contour features of each visible box;

[0052] Target generation module: Based on the extracted opening contour features, determine whether the open-top box is empty or whether there are obstructions, and generate a set of candidate box grabbing points according to the grabbing priority rules;

[0053] Scoring and Ranking Module: Scores the credibility of each capture point in the candidate capture point set to determine the capture order;

[0054] Command generation module: Inputs the coordinates of the highest-scoring grabbing point and its corresponding box edge features into the control system to generate grabbing control commands;

[0055] Grab and release execution module: Executes grab and release actions according to the grab box control command, and monitors the grab status in real time;

[0056] Anomaly handling module: When a grabbing failure or pose deviation exceeding a set threshold is detected, the next grabbing target is selected from the candidate grabbing point set and the control process is repeated until the grabbing is successful or the open-top box is marked as ungrabable.

[0057] In this invention, the image acquisition module is used to acquire image information of open-top containers in the port, and is the front-end foundation of the system for realizing automated identification and control.

[0058] Specifically, the image acquisition module includes multiple industrial cameras, depth cameras, LiDAR, or combinations thereof mounted on fixed supports or mobile platforms (such as unmanned vehicles, rail-mounted gantry cranes, and forklifts) to acquire real-time images of the port area from multiple perspectives. The acquired image information includes the overall outline of open-top containers, their stacking location, edge features, and the state of their internal openings, which are then used by the subsequent image processing module for target recognition and feature extraction. The image acquisition module can perform area scanning or point-to-point snapshots according to task scheduling instructions and transmit the acquired image data to the backend processing system in real time via wired or wireless communication.

[0059] In this invention, the image preprocessing module is used to perform preliminary processing on the image information acquired by the image acquisition module in order to improve image clarity and enhance the recognizability of key target areas.

[0060] This module first normalizes the size of the input image to ensure consistent resolution and smooth processing. Next, it automatically adjusts the image's brightness and contrast to adapt to variations in lighting conditions, ensuring the open-top box's outline and edge structure are clearly represented. Simultaneously, the image preprocessing module masks interfering elements such as background clutter, ground markings, and vehicle components by defining region boundaries and identifying color ranges to retain valid target areas. Furthermore, to eliminate blur or noise caused by equipment vibration or environmental changes, the preprocessing module performs multi-frame comparison and fusion, extracting stable frames for subsequent analysis. Finally, the preprocessed image is divided into multiple regions for analysis, each containing at least one possible open-top box structure.

[0061] The feature extraction module in this invention first performs region segmentation on the image output by the image preprocessing module. Specifically, the entire image is divided into blocks according to a fixed grid size (e.g., each grid region is 128×128 pixels), defining multiple analysis regions. Within each analysis region, the brightness gradient change amplitude between adjacent pixels is calculated, and a pixel brightness difference greater than a set threshold (e.g., 25) is used as the initial edge judgment criterion, thereby forming an initial edge contour line within the region. The initial edge contour line is used to indicate edge features that may contain open-top box structures.

[0062] After obtaining the initial edge contour lines, the feature extraction module further performs directional analysis to extract boundary extension features in the vertical and horizontal directions. Specifically, the system adopts a line segment growth strategy, starting from the endpoint of each initial edge contour line and extending the boundary to both sides along the horizontal direction (X-axis) and the vertical direction (Y-axis) until the brightness change is lower than a set termination threshold (e.g., gradient difference less than 10) or an image edge is encountered. All boundary lines that meet the requirement of a length greater than a set minimum value (e.g., 30 pixels) are included in the box edge candidate set.

[0063] For the feature line segments in the candidate set of box edges, structural connectivity analysis is performed to determine whether a closed rectangular or approximately rectangular contour structure can be formed. Specifically, the intersection points and included angles between candidate edge line segments are calculated. When four line segments form a closed polygon with included angles between 85° and 95° and side length ratios between 1:1 and 1:3, it is considered a valid open-top box contour structure. Based on the verified contour, the average brightness of its internal area is checked to see if it is lower than the surrounding area, further confirming it as a concave structure. All contours meeting the above conditions are marked as valid target box contours and used for subsequent processing.

[0064] Based on the target box contour extracted in the above steps, the continuity of its upper boundary and the internal brightness distribution are further analyzed to identify the opening area at the top of the box. Specifically, the system first extracts the pixels of the upper boundary of the box and detects its continuous length in the horizontal direction. When the number of continuous pixels exceeds a set threshold (e.g., 60 pixels) and there are no obvious breaks at the edge (the interval between broken pixels does not exceed 5 pixels), the boundary is determined to be the upper edge of a complete opening. Subsequently, the brightness of the internal region of the contour is statistically analyzed. If the average internal brightness is higher than a set average background brightness (e.g., greater than 1.2 times the average brightness of the image), it is inferred that the box is in an open state with no internal obstruction. Finally, the boundary coordinates and brightness features of the opening area are integrated into an opening contour feature set for subsequent modules to perform empty box recognition and optimal grasping point selection.

[0065] In this invention, the target generation module first performs statistical analysis of the internal brightness features of the identified opening contour regions to determine whether the open-top box is empty. Specifically, the image region enclosed by each opening contour is selected as the analysis object, and the average brightness value and brightness variance of all pixels within that region are calculated. The average brightness value reflects the overall brightness of the region, while the brightness variance determines whether the content of the region exhibits strong variations. The system pre-sets an empty box judgment threshold range; for example, if the average brightness value is higher than 1.2 times the global average brightness of the image, and the variance is lower than a set value (e.g., less than 50), it is determined to be an empty box with no internal contents and a simple structure. This judgment is based on statistical feature modeling from a large number of actual images of open-top boxes, and sample verification enables effective differentiation between empty and full boxes.

[0066] When the container is determined to be non-empty, the target generation module continues with occlusion analysis to identify any obstructions inside the container. First, the brightness of each pixel within the opening contour area is scanned in sections to identify dark areas with continuous brightness lower than 0.8 times the background mean, which are considered potential occlusion areas. The background mean is the average brightness of the non-container areas in the entire image, calculated by the image preprocessing module. If the area of ​​this dark area exceeds a set threshold (e.g., 30%), the open-top container is marked as having an obstruction. The occlusion analysis also considers edge irregularities and texture interference. By calculating the continuity and area density of the occluded area's edge contour, false positives are further eliminated, improving occlusion detection accuracy.

[0067] After confirming that the target container is empty or partially unobstructed, candidate grab points are selected within the opening area. Candidate point selection is based on edge smoothness and spatial symmetry. Specifically, the intersection of the upper edge of the opening and the left and right boundaries is selected, and a fixed depth (e.g., 30 pixels) is set vertically below it as the grab point evaluation zone. The continuity and smoothness of the edge lines within this area are analyzed. If the pixel continuity of the edge lines within this grab area is higher than 90%, and the difference in left and right symmetry does not exceed a set value (e.g., the difference in left and right edge lengths does not exceed 10 pixels), it is considered a structurally complete candidate area suitable for grabbing, and its center point coordinates are recorded as a candidate grab point.

[0068] All selected grab points are prioritized to generate a final candidate grab point set. The priority ranking rules are set according to actual operational needs and may include the following categories:

[0069] Those located in the center of the container will be given priority due to their high gripping stability;

[0070] Prioritize images located on the upper layer of the stack or near the upper part of the image to avoid occlusion and conflict;

[0071] Prioritize those with higher edge integrity scores to improve the success rate of one-time capture.

[0072] The location information, corresponding score value, and container number of each candidate grab point are encapsulated into a set of candidate grab points, which are then called by the scoring and control module to execute the grab and release action.

[0073] In this invention, the scoring and ranking module first scores the structural integrity of the edge region where the candidate grab point is located. This score is used to evaluate whether the grab position has sufficient structural stability and continuity. A fixed-size edge analysis window (e.g., 50 pixels wide and 30 pixels high) is set centered on the candidate grab point, and the percentage of continuous pixels on the edge line is counted within this window. If the proportion of continuous pixels to the total number of edge line pixels is higher than 90%, and the offset of the edge in the horizontal or vertical direction does not exceed 5 pixels, the boundary of the region is determined to be straight. Based on the continuity rate and the edge fluctuation range, a structural integrity score is given out out of 100, with the continuity rate accounting for 60% and straightness accounting for 40%.

[0074] Further calculations are made to determine the symmetry score of the candidate capture points. Using the geometric center of the box as a reference point, the offset distance of each capture point relative to the center point in the image coordinate system is calculated, and the degree of left-right symmetry is judged. If the difference in offset distance between the capture points on either side of the center is less than 10 pixels, the symmetry is considered good. During scoring, a maximum allowable offset of 50 pixels is set. The symmetry score is linearly distributed according to an inverse relationship between the offset distance and the score; the smaller the offset, the higher the score. For example, an offset of 0 pixels results in a score of 100, while an offset of 50 pixels results in a score of 0.

[0075] To prioritize grabbing boxes located at the top of the stack or in easily accessible positions, a stacking priority scoring mechanism is introduced. First, the image acquisition module obtains the vertical stacking height of the boxes corresponding to the grabbing point. This height is divided into several levels, for example, the top layer is layer 1, and so on, down to layer 2, layer 3, and so on. The highest priority is set at 100 points, and the score decreases by 20 points for each additional stacking layer, with a minimum score of 0. If the grabbing point is on the top layer (e.g., layer 1), the score is 100 points; if it's on layer 3, the score is 60 points.

[0076] The scoring and ranking module weights and fuses the three sub-scoring results to generate a comprehensive credibility score for each candidate grabbing point. Weight allocation can be set based on experience or pre-trained parameters, for example: structural integrity score weight 0.4, symmetry score weight 0.3, and stacking priority score weight 0.3. The comprehensive credibility score = (structural integrity score × 0.4) + (symmetry score × 0.3) + (stacking priority score × 0.3). All grabbing points are sorted from highest to lowest credibility score to generate a final grabbing order list. This grabbing order is directly transmitted to the control module to control the robotic arm or grabbing device to perform priority grabbing.

[0077] In this invention, the instruction generation module transmits the optimal grabbing point coordinates and corresponding box edge feature information output by the scoring and ranking module to the grabbing control system, and generates grabbing control instructions that can be recognized by the actuator. Specifically, the grabbing point with the highest comprehensive confidence score is first extracted. Its two-dimensional coordinate values ​​are based on the image coordinate system and converted into three-dimensional spatial pose information through calibration parameters and depth information, including position coordinates (X, Y, Z) and grabbing posture angles (Pitch, Yaw, Roll). Simultaneously, the edge structure features of the box where the grabbing point is located are extracted, including the four corner positioning points, side length parameters, and box opening plane normal vector, to assist in posture adjustment and obstacle avoidance path planning. The instruction generation module constructs a control instruction data packet based on the above parameters. The instruction content includes the three-dimensional coordinates of the target grabbing point, the opening angle of the grabbing fixture, the descent depth, the grabbing force level, and the execution timing identifier. This control instruction is transmitted to the control execution unit in real time through a communication interface (such as CAN bus or Ethernet), achieving seamless integration from visual recognition to motion control.

[0078] In this invention, the gripping and releasing execution module receives and executes the gripping control commands issued by the command generation module, and monitors the entire gripping process in real time through a sensing and feedback mechanism to ensure the accuracy and safety of the gripping and releasing operation. Specifically, the execution unit (such as an industrial robot, intelligent forklift, or gripping device) first parses the three-dimensional gripping coordinates, posture angles, descent depths, and fixture parameters in the control commands, and automatically adjusts its posture according to the gripping path. During the gripping process, the gripping and releasing execution module collects gripping status data in real time through a multimodal sensing device mounted on the end effector, including the closing angle of the gripping fixture, contact pressure value, vibration feedback, and weight change data after the object is gripped.

[0079] In this invention, the anomaly handling module monitors the gripping status in real time during the gripping and releasing process, collecting key parameters including clamp pressure, end effector pose change, and gripping path displacement to determine the effectiveness of the gripping action. Clamp pressure is collected in real time by a pressure sensor integrated inside the mechanical gripper, measured in Newtons, with a refresh rate of no less than 10 times per second; pose change is provided by a six-axis inertial sensor or encoder mounted on the clamp, including changes in attitude angles (pitch, yaw, and roll); gripping path displacement is fed back from the servo system, measured in millimeters.

[0080] If the gripper fails to close to the target distance within a predetermined time window (e.g., 2 seconds) and the gripper pressure is lower than the set minimum effective gripping threshold (e.g., 20 Newtons), a preliminary gripping failure is determined. Simultaneously, if the deviation between the actual pose and the commanded pose of the end effector after attitude adjustment exceeds a set angle range (e.g., pitch or yaw angle exceeding 5 degrees), this is also considered an attitude deviation exceeding the limit, resulting in gripping failure. The system uses a logic checker to treat "insufficient gripping force" or "attitude deviation exceeding the limit" as independent failure conditions; if either condition is met, an exception handling process is triggered to ensure timely fault response.

[0081] After a capture failure is determined, the failed capture point is immediately removed from the candidate capture point set to avoid duplicate operations. The candidate capture point set is a set of capture targets generated and sorted by the scoring and sorting module. The system records the execution status of each capture point. When a capture point is marked as "failed," its priority in the set is set to the lowest, or it is removed directly. Subsequently, a comprehensive calculation is performed again from the remaining candidate capture points according to the original scoring weights (such as structural integrity, symmetry, and stacking priority). The new capture point with the highest score is selected and sent to the instruction generation module to trigger a new round of capture control instruction generation.

[0082] After executing a new round of capture, the capture results are continuously monitored. When the number of consecutive capture failures reaches the set limit (e.g., 3 times), meaning that all capture points within the open-top container have been attempted without success, the system marks the current open-top container as "uncaptureable." This status is achieved by adding an anomaly flag field to the container in the system task list, and simultaneously recording the container ID in the anomaly container queue for subsequent manual intervention by the operator or for the system to skip. After terminating the current task, the system automatically schedules the next target for identification, ensuring uninterrupted workflow.

[0083] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A vision-based automated open-top container grabbing and placing control system, characterized in that: include: Image acquisition module: Acquires image information of open-top containers in the port; Image preprocessing module: performs preprocessing on the image information; Feature extraction module: Uses the trained neural network model to extract open-top box features from the preprocessed image, obtaining the edge feature set and opening contour features of each visible box; Target generation module: Based on the extracted opening contour features, determine whether the open-top box is empty or whether there are obstructions, and generate a set of candidate box grabbing points according to the grabbing priority rules; Scoring and Ranking Module: Scores the credibility of each capture point in the candidate capture point set to determine the capture order; Command generation module: Inputs the coordinates of the highest-scoring grabbing point and its corresponding box edge features into the control system to generate grabbing control commands; Grab and release execution module: Executes grab and release actions according to the grab box control command, and monitors the grab status in real time; Anomaly handling module: When a grabbing failure or pose deviation exceeding a set threshold is detected, the next grabbing target is selected from the candidate grabbing point set and the control process is repeated until the grabbing is successful or the open-top box is marked as ungrabable.

2. The vision-based automated open-top container grabbing and placing control system according to claim 1, characterized in that: The feature extraction module includes: The preprocessed image is divided into multiple analysis regions, and the initial edge contour lines are determined based on the brightness gradient changes within the regions. Based on the edge contour line, extract the boundary extension features in the vertical and horizontal directions to construct a candidate set of box edges; Structural connectivity analysis is performed on the feature points in each candidate set to select regions that form closed rectangles or approximate rectangles as the target box contour; Based on the continuity of the upper boundary of the target box profile and the internal brightness distribution characteristics, the opening area at the top of the box is identified, and a corresponding opening profile feature set is generated.

3. The vision-based automated open-top container grabbing and placing control system according to claim 1, characterized in that: The target generation module includes: The mean and variance of the brightness of the internal area of ​​the box are calculated based on the opening contour features and compared with the set empty box threshold range to determine whether the open-top box is empty. Occlusion analysis is performed on image regions that are detected as non-empty boxes, and regions whose brightness is less than 0.8 times the average of the background and whose area exceeds a set ratio are identified as occluded regions. Within the unobstructed opening area, candidate areas for grabbing points that meet the grabbing conditions are selected based on edge smoothness and spatial symmetry. A set of candidate capture points is generated by combining preset priority rules.

4. The vision-based automated open-top container grabbing and placing control system according to claim 3, characterized in that: The process of selecting candidate regions for grabbing points that meet the grabbing conditions includes: selecting the intersection of the upper edge of the opening and the left and right boundaries, and setting a fixed depth range vertically below it as the grabbing point evaluation zone, and analyzing the continuity and smoothness of the edge lines within the region; if the pixel continuity rate of the edge lines within the grabbing region is higher than 90%, and the difference in length between the left and right edges does not exceed 10 pixels, it is considered a candidate region with a complete structure and suitable for performing grabbing operations, and its center point coordinates are recorded as candidate grabbing points.

5. The vision-based automated open-top container grabbing and placing control system according to claim 1, characterized in that: The scoring and ranking module includes: For each candidate grab point, a structural integrity score is calculated, based on the continuous pixel rate and boundary straightness of its edge region; Extract the spatial position of each grab point in the image coordinate system, and calculate the symmetry score based on the offset distance from the geometric center of the box; Priority levels are assigned based on the stacking height information of the containers at the grab point; The three indicators of structural integrity score, symmetry score and stacking priority are weighted and fused to output the comprehensive credibility score of each grab point, and the grab order is arranged from high to low according to the score.

6. The vision-based automated open-top container grabbing and placing control system according to claim 5, characterized in that: The method for obtaining the comprehensive credibility score is as follows: The overall credibility score is calculated as follows: (structural integrity score × 0.4) + (symmetry score × 0.3) + (stack priority score × 0.3). All grabbing points are sorted from highest to lowest credibility score to generate the final grabbing order list. The method for obtaining the symmetry score is as follows: taking the geometric center of the box as the reference point, calculate the offset distance of the grab point relative to the center point in the image coordinate system, and determine its left and right symmetry. If the difference in offset distance between the grab point and the center is less than 10 pixels, the maximum allowable offset is set to 50 pixels when scoring. The symmetry score is linearly distributed according to the inverse proportional relationship between the offset distance from small to large. The method for obtaining the stacking priority score is as follows: obtain the vertical stacking height of the boxes corresponding to the grab point, divide the height into several levels, set the highest priority to 100 points, and decrease the score by 20 points for each additional layer of stacking height, with a minimum score of 0 points.

7. The vision-based automated open-top container grabbing and placing control system according to claim 6, characterized in that: The capture and release execution module includes: Collect grasping status data, including the closing angle of the grasping fixture, contact pressure value, vibration feedback, and weight change data after the object is grasped.

8. The vision-based automated open-top container grabbing and placing control system according to claim 7, characterized in that: The exception handling module includes: During the gripping and releasing process, the clamp pressure, posture changes and gripping displacement data are collected in real time to determine whether the gripping is successful. When the clamp pressure is detected to be lower than the preset gripping threshold or the pose deviation exceeds the set angle range, the gripping is deemed to have failed. Based on the previous capture results, the failed capture points are removed from the candidate capture point set, and the next optimal capture point is selected again based on the original score; Regenerate the corresponding grab box control command to execute a new round of grabbing. If the number of consecutive grabbing failures reaches the set limit, mark the current open-top box as ungrabable and terminate the task.

Images