An unmanned vehicle precise parking method for a port crane scene

Abstract

The application discloses a kind of unmanned vehicle accurate parking methods of port crane scene, through the setting of desired accurate parking position, adjust the distance and direction of vehicle, guide vehicle accurate parking, through the position that template map has saved desired accurate parking, and the plane parameter of crane, improve the accuracy of vehicle stop, can improve the efficiency of loading and unloading box operation, increase the economic benefit of single vehicle unit time, simultaneously, through laser radar, the point cloud data of crane can be directly scanned and obtained, realize the accurate parking of port crane scene, do not need to make any modification to crane equipment and install any equipment, there is no special requirement to crane structure, size etc., with good universality, the characteristics such as low introduction cost.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology for unmanned vehicles in ports, specifically relating to a method for precise parking of unmanned vehicles in port crane scenarios. Background Technology

[0002] A port is a transportation hub located along the coast of the sea, river, lake, or reservoir, equipped with water-land intermodal transport facilities and conditions to allow ships to safely enter, exit, and berth. Furthermore, ports are the gathering points and hubs of water and land transportation, distribution centers for industrial and agricultural products and foreign trade import and export goods, and places where ships berth, load and unload cargo, embark and disembark passengers, and replenish supplies.

[0003] With the rapid development of smart port construction, port automation and intelligence are continuously upgrading. In particular, the requirements for precise positioning of unmanned vehicles in automated terminals are becoming increasingly stringent. Unmanned vehicles need to adjust their parking positions in real time according to different types of operations to reduce the positioning time of quay crane spreaders and improve work efficiency. However, in the current technology, the positioning of container transport vehicles in port crane scenarios mostly relies on the spreader driver and vehicle driver communicating through communication equipment to adjust the vehicle position, or on some degree of modification to the spreader yard to achieve the requirement of precise positioning. The former increases the time and labor costs of container transportation, while the latter requires regular maintenance of the equipment at the yard, which increases certain human and material costs.

[0004] Aside from simple manual communication adjustments, existing technologies typically address the positioning and orientation of unmanned trucks during container loading and unloading operations by deploying UWB equipment on the port quay crane surface, quay cranes, and unmanned trucks. However, UWB is currently expensive and has high operating costs. Another approach involves scanning reflectors installed at the bottom of the quay crane. Once the unmanned truck enters the effective positioning area, the scanning module is activated to scan and identify the reflector's position information, calculating the required movement position of the unmanned truck in real time for active alignment. However, this requires modification of existing cranes, resulting in high introduction costs. Summary of the Invention

[0005] To address the aforementioned problems, the main objective of this invention is to design a method for precise parking of unmanned vehicles in port crane scenarios, thereby improving the intelligence level of the vehicles and solving the problem of rapid and precise parking of unmanned flatbed trucks in port crane scenarios.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for precise parking of unmanned vehicles in port crane scenarios includes:

[0008] Step 1: Template map creation. Collect point cloud data for the panoramic view of the crane and create a template map based on the point cloud data.

[0009] Step 2: Align the origin of the target point on the template map. The vehicle is parked at the desired precise parking position under the crane. The point cloud data of the vehicle's location is collected and aligned with the point cloud of the template map.

[0010] Step 3: Precise parking based on template map. When the unmanned vehicle approaches the crane, it receives a request for precise parking signal from the task management module, triggering the precise parking task. The laser scans the crane to obtain point cloud data, and matches this data with the template map to obtain the distance of the unmanned vehicle relative to the target origin and the required direction of movement in real time. Then, the position information of the unmanned vehicle is adjusted to achieve precise parking of the unmanned vehicle in the crane scenario.

[0011] The method for creating a template map includes the following steps:

[0012] S101: Scan the panoramic view of the crane to obtain point cloud data of the crane, and perform post-processing on the point cloud data to filter out noise points;

[0013] S102: Extract different planar structures in the crane using point cloud data;

[0014] S103: Perform plane fitting on discrete point clouds belonging to the same plane to obtain the plane parameters of the plane;

[0015] The plane parameters include the centroid of the plane and the normal vector of the plane;

[0016] S104: Set the intensity value of all points on the plane to the plane number id;

[0017] S105: Save the centroid of the plane, the normal vector of the plane, and the plane's index id to obtain the template map.

[0018] As a further description of the present invention, in step two, the alignment of the template map target origin includes the acquisition of the reference point cloud and the registration of the template point cloud.

[0019] As a further description of the present invention, the reference point cloud acquisition includes: bringing the vehicle to the desired precise parking position under the crane and actually completing a loading and unloading operation, and acquiring crane point cloud data, i.e., reference point cloud, while both the crane and the vehicle remain stationary.

[0020] Template point cloud registration includes: using point cloud algorithms or manual operation to register the template map and the reference point cloud, transforming the origin of the template map's coordinate system to the actual desired precise parking position.

[0021] As a further description of the present invention, in step three, the fine stopping based on the template map includes a matching algorithm based on the fine stopping module; the matching algorithm is to find the plane corresponding to the real-time point cloud on the template map based on the kdtree, and to achieve the matching of the real-time point cloud and the template map by constructing the constraint of the minimum distance from the point to the plane.

[0022] As a further description of the present invention, the specific steps of the matching algorithm include:

[0023] S301: For real-time acquired crane point cloud data For each point in the map, a kd-tree is used to find its three nearest neighbors in the template map, and the strength values ​​of the three points are compared to determine if they are consistent.

[0024] S302: If the intensity values ​​of the three points obtained in S301 are consistent, then the three points are determined to be on the same plane. By looking up the intensity values ​​in the table, the centroid and normal vector of the corresponding plane are obtained.

[0025] S303: Using the distance from the point to the plane, construct the residual function, and solve it through CERES optimization to minimize the distance from the residual function point to the plane, thereby obtaining the rotation transformation matrix between the real-time point cloud and the template map;

[0026] The first two elements of the last column of the rotation transformation matrix represent the vehicle's position (x, y) relative to the origin of the template map target, and the yaw angle corresponding to the rotation matrix represents the vehicle's heading angle (yaw) relative to the origin of the template map target.

[0027] S304: Using the vehicle's position relative to the origin of the template map and the yaw angle information of the vehicle relative to the origin of the template map obtained from S303, adjust the distance and direction of the unmanned vehicle.

[0028] An interactive method for precise parking, comprising a precise parking module, a task management module, and a planning and control module, wherein the precise parking module includes a method for implementing precise parking;

[0029] The interaction process is as follows:

[0030] 1) The task management module sends a request for a precise stop signal and information to the precise stop module;

[0031] 2) The precision stopping module outputs the current distance and direction of the vehicle from the target origin to the task management module;

[0032] 3) The task management module sends the distance and direction from the target origin to the planning and control module to control the vehicle's movement;

[0033] 4) Once the vehicle has moved to the target origin, it sends a notification to the task management module indicating that the movement is complete.

[0034] 5) The task management module sends a fine-stop end signal to the fine-stop module, and the fine-stop module stops working.

[0035] As a further description of the present invention, the information sent from the task management module to the precision stop module includes: whether to activate the precision stop signal, the target crane number, the lane number of the current vehicle, and the size of the container.

[0036] As a further description of the present invention, after receiving the information from the task management module, the precision stopping module performs x-direction and y-direction compensation on the template map;

[0037] The compensation amount in the x-direction is the length difference between different box gauges, and the compensation amount in the y-direction is the difference between the current lane number and the lane number on the template map multiplied by the lane width.

[0038] Compared with the prior art, the technical advantages of the present invention are as follows:

[0039] This invention provides a method for precise parking of unmanned vehicles in port crane scenarios. By setting the desired precise parking position, the distance and direction of the vehicle are adjusted to guide it to precise parking. In addition, by storing the desired precise parking position and the planar parameters of the crane in the template map, the accuracy of vehicle stopping is improved, which can improve the efficiency of loading and unloading operations and increase the economic benefits per vehicle per unit time. At the same time, this invention can directly scan and obtain the point cloud data of the crane through LiDAR to achieve precise parking in port crane scenarios. No modification or addition of any equipment is required to the crane equipment, and there are no special requirements for the crane structure, size, etc. It has the characteristics of good universality and low introduction cost. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the overall process of the precise parking method of the present invention;

[0041] Figure 2 This is a schematic diagram of the workflow of the precision stop module in this invention;

[0042] Figure 3 This is a schematic diagram of the precise parking interaction method of the present invention;

[0043] Figure 4 This is a schematic diagram illustrating the acquisition of the centroid and normal vector of the crane template map plane according to the present invention;

[0044] Figure 5 This is a schematic diagram of the reference point cloud of the present invention;

[0045] Figure 6 This is a schematic diagram illustrating the registration effect between the template map and the reference point cloud according to the present invention;

[0046] Figure 7 Comparison of different template maps of this invention;

[0047] Figure 8 This is a schematic diagram comparing the distance curves of precise parking and manual parking according to the present invention. Detailed Implementation

[0048] The present invention will now be described in detail with reference to the accompanying drawings:

[0049] A method for precise parking of unmanned vehicles in port crane scenarios, referenced Figure 1 As shown, it includes:

[0050] Step 1: Template map creation. Collect point cloud data for the panoramic view of the crane and create a template map based on the point cloud data.

[0051] Step 2: Align the origin of the target point on the template map. The vehicle is parked at the desired precise parking position under the crane. The point cloud data of the vehicle's location is collected and aligned with the point cloud of the template map.

[0052] Step 3: Precise parking based on template map. When the unmanned vehicle approaches the crane, it receives a request for precise parking signal from the task management module, triggering the precise parking task. The laser scans the crane to obtain point cloud data, and matches this data with the template map to obtain the distance of the unmanned vehicle relative to the target origin and the required direction of movement in real time. Then, the position information of the unmanned vehicle is adjusted to achieve precise parking of the unmanned vehicle in the crane scenario.

[0053] It should be noted that since there are many types of cranes suitable for handling containers in ports, such as quay cranes and rubber-tired gantry cranes, this embodiment uses a quay crane as an example for detailed explanation. However, this method is also applicable to similar scenarios such as rubber-tired gantry cranes and locking twisting machines. The specific implementation method is disclosed as follows:

[0054] Specifically, step one, the method for creating a template map, includes the following steps:

[0055] S101: The panoramic view of the quay crane is scanned by LiDAR to obtain the point cloud data of the quay crane. The point cloud data is then post-processed to filter out noise points, which are other point clouds detected by the LiDAR other than the quay crane structure.

[0056] S102: Extract different planar structures in the crane using point cloud data, that is, divide the various planes of the quay crane according to the drawings of the quay crane;

[0057] S103: By performing plane fitting on discrete point clouds belonging to the same plane, the plane parameters of that plane can be obtained, such as... Figure 4 As shown, RANSAC plane fitting is usually used for fitting, but it is not limited to this fitting method. Other methods that can be used for plane fitting of discrete point clouds are also included.

[0058] Among them, the plane parameters include the centroid of the plane. and the normal vector of the plane ;

[0059] S104: Set the intensity value of all points on the plane to the plane number id;

[0060] S105: Save the centroid of the plane, the normal vector of the plane, and the plane index id to obtain the template map. .

[0061] In step two, the origin of the template map is aligned, including the acquisition of the baseline point cloud and the registration of the template point cloud;

[0062] a. Baseline point cloud acquisition includes: manually adjusting the vehicle to the desired precise stopping position under the quay crane (the target position for loading and unloading containers), and actually completing one container loading or unloading operation. Then, with both the quay crane and the vehicle stationary, the laser point cloud data of the quay crane is acquired using LiDAR, i.e., the baseline point cloud. Figure 5 As shown;

[0063] b. Template point cloud registration, including: using point cloud algorithms or manual operation, registering the template map and the reference point cloud, transforming the origin of the template map's coordinate system to the actual desired precise parking position, such as... Figure 6 As shown.

[0064] Point cloud registration is not limited to a certain algorithm, including any publicly available algorithm that can achieve point cloud registration, such as common point cloud algorithms such as ICP and NDT.

[0065] It should be noted that for the same model of quay crane, the relative position of the vehicle and the quay crane is fixed during the loading and unloading of containers by the unmanned vehicle. This invention records the precise parking position in the template map. When the vehicle needs to park precisely under the bridge, the point cloud of the quay crane is perceived in real time by LiDAR and matched with the template map to obtain the relative relationship between the current vehicle position and the expected precise parking position, thereby knowing the adjustment direction and adjustment distance of the vehicle.

[0066] Step three involves precise stopping based on the template map, including a matching algorithm based on the precise stopping module.

[0067] The matching algorithm is used to find the plane corresponding to the real-time point cloud on the template map based on the kdtree. By constructing the constraint of the minimum distance from the point to the plane, the matching between the real-time point cloud and the template map is achieved.

[0068] The specific algorithm flow of the matching algorithm includes:

[0069] S301: For point cloud data acquired in real time by the quay crane For each point in the map, a kd-tree is used to find its three nearest neighbors in the template map, and the strength values ​​of the three points are compared to determine if they are consistent.

[0070] When creating the template map, the intensity value of different planar points is assigned to the corresponding planar index ID. If the planar index IDs are the same, the intensity values ​​will be the same.

[0071] S302: If the intensity values ​​of the three points obtained in S301 are consistent, then the three points are determined to be on the same plane. The centroid and normal vector of the corresponding plane can be obtained by looking up the intensity values ​​in a table.

[0072] When the template map is created, a corresponding table is obtained. The table includes the centroid and direction vector of the plane corresponding to all plane number IDs. Therefore, after obtaining the plane number ID, the corresponding plane centroid and plane normal vector can be obtained.

[0073] S303: Using the formula for the distance from a point to the plane, a residual function is constructed. The distance from the point in the residual function to the plane is minimized through Ceres optimization, thus obtaining the rotation transformation matrix between the real-time point cloud and the template map. Ceres optimization is a commonly used optimization method in existing technologies.

[0074] The first two elements of the last column of the rotation transformation matrix represent the vehicle's position (x, y) relative to the origin of the template map, and the yaw angle corresponding to the rotation matrix represents the vehicle's heading angle (yaw) relative to the origin of the template map.

[0075] S304: Using the vehicle's position relative to the target origin of the template map and the yaw angle information of the vehicle relative to the target origin of the template map obtained by S303, the distance and direction of the unmanned vehicle are adjusted to achieve precise parking of the unmanned vehicle in the quay crane scenario.

[0076] In another embodiment of the present invention, an interactive method for precise parking is provided, the method comprising a precise parking module, a task management module, and a planning control module, wherein the precise parking module includes a method for implementing precise parking;

[0077] The interaction process is as follows Figure 2-3 As shown, specifically:

[0078] 1) The task management module sends a request for precise stop signal to the precise stop module, along with the target quay crane number, the lane number of the current vehicle, and the size of the container.

[0079] 2) The precision stopping module outputs the current distance and direction of the vehicle from the target origin to the task management module;

[0080] 3) The task management module sends the distance and direction from the target origin to the planning and control module, thereby adjusting the movement of the vehicle;

[0081] 4) After the vehicle moves to the target origin, the planning and control module reports the completion of the movement to the task management module;

[0082] 5) The task management module sends a fine-stop end signal to the fine-stop module, and the fine-stop module stops working.

[0083] It should be noted that during the precise parking interaction process, a separate template map is built for different models of quay cranes, and the map is automatically switched to complete precise parking for different models of quay cranes. The task management module receives the quay crane number, then looks up the storage path of the corresponding template map from a table and loads it. Different model template maps are as follows: Figure 7 As shown. Furthermore, because the lane and container location of the vehicle differ between the template map and actual operation, the actual map may show discrepancies. For example, the vehicle might be in lane 3 and parked at the 40-foot container loading / unloading position when the template map is built, while in actual operation, the vehicle might be parked in lane 5 at the 20-foot container front position. Therefore, after the precise parking module receives the lane number and container location information from the task management module, it needs to compensate the template map in the x and y directions. The x-direction compensation is the difference in container length between the different container sizes, and the y-direction compensation is the difference between the current lane number and the template map lane number multiplied by the lane width.

[0084] In addition, in the interactive process of precise parking, such as Figure 3 As shown, when the unmanned vehicle approaches the crane, the fine-stop module receives a fine-stop request signal from the task management module and triggers a fine-stop task. At this time, the fine-stop module starts working, using laser scanning to obtain point cloud data from the crane and matching this data with the template map to obtain the distance of the unmanned vehicle relative to the target origin and the required movement direction (forward or backward) in real time. The distance and direction are then sent back to the task management module, which makes a judgment and sends it to the planning and control module. After the planning and control module completes the movement task, it sends a movement completion signal back to the task management module. When the task management judgment module determines that the fine-stop sending distance is less than the threshold, it sends a fine-stop end signal to the laser fine-stop module, and the fine-stop module stops working.

[0085] Among them, the precision stopping module, such as Figure 2As shown, the precision stopping module first needs to acquire the real-time point cloud data of the crane and the corresponding template map of the crane. For each point in the real-time point cloud, it uses a kdtree to find the three nearest neighbors in the template map, determining whether the three points are on the same plane. This is because when the template map is created, the intensity value of points on different planes is assigned the corresponding plane index ID. If the plane index IDs are the same, then the intensity values ​​are the same. Therefore, the essence is to determine whether the intensity values ​​of the three points are consistent. If they are consistent, the centroid and normal vector of the corresponding plane are obtained by looking up the intensity value in a table. Then, using the distance formula from a point to the plane, a residual function is constructed. This residual function is optimized using Ceres to minimize the distance from the point to the plane, thus obtaining the rotation transformation matrix between the real-time point cloud and the template map. The first two elements of the last column of the rotation transformation matrix represent the vehicle's position (x, y) relative to the origin of the template map, and the yaw angle corresponding to the rotation matrix is ​​the vehicle's heading angle (yaw) relative to the origin of the template map. The precision stopping module then completes its work.

[0086] Based on the technical solution disclosed in this invention, a comparison is made with the curves of manual parking distance in existing technologies. Figure 8 As shown, manually adjusting the distance between the vehicle and the spreader under the quay crane requires the driver to visually judge the distance between the vehicle and the spreader after each adjustment. However, the automatic and precise parking method disclosed in this invention can realize the autonomous, real-time, and continuous adjustment of the vehicle and stop it in one go. Through the curve comparison chart, it can be found that manual adjustment will inevitably exceed the target distance by a certain amount.

[0087] Compared with the prior art, the present invention has four advantages, which are disclosed in detail below:

[0088] 1. The unmanned vehicle precise parking method disclosed in this invention can solve the problem of rapid and precise parking of unmanned vehicles in port crane scenarios by improving the intelligence level of the vehicle, thereby improving the efficiency of loading and unloading operations and increasing the economic benefits of a single vehicle per unit time.

[0089] 2. The unmanned vehicle precision parking method disclosed in this invention can directly scan and obtain the point cloud data of the crane through lidar, realizing precise parking in port crane scenarios. It does not require any modification or addition of any equipment to the crane, and has no special requirements for the crane structure, size, etc. It has the characteristics of good universality and low introduction cost.

[0090] 3. The unmanned vehicle precise parking method disclosed in this invention constructs different template maps for different cranes. The precise parking module can obtain the corresponding template map for the crane by looking up the crane number information issued by the task management module, and compensate for the offset of the template map in the x and y directions by the current container size signal and lane line signal issued by the task management module, so as to meet the parking needs of all lanes in the crane scenario.

[0091] 4. The unmanned vehicle precise parking method disclosed in this invention constructs a template map, extracts all planes of the crane, and saves the plane ID, plane centroid, and plane normal vector of each plane to guide the vehicle to park precisely, improving the accuracy of vehicle stopping within 3cm, which is a significant improvement compared to the positioning accuracy of the prior art which is within 1m.

[0092] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solutions of the present invention, as long as they do not depart from the spirit and scope of the technical solutions of the present invention, should be covered within the scope of the claims of the present invention.

Claims

1. A method for precise parking of unmanned vehicles in a port crane scenario, characterized in that, include: Step 1: Template map creation. Collect point cloud data for the panoramic view of the crane and create a template map based on the point cloud data. The method for creating the template map includes the following steps: S101: Scan the panoramic view of the crane to obtain point cloud data of the crane, and perform post-processing on the point cloud data to filter out noise points; S102: Extract different planar structures in the crane using point cloud data; S103: Perform plane fitting on discrete point clouds belonging to the same plane to obtain the plane parameters of the plane; The plane parameters include the centroid of the plane and the normal vector of the plane; S104: Set the intensity value of all points on the plane to the plane number id; S105: Save the centroid of the plane, the normal vector of the plane, and the plane's index id to obtain the template map; Step 2: Align the origin of the target point on the template map. The vehicle is parked at the desired precise parking position under the crane. The point cloud data of the vehicle's location is collected and aligned with the point cloud of the template map. Step 3: Precise parking based on template map. When the unmanned vehicle approaches the crane, it receives a request for precise parking signal from the task management module, triggering the precise parking task. The laser scans the crane to obtain point cloud data, and matches this data with the template map to obtain the distance of the unmanned vehicle relative to the target origin and the required direction of movement in real time. Then, the position information of the unmanned vehicle is adjusted to achieve precise parking of the unmanned vehicle in the crane scenario.

2. The method for precise parking of unmanned vehicles in a port crane scenario according to claim 1, characterized in that: In step two, the alignment of the template map target origin includes the acquisition of the reference point cloud and the registration of the template point cloud.

3. The method for precise parking of unmanned vehicles in a port crane scenario according to claim 2, characterized in that: Baseline point cloud acquisition includes: The vehicle is brought to the desired precise stopping position under the crane, and a loading and unloading operation is actually completed. While both the crane and the vehicle are stationary, the crane point cloud data, i.e. the reference point cloud, is collected. Template point cloud registration, including: The template map and the reference point cloud are registered using point cloud algorithms or manual operation to transform the origin of the coordinate system of the template map to the actual desired precise parking position.

4. The method for precise parking of unmanned vehicles in a port crane scenario according to claim 1, characterized in that: Step three involves precise stopping based on the template map, including a matching algorithm based on the precise stopping module. The matching algorithm is used to find the plane corresponding to the real-time point cloud on the template map based on the kdtree. By constructing the constraint of the minimum distance from the point to the plane, the matching between the real-time point cloud and the template map is achieved.

5. The method for precise parking of unmanned vehicles in a port crane scenario according to claim 4, characterized in that: The specific steps of the matching algorithm include: S301: For each point in the real-time acquired crane point cloud data, use kdtree to find its three nearest neighbors in the template map and determine whether the intensity values ​​of the three points are consistent. S302: If the intensity values ​​of the three points obtained in S301 are consistent, then the three points are determined to be on the same plane. By looking up the intensity values ​​in the table, the centroid and normal vector of the corresponding plane are obtained. S303: Using the distance from the point to the plane, construct the residual function, and solve it through CERES optimization to minimize the distance from the residual function point to the plane, thereby obtaining the rotation transformation matrix between the real-time point cloud and the template map; The first two elements of the last column of the rotation transformation matrix represent the vehicle's position (x, y) relative to the origin of the template map target, and the yaw angle corresponding to the rotation matrix represents the vehicle's heading angle (yaw) relative to the origin of the template map target. S304: Using the vehicle's position relative to the origin of the template map and the yaw angle information of the vehicle relative to the origin of the template map obtained from S303, adjust the distance and direction of the unmanned vehicle.

6. A precise parking interaction method, comprising the precise parking method for unmanned vehicles in a port crane scenario as described in any one of claims 1-5, characterized in that: The method includes a precision parking module, a task management module, and a planning and control module, wherein the precision parking module includes a method for achieving precise parking; The interaction process is as follows: 1) The task management module sends a request for a precise stop signal and information to the precise stop module; 2) The precision stopping module outputs the current distance and direction of the vehicle from the target origin to the task management module; 3) The task management module sends the distance and direction from the target origin to the planning and control module to control the vehicle's movement; 4) Once the vehicle has moved to the target origin, it sends a notification to the task management module indicating that the movement is complete. 5) The task management module sends a fine-stop end signal to the fine-stop module, and the fine-stop module stops working.

7. The interactive method for precise parking according to claim 6, characterized in that: The information sent from the task management module to the precision stop module includes: whether to activate the precision stop signal, the target crane number, the lane number of the current vehicle, and the size of the container.

8. The interactive method for precise parking according to claim 7, characterized in that: After receiving the information from the task management module, the precision stopping module performs x- and y-direction compensation on the template map; The compensation amount in the x-direction is the length difference between different box gauges, and the compensation amount in the y-direction is the difference between the current lane number and the lane number on the template map multiplied by the lane width.

Images