FTR lock identification method and system

Abstract

The invention relates to the technical field of target recognition, and provides an FTR lock recognition method and system, and the method comprises the steps: obtaining point cloud data, and enabling the point cloud data to be obtained through scanning a flat car; and identifying a suspected lock target from the point cloud data, and then performing trajectory tracking on the suspected lock target so as to confirm three-dimensional position information of the FTR lock in a preset form from the suspected lock target. According to the method, the suspected lock target is recognized through the point cloud data, trajectory tracking is performed, the three-dimensional position of the FTR lock is automatically confirmed, manual inspection is replaced, and the recognition automation degree and accuracy are improved.

Description

Technical Field

[0001] This application relates to the field of target recognition technology, and in particular to an FTR lock recognition method and system. Background Technology

[0002] FTR (Flip-Top Rail) locks are specialized mechanical devices used to secure containers on flatcars of freight trains. In automated loading and unloading systems for railway containers, track-type gantry cranes transfer containers to flatcars, requiring the corner fitting holes to align with the lock heads as they descend to insert the lock heads and ensure transportation safety. This system requires high-precision alignment to meet the needs of automated loading and unloading.

[0003] Container loading is done manually. Ground personnel check and adjust the FTR lock status, the driver operates the gantry crane to move the container, and ground personnel use iron hooks to assist the corner fitting holes in inserting the lock heads. The entire process relies on on-site personnel cooperation to complete the alignment and fixation.

[0004] This method requires multiple people to work together, and ground personnel face safety risks. It requires a large amount of manpower, resulting in low work efficiency and insufficient automation. Summary of the Invention

[0005] This application provides an FTR lock identification method and system to solve the problems of low operation efficiency and insufficient automation.

[0006] In a first aspect, this application provides an FTR lock identification method, including: Point cloud data is acquired by scanning the flatbed vehicle. Identify suspected lock targets from the point cloud data; The trajectory of the suspected lock target is tracked to confirm the three-dimensional position information of the FTR lock of the preset shape from the suspected lock target.

[0007] By identifying suspected lock targets using point cloud data and tracking their trajectories, the three-dimensional position of the FTR lock can be automatically confirmed, replacing manual inspection and improving the automation and accuracy of identification.

[0008] In some feasible embodiments, acquiring point cloud data includes: The flatbed truck is vertically scanned using a single-line lidar to obtain scan data; The encoder is used to obtain the real-time position information of the gantry crane; The point cloud data is generated based on the scan data and the real-time location information.

[0009] By combining single-line lidar scanning data and encoder position information, point cloud data is generated, providing a complete and reliable data foundation for FTR lock recognition.

[0010] In some feasible embodiments, identifying suspected lock targets from the point cloud data includes: Rotate the orbital vertices in the point cloud data to a preset position, the preset position being parallel to a preset coordinate axis; After the track vertex reaches a preset position, a spatial filtering region is set, which is defined based on the track center position and height range; Identify suspected lock targets from point cloud data located within the spatial filtering area.

[0011] By rotating the point cloud data and setting spatial filtering areas, the point cloud data structure is optimized, thereby improving the processing efficiency and accuracy of subsequent suspected target identification.

[0012] In some feasible embodiments, identifying suspected lock targets from the point cloud data includes: Calculate the position of the center point of the point cloud data in the width direction; Using the center point as a reference, find the local highest point at symmetrical positions on both sides perpendicular to the direction of track extension; Calculate the height difference between the local highest point and the lock base; If the height difference is within a preset range, the highest local point within the preset range is identified as a suspected target.

[0013] By finding the highest local point based on the center point symmetry and verifying the height difference, the suspected target can be accurately screened from the point cloud data.

[0014] In some feasible embodiments, the step of trajectory tracking of the suspected lock target to confirm the three-dimensional position information of the FTR lock of a preset shape from the suspected lock target includes: The suspected target is correlated and matched with historical tracking trajectories to output the correlation and matching results; Update the status information of the historical tracking trajectory based on the association matching results; Based on the updated status information, the three-dimensional position information of the FTR lock with the preset shape is confirmed from the suspected lock targets.

[0015] By using correlation matching and status updates, trajectory tracking is performed on suspected target locks, effectively distinguishing between real target locks and transient interference targets.

[0016] In some feasible embodiments, the step of associating and matching the suspected target with historical tracking trajectories to output association and matching results includes: Calculate the spatial distance between the suspected target and the midpoint of the historical tracking trajectory; Historical tracking trajectories with a spatial distance less than a preset distance threshold are identified as matching trajectories; The suspected target is added to the point sequence of the matching trajectory to output the associated matching result.

[0017] By calculating spatial distance and matching trajectory updates, the system can accurately correlate suspected target locations with historical trajectories, ensuring the continuity of trajectory tracking.

[0018] In some feasible embodiments, the step of confirming the three-dimensional position information of an FTR lock of a preset shape from suspected lock targets based on the updated state information includes: Obtain the number of consecutive non-matches of the historical tracking trajectory; When the number of consecutive unmatched occurrences reaches a preset threshold, an confirmation process is triggered. After the confirmation process, the number of points in the historical tracking trajectory and the spatial distribution concentration of the point sequence are obtained. When the number of points reaches a preset threshold and the spatial distribution concentration meets a preset requirement, the centroid of the point sequence is calculated, and the centroid is determined as the three-dimensional position information of the FTR lock of the preset shape.

[0019] The three-dimensional position information is confirmed based on the number of consecutive unmatches and the spatial distribution of the point sequence, ensuring that the output FTR lock position has sufficient stability and reliability.

[0020] In some feasible embodiments, after confirming the three-dimensional position information of the FTR lock in the preset configuration, the process includes: Acquire historical point cloud data; In the historical point cloud data, the number of target points within a preset range around the three-dimensional location information is obtained, where the target points are the number of points whose height coordinates are greater than the height coordinates of the three-dimensional location information; When the number of target points exceeds a preset threshold, the three-dimensional location information is marked as misidentified information; Delete the misidentified information in the three-dimensional location information.

[0021] By examining the number of target points in historical point cloud data, misidentified 3D position information can be effectively identified and deleted, thereby improving the accuracy of the output results.

[0022] In some feasible embodiments, the method further includes: Obtain the spacing of the corner fitting holes in the container; Based on the aforementioned spacing, the three-dimensional position information of the confirmed preset FTR lock is verified by spacing verification, and the verification result is output. If the verification result does not match the spacing, the three-dimensional position information is deleted to obtain updated three-dimensional position information; Based on the updated three-dimensional position information, the gantry crane is controlled to move the container to the target position and drop it, where the target position is the position corresponding to the updated three-dimensional position information.

[0023] The three-dimensional position information is verified and filtered based on the spacing of the corner fitting holes of the container to ensure that the final output position information conforms to the physical constraints of the actual loading operation.

[0024] Secondly, this application provides an FTR lock identification system for performing the FTR lock identification method described in the first aspect, comprising: A point cloud acquisition unit is used to acquire point cloud data, which is acquired by scanning the flatbed vehicle; The identification unit is used to identify suspected lock targets from the point cloud data; and to perform trajectory tracking on the suspected lock targets to confirm the three-dimensional position information of the FTR lock of the preset shape from the suspected lock targets.

[0025] As can be seen from the above technical solutions, this application provides an FTR lock identification method and system. The method includes: acquiring point cloud data, which is obtained by scanning a flatbed truck; identifying suspected lock targets from the point cloud data; and then performing trajectory tracking on the suspected lock targets to confirm the three-dimensional position information of an FTR lock of a preset shape from the suspected lock targets. This method automatically confirms the three-dimensional position of the FTR lock by identifying suspected lock targets and performing trajectory tracking using point cloud data, replacing manual inspection and improving the automation and accuracy of identification. Attached Figure Description

[0026] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A flowchart illustrating the FTR lock recognition method provided in this application embodiment; Figure 2 A schematic diagram of a preset coordinate system provided for embodiments of this application; Figure 3 This is a schematic diagram of the point cloud data before rotation provided in an embodiment of this application; Figure 4 This is a schematic diagram of the rotated point cloud data provided in an embodiment of this application; Figure 5 This is a schematic diagram of the filtering area provided in an embodiment of this application; Figure 6 A flowchart illustrating the matching process provided in an embodiment of this application; Figure 7This is a schematic diagram of the process for confirming three-dimensional position information provided in an embodiment of this application. Detailed Implementation

[0028] The embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following examples do not represent all embodiments consistent with this application.

[0029] In railway container loading and unloading operations, accurately identifying FTR locks on flatcars is a prerequisite for automated loading. An FTR lock is a small mechanical device with a bolt height of approximately 100mm and a diameter of less than 80mm.

[0030] In the fields of industrial automation and object recognition, this also includes solutions for the stable detection of specific small targets from complex backgrounds. For example, in the field of autonomous driving, vehicles use LiDAR and visual sensors to identify small obstacles on the road, such as scattered tires, bricks, or other debris, with a size comparable to an FTR lock. The processing typically involves segmenting potential targets from point clouds or images and then identifying them using a classifier.

[0031] Unlike FTR lock recognition, the autonomous driving environment is more open and the target type is uncertain, while the railway flatcar environment has a relatively fixed background and a clear target type. The surface of the flatcar is not an ideal smooth plane, but has welds, rivets, reinforcing ribs and pipeline interfaces of various sizes. These structures will form a large number of geometric features in the point cloud data that are highly similar to small-sized FTR locks. If conventional point cloud segmentation methods are used, these structural features will generate a large number of false targets, which will seriously interfere with the recognition process.

[0032] Some solutions typically employ general strategies when dealing with similar problems. For example, in the field of autonomous driving, when identifying small obstacles, all potential obstacles are marked due to the uncertainty of the target type, and these areas are then avoided through subsequent path planning. This strategy is not feasible in the railway loading scenario because the gantry crane must know the exact position of each real FTR lock in order to complete the alignment. The template matching method commonly used in industrial vision inspection also faces difficulties in this scenario. Although the FTR locks are of standard size, their specific performance in the point cloud will vary significantly with changes in scanning angle and distance.

[0033] This application provides an FTR lock identification method in some embodiments. The method utilizes the prior knowledge of the FTR lock system, namely that the locks are symmetrically distributed on the flatbed at standard intervals. This feature does not exist in other open environments. By calculating the center point in the width direction of the point cloud and finding the local highest point at specific positions on both sides, this regular layout information can be effectively utilized to directly eliminate interference caused by other irregular protrusions on the surface of the flatbed.

[0034] During the uniform speed travel of the gantry crane, the real FTR lock will form a stable motion trajectory in the continuous point cloud frame, while false targets generated by fixed structures such as weld seams will appear randomly. By analyzing the continuity of the target in the time dimension, the real lock can be reliably distinguished from fixed interference. This spatiotemporal analysis process is not needed in static industrial inspection and cannot be effectively implemented in autonomous driving where the target appears randomly.

[0035] Setting up a spatial filtering region based on the track center and height directly utilizes the semi-structured characteristics of the railway environment. This process limits the search range to the space where FTR locks may appear, which not only improves processing efficiency but also avoids interference from similar structures in other areas. The operation of rotating the point cloud to make the track vertices parallel to the coordinate axes makes full use of the deterministic features of track geometry.

[0036] Regarding security requirements, this application ensures the reliability of the output results through multiple verification mechanisms. False identification filtering checks whether there are target points around the lock. This addresses the physical constraint that the corner fitting holes of the container must be unobstructed for insertion. Verification based on standard spacing ensures that the identified lock position conforms to the actual size requirements of the container.

[0037] like Figure 1 As shown, the method includes the following steps S110-S130.

[0038] S110: Acquire point cloud data, which is obtained by scanning the flatbed vehicle.

[0039] Point cloud data is a collection of data obtained by scanning the physical world in three dimensions. Point cloud data consists of a large number of points, each of which includes its three-dimensional coordinate information in space. Point cloud data can characterize the geometric shape and spatial position of the surface of the scanned object.

[0040] In this embodiment, point cloud data is used to characterize the surface three-dimensional morphology of the freight flatbed truck and its FTR lock. The device for acquiring point cloud data can be a lidar, which measures the distance and angle of a point by emitting a laser beam and receiving the reflected signal, thereby calculating the three-dimensional coordinates of the point.

[0041] In some embodiments, acquiring point cloud data includes: vertically scanning the flatbed truck using a single-line lidar to obtain scan data; acquiring the real-time position information of the gantry crane using an encoder; and generating the point cloud data based on the scan data and the real-time position information.

[0042] A single-line lidar, acting as the scanning actuator, is positioned at a specific location on the gantry crane. It only needs to ensure that the scanning plane remains perpendicular to the crane's travel direction. During operation, the lidar continuously emits a laser beam. This beam is reflected upon contact with objects such as the flatbed truck surface, FTR locks, and rails. The lidar measures distance by receiving these reflected signals. Each scan generates a scan line composed of several points. These points record the contour information of the scanned object in a two-dimensional plane at the instant of the scan. The scan data itself only includes the lateral offset and vertical height of the points, lacking positional coordinates along the rail extension direction.

[0043] To supplement the missing dimensions in the scanned data, the real-time position information of the gantry crane is synchronously acquired via an encoder mounted on the gantry crane's traveling mechanism. As the gantry crane travels along the track, the encoder calculates the linear displacement of the gantry crane from its starting position by detecting the rotation of the traveling motor or the rotation of the wheels. This displacement value is recorded as the real-time position information, defining the specific coordinates of the current scan line in the track's extension direction. The real-time position information and the scanned data are synchronized in time; each frame of scanned data corresponds to a specific real-time position.

[0044] After receiving the scanning data from the single-line lidar, the system reads the real-time position information collected synchronously and assigns this real-time position information as its longitudinal coordinate to each point in the scanning data. After the coordinate assignment, the original two-dimensional scanning line with only horizontal and height information is converted into point cloud data with complete three-dimensional coordinates. These three-dimensional points constitute a data set that can characterize the three-dimensional shape of the flatbed vehicle surface.

[0045] like Figure 2 As shown, in this embodiment, the X-axis of the coordinate system is perpendicular to the extension direction of the steel rail, that is, the travel direction of the trolley in the gantry crane; the Y-axis is parallel to the extension direction of the steel rail, that is, the travel direction of the main trolley in the gantry crane; and the Z-axis is perpendicular to the steel rail, pointing to the sky, with the rail surface at point 0.

[0046] S120: Identify suspected lock targets from the point cloud data.

[0047] Before FTR lock recognition, the point cloud of the single-line LiDAR needs to be rotated and a scanning filter area needs to be set. This step only needs to be done once during the installation and commissioning phase and will not be required before each subsequent FTR lock recognition. The requirement for rotating the single-line point cloud is that the vertices of the two tracks are parallel to the X-axis. Rotation simplifies the algorithm and improves its accuracy. The filter area is set 2 meters to the left and right of the track center, with a height ranging from 900mm to 5500mm. The filter area limits the detection range, allowing FTR lock recognition only on the point cloud within the filter area, thus improving the accuracy of FTR lock recognition.

[0048] Before identification, in some embodiments, the orbital vertices in the point cloud data are rotated to a preset position, which is parallel to a preset coordinate axis, such as the X-axis; after the orbital vertices are rotated to the preset position, a spatial filtering region is set, which is defined based on the orbital center position and height range.

[0049] Specifically, such as Figure 3 As shown, the track vertices in the point cloud data may have an angle with the desired coordinate axis direction due to factors such as scanning equipment installation deviation or slight track deformation. By calculating the spatial distribution characteristics of the point cloud at the top of the track, the actual direction of the track vertices is identified. Then, a coordinate transformation algorithm is used to rigidly rotate the entire point cloud dataset, such as... Figure 4 As shown, the rotation operation keeps the orbit vertex parallel to the preset coordinate axis direction in the system.

[0050] After completing the coordinate rotation, set the spatial filtering region, such as... Figure 5 As shown, the spatial filtering region is a clearly defined cubic region in three-dimensional space. This region is set based on the standard dimensions of the flatbed vehicle, specifically extending a fixed width to both sides from the center of the track, while also defining a specific height range in the vertical direction from the track surface. Based on these parameters, a closed cubic boundary is defined in the rotated point cloud coordinate system. The spatial filtering region is set based on the understanding that all FTR locks to be identified are located within this spatial volume, while point cloud data outside the region are considered irrelevant background or interfering objects.

[0051] After the spatial filtering region is defined, the rotated point cloud data is compared with the preset spatial filtering region. Each point is checked to see if it is inside the filtering region. All point cloud data that fall outside the spatial filtering region are removed, and only point cloud data that are completely inside the region are retained. Through this spatial filtering mechanism, irrelevant point clouds in the environment around the flatbed truck are removed, including point clouds of objects that may cause interference, such as ground debris and adjacent equipment. At the same time, the total amount of data that needs to be processed is significantly reduced.

[0052] After the above preprocessing process, the point cloud data at this time has a unified coordinate direction and only includes the spatial area where the target may appear. This allows the recognition algorithm to focus more on the feature analysis of the FTR lock, without having to process coordinate data with inconsistent directions or search for the target in a large number of irrelevant point clouds.

[0053] Suspected lock targets are sets or spatial locations of points that may be FTR locks, initially screened from point cloud data. Suspected lock targets are not the final confirmed FTR locks, but rather candidate targets that have certain FTR lock characteristics and need further verification. The process of identifying suspected lock targets is a preliminary filtering process, the purpose of which is to narrow down the detection range and improve the efficiency of subsequent processing.

[0054] Suspected lock targets possess local features of FTR locks, such as being a raised cluster of points in a specific region, or having a height that differs significantly from the surrounding point cloud.

[0055] In some embodiments, identifying a suspected lock target from the point cloud data includes: calculating the position of the center point of the point cloud data in the width direction; using the center point position as a reference, finding local highest points at symmetrical positions on both sides perpendicular to the track extension direction; calculating the height difference between the local highest point and the lock base; and if the height difference is within a preset range, determining the local highest point within the preset range as a suspected lock target.

[0056] Specifically, the spatial distribution of the filtered point cloud data in the horizontal dimension is analyzed first. By traversing the horizontal coordinate values ​​of all points, the minimum and maximum values ​​are found. The minimum and maximum values ​​in the horizontal direction are added together and then divided by two to obtain the center point position value of the point cloud data in the width direction. This center point position represents the geometric center of the flatbed truck in the horizontal direction. The calculation of the center point position utilizes the symmetrical characteristics of the flatbed truck structure in the horizontal direction.

[0057] After determining the center point location, since the FTR locks appear in pairs on the flatbed and are symmetrically distributed about the vehicle's centerline, the center point location is used as a reference to search at symmetrical locations on both sides perpendicular to the direction of track extension. The search area is a circular range with a specific radius centered on the aforementioned symmetrical location. The height coordinates of all points are compared within these circular ranges to find the point with the largest height value in each area. These points are marked as the local highest points.

[0058] The height of the lock base is determined by analyzing point cloud data of the area near the local highest point. Within a large circular area centered on the local highest point, dense clusters of points with lower heights are identified. These clusters represent the top plane of the lock base. The height of the lock base is subtracted from the height of the local highest point to obtain the specific height difference. The height difference calculation can capture the typical geometric features of the FTR lock's raised bolt.

[0059] Finally, the calculated height difference is compared with a preset range, which is a threshold range set according to the actual size of the FTR lock's bolt. If the height difference of a certain local highest point is within this preset range, the local highest point is identified as a suspected lock target. For local highest points whose height difference is not within the preset range, they are regarded as interference points that do not conform to the characteristics of FTR locks and are excluded. This verification step can ensure that the suspected lock target conforms to the physical characteristics of FTR locks in the vertical dimension.

[0060] By utilizing the center point calculation and symmetric search mechanism, the structural symmetry of the flatbed truck is effectively taken advantage of. The geometric features of the FTR lock are accurately captured through height difference verification. This allows for the rapid selection of candidate targets that conform to the spatial distribution pattern and morphological characteristics of the FTR lock from complex point cloud data.

[0061] S130: Track the trajectory of the suspected lock target to confirm the three-dimensional position information of the FTR lock of the preset shape from the suspected lock target.

[0062] Trajectory tracking is the process of associating and continuously observing the state of the same target at different times in sequence data. In FTR lock identification, trajectory tracking is performed on suspected lock targets in sequence point cloud data. Trajectory tracking determines whether they belong to the same physical entity by associating and matching the suspected lock targets identified at the current time with the tracking trajectories at historical times.

[0063] The trajectory tracking process maintains and updates the status information of each tracked target, such as location history and the number of consecutive unmatched times. By tracking the trajectory of suspected lock targets, it is possible to effectively eliminate false identifications that may occur in a single scan and confirm a stable and real FTR lock target.

[0064] FTR locks have multiple working states, such as upright and folded. The preset FTR lock is the FTR lock in a specific working state on the freight flatbed truck. During container loading operations, the FTR lock that needs to be identified is in the upright state and ready to receive the container corner fitting holes.

[0065] In this embodiment, the preset form is the upright state. The FTR lock in this form exhibits specific three-dimensional geometric features in the point cloud data, such as having a bolt portion that is significantly higher than the lock base. The target is clearly identified as the FTR lock in the preset form, making the identification process specific and excluding the detection of other irrelevant lock forms.

[0066] After identifying a suspected target, the trajectory tracking process compares the suspected target found in the current scan cycle with the tracking trajectory formed in the previous scan cycles. A real and stable FTR target will appear continuously in the continuous scan data, and its position change conforms to the movement pattern of the scanning platform. Through trajectory tracking, false targets caused by instantaneous noise, foreign object interference, or missing point clouds can be identified. These false targets may only appear in a single or a few scans and cannot form a stable and continuous trajectory.

[0067] Trajectory tracking identifies FTR locks that are determined to be true and have a preset shape from all tracked suspected lock targets and outputs their three-dimensional position information. For FTR locks that are confirmed to have a preset shape, the centroid or latest position of their trajectory is calculated as the three-dimensional position information of the FTR lock.

[0068] Three-dimensional position information is data used to describe the spatial location of a pre-defined FTR lock. It includes coordinate values ​​in at least three mutually perpendicular directions, defining the specific spatial position of the FTR lock within a coordinate system. This three-dimensional position information is used by the gantry crane control system to calculate the target location where the container needs to be moved.

[0069] For example, a gantry crane equipped with a scanning device travels at a constant speed along a track, scanning empty freight flatbeds parked on the track. The scanning device generates several frames of point cloud data per second. After receiving the first frame of point cloud data, it finds several suspected points that match the local features of an FTR lock and initializes them into several tracking trajectories. The gantry crane continues to move forward and receives subsequent second and third frames of point cloud data. It continuously matches newly identified suspected lock targets with existing trajectories. One trajectory is matched successfully for several consecutive frames, and the point sequence recorded inside it continues to grow, while another trajectory does not have any new suspected targets to match in several consecutive frames.

[0070] When the gantry crane passes the flatbed truck and the scanning ends, all tracking trajectories are confirmed. Trajectories that are continuously matched are confirmed as real, upright FTR locks because of the sufficient number of trajectory points and their concentrated spatial distribution. Their three-dimensional position information is calculated and output. Trajectories that are not continuously matched are discarded because they fail to meet the confirmation conditions, thus completing the FTR lock identification process.

[0071] Specifically, in some embodiments, trajectory tracking is performed on the suspected lock target to confirm the three-dimensional position information of the FTR lock of a preset shape from the suspected lock target, including: associating and matching the suspected lock target with historical tracking trajectories to output association and matching results; updating the status information of the historical tracking trajectories according to the association and matching results; and confirming the three-dimensional position information of the FTR lock of a preset shape from the suspected lock target based on the updated status information.

[0072] During the association matching process, suspected target locks are processed, and a set of historical tracking trajectories established in previous scanning cycles is also included. For each suspected target lock, the closest historical tracking trajectory is found, and the spatial relationship between them is calculated to determine whether they belong to the same physical entity.

[0073] After completing the association matching, the state of each historical tracking trajectory is maintained according to the matching results. For a successfully matched trajectory, its internal state counter is reset, and the spatial coordinates of the current suspected target are added to the point sequence of that trajectory. For a trajectory that does not match successfully, its state counter is incremented. The state information records the continuity and stability of each tracking trajectory in the time dimension. The state update process ensures that the tracking trajectory can reflect the latest dynamics of the target in real time.

[0074] Based on the updated state information, the state information of each tracking trajectory is analyzed, including trajectory duration, point sequence length, and continuous matching status. Tracking trajectories that meet specific state conditions are determined to be real FTR lock targets. The spatial features of the trajectory point sequence are then calculated as the final three-dimensional position information. Tracking trajectories that do not meet the state conditions are excluded.

[0075] For the matching process, such as Figure 6 As shown, in some embodiments, the suspected lock target is associated with and matched with historical tracking trajectories to output an association matching result, including: calculating the spatial distance between the midpoint of the suspected lock target and the historical tracking trajectory; determining the historical tracking trajectory whose spatial distance is less than a preset distance threshold as the matching trajectory; and adding the suspected lock target to the point sequence of the matching trajectory to output the association matching result.

[0076] The straight-line distance between a suspected target and the midpoint of each historical tracking trajectory is the Euclidean distance in three-dimensional space. Spatial distance is an indicator for quantitatively assessing the spatial proximity between targets. After obtaining the spatial distance, the spatial distance value is compared with a preset distance threshold, which is a parameter determined based on the gantry crane's travel speed, scanning frequency, and the actual size of the FTR lock.

[0077] When the spatial distance between a historical tracking trajectory and the current suspected lock target is less than or equal to a preset distance threshold, the historical tracking trajectory is determined as a matching trajectory. This judgment process is based on the physical fact that, in a continuous scanning cycle, the position change of the same FTR lock in the point cloud data will not exceed a specific range.

[0078] After determining the matching trajectory, the three-dimensional coordinates of the current suspected lock target are added to the end of the point sequence of the matching trajectory. This point sequence records the spatial location history of the same FTR lock at different scan times. The update of the point sequence maintains the spatiotemporal continuity of the tracking trajectory, enabling the trajectory to reflect the target's motion state over time.

[0079] Through the above steps, the association and matching between the current suspected target and the historical tracking trajectory is completed. The association and matching process establishes the correspondence between the current observation data and the historical trajectory data, and maintains the continuity of the tracking trajectory in the time dimension.

[0080] Based on the updated state information, the three-dimensional position information of the FTR lock with a preset shape is confirmed from the suspected lock targets. In some embodiments, such as... Figure 7 As shown, the process includes: obtaining the number of consecutive unmatches in the historical tracking trajectory; triggering an confirmation process when the number of consecutive unmatches reaches a preset threshold; after the confirmation process, obtaining the number of points in the historical tracking trajectory and the spatial distribution concentration of the point sequence; when the number of points reaches a preset quantity threshold and the spatial distribution concentration meets a preset requirement, calculating the centroid of the point sequence and determining the centroid as the three-dimensional position information of the FTR lock of a preset shape.

[0081] The parameter in the tracking trajectory status information, which indicates the number of times the trajectory has failed to match any suspected target in the most recent consecutive scanning cycle, is used to count the number of times the trajectory has failed to match any suspected target. The value is obtained by querying the status record of the tracking trajectory.

[0082] The preset threshold number is a value set in advance based on the gantry crane's travel speed and scanning frequency. Reaching this threshold means that the tracking trajectory has not received any new observation data for several consecutive scanning cycles, indicating that the gantry crane may have already passed the area where the FTR lock is located.

[0083] The response confirmation process obtains the number of points in the historical tracking trajectory and the spatial distribution concentration of the point sequence. The number of points is the total number of spatial location points included in the point sequence of the tracking trajectory, reflecting the amount of observation data accumulated by the trajectory. The spatial distribution concentration is a quantitative index obtained by calculating the dispersion of the spatial location of each point in the point sequence, which characterizes the degree of aggregation of these observation points in three-dimensional space.

[0084] The system checks whether the number of points has reached a preset threshold and verifies whether the spatial distribution concentration meets preset requirements. The preset threshold sets the minimum amount of observation data required for trajectory confirmation to ensure that the judgment is based on sufficient data. The preset requirements specify the degree of density that the spatial distribution concentration should reach to ensure that the identified locations have sufficient spatial consistency, so that only trajectories with sufficient observation data and stable spatial locations will be confirmed.

[0085] When the number of points reaches the preset threshold and the spatial distribution concentration meets the preset requirements, the centroid of the point sequence is calculated and the centroid is determined as the three-dimensional position information of the FTR lock with the preset shape. The centroid calculation is achieved by averaging the three-dimensional coordinates of all points in the point sequence. This centroid point represents the spatial center position of the trajectory. The centroid point is used as the final three-dimensional position information output, which takes into account the historical data of the trajectory and provides a single spatial coordinate.

[0086] By determining the confirmation timing through the number of consecutive non-matches, verifying the trajectory quality through the number of points and the concentration of spatial distribution, and determining the location through centroid calculation, this confirmation mechanism based on multi-dimensional evaluation ensures that only those trajectories with sufficient observation data and stable spatial positions will be output, effectively improving the accuracy and reliability of FTR lock position information.

[0087] As can be seen from the above technical solution, the FTR lock identification method provided in this embodiment transforms the lock identification work that relies on manual visual inspection and judgment into a process that is automatically completed by scanning equipment and processing algorithms. This eliminates the need for ground personnel to approach moving containers and flatbed trucks for manual inspection and assisted alignment, thus eliminating the safety risks associated with this work. At the same time, automated identification reduces the need for ground personnel and lowers the overall manpower input for the operation.

[0088] In some embodiments, after confirming the three-dimensional position information of the FTR lock with a preset shape, the method includes: acquiring historical point cloud data; acquiring the number of target points within a preset range around the three-dimensional position information in the historical point cloud data, wherein the target points are the number of points whose height coordinates are greater than the height coordinates of the three-dimensional position information; when the number of target points is greater than a preset number threshold, marking the three-dimensional position information as misidentified information; and deleting the misidentified information in the three-dimensional position information.

[0089] Historical point cloud data is a collection of point cloud data acquired and stored during previous scanning cycles. This data is stored in the storage unit in chronological order, and the historical point cloud dataset corresponding to the current processing time period is retrieved from the storage unit. After acquiring the historical point cloud data, a cylindrical or cubic spatial detection area is established around the confirmed 3D position information within a preset range.

[0090] Traverse all points within the detection area in the historical point cloud data, compare the height coordinates of these points with the confirmed 3D location information, count the number of all points whose height coordinates are greater than the confirmed 3D location information, and obtain the target point count value to detect whether there are obstructions or other high-altitude structures above the confirmed location.

[0091] When the number of target points obtained by statistics exceeds the preset threshold, the three-dimensional position information is marked as misidentified information. The preset threshold is an integer value set according to the number of noise points allowed in the actual scene. The basis for marking misidentified information is that there should not be a dense cluster of high points above the normally functioning FTR lock, otherwise it will hinder the smooth insertion of the container corner fitting holes.

[0092] For all 3D location information marked as misidentified, these records are removed from the list of confirmed FTR lock locations, ensuring that the 3D location information list only includes valid locations that conform to physical constraints. After deleting misidentified information, the correctly identified results that were not marked are retained. By analyzing the spatial relationships in historical point cloud data, it is possible to detect whether there are obstructions above the identified locations, effectively identifying and removing misidentified results that do not conform to physical constraints.

[0093] In some embodiments, the method further includes: obtaining the spacing of the corner fitting holes of the container; performing spacing verification on the three-dimensional position information of the confirmed preset FTR lock based on the spacing, and outputting the verification result; if the verification result does not conform to the spacing, deleting the three-dimensional position information to obtain updated three-dimensional position information; and controlling the gantry crane to move the container to a target position and drop it based on the updated three-dimensional position information, wherein the target position is the position corresponding to the updated three-dimensional position information.

[0094] The spacing between corner fitting holes varies for containers of different specifications. After obtaining the standard spacing data, the three-dimensional position information of the FTR lock with the confirmed preset shape is analyzed. That is, the relative distance of each position point in the direction of track extension is calculated, and the actual measured distance is compared with the standard spacing value to determine whether they conform to the distribution pattern of corner fitting holes of standard containers.

[0095] When the verification results do not conform to the standard spacing, the location points that cannot match any standard spacing are marked as abnormal data and removed from the list of valid results. After the deletion operation is completed, updated 3D location information is obtained. This updated location information set only includes FTR lock positions that meet the standard spacing requirements, forming a spatially reasonable lock layout. By comparing the standard spacing of the container with the identified lock position distribution, the spatial rationality of the identification results is effectively verified.

[0096] Based on the above-described FTR lock recognition method, some embodiments of this application provide an FTR lock recognition system, including: A point cloud acquisition unit is used to acquire point cloud data, which is acquired by scanning the flatbed vehicle; The identification unit is used to identify suspected lock targets from the point cloud data; and to perform trajectory tracking on the suspected lock targets to confirm the three-dimensional position information of the FTR lock of the preset shape from the suspected lock targets.

[0097] For example, for the point cloud acquisition unit, the acquisition frequency is 50Hz, and the data is stored in the data_line variable. The data_line is a custom structure of type RADAR_LINE_DATA, and its data structure is as follows: struct RADAR_LINE_DATA { int y; / / Encoder acquisition value—trolley position vector<POINT2D_XZ> points; / / Single-line radar point cloud }; The POINT2D_XZ structure is as follows: struct STRUCTPACKED POINT2D_XZ { int x; / / Direction of the car int z; / / Height coordinates }; The generated rotation and filtering parameters are used to perform rotation transformation and point cloud filtering on the single-line radar point cloud in data_line, and stored in the data_line_filtered variable. At the same time, the data_line_filtered variable is added to the end of the v_data_line_filtered vector. v_data_line_filtered is used to store the most recent 50 historical point clouds for subsequent FTR lock misidentification filtering.

[0098] data_line_filtered single-line point cloud center point x along the X-axis (car direction) mid Calculate: Find the minimum value of x in the points member variable of data_line_filtered. min and maximum value x max x mid =(x min +x max ) / 2.

[0099] Based on the size and location characteristics of FTR locks, a search for suspected FTR locks is conducted in x. midSearching on both the left and right sides separately; when searching on one side only, in relation to x... mid Find the highest point P within a radius of 100mm, 1133mm apart (the distance between the two FTR locks is 2266mm). max _z; at point P max Search for the FTR lock base within a radius of 100mm centered on _z. If P max If the height difference between _z and the FTR lock base is between 50mm and 100mm, then P is considered... max _z is a suspected FTR lock, recorded using the point PFTRMaybe. The structure of PFTRMaybe is as follows: struct POINT3D_XYZ { int x; / / Coordinates of the car int y; / / Coordinates of the main vehicle int z; / / Height coordinates }; The current PFTRMaybe point found in the current single-line point cloud data_line_filtered needs to be tracked and matched with the historical PFTRMaybe points generated in the previously collected historical single-line point clouds. The historical PFTRMaybe points identified in the previously collected single-line point clouds are stored in a vector.<FTR_TRACK> In the v_ftr_track vector, the FTR_TRACK structure is as follows: struct FTR_TRACK { vector<POINT3D_XYZ> ftr_points; / / Stores PFTRMaybe points for the same FTR lock.

[0100] int unmatched_times; / / Number of consecutive unmatched times }; Tracking and matching principle: In v_ftr_track, find the element closest to the current PFTRMaybe point. If the closest distance is ≤100mm, the tracking and matching is considered successful. The PFTRMaybe point is stored at the end of the ftr_points member variable of that element in v_ftr_track, and unmatched_times is set to 0. If no element with a distance ≤100mm is found in v_ftr_track, a new FTR_TRACK element is added to v_ftr_track, and the current point PFTRMaybe is added as the first point in the ftr_points member variable of that element. For elements in the v_ftr_track vector that do not match, the value of their unmatched_times member variable is incremented by 1.

[0101] FTR lock identification is performed based on the data in v_ftr_track. If the unmatched_times value of an element in v_ftr_track is ≥2, meaning that no match has been found twice, then it is considered that the FTR lock has just been scanned and needs to be identified and confirmed. It is considered an FTR lock if the following conditions are met: Condition 1: The number of points in the ftr_points member variable of the element in v_ftr_track is ≥2; Condition 2: The minimum outer cube of the ftr_points member variable has a length, width, and height of less than 100mm.

[0102] If the above two conditions are met, it is considered an FTR lock. Centroids are generated from each point in ftr_points and stored in v_ftr. The elements of v_ftr are points of the POINT3D_XYZ structure.

[0103] Real-time filtering of FTR false recognition: There should be no point higher than the FTR lock within a 100mm radius around the FTR lock.

[0104] If the distance between the Y-value of the current single-line point cloud `data_line_filtered` and the Y-value of an element in `v_ftr` is exactly greater than 100mm, a false positive filtering should be performed, and only one false positive filtering is required. The filtering method is as follows: In the historically scanned single-line point cloud set `v_data_line_filtered`, if the number of points whose Y-value distance from the FTR lock in `v_ftr` is less than 100mm and whose height exceeds the height of the FTR lock is ≥2, then it is considered a falsely identified FTR lock, and the element is deleted from `v_ftr`.

[0105] The FTR lock recognition result is uploaded to the gantry crane travel control system via the network. The point cloud acquisition unit and the recognition unit enter the sleep state from the scanning state. The gantry crane travel control system places the container in the correct position on the train to be loaded based on the received FTR lock position information, i.e., the three-dimensional position information.

[0106] Similar parts between the embodiments provided in this application can be referred to mutually. The specific implementation methods provided above are only a few examples under the overall concept of this application and do not constitute a limitation on the scope of protection of this application. For those skilled in the art, any other implementation methods extended from the solution of this application without creative effort shall fall within the scope of protection of this application.

Claims

1. An FTR lock identification method, characterized in that, include: Point cloud data is acquired by scanning the flatbed vehicle. Identify suspected lock targets from the point cloud data; The trajectory of the suspected lock target is tracked to confirm the three-dimensional position information of the FTR lock of the preset shape from the suspected lock target.

2. The FTR lock identification method according to claim 1, characterized in that, The acquisition of point cloud data includes: The flatbed truck is vertically scanned using a single-line lidar to obtain scan data; The encoder is used to obtain the real-time position information of the gantry crane; The point cloud data is generated based on the scan data and the real-time location information.

3. The FTR lock identification method according to claim 1, characterized in that, The step of identifying suspected lock targets from the point cloud data includes: Rotate the orbital vertices in the point cloud data to a preset position, the preset position being parallel to a preset coordinate axis; After the track vertex reaches a preset position, a spatial filtering region is set, which is defined based on the track center position and height range; Identify suspected lock targets from point cloud data located within the spatial filtering area.

4. The FTR lock identification method according to claim 1, characterized in that, The step of identifying suspected lock targets from the point cloud data includes: Calculate the position of the center point of the point cloud data in the width direction; Using the center point as a reference, find the local highest point at symmetrical positions on both sides perpendicular to the direction of track extension; Calculate the height difference between the local highest point and the lock base; If the height difference is within a preset range, the highest local point within the preset range is identified as a suspected target.

5. The FTR lock identification method according to claim 1, characterized in that, The step of tracking the trajectory of the suspected lock target to confirm the three-dimensional position information of the FTR lock of the preset shape from the suspected lock target includes: The suspected target is correlated and matched with historical tracking trajectories to output the correlation and matching results; Update the status information of the historical tracking trajectory based on the association matching results; Based on the updated status information, the three-dimensional position information of the FTR lock with the preset shape is confirmed from the suspected lock targets.

6. The FTR lock identification method according to claim 5, characterized in that, The step of associating and matching the suspected target with historical tracking trajectories to output the association and matching results includes: Calculate the spatial distance between the suspected target and the midpoint of the historical tracking trajectory; Historical tracking trajectories with a spatial distance less than a preset distance threshold are identified as matching trajectories; The suspected target is added to the point sequence of the matching trajectory to output the associated matching result.

7. The FTR lock identification method according to claim 5, characterized in that, The step of confirming the three-dimensional position information of an FTR lock of a preset shape from the suspected lock targets based on the updated state information includes: Obtain the number of consecutive non-matches of the historical tracking trajectory; When the number of consecutive unmatched occurrences reaches a preset threshold, an confirmation process is triggered. After the confirmation process, the number of points in the historical tracking trajectory and the spatial distribution concentration of the point sequence are obtained. When the number of points reaches a preset threshold and the spatial distribution concentration meets a preset requirement, the centroid of the point sequence is calculated, and the centroid is determined as the three-dimensional position information of the FTR lock of the preset shape.

8. The FTR lock identification method according to claim 1, characterized in that, After confirming the three-dimensional position information of the FTR lock in the preset configuration, the process includes: Acquire historical point cloud data; In the historical point cloud data, the number of target points within a preset range around the three-dimensional location information is obtained, where the target points are the number of points whose height coordinates are greater than the height coordinates of the three-dimensional location information; When the number of target points exceeds a preset threshold, the three-dimensional location information is marked as misidentified information. Delete the misidentified information in the three-dimensional location information.

9. The FTR lock identification method according to claim 1, characterized in that, The method further includes: Obtain the spacing of the corner fitting holes in the container; Based on the aforementioned spacing, the three-dimensional position information of the confirmed preset FTR lock is verified by spacing verification, and the verification result is output. If the verification result does not match the spacing, the three-dimensional position information is deleted to obtain updated three-dimensional position information; Based on the updated three-dimensional position information, the gantry crane is controlled to move the container to the target position and drop it, where the target position is the position corresponding to the updated three-dimensional position information.

10. An FTR lock identification system, characterized in that, The method for performing the FTR lock identification method according to any one of claims 1-9 includes: A point cloud acquisition unit is used to acquire point cloud data, which is acquired by scanning the flatbed vehicle; The identification unit is used to identify suspected lock targets from the point cloud data; and to perform trajectory tracking on the suspected lock targets to confirm the three-dimensional position information of the FTR lock of the preset shape from the suspected lock targets.

Images