Container lock head disorder sorting and lock head management system

Abstract

The application discloses a container lock head disordered sorting and lock head management system and relates to the technical field of port logistics automation. The system establishes a lock head and lock frame model library through 3D scanning modeling, combines AI video recognition to trigger laser radar scanning, obtains a point cloud map and matches a lock head coarse coordinate; a robot uses a No. 1 electric control magnetic suction head to move the lock head to an electric primary sorting table, then a 3D camera finely identifies the type and calculates a magnetic suction point, if the identification fails, the primary sorting table is rolled to change the posture; then a No. 2 electric control magnetic suction head is used to move the lock head to a self-adaptive positioning disc composed of spring needles to keep vertical, then a 3D camera obtains an accurate grabbing point and posture, an electric clamp grabs and places the lock head in an electric ring-shaped lock rack, and an upper computer records position information throughout the process. The application combines 3D vision, point cloud matching, controllable magnetic adsorption and self-adaptive positioning technology, realizes lock head disordered sorting, type identification, accurate positioning and intelligent management, and significantly improves sorting efficiency and safety.

Description

Technical Field

[0001] This invention relates to the fields of port logistics automation, intelligent warehousing and robot application technology, specifically a container lock disorder sorting and lock management system. Background Technology

[0002] In container loading and unloading operations, locks (also known as latches or corner locks) are used to secure the connection between containers and are a key component to ensure transportation safety. Traditional methods rely on manual sorting, handling, and locking of locks, which presents the following problems: The locks are of various types and irregular shapes, making manual identification and sorting inefficient; manual operation poses safety hazards and is prone to accidents such as pinching and crushing injuries; the large number of locks and their disorderly stacking make automated management difficult; and there is a lack of systematic recording and tracking of the location, quantity, and type of locks.

[0003] Therefore, there is an urgent need for a system that can achieve disordered sorting, automatic identification, precise positioning, and intelligent management of lock heads in order to improve operational efficiency, reduce labor costs, and ensure operational safety. Summary of the Invention

[0004] The purpose of this invention is to provide a container lock sorting and lock management system to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: the container lock disorder sorting and lock management system includes the following steps: Step 1: Perform 3D scanning and modeling of all lock heads and lock frames, and save the model data to the model library; Step 2: Collect video in real time through the camera and perform AI recognition. When a lock frame is detected in the area where the lock frame is placed, notify the host computer to start the multi-line LiDAR to scan the lock frame. Step 3: Match the point cloud map obtained by the LiDAR with the saved lock frame model, analyze and determine the position of the lock head, obtain the three-dimensional coordinates of the lock head, and then notify the host computer, which in turn notifies the robot. Step 4: Based on the coordinate information, the robot uses the No. 1 electrically controlled magnetic suction head to pick up the lock from the lock frame and place it on the electric initial sorting table; Step 5: The host computer notifies the 3D camera to obtain the point cloud image data of the lock head, matches it with the model library, determines the lock head type, and calculates the magnetic point coordinates of the lock head based on the magnetic point information in the model library, and notifies the robot to perform grasping. Step 6: If the 3D camera can identify the lock type but cannot obtain the magnetic point coordinates, start the electric initial sorting table to roll for 2 seconds and then stop to change the lock posture; Step 7: The host computer instructs the 3D camera to acquire point cloud data again. If the magnetic point coordinates still cannot be acquired, step 6 can be repeated. Step 8: If the 3D camera detects that the No. 1 electronically controlled magnetic head has grabbed two small locks (the large locks cannot be grabbed due to their weight and suction force), the host computer will notify the robot to grab them one by one. Step 9: After the host computer obtains the coordinates of the magnetic attraction point of the lock head, it instructs the robot to use the No. 2 electronically controlled magnetic attraction head to pick up the magnetic attraction point of the lock head and place it on the adaptive positioning disk; Step 10: The host computer notifies the 3D camera again to obtain the point cloud data of the lock head, obtains the coordinates (xyz) and orientation (ABC) of the grasping point of the lock head through template matching, and notifies the robot to use the electric gripper to grasp and place it on the electric ring lock frame. Step 11: The host computer simultaneously instructs the electric ring lock frame to rotate the lock frame position of the cavity to the dedicated position for the robot to place the lock head; Step 12: The host computer records the specific position of the lock head on the electric ring lock frame, completing the entire workflow.

[0006] The system includes the following components: Electric ring lock frame robot Electric clamps Electromagnetic suction heads (No. 1, No. 2) Electric sorting table Lock frame Adaptive positioning disk 3D camera Camera (video recognition) Multi-line lidar Host computer (control and data processing center).

[0007] The beneficial effects of this invention are: 1. Multi-technology integration design: Integrates 3D vision, AI recognition, point cloud matching, robot control, electromagnetic adsorption and other technologies to solve the problem of fully automatic assembly and unlocking; 2. 3D point cloud map + model matching and recognition: The point cloud map of the lock head is acquired by a 3D camera, matched with the model library, the lock head type is identified and the capture coordinates are calculated; 3. Controllable magnetic suction head design: It adopts a special shaped magnet and works with a servo motor to control the magnetic force, so as to achieve vertical gripping of the lock head and ensure the sorting success rate; 4. Adaptive positioning plate: Composed of hundreds of spring pins, the locking head is placed and kept vertically stable by the spring force, which facilitates the gripping of electric clamps and is the key to realizing disordered sorting. 5. Electric intelligent lock rack management: The electric ring lock rack can effectively manage the position and quantity of different types of lock heads, improve lock installation efficiency, and significantly increase production efficiency. Attached Figure Description

[0008] Figure 1 Overall system diagram. Detailed Implementation

[0009] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0010] I. System Initialization and Modeling Phase When the system is first put into operation or a new lock type is added, an initialization modeling step must be performed: Lock head modeling: Various standard container lock heads (such as semi-automatic locks, fully automatic locks, intermediate locks, etc.) are placed in the modeling station in sequence. High-precision 3D cameras (such as structured light or laser stripe cameras) are used to collect their point cloud data from multiple angles. The geometric contours, size features, magnetic attraction point positions, clamp gripping points and posture information of the lock heads are extracted to generate standardized templates and store them in the model library.

[0011] Lock frame modeling: Perform 3D scanning on the actual lock frame (usually a mesh or grid structure) to obtain its external dimensions, mesh distribution, support surface height and other features, which are used to distinguish the frame from the lock head during subsequent point cloud matching.

[0012] Each record in the model library includes: a unique lock ID, a type name, a 3D point cloud template, recommended magnetic attraction point coordinates (x, y, z), recommended grab point coordinates and pose (x, y, z, A, B, C), and suggested magnetic attraction level values.

[0013] II. Lock Frame Inspection and Lock Cylinder Coarse Positioning Stage After the system starts up, each module enters standby mode: Video surveillance and AI recognition: A camera mounted above the lock frame (either a standard RGB camera or a low-light camera with infrared capability) captures images in real time. A built-in deep learning object detection model (such as YOLOv8 or Faster R-CNN) determines whether the lock frame is in place. Recognition includes: frame outline, positional offset, and tilt angle.

[0014] Triggering LiDAR Scan: When the AI ​​detects that the lock frame is stably present and remains there for more than 3 seconds (to prevent accidental triggering by personnel passing by), the camera sends a "frame in position" signal to the host computer. The host computer then activates a multi-line LiDAR (such as a 16-line or 32-line LiDAR, mounted on the robot or a fixed bracket) to scan the lock frame area.

[0015] Point cloud acquisition and filtering: The lidar acquires raw point cloud data, and the host computer preprocesses the point cloud, including: Outlier removal (radius filtering or statistical filtering); Downsampling (voxel filtering, voxel size can be set to 1cm); Ground and background removal (ground points are removed by RANSAC plane fitting).

[0016] Coarse localization matching: The preprocessed point cloud is registered with the lock frame template in the model library using ICP (Iterative Nearest Point) to identify candidate point cloud clusters for all lock heads within the frame. The centroid coordinates of each point cloud cluster are calculated as the coarse 3D position (x, y, z) of the lock head and sent to the host computer. If matching fails (e.g., due to mismatched lock frame type or incomplete point cloud), the system will repeat the scan and, if it still fails, will issue an alarm prompting manual intervention.

[0017] III. Initial Grabbing and Sorting Stage Robot path planning: After receiving the coarse positioning coordinates, the host computer combines the robot's current pose and uses the RRT algorithm to plan a collision-free picking path, and controls the robot to move to the first lock position.

[0018] Electro-controlled magnetic suction head No. 1: The robot's end effector is equipped with an electro-controlled magnetic suction head No. 1, which internally contains an electromagnet and a servo pressure control mechanism. The host computer issues magnetic force level commands based on the lock type (estimated through preliminary point cloud matching results), for example: Small padlock (weight <1kg): Magnetic force 30%; Medium-sized lock (1-3kg): Magnetic force 60%; Large padlock (>3kg): Magnetic force 90%.

[0019] After the magnetic head contacts the surface of the lock, the built-in contact sensor confirms successful adsorption. Then, the robot lifts the lock vertically from the lock frame and moves it above the electric initial sorting table.

[0020] Placement on the initial sorting table: The robot places the lock in the starting area of ​​the motorized initial sorting table (consisting of multiple parallel motorized rollers) and then retreats to a safe position. The motorized initial sorting table has independent control capabilities and can roll in both forward and reverse directions.

[0021] IV. Fine Identification and Magnetic Attraction Point Calculation Stage 3D camera acquisition: The host computer instructs the 3D camera (such as a binocular structured light camera with an accuracy of ±0.5mm) located above the initial sorting table to perform a fine scan of the lock head and acquire high-density point cloud data.

[0022] Type matching: The system matches the detailed point cloud against all templates in the model library, calculates the similarity using point-to-point feature (PPF) or deep learning methods, and selects the template with the highest similarity and above a threshold (e.g., 85%) as the recognition result. If the similarity is below the threshold, the system determines it as an unknown lock and issues an alarm for manual handling.

[0023] Magnetic point localization: Based on the preset magnetic point position in the matched template (the offset relative to the center of mass of the lock head), the coordinates of the template magnetic point are transformed into the actual coordinate system of the current lock head to obtain the precise xyz coordinates of the magnetic point. If the template contains multiple candidate magnetic points (e.g., both sides of the lock head are suitable for adsorption), the system prioritizes the one in the point cloud that is not occluded and has a relatively flat surface.

[0024] Exception handling process: Scenario A: If the lock type is identified but the magnetic attraction point cannot be calculated (e.g., the magnetic attraction point is obstructed or located in a void area of ​​the point cloud), the motorized initial sorting platform is activated, causing it to roll at a speed of 0.2 m / s for 2 seconds to change the lock's orientation. Then, the rolling stops, and steps 1-3 are repeated. If the process fails after two repetitions, the lock is considered abnormal and is picked up by the robot and placed in the abnormal item collection bin.

[0025] Scenario B: The point cloud map shows that two small locks are simultaneously attracted to the magnetic head 1 (determined by detecting the number or area of ​​point cloud clusters), indicating that the two locks are being attracted together due to stacking or magnetic adhesion. At this point, the host computer does not perform subsequent fine-tuning; instead, it directly instructs the robot to move above the initial sorting table, use the magnetic head 1 to release one of the locks onto the initial sorting table, then rotate or translate the robot to release the second lock. The two locks are then processed one by one.

[0026] Scenario C: If the lock head is found to be severely deformed or damaged during fine recognition (the point cloud and template matching degree is <50%), the system will automatically remove it to the waste area.

[0027] V. Precise Transfer and Adaptive Positioning Stage Electro-controlled magnetic head #2 gripping: After the host computer obtains the coordinates of the effective magnetic attraction point, it switches to the No. 2 electro-controlled magnetic attraction head on the robot's end effector (which has the same structure as No. 1 but can be controlled independently to avoid cross-contamination). The robot moves to directly above the magnetic attraction point, descends vertically to make contact, and uses a medium magnetic force (e.g., 50%) to attract the lock head.

[0028] Placement onto the adaptive positioning disk: The robot moves the lock head above the adaptive positioning disk. The adaptive positioning disk consists of a densely arranged matrix of spring pins (e.g., a 20×20 array, each spring pin with a diameter of 3mm, a stroke of 15mm, and an initial spring force of 0.5N). The robot gently presses the lock head onto the positioning disk, causing the spring pins to automatically compress according to the shape of the bottom of the lock head, forming a matching indentation. Under the action of the spring restoring force, the lock head is clamped and kept in a vertically upward posture, without tilting or falling over.

[0029] Attitude confirmation: The host computer again instructs the 3D camera to quickly scan the lock head on the positioning disk to confirm that the verticality error of the lock head is less than 2°. If there is a deviation, it can be finely adjusted by the soft push rod at the end of the robot.

[0030] VI. Precise Positioning and Electric Gripper Grabbing Stage Grasp point calculation: The host computer calls the 3D camera again to obtain the point cloud map of the lock head on the adaptive positioning disk, performs fine registration with the model library, and calculates the most suitable gripping point coordinates (xyz) and gripping posture (ABC, i.e., rotation angles around the X, Y, and Z axes) for the electric clamp.

[0031] Electric gripper gripping: The robot switches to an electric gripper (such as a two-finger parallel gripper with force feedback control), and grips the lock head with a flexible force control method based on the calculated gripping point and posture (the gripping force is preset to 20N to prevent damage to the surface coating of the lock head).

[0032] Placement into the electric ring lock frame: The host computer queries the position of the empty slots in the electric ring lock frame in real time (i.e., the slots on the lock frame that do not yet contain lock heads) and sends the nearest empty slot number and position (polar coordinates or Cartesian coordinates) to the robot. The robot carries the lock head to directly above the empty slot, descends vertically and locks the lock head into the slot, the electric gripper releases and retracts.

[0033] VII. Locker Management and Full Process Recording Lock frame rotation and alignment: The electric ring lock frame has a multi-layer turntable structure, with each layer corresponding to a different type of lock head. The host computer controls the servo motor to drive the lock frame to rotate according to the current lock head type, rotating the target cavity layer of that type of lock head to the robot's dedicated lock placement station, and confirming its positioning through photoelectric sensors.

[0034] Location Recording: After each successful placement of the lock, the host computer records the following information in the database: The lock has a unique ID (if it is not pre-coded, a hash value can be generated based on point cloud features). Lock type; Placement time; Lock frame number and slot coordinates; Current status of the lock (ready for use / used / under maintenance).

[0035] Visualization and Query: The host computer provides a human-machine interface to display the occupancy status of each slot in the lock rack in a graphical way. It supports querying by type, time, and location, and has an inventory warning function (such as reminding to replenish when the quantity of a certain type of lock is lower than the threshold).

[0036] 8. Complete Loop and Abnormal Recovery Normal cycle: The system repeats steps two through seven until all locks in the lockbox have been sorted, or the user issues a stop command.

[0037] Anomaly recovery mechanism: If any communication timeout, robot collision, magnetic attraction failure, or fixture slippage occurs at any step, the system will immediately stop, lock the site, display the fault code and possible causes on the host computer interface, and issue an audible and visual alarm.

[0038] After the administrator resolves the fault, the system can resume execution from the point of interruption (the current step number is recorded by the state machine) without having to start over.

[0039] Safety protection: The entire work area is equipped with safety light curtains and emergency stop buttons. Once personnel are detected entering, the system will immediately shut down.

[0040] IX. Parameter Examples and Performance Indicators In one embodiment: Lock types: 3 types (semi-automatic lock, fully automatic lock, intermediate lock); Average sorting time per lock: 18 seconds; System positioning accuracy: ±1mm; Magnetic attraction success rate: >99.5%; The success rate of the adaptive positioning disk maintaining verticality is 100%. Electric ring lock frame capacity: 120 lock heads (6 layers × 20 slots).

Claims

1. A container lock sorting and lock management system, characterized in that, Includes the following steps: Step 1: Perform 3D scanning and modeling of all lock heads and lock frames, and save them to the model library; Step 2: Real-time video acquisition and AI recognition of the lock frame, and notification to the host computer to start LiDAR scanning; Step 3: Analyze the point cloud map to determine the lock position, obtain the xyz coordinates, and notify the robot. Step 4: The robot uses the No. 1 electrically controlled magnetic suction head to pick up the lock and place it on the electric initial sorting table; Step 5: The 3D camera acquires a point cloud image, matches it with the model library to identify the lock type, calculates the coordinates of the magnetic attraction point, and notifies the robot. Step 6: If the type is identified but the magnetic suction point cannot be obtained, start the initial picking station to roll for 2 seconds; Step 7: Repeat the attempt to obtain the coordinates of the magnetic attraction point; Step 8: If multiple small padlocks are detected, the robot will pick them up one by one; Step 9: Use the No. 2 electronically controlled magnetic suction head to pick up the magnetic suction point and place it on the adaptive positioning plate; Step 10: Obtain the point cloud map again, match the coordinates and attitude of the grab point, use the electric gripper to grab and place it onto the electric ring lock frame; Step 11: Rotate the electric ring lock frame to the release position; Step 12: The host computer records the position of the lock head on the lock frame, completing the process.

2. The container lock sorting and lock management system according to claim 1, characterized in that, The adaptive positioning disk consists of several spring pins, which are used to keep the lock head in a vertical position.

3. The container lock disorder sorting and lock management system according to claim 1, characterized in that, The electronically controlled magnetic suction head is equipped with a servo motor, which can control the magnetic force to ensure that the lock head grips vertically.

4. The container lock sorting and lock management system according to claim 1, characterized in that, The electric ring lock frame can rotate and record the position of each lock head, enabling orderly management of the lock heads.

5. The container lock sorting and lock management system according to claim 1, characterized in that, The host computer is responsible for coordinating the collaborative work of modules such as LiDAR, 3D camera, robot, locking frame, and initial sorting station.

Images