Gantry crane boom anti-collision method and device based on millimeter wave radar

Abstract

The invention provides a gantry crane boom anti-collision method and device based on millimeter wave radar, and relates to the technical field of industrial automation and safety monitoring. The method comprises the following steps: performing coordinate conversion on data acquired by a millimeter wave radar to obtain initial point cloud data, and performing irrelevant point filtering and noise point suppression on the initial point cloud data to obtain effective target point cloud data; performing clustering analysis and multi-frame time sequence consistency verification on the effective target point cloud data, and determining target point cloud clusters which exist continuously and stably; for each acquisition moment, calculating the minimum distance value between the target point cloud cluster and the axis of the boom to form a minimum distance value sequence; performing multi-frame time sequence filtering and robust estimation on the minimum distance value sequence to obtain a nearest distance evaluation value between the target object and the boom; and according to the nearest distance evaluation value, determining a safety alarm area where the target object is located currently, and performing anti-collision control. According to the invention, the accuracy of anti-collision monitoring and control of the gantry crane boom is improved.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and safety monitoring technology, and in particular to a method and device for preventing collisions with the boom of a gantry crane based on millimeter-wave radar. Background Technology

[0002] In modern port bulk cargo handling operations, large equipment such as gantry cranes and bridge unloaders have widely adopted semi-automatic or automatic control. With the continuous acceleration of the operation pace and the increasing density of equipment on site, the risk of spatial interference between crane booms and between booms and ship hulls is becoming increasingly prominent, placing higher demands on the reliability and real-time performance of collision avoidance monitoring technology.

[0003] For example, patent CN117657968A discloses a collision avoidance method for gantry crane booms. This method uses monitoring equipment to acquire the gantry crane trolley position, boom slewing angle, and luffing parameters, establishes a geometric model of the gantry crane group, calculates the height difference and projected distance difference between adjacent booms, and compares these with preset thresholds to control deceleration or stopping. This solution primarily relies on idealized geometric parameters for one-dimensional distance threshold judgment, making it difficult to accurately depict the actual envelope shape of the boom in complex three-dimensional space, and also difficult to reflect the true closest distance relationship between the boom and external irregular obstacles in a timely manner.

[0004] Patent CN222556428U discloses a gantry crane boom anti-collision alarm device. This device uses detection equipment installed between gantry cranes to trigger an infrared alarm when the boom passes through a localized area near the detection equipment, thus indicating a potential risk of proximity between two gantry cranes. This type of device employs a localized "gate-type" detection method, resulting in a limited alarm area. Furthermore, infrared detection is susceptible to environmental factors such as smoke, dust, rain, and fog in ports, making it difficult to achieve continuous distance monitoring and tiered early warning for the entire operating space.

[0005] In existing technologies, stable and reliable distance assessment cannot be achieved and a closed-loop anti-collision control cannot be supported in complex dynamic environments, resulting in poor accuracy of anti-collision monitoring and control of gantry crane booms. Summary of the Invention

[0006] This invention provides a method and device for collision prevention of gantry crane boom based on millimeter-wave radar, in order to solve the problem of poor accuracy in collision prevention monitoring and control of gantry crane boom.

[0007] In a first aspect, embodiments of the present invention provide a method for preventing collisions with the boom of a gantry crane, comprising: A dynamic three-dimensional safety protection model for the boom of a gantry crane is constructed. The dynamic three-dimensional safety protection model includes a three-dimensional capsule based on the real-time dynamic update of the boom's posture, and multi-level safety alarm zones obtained by dividing the spatial regions of the three-dimensional capsule. The point cloud data acquired by millimeter-wave radar is transformed into coordinates to obtain initial point cloud data. Based on a dynamic three-dimensional security protection model, irrelevant points are filtered out and noise points are suppressed in the initial point cloud data to obtain effective target point cloud data. Among them, irrelevant point filtering and noise point suppression include self-echo suppression based on black-white and gray lists and evidence accumulation mechanism. An adaptive density clustering algorithm is used to perform cluster analysis and multi-frame temporal consistency verification on effective target point cloud data to determine the target point cloud clusters that exist continuously and stably. For each acquisition time, based on the linear feature model, the area feature model and the robust feature model, multiple candidate distance values ​​between the target point cloud cluster and the boom axis are calculated, and the minimum value among the multiple candidate distance values ​​is selected as the minimum distance value between the target point cloud cluster and the boom axis, forming a sequence of minimum distance values ​​arranged in chronological order; Multi-frame temporal filtering and robust estimation are performed on the minimum distance value sequence to obtain the nearest distance assessment value between the target object and the boom. The target object is the object corresponding to the target point cloud cluster. Based on the nearest distance assessment value, determine the current safety alarm zone where the target object is located, and implement anti-collision control based on the safety alarm zone where the target object is located.

[0008] Secondly, embodiments of the present invention provide a gantry crane boom anti-collision device, the device comprising: The construction module is used to build a dynamic three-dimensional safety protection model of the boom of the gantry crane. The dynamic three-dimensional safety protection model includes a three-dimensional capsule that is dynamically updated based on the real-time posture of the boom, and a multi-level safety alarm area obtained by dividing the spatial region of the three-dimensional capsule. The suppression module is used to transform the point cloud data acquired by the millimeter-wave radar to obtain initial point cloud data, and based on the dynamic three-dimensional security protection model, to filter out irrelevant points and suppress noise points in the initial point cloud data to obtain effective target point cloud data; among them, the filtering out irrelevant points and suppressing noise points includes self-echo suppression based on black-white and gray lists and evidence accumulation mechanism. The clustering module is used to perform cluster analysis and multi-frame temporal consistency verification on effective target point cloud data using an adaptive density clustering algorithm to determine the target point cloud clusters that exist continuously and stably. The calculation module is used to calculate multiple candidate distance values ​​between the target point cloud cluster and the boom axis based on the linear feature model, the area feature model and the robust feature model for each acquisition time, and select the minimum value from the multiple candidate distance values ​​as the minimum distance value between the target point cloud cluster and the boom axis, forming a sequence of minimum distance values ​​arranged in chronological order; The evaluation module is used to perform multi-frame temporal filtering and robust estimation on the minimum distance value sequence to obtain the minimum distance evaluation value between the target object and the boom. The target object is the object corresponding to the target point cloud cluster. The control module is used to determine the current safety alarm zone of the target object based on the nearest distance assessment value, and to perform anti-collision control based on the safety alarm zone where the target object is located.

[0009] In this embodiment of the invention, a dynamic three-dimensional safety protection model is constructed to accurately represent the spatial envelope of the gantry crane boom. This model can dynamically update with the real-time attitude of the boom, effectively covering the maximum outer dimensions of the boom and its associated components. Through multi-level alarm zone division, it provides clear threshold criteria for collision avoidance decisions, significantly improving adaptability to complex operating scenarios. Furthermore, irrelevant point filtering and noise suppression are performed on the initial point cloud data, effectively eliminating irrelevant points outside the monitoring area, strong reflection interference from the gantry crane's own structure, and environmental noise, resulting in valid target point cloud data and significantly improving the reliability of the target point cloud data.

[0010] Furthermore, cluster analysis and multi-frame temporal consistency verification are performed on the effective target point cloud data to identify continuously stable target point cloud clusters, solving the problems of traditional clustering algorithms being sensitive to noise and unable to adapt to dynamic scenes. Further, for each acquisition moment, the minimum distance between the target point cloud cluster and the boom axis is calculated, forming a sequence of minimum distance values ​​arranged in chronological order, achieving high-precision assessment of the closest distance between the target point cloud cluster and the boom axis. Further, multi-frame temporal filtering and robust estimation are applied to the minimum distance value sequence, effectively suppressing the interference of instantaneous distance fluctuations on control decisions and obtaining the closest distance assessment value between the target object and the boom. Based on this, hierarchical collision avoidance control is implemented according to the closest distance assessment value, realizing intelligent and hierarchical safety decision-making and execution.

[0011] Therefore, the solution proposed in this application can obtain stable and reliable closest distance assessment in complex dynamic environments and support the anti-collision control closed loop, realizing intelligent anti-collision and safety control between gantry crane booms and improving the accuracy of anti-collision monitoring and control of gantry crane booms. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating the implementation of the anti-collision method for the boom of a gantry crane provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the capsule safety boundary model and alarm classification provided in the embodiments of the present invention; Figure 3 This is a bilateral initial point cloud map acquired by radar according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the radar installation location and scanning range provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of a continuously and stably existing target point cloud cluster provided in an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the principle of nearest distance calculation provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the system GUI alarm display interface provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the anti-collision device for the boom of a gantry crane provided in an embodiment of the present invention. Detailed Implementation

[0013] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0014] Figure 1 This is a flowchart illustrating the implementation of the anti-collision method for the boom of a gantry crane provided in this embodiment of the invention. Figure 1 As shown, the method includes the following steps: Step 11: Construct a dynamic three-dimensional safety protection model for the boom of the gantry crane. The dynamic three-dimensional safety protection model includes a three-dimensional capsule based on the real-time dynamic update of the boom's posture, and a multi-level safety alarm area obtained by dividing the spatial region of the three-dimensional capsule. Step 12: Perform coordinate transformation on the point cloud data acquired by the millimeter-wave radar to obtain initial point cloud data, and based on the dynamic three-dimensional security protection model, filter out irrelevant points and suppress noise points in the initial point cloud data to obtain effective target point cloud data. Step 13: Use an adaptive density clustering algorithm to perform cluster analysis on the effective target point cloud data and perform multi-frame temporal consistency verification to determine the target point cloud clusters that exist continuously and stably. Step 14: For each acquisition time, based on the linear feature model, the area feature model, and the robust feature model, calculate multiple candidate distance values ​​between the target point cloud cluster and the boom axis, and select the minimum value from the multiple candidate distance values ​​as the minimum distance value between the target point cloud cluster and the boom axis, forming a sequence of minimum distance values ​​arranged in chronological order. Step 15: Perform multi-frame temporal filtering and robust estimation on the minimum distance value sequence to obtain the nearest distance evaluation value between the target object and the boom. The target object is the object corresponding to the target point cloud cluster. Step 16: Based on the nearest distance assessment value, determine the current safety alarm zone of the target object, and perform anti-collision control based on the safety alarm zone of the target object.

[0015] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0016] See Figure 1Step 11 includes: constructing a dynamic three-dimensional safety protection model of the gantry crane boom. The dynamic three-dimensional safety protection model includes a three-dimensional capsule based on the real-time dynamic update of the boom's posture, and multi-level safety alarm areas obtained by dividing the spatial regions of the three-dimensional capsule.

[0017] Among them, the dynamic three-dimensional safety protection model is a real-time updated spatial model used to accurately characterize the area actually occupied by the boom of the gantry crane in three-dimensional space and its classification level. The model is dynamically updated with the real-time posture of the boom.

[0018] The three-dimensional capsule is composed of a cylinder (the main body of the boom) and two hemispheres (the head of the boom) joined together to form a capsule-shaped closed curved surface. Its centerline coincides with the axis of the boom, forming the spatial framework for dividing the safety alarm zone.

[0019] The multi-level safety alarm zone refers to the functional subspaces nested within the three-dimensional capsule, based on their distance from the boom axis. In practical applications, corresponding equivalent radii can be set for different levels of safety alarm zones.

[0020] In practical applications, the specific dimensions of the three-dimensional capsule, the equivalent radius of each level of safety alarm zone, and the number of safety alarm zones can be adjusted or reset according to the actual structural dimensions, external outlines of auxiliary components, and safety strategies of different models of gantry cranes. Those skilled in the art can adapt and optimize the relevant parameters based on the above settings without changing the basic principle of this invention of constructing graded safety protection zones for the boom through a dynamic three-dimensional capsule model.

[0021] Specifically, the setting of the equivalent radius of each level of safety alarm zone is based on: measuring the maximum half width of the boom structure, identifying the maximum overhang of its auxiliary components (such as walkways, railings, and pulley blocks), and then adding different levels of safety margins that comply with industry safety standards. The safety margins need to be determined by comprehensively considering the equipment's positioning error, system response delay, and on-site environmental factors (such as wind speed).

[0022] Optionally, in one possible implementation, millimeter-wave radars are installed on both sides of the gantry crane boom. The dynamic three-dimensional safety protection model of the gantry crane boom constructed in step 11 above includes: Based on the structural dimensions of the boom, the position of the rotation center, the pitch angle and the telescopic length, a coordinate system for the gantry crane body is established with the millimeter-wave radar installation position as the origin. In the body coordinate system, a three-dimensional capsule is constructed with the boom axis as the center line; Based on the boom's structural dimensions, the outline of its auxiliary components, and the required safety strategy, a first equivalent radius is set. Second equivalent radius and the third equivalent radius The three-dimensional capsule is divided into three different levels of safety alarm zones; in, The three different levels of safety alarm zones include the attention zone, the deceleration zone, and the stop zone. , , These are the equivalent radii corresponding to the attention zone, deceleration zone, and stopping zone, respectively.

[0023] Specifically, a coordinate system for the gantry crane body is established based on parameters such as the structural dimensions of the boom, the position of the slewing center, the pitch angle, and the extension length. For example, this coordinate system can be a right-handed rectangular coordinate system. Specifically, with the millimeter-wave radar installation position as the origin, the boom extension direction is defined as the X-axis, the width direction as the Y-axis, and the vertical direction as the Z-axis, thus forming the gantry crane body coordinate system.

[0024] Based on this, the geometry of the boom is modeled by combining the real-time attitude parameters of the boom. Figure 2 This is a schematic diagram of the capsule safety boundary model and alarm classification provided in an embodiment of the present invention, such as... Figure 2 As shown, a three-dimensional capsule corresponding to its shape is generated. The three-dimensional capsule can be dynamically updated with the boom's rotation, pitch, and extension movements. The three-dimensional capsule represents the spatial detection alarm area of ​​the boom at the current moment, and this area is divided into three alarm zones: yellow, orange, and red, corresponding to the attention zone, deceleration zone, and stop zone, respectively.

[0025] In practical applications, based on the gantry crane's body coordinate system, the dynamic three-dimensional safety protection model can be parametrically described using the shortest distance from a point to the boom axis. Let any point in the gantry crane's body coordinate system be... Its closest distance to the boom axis is denoted as Then the first j Level alarm zones can be represented as:

[0026] Where j = 1, 2, 3, 、 、 The equivalent radii for the attention zone, deceleration zone, and stopping zone are respectively, and satisfy the following conditions: .

[0027] In a typical embodiment of the present invention, for a gantry crane with a model of 40t-45m, the equivalent radii of the three alarm zones are set to 4m, 3m and 2m respectively, based on the above-mentioned setting criteria, and the three-dimensional capsule body is divided into three different levels of safety alarm zones: attention zone, deceleration zone and stopping zone.

[0028] It should be noted that the above radius values ​​are the preferred configuration based on a specific gantry crane model and operating scenario in this embodiment. Those skilled in the art can adjust or reset the equivalent radius based on the above settings according to the actual structural dimensions and specific safety requirements of the target equipment, without affecting the substantive content and scope of application of the method of the present invention.

[0029] Understandably, by constructing a dynamic three-dimensional safety protection model, a precise representation of the spatial envelope of the gantry crane boom is achieved. This model can be dynamically updated with the real-time posture of the boom, effectively covering the maximum outer dimensions of the boom and its attached components. Furthermore, through multi-level alarm zone division, it provides clear threshold criteria for collision avoidance decisions, significantly improving adaptability to complex operating scenarios.

[0030] See also Figure 1 Step 12 includes: performing coordinate transformation on the point cloud data acquired by the millimeter-wave radar to obtain initial point cloud data, and based on the dynamic three-dimensional security protection model, filtering out irrelevant points and suppressing noise points on the initial point cloud data to obtain effective target point cloud data. The filtering out irrelevant points and suppressing noise points includes self-echo suppression based on blacklists, whitelists, and graylists and an evidence accumulation mechanism.

[0031] In practical applications, point cloud data can be acquired using a millimeter-wave radar installed on the gantry crane body, and the point cloud data acquired by the millimeter-wave radar can be transformed into coordinates to obtain initial point cloud data. Optionally, in one possible implementation, the step 12 above, in which the point cloud data acquired by the millimeter-wave radar is transformed into coordinates to obtain initial point cloud data, specifically includes: The point cloud data is transformed from the millimeter-wave radar coordinate system to the body coordinate system to obtain the initial point cloud data.

[0032] Figure 3 This is a bilateral initial point cloud map acquired by radar according to an embodiment of the present invention. Combined with... Figure 3 Point cloud data of the work area can be collected using millimeter-wave radar (such as 4D millimeter-wave radar). Furthermore, the point cloud data collected by the millimeter-wave radar is transformed from the millimeter-wave radar coordinate system to the body coordinate system to obtain initial point cloud data.

[0033] Figure 4 This is a schematic diagram showing the radar installation location and scanning range provided in an embodiment of the present invention. Figure 4As shown, the millimeter-wave radar is installed on both sides of the boom of the gantry crane, with the installation position fixed relative to the body of the gantry crane, and its beam is directed towards the working area of ​​the gantry crane.

[0034] It should be noted that the position of the millimeter-wave radar installed on the gantry crane is fixed. Therefore, the transformation matrix from the radar coordinate system to the gantry crane body coordinate system is a fixed rigid body transformation, which can be obtained in advance by measuring the installation position.

[0035] Optionally, in one possible implementation, the aforementioned irrelevant point filtering and noise point suppression further include: spatial range filtering, reflection intensity filtering, and sparse point filtering; in step 12 above, based on the dynamic three-dimensional security protection model, irrelevant point filtering and noise point suppression are performed on the initial point cloud data to obtain effective target point cloud data, including: Step 121: Perform spatial range filtering on the initial point cloud data to retain the point cloud data located within the preset three-dimensional monitoring area, and obtain the first intermediate point cloud data; Step 122: Perform auto-echo suppression on the first intermediate point cloud data based on black-and-white lists and evidence accumulation mechanism to filter out auto-echo points from the boom of the gantry crane and the fixed structure, and obtain the second intermediate point cloud data. Step 123: Perform reflection intensity filtering on the second intermediate point cloud data to filter out abnormal echo points with reflection intensity lower than the reflection threshold or higher than the saturation value, and obtain the third intermediate point cloud data. Step 124: Perform sparse point filtering based on local point density on the third intermediate point cloud data to filter out sparse noise points with local point density below the density threshold, and obtain effective target point cloud data.

[0036] In step 121, spatial range filtering refers to filtering out points outside the three-dimensional monitoring area. In one example, any point in the initial point cloud data is determined as the current point; it is then determined whether the point falls within the three-dimensional monitoring area; if the point exceeds the three-dimensional monitoring area, it is deleted; if it falls within the preset three-dimensional monitoring area, it is retained.

[0037] The preset spatial range (three-dimensional monitoring area) can be determined based on the maximum working radius of the gantry crane, the boom pitch limit, and the effective range of the radar. This is used to focus processing resources on the effective collision avoidance monitoring area. In one example, the spatial range is set with the radar installation position as the coordinate origin, with the range along the X-axis set to [-30m, 30m], the range along the Y-axis set to [-30m, 30m], and the range along the Z-axis set to [-30m, 30m], corresponding to a three-dimensional monitoring area of ​​approximately 60m × 60m × 60m around the gantry crane body.

[0038] Further, step 122 includes: performing auto-echo suppression processing on the first intermediate point cloud data, and using an auto-echo management method based on black-and-white lists and evidence accumulation mechanism to filter out auto-echo points from the boom of the gantry crane and the fixed structure to obtain the second intermediate point cloud data.

[0039] In one example, any point in the first intermediate point cloud data is determined as the current point; if the point meets the self-echo point determination condition, the point is determined as the self-echo point and deleted; if the point does not meet the self-echo condition, the point is retained.

[0040] Optionally, in one embodiment, step 122 above specifically includes the following sub-steps: Step 1221: Determine the spatial relationship between the current point and the 3D capsule model; based on the coordinates of the current point, determine whether it is located inside the 3D capsule model or within the adjacent threshold range of the outer surface of the capsule.

[0041] In this embodiment, the three-dimensional capsule model is arranged along the central axis of the boom, with its axial length consistent with the effective length of the boom. Its cross-section is circular, with an equivalent radius of 4m, equivalent to a cylindrical domain. The radius of the spherical domain at the boom head is also 4 meters. The two are joined to form the three-dimensional capsule model. A 0.3m buffer zone is set outward from the outer surface of the capsule to determine the point cloud near the boom's outer contour. When a current point falls within the capsule and its buffer zone within a radius of 4.3m, it is considered to have an adjacent relationship with the boom structure.

[0042] Step 1222: Set the static velocity threshold to 0.10 m / s, determine the velocity magnitude of the current point, and mark the current point as a candidate static point when the velocity magnitude is lower than the static velocity threshold.

[0043] For example, the stationary velocity threshold can be obtained by collecting and statistically analyzing the point cloud velocity modulus of the gantry crane's own structure (such as the boom and machine room) due to environmental vibration over a long period of time while the crane is stationary, and taking its statistical upper limit (such as the 99th percentile) as the stationary velocity threshold to distinguish between stationary structures and low-speed moving targets.

[0044] Step 1223: Obtain the radar cross section (RCS) value of the current point and make a judgment based on the RCS distribution of its local neighborhood. In this embodiment, the RCS stability judgment window length is set to 60 consecutive frames. If the standard deviation of the RCS sequence of the current point within the time window does not exceed 1.0, and the mean RCS of the current point is significantly higher than (for example, the RCS value of the current point is 2 to 5 times higher than the mean RCS of its local neighborhood by more than 1.0 standard deviation) the mean RCS of its local neighborhood, then the current point is marked as a candidate for metal reflection to reliably identify the strong reflection characteristics of the metal structure.

[0045] Step 1224: Set the position stability threshold to 0.08m and determine the position change of the current point over 60 consecutive frames. If the standard deviation of the distance sequence from the current point to the radar or boom axis within this time window does not exceed 0.08m, its spatial position is considered to remain basically unchanged, and the current point is marked as a candidate for a spatial fixed point.

[0046] For example, the position stability threshold can be determined by statistically analyzing the amount of coordinate drift caused by measurement errors in the structure of the gantry crane itself across multiple frames of data, and taking its statistical upper limit (such as the 95th percentile) to identify points that are truly fixed in space.

[0047] Step 1225: When a point simultaneously meets the four conditions for determining the automatic echo point in steps 1221 to 1224, the point is determined as an automatic echo point, added to the blacklist, and deleted.

[0048] Preferably, this application provides a blacklist / whitelist / graylist mechanism, which is as follows: If a point satisfies step 1225, it is added to the blacklist and deleted. If a point does not satisfy step 1225 and is clearly located outside the capsule or has motion characteristics, it is added to the whitelist and retained. If a point only satisfies some conditions or is unstable in cross-frame judgment, it is added to the graylist for temporary storage. The graylist is dynamically updated as follows: if a point in the graylist satisfies all echo judgment conditions for several consecutive frames, it is converted from the graylist to the blacklist and deleted; if a point in the graylist does not satisfy the echo judgment conditions for several consecutive frames, it is converted from the graylist to the whitelist and retained.

[0049] The "several consecutive frames" in the gray list update rules can be set according to the system's data update frequency and decision response time, usually 3 to 10 frames, to ensure the stability and timeliness of the judgment.

[0050] Step 1226: If there are unprocessed points, return to step 1221 until all points have been traversed, and the automatic echo point filtering is completed.

[0051] Preferably, the four judgment conditions in steps 1221 to 1224 can be jointly judged in parallel to improve the accuracy of the self-echo point identification. The parallel judgment of the above four conditions is a preferred implementation scheme. Those skilled in the art can adjust or equivalently replace any of the thresholds, weights or condition combinations without departing from the basic principles of the present invention to adapt to different radar installation methods, beam shapes or field conditions.

[0052] Further, step 123 includes: performing reflection intensity filtering on the second intermediate point cloud data to filter out abnormal echo points with reflection intensity lower than the reflection threshold or higher than the saturation value, thereby obtaining the third intermediate point cloud data.

[0053] In one example, any point in the second intermediate point cloud data is determined as the current point; the radar cross-section (RCS) value of that point is obtained, and it is determined whether the RCS value falls within the preset radar cross-section range.

[0054] In this embodiment, the lower limit threshold of RCS is set to 10dBsm, with an upper limit set to 25dBsm, when the RCS value at that point is less than If the RCS value is 10 dBsm or greater than 25 dBsm, the point is identified as an abnormally strong noise point and deleted; if the RCS value of the point is greater than... The point is retained if the RCS value is between 10 dBsm and less than 25 dBsm. The above RCS threshold can be adjusted according to the radar equipment's range and the ambient noise level.

[0055] Specifically, the radar scattering intensity range can be determined by statistically analyzing the RCS echo distribution of real obstacles (such as ships, vehicles, and buildings) in typical operating scenarios and removing outliers that are too high (possibly due to specular reflection) or too low (possibly due to noise).

[0056] Further, step 124 includes: performing sparse point filtering on the third intermediate point cloud data based on local point density to filter out sparse noise points with local point density below the density threshold, thereby obtaining effective target point cloud data.

[0057] In one example, any point in the third intermediate point cloud data is determined as the current point; the set of nearest points of the current point is determined based on the nearest point search window, and the local point density of the current point is determined based on the set of nearest points.

[0058] In this embodiment, an adaptive local density filtering strategy is adopted: when the number of remaining points in the current frame of the third intermediate point cloud data is not less than 300, local density determination is initiated; in the local density determination, the neighborhood of the current point on the XY plane is used as the nearest neighbor search window, with a nearest neighbor radius r = 1.0 meter. Within a given radius r, if the number of nearest neighbor points of the current point is less than 6, the point is determined as a local sparse point and deleted; when the number of nearest neighbor points is not less than 6, the point is retained.

[0059] To avoid accidental deletion, if the proportion of retained points after a local density filter is less than 25% of the number of points before filtering, the result of this density filter is automatically discarded and the data is rolled back to the point cloud data before filtering.

[0060] The density threshold can be set by analyzing the statistical differences in local neighborhood density between real obstacle point clouds and random noise point clouds. For example, the density threshold can be set to 1.5 to 3 times the average density of the noise point cloud.

[0061] Understandably, filtering out irrelevant points and suppressing noise points in the initial point cloud data effectively eliminates irrelevant points outside the monitoring area, strong reflection interference from the gantry crane's own structure, and environmental noise, thus obtaining effective target point cloud data and significantly improving the reliability of the target point cloud data.

[0062] See also Figure 1 Step 13 includes: using an adaptive density clustering algorithm to perform cluster analysis on the effective target point cloud data, and performing multi-frame temporal consistency verification to determine the target point cloud clusters that exist continuously and stably.

[0063] Optionally, in one possible implementation, step 13 above includes: Step 131: Based on the k-nearest neighbor distances of each point in the effective target point cloud data of the current frame, calculate the k-nearest neighbor distance set of the current frame, and calculate the density index of the current frame based on the set; Step 132: Based on the density index and the preset mapping relationship, dynamically determine the candidate clustering radius parameter and the candidate minimum neighborhood point parameter for the current frame; Step 133: If the absolute value of the difference between the density index of the current frame and the density index of the previous frame does not exceed the update threshold, then the clustering radius parameter and the minimum number of neighborhood points parameter of the previous frame are used as the clustering parameters of the current frame; otherwise, the candidate clustering radius parameter and the candidate minimum number of neighborhood points parameter of the current frame are used as the clustering parameters of the current frame, and the density index is updated. Step 134: Determine whether the number of valid target point cloud data points in the current frame is less than the threshold for valid clustering points; if it is less, determine that there is no stable target point cloud cluster in the current frame and skip clustering analysis; otherwise, use the clustering parameters of the current frame to perform density clustering on the valid target point cloud data of the current frame to obtain the clustering results. Step 135: Calculate the proportion of non-noise points in the clustering results to the number of valid target point cloud data points in the current frame; if the proportion is lower than the retention proportion threshold, the clustering results of the current frame are deemed invalid and discarded; otherwise, the clustering results are retained and used as candidate target point cloud clusters for the current frame. Step 136: For candidate target point cloud clusters retained in multiple consecutive frames, perform temporal consistency determination based on spatial location, scale and number association. Determine the candidate target point cloud clusters that appear in all consecutive preset frames and meet the conditions of spatial location and scale consistency as stable target point cloud clusters.

[0064] Specifically, the valid target point cloud data output in step 12 is labeled according to the radar source, forming a left point cloud dataset and a right point cloud dataset respectively. Adaptive density clustering analysis and subsequent multi-frame temporal consistency verification are then performed independently on both point cloud datasets. This splitting method is beneficial for adaptively adjusting clustering parameters for radars with different installation angles and avoids mutual interference between the left and right fields of view.

[0065] Further, step 131 includes: calculating the set of k-nearest neighbor distances for the current frame based on the k-nearest neighbor distances of each point in the valid target point cloud data of the current frame. .

[0066] In this embodiment, the nearest neighbor order is set to k=8. For each point in the current frame, the Euclidean distance from its 8th nearest neighbor to itself is calculated to obtain the set of 8th nearest neighbor distances. Density index Defined as The average value is used to characterize the overall density of the point cloud in the current frame.

[0067] Further, step 132 includes: dynamically determining the candidate clustering radius parameter of the current frame based on the density index and a preset mapping relationship. and the minimum number of neighborhood points parameter Among them, the cluster radius parameter The quantile relationships are given as follows: , in, This indicates the quantile operation performed on the 8 nearest neighbor distance set according to a preset quantile q=90%, and the result... Perform amplitude limiting to constrain it to Within a certain range, to avoid clustering radii that are too small or too large.

[0068] In candidate cluster radius Next, count the number of neighboring points of each point within the radius to obtain the set of neighboring points. In this embodiment, The 15th percentile is used as the parameter for the minimum number of candidate neighborhood points. And set a lower limit for it, requiring This ensures that each cluster contains at least three point cloud samples.

[0069] Further, step 133 includes: setting the current frame density index Density index compared to the previous frame Compare. When satisfied... That is, if the density change does not exceed the preset 30% update threshold, the cluster radius parameter determined in the previous frame will be used. and the minimum number of neighborhood points parameter Used as the clustering parameter for the current frame. When If the update threshold is exceeded, or if no valid density index has been recorded in the previous frame, then the candidate clustering parameters of the current frame are used. and the parameter of minimum number of candidate neighborhood points As the clustering parameter for the current frame, and It is updated as a density metric for the current frame.

[0070] Furthermore, small sample protection and clustering are performed, and step 134 above includes: Before performing adaptive clustering, a small sample protection mechanism is set for the current point cloud data volume. In this embodiment, the number of valid points on the current side is set to... The nearest neighbor order k is set to 8, and the threshold for the number of effective cluster points is determined according to the following formula. :

[0071] That is, when k=8, Not less than 16.

[0072] when If the number of points in the current frame is insufficient to support reliable density estimation and clustering structure, the density clustering analysis of the current frame is skipped and no candidate target point cloud clusters are generated. In this case, the effective point cloud obtained by the aforementioned filtering of the current frame is conservatively passed through to the subsequent processing steps to avoid mistakenly deleting effective points as noise when there is insufficient data.

[0073] when Using the cluster radius determined in step 132 and minimum number of neighborhood points Perform density-based spatial clustering of applications with noise (DBSCAN) on the current side point cloud to obtain one or more non-noise point clusters and points marked as noise.

[0074] Furthermore, retain the percentage threshold and roll back the result. Step 135 above includes: After density clustering is completed, the proportion of non-noise points retained is calculated:

[0075] in, The proportion of non-noise points retained. The number of non-noise points. This represents the total number of points in the current side point cloud.

[0076] In this embodiment, the threshold for retaining the proportion of non-noise points is set to 0.15. Specifically, when If the current clustering result is considered unstable or the clustering parameters are improperly set, no hard deletion operation is performed on noisy points. Instead, the process reverts to the valid target point cloud before clustering, retaining all points for subsequent processing. Furthermore, when the clustering process is completed normally, the set of all non-noise points is used as the candidate target point cloud cluster set on this side, and noise points are deleted in the clustering filtering link to suppress isolated points and obvious outliers.

[0077] By using the aforementioned retention ratio gating and backoff mechanism, we can ensure the clustering denoising effect while avoiding excessive deletion of points in high-noise or abnormal scenarios.

[0078] Furthermore, step 136 above includes: screening stable target point cloud clusters based on multi-frame temporal consistency, and extracting representative point sets of stable target clusters.

[0079] For the candidate target point cloud clusters obtained on the left and right sides respectively, multi-frame consistency determination is performed in the time dimension. In this embodiment, it can be implemented in the following way: For each candidate point cloud cluster, its spatial location (e.g., cluster centroid or envelope center), scale features (e.g., number of points within the cluster or spatial expansion range), and cluster number matching relationship are tracked in consecutive frames. If a cluster can find a corresponding cluster in adjacent frames whose position and scale are within the tolerance range within a consecutive preset number of frames, then the cluster is considered to have sufficient stability on the time axis and is determined to be a stable target point cloud cluster. Clusters that only appear briefly, disappear frequently in consecutive frames, or whose position / scale fluctuations exceed the preset tolerance are marked as unstable clusters and removed.

[0080] The preset frame count range, position, and scale tolerance can be configured according to the system sampling frequency and control response time. For example, at a sampling frequency of 20Hz, the "number of consecutive occurrences" can be set in the range of 3 to 10 frames to balance response speed and stability. In this embodiment, it is preferable to confirm a cluster as a stable target only when the consistency condition is met within a certain number of consecutive frames. However, those skilled in the art can adjust the above frame count and tolerance parameters according to actual working conditions.

[0081] Furthermore, for the stable target point cloud clusters selected above, to facilitate subsequent distance calculation and early warning level determination, this embodiment further extracts the core region and calculates representative positions within the cluster point cloud, specifically including: Within each stable target cluster, based on the local density or proximity of points within the cluster, a more spatially dense core point set is selected to suppress the interference of sparse points at the cluster edge on subsequent distance estimation. The geometric center (e.g., centroid) or the median value of the coordinate components of the core point set is calculated to obtain the representative position point of the cluster, which is used in subsequent steps to participate in the distance calculation with the boom axis and the determination of different collision avoidance levels. For stable target clusters where the left and right radars may correspond to the same obstacle in space, the overlap between their representative positions and the point cloud can be further compared in the body coordinate system. If necessary, they can be merged into a unified target cluster to improve the consistency of target description.

[0082] Figure 5 This is a schematic diagram of a continuously and stably existing target point cloud cluster provided in an embodiment of the present invention, such as... Figure 5 As shown, through the processing of steps 131 to 136 above, this embodiment can adaptively extract a continuously stable target point cloud cluster from the effective target point cloud data under complex background noise and dynamic scenes, which can provide reliable input data for the subsequent nearest distance calculation of the boom-based three-dimensional geometric model.

[0083] The aforementioned adaptive update can be executed frame by frame or triggered when a density abrupt change event is detected. Those skilled in the art can adaptively adjust or equivalently replace the selection of statistics, the setting of update thresholds, and the specific magnitude relationships of parameter adjustments without departing from the basic principles of this invention.

[0084] Understandably, clustering analysis and multi-frame temporal consistency verification of effective target point cloud data can identify continuously stable target point cloud clusters, thus solving the problems of traditional clustering algorithms being sensitive to noise and difficult to adapt to dynamic scenes.

[0085] See also Figure 1 Step 14 includes: for each acquisition time, based on the linear feature model, the area feature model and the robust feature model, calculate multiple candidate distance values ​​between the target point cloud cluster and the boom axis, and select the minimum value from the multiple candidate distance values ​​as the minimum distance value between the target point cloud cluster and the boom axis, forming a sequence of minimum distance values ​​arranged in chronological order.

[0086] Alternatively, in one possible implementation, step 14 above includes: Based on the geometric features of the target point cloud cluster and the three-dimensional geometric model of the boom, linear feature models, surface feature models and robust feature models are constructed respectively. For each acquisition time, the linear feature model, the area feature model, and the robust feature model are applied respectively to calculate the first candidate distance value, the second candidate distance value, and the third candidate distance value at the acquisition time. The minimum value among the first, second, and third candidate distance values ​​at the time of acquisition is taken as the minimum distance between the target point cloud cluster and the boom axis at the time of acquisition. The minimum distance values ​​between the target point cloud clusters and the boom axis at each sampling time are arranged in chronological order to form a minimum distance value sequence.

[0087] It should be noted that in this invention, the three-dimensional capsule model is mainly used to construct graded safety alarm zones and perform spatial determination of the self-echo points. Its function is to represent the extended envelope of the boom in space in an equivalent manner. In the calculation of the nearest distance value, in order to improve the accuracy of distance assessment, this invention uses a three-dimensional geometric model (simplified geometric model) of the boom based on the boom's central axis and a fixed equivalent radius for accurate calculation, rather than using the outer surface of the capsule used for alarm zone division.

[0088] Figure 6 This is a schematic diagram illustrating the principle of nearest distance calculation provided in an embodiment of the present invention. In this embodiment, the distance calculation adopts the following... Figure 6 The nearest distance calculation scheme is shown below. Figure 6 As shown, in the coordinate system of the gantry crane, the boom axis is represented by line segment AB, point A is the boom head, and point B is the boom joint connection. The coordinates of the two points are A( , , ) and B( , , The boom axis direction vector is For any point P in the stable target point cloud cluster, with coordinates P(x,y,z), calculate the projection parameter λ of this point on the boom axis:

[0089] Based on the value of the projection parameter λ, determine the closest distance of point P relative to the boom axis. :

[0090] in, This represents the distance from the point to the boom axis. When the projection point is on the boom axis, the value being calculated is the perpendicular distance from point P to line AB; when When the projection point is outside point A at the head of the boom axis, the Euclidean distance from point P to point A is calculated; when When the projection point is outside point B on the boom axis, the collision does not occur in this area according to the four-bar linkage structure of the gantry crane, so it is not considered.

[0091] Preferably, the distance calculation is based on feature modeling of the representative point set of the target cluster extracted in step 136 to reduce the impact of sparse points at the cluster edge on the distance estimation; when the representative point set of the target cluster is not explicitly extracted, the linear feature model, the area feature model and the robust feature model can be applied to all points in the stable target point cloud cluster to calculate the first candidate distance value, the second candidate distance value and the third candidate distance value at the acquisition time.

[0092] Specifically, in one possible implementation, the above-mentioned application of a linear feature model, a planar feature model, and a robust feature model to calculate the first candidate distance value, the second candidate distance value, and the third candidate distance value at each acquisition time includes: The principal component analysis of the target point cloud cluster is performed using a linear feature model to determine the main direction, and a robust straight line fitting is performed using a random sampling consensus algorithm to obtain the feature line and the set of interior points of the feature line. The nearest distance from each point in the set of interior points to the boom axis is calculated, and the minimum value is taken as the first candidate distance value at the acquisition time. The geometric outer envelope corresponding to the target point cloud cluster after noise and outliers are constructed by applying the planar feature model. The nearest distance from each vertex or sampling point on the surface of the geometric outer envelope to the boom axis is calculated, and the minimum value is taken as the second candidate distance value at the acquisition time. A robust feature model is applied to divide the target point cloud cluster into multiple sectors according to the azimuth angle. Core points that meet the local point density conditions are selected in each sector, and the nearest distance between each core point and the boom axis is calculated. Quantile or median statistics are performed on the nearest distance between each core point and the boom axis in each sector to obtain the representative distance value of each sector. The minimum value among the representative distance values ​​of each sector is used as the third candidate distance value at the acquisition time.

[0093] In this embodiment, for each acquisition time, a linear feature model, a planar feature model, and a robust feature model are applied respectively to calculate the first, second, and third candidate distance values ​​at that acquisition time. This provides a multi-dimensional, complementary distance estimation framework that effectively overcomes the limitations of single geometric models in dealing with complex, irregular, partially occluded, or noisy target point clouds. It ensures that under any conditions, it can output a minimum distance value that is as close as possible to the actual collision risk, providing an extremely reliable basis for collision avoidance decisions.

[0094] Furthermore, the minimum value among the first, second, and third candidate distance values ​​at the sampling time is taken as the minimum distance between the target point cloud cluster and the boom axis at the sampling time; the minimum distance values ​​between the target point cloud cluster and the boom axis at each sampling time are arranged in chronological order to form a minimum distance value sequence.

[0095] Understandably, for each acquisition time, the minimum distance between the target point cloud cluster and the boom axis is calculated, forming a sequence of minimum distance values ​​arranged in chronological order, thus achieving a high-precision assessment of the closest distance between the target point cloud cluster and the boom axis.

[0096] See also Figure 1 Step 15 includes: performing multi-frame temporal filtering and robust estimation on the minimum distance value sequence to obtain the nearest distance assessment value between the target object and the boom. The target object is the object corresponding to the target point cloud cluster, that is, the object that may collide with the boom.

[0097] Understandably, multi-frame temporal filtering and robust estimation of the minimum distance value sequence effectively suppresses the interference of instantaneous distance fluctuations on control decisions and obtains the closest distance assessment value between the target object and the boom.

[0098] Optionally, in one possible implementation, step 15 above includes: The distance estimate at the current time is obtained by performing time-series filtering on the minimum distance value sequence based on a time sliding window. Determine whether the target point cloud cluster is in a low-speed stable state; If the target point cloud cluster is in a low-speed stable state, the nearest distance assessment value of the previous moment will be used as the nearest distance assessment value of the current moment. Otherwise, the distance estimate at the current moment is used as the nearest distance assessment at the current moment.

[0099] Specifically, let the sequence of minimum distance values ​​arranged in chronological order be... ,in, i The frame number is the length of the frame. L Distance estimate within the time sliding window The weighted average can be calculated using the following formula: , in, To meet The non-negative weighting coefficients.

[0100] In this embodiment, the time sliding window length is L = 3 frames, and the weighting coefficient is... Take equal weights, and preferably within the window Take the median directly as This enhances the ability to suppress outliers.

[0101] Furthermore, if the target point cloud cluster is in a low-speed stable state, the nearest distance assessment value of the previous moment is used as the nearest distance assessment value of the current moment; if the target point cloud cluster is not in a low-speed stable state, the distance estimate of the current moment is used as the nearest distance assessment value of the current moment.

[0102] Among them, the low-speed stable state refers to the velocity magnitude of the target point cloud cluster. Below the speed threshold And the change in the minimum distance value between adjacent frames does not exceed the distance tolerance. In this embodiment, the speed threshold is taken as... Distance tolerance That is, when the velocity of the target point cloud cluster is less than 0.05 m / s, and the following conditions are met... At that time, it is determined that it is in a low-speed stable state.

[0103] See also Figure 1 Step 16 includes: determining the current safety alarm zone of the target object based on the nearest distance assessment value, and performing anti-collision control based on the safety alarm zone of the target object.

[0104] Understandably, by implementing graded collision avoidance control based on the closest distance assessment value, intelligent and graded safety decision-making and execution are achieved.

[0105] In one example, based on the nearest distance assessment value, the current safety alarm zone of the target object is determined, which in turn determines the current collision avoidance status of the gantry crane boom. Collision avoidance status includes safe status, warning status, deceleration status, and stop status. The collision avoidance status can be provided to the display module, alarm module, and control module to trigger interface prompts, audible and visual alarms, or gantry crane linkage control.

[0106] Figure 7 This is a schematic diagram of the system GUI alarm display interface provided in an embodiment of the present invention, as shown below. Figure 7 As shown, the operation interface can intuitively display the current nearest distance value and its corresponding anti-collision level, making it easy for operators to keep abreast of the safety conditions around the gantry crane.

[0107] For example, the display module is used to display the nearest distance assessment value, the current collision avoidance level, and the status indication in real time on the operation interface; the alarm module is used to trigger audible and visual alarm signals when entering the warning, deceleration, or shutdown state; and the control module is used to issue graded control commands to the crane control system according to the collision avoidance status, such as performing deceleration, limiting the range of motion, or emergency shutdown.

[0108] Optionally, in one possible implementation, step 16 above includes: If the nearest distance assessment value is greater than the first equivalent radius If the target object is located in the safe zone, no warning signal or control command will be output. If the nearest distance assessment value is less than or equal to the first equivalent radius And greater than the second equivalent radius If the target object is located in the attention zone, a warning signal is output. If the nearest distance assessment value is less than or equal to the second equivalent radius And greater than the third equivalent radius If the target object is located in the deceleration zone, a deceleration control command is output to cause the gantry crane to decelerate. If the nearest distance assessment value is less than or equal to the third equivalent radius If the target object is located in the stopping area, a stopping control command is output to make the gantry crane perform an emergency stop.

[0109] In practical applications, the three-dimensional capsule is divided into a attention zone, a deceleration zone, and a stopping zone. In addition, the area outside the three-dimensional capsule is defined as a safety zone.

[0110] In this embodiment, by comparing the nearest distance assessment value with three equivalent radii, the area where the target object is located can be determined, and graded anti-collision control can be performed according to the area where the target object is located, thereby improving the accuracy and intelligence of anti-collision control of the gantry crane boom.

[0111] Furthermore, to provide a more stable and intuitive display in complex operating environments and avoid frequent fluctuations in the anti-collision status due to minor data fluctuations, a stable anti-collision status (including at least safety, warning, deceleration, and shutdown) is output to avoid frequent state switching caused by slight fluctuations in the measured value near the threshold. In one possible implementation, during the anti-collision status determination process, sticky processing can also be performed on the displayed distance value, specifically including: Based on the nearest distance assessment value, the speed of the target point cloud cluster, and the distance change trend over multiple consecutive frames, the system switches between three display states: tracking, holding, and reacquisition. In tracking state, the system tracks the target point cloud cluster in real time and updates the displayed distance value. In holding state, the displayed distance value output externally is the quantized locked distance value. The holding state will exit and the lock will be released when any of the following conditions are met: (1) The speed of the target point cloud cluster exceeds the preset speed threshold; (2) The cumulative value of the distance change trend over multiple consecutive frames exceeds the preset change threshold; (3) The preset timeout condition is met, that is, the target point cloud cluster does not change significantly within the set time window; When any breaking condition is triggered, the system enters a reacquisition state and recalculates the nearest distance assessment value based on the new input data.

[0112] Preferably, to achieve continuous protection, the system cyclically executes the aforementioned steps at a set frequency within the operation cycle, performing closed-loop processing of data acquisition, model updating, target extraction, distance assessment, and collision avoidance decision-making, thereby continuously tracking the boom movement and changes in the surrounding environment, updating the collision avoidance status in real time, and outputting corresponding instructions to ensure the safe operation of the gantry crane throughout the entire operation process.

[0113] The solution proposed in this application can obtain stable and reliable closest distance assessment in complex dynamic environments and support anti-collision control closed loop, realizing intelligent anti-collision and safety control between gantry crane booms and improving the accuracy of anti-collision monitoring and control of gantry crane booms.

[0114] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0115] Figure 8 This is a structural schematic diagram of the anti-collision device for the boom of a gantry crane provided in an embodiment of the present invention, as shown below. Figure 8 As shown, the device includes: Module 81 is used to construct a dynamic three-dimensional safety protection model of the boom of a gantry crane. The dynamic three-dimensional safety protection model includes a three-dimensional capsule that is dynamically updated based on the real-time posture of the boom, and a multi-level safety alarm area obtained by dividing the spatial region of the three-dimensional capsule. The suppression module 82 is used to perform coordinate transformation on the point cloud data acquired by the millimeter-wave radar to obtain initial point cloud data, and based on the dynamic three-dimensional security protection model, to filter out irrelevant points and suppress noise points on the initial point cloud data to obtain effective target point cloud data; wherein, the filtering out irrelevant points and suppressing noise points includes self-echo suppression based on black-white and gray lists and evidence accumulation mechanism; Clustering module 83 is used to perform clustering analysis on effective target point cloud data using an adaptive density clustering algorithm, and to perform multi-frame temporal consistency verification to determine the target point cloud clusters that exist continuously and stably. The calculation module 84 is used to calculate multiple candidate distance values ​​between the target point cloud cluster and the boom axis based on the linear feature model, the area feature model and the robust feature model for each acquisition time, and select the minimum value from the multiple candidate distance values ​​as the minimum distance value between the target point cloud cluster and the boom axis, forming a sequence of minimum distance values ​​arranged in chronological order. Evaluation module 85 is used to perform multi-frame temporal filtering and robust estimation on the minimum distance value sequence to obtain the nearest distance evaluation value between the target object and the boom. The target object is the object corresponding to the target point cloud cluster. The control module 86 is used to determine the current safety alarm zone of the target object based on the nearest distance assessment value, and to perform anti-collision control based on the safety alarm zone of the target object.

[0116] In this embodiment of the invention, the construction module 81 constructs a dynamic three-dimensional safety protection model, achieving accurate representation of the spatial envelope of the gantry crane boom. This model can dynamically update with the real-time posture of the boom, effectively covering the maximum outer dimensions of the boom and its associated components. Through multi-level alarm zone division, it provides clear threshold criteria for collision avoidance decisions, significantly improving adaptability to complex operating scenarios. Furthermore, the suppression module 82 filters out irrelevant points and suppresses noise points in the initial point cloud data, effectively eliminating irrelevant points outside the monitoring area, strong reflection interference from the gantry crane's own structure, and environmental noise, obtaining valid target point cloud data and significantly improving the reliability of the target point cloud data.

[0117] Furthermore, the clustering module 83 performs clustering analysis and multi-frame temporal consistency verification on the effective target point cloud data to identify continuously stable target point cloud clusters, solving the problems of traditional clustering algorithms being sensitive to noise and difficult to adapt to dynamic scenes. Further, for each acquisition moment, the calculation module 84 calculates the minimum distance between the target point cloud cluster and the boom axis, forming a sequence of minimum distance values ​​arranged in chronological order, achieving high-precision assessment of the closest distance between the target point cloud cluster and the boom axis. Further, the evaluation module 85 performs multi-frame temporal filtering and robust estimation on the minimum distance value sequence, effectively suppressing the interference of instantaneous distance fluctuations on control decisions and obtaining the closest distance assessment value between the target object and the boom. Based on this, the control module 86 performs hierarchical anti-collision control according to the closest distance assessment value, realizing intelligent and hierarchical safety decision-making and execution.

[0118] Therefore, the solution proposed in this application can obtain stable and reliable closest distance assessment in complex dynamic environments and support the anti-collision control closed loop, realizing intelligent anti-collision and safety control between gantry crane booms and improving the accuracy of anti-collision monitoring and control of gantry crane booms.

[0119] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0120] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A collision avoidance method for the boom of a gantry crane based on millimeter-wave radar, characterized in that, The method includes: A dynamic three-dimensional safety protection model for the boom of a gantry crane is constructed. The dynamic three-dimensional safety protection model includes a three-dimensional capsule that is dynamically updated based on the real-time posture of the boom, and a multi-level safety alarm area obtained by dividing the spatial region of the three-dimensional capsule. The point cloud data acquired by millimeter-wave radar is transformed into coordinates to obtain initial point cloud data. Based on the dynamic three-dimensional security protection model, irrelevant points are filtered out and noise points are suppressed in the initial point cloud data to obtain effective target point cloud data. The irrelevant point filtering and noise point suppression include self-echo suppression based on black-and-white lists and evidence accumulation mechanisms. An adaptive density clustering algorithm is used to perform cluster analysis on the effective target point cloud data, and multi-frame temporal consistency verification is performed to determine the target point cloud clusters that exist continuously and stably. For each acquisition time, based on the linear feature model, the area feature model, and the robust feature model, multiple candidate distance values ​​between the target point cloud cluster and the boom axis are calculated, and the minimum value is selected from the multiple candidate distance values ​​as the minimum distance value between the target point cloud cluster and the boom axis, forming a sequence of minimum distance values ​​arranged in chronological order; Multi-frame temporal filtering and robust estimation are performed on the minimum distance value sequence to obtain the nearest distance evaluation value between the target object and the boom, wherein the target object is the object corresponding to the target point cloud cluster; Based on the nearest distance assessment value, the current safety alarm zone of the target object is determined, and anti-collision control is performed based on the safety alarm zone of the target object.

2. The anti-collision method for gantry crane boom based on millimeter-wave radar according to claim 1, characterized in that, Millimeter-wave radars are installed on both sides of the gantry crane boom; the dynamic three-dimensional safety protection model for constructing the gantry crane boom includes: Based on the structural dimensions, rotation center position, pitch angle and telescopic length of the boom, a gantry crane body coordinate system is established with the millimeter-wave radar installation position as the origin. The three-dimensional capsule body is constructed with the boom axis as the center line in the body coordinate system. Based on the structural dimensions of the boom, the outline of its auxiliary components, and the required safety strategy, a first equivalent radius is set. Second equivalent radius and the third equivalent radius The three-dimensional capsule is divided into three different levels of safety alarm zones; in, The three different levels of safety alarm zones include a warning zone, a deceleration zone, and a stop zone. , , These are the equivalent radii corresponding to the attention zone, the deceleration zone, and the stopping zone, respectively.

3. The anti-collision method for gantry crane boom based on millimeter-wave radar according to claim 2, characterized in that, The process of transforming the point cloud data acquired by millimeter-wave radar to obtain initial point cloud data specifically includes: The point cloud data is transformed from the millimeter-wave radar coordinate system to the body coordinate system to obtain the initial point cloud data.

4. The anti-collision method for gantry crane boom based on millimeter-wave radar according to claim 1, characterized in that, The irrelevant point filtering and noise point suppression also include: spatial range filtering, reflection intensity filtering, and sparse point filtering; Based on the dynamic three-dimensional security protection model, the initial point cloud data is processed by filtering out irrelevant points and suppressing noise points to obtain effective target point cloud data, including: Spatial range filtering is performed on the initial point cloud data to retain the point cloud data located within the preset three-dimensional monitoring area, thereby obtaining the first intermediate point cloud data; The first intermediate point cloud data is subjected to auto-echo suppression based on black-and-white lists and evidence accumulation mechanism to filter out auto-echo points from the boom and fixed structure of the gantry crane, thereby obtaining the second intermediate point cloud data. The second intermediate point cloud data is filtered by reflection intensity to remove abnormal echo points with reflection intensity below the reflection threshold or above the saturation value, thereby obtaining the third intermediate point cloud data. The third intermediate point cloud data is subjected to sparse point filtering based on local point density to filter out sparse noise points with local point density below the density threshold, thereby obtaining the effective target point cloud data.

5. The anti-collision method for gantry crane boom based on millimeter-wave radar according to claim 1, characterized in that, The step of using an adaptive density clustering algorithm to perform cluster analysis on the effective target point cloud data and verifying multi-frame temporal consistency to determine the continuously stable target point cloud clusters includes: Based on the k-nearest neighbor distances of each point in the effective target point cloud data of the current frame, calculate the k-nearest neighbor distance set of the current frame, and calculate the density index of the current frame based on the set; Based on the density index and the preset mapping relationship, the candidate clustering radius parameter and the candidate minimum neighborhood point parameter of the current frame are dynamically determined; If the absolute value of the difference between the density index of the current frame and the density index of the previous frame does not exceed the update threshold, then the clustering radius parameter and the minimum number of neighborhood points parameter of the previous frame are used as the clustering parameters of the current frame; otherwise, the candidate clustering radius parameter and the candidate minimum number of neighborhood points parameter of the current frame are used as the clustering parameters of the current frame, and the density index is updated. Determine whether the number of valid target point cloud data points in the current frame is less than the threshold for valid clustering points; if it is less, determine that there is no stable target point cloud cluster in the current frame and skip clustering analysis; otherwise, use the clustering parameters of the current frame to perform density clustering on the valid target point cloud data of the current frame to obtain the clustering results. Calculate the proportion of non-noise points in the clustering results to the number of valid target point cloud data points in the current frame; if the proportion is lower than the retention proportion threshold, the clustering results of the current frame are deemed invalid and discarded; otherwise, the clustering results are retained and used as candidate target point cloud clusters for the current frame. For candidate target point cloud clusters retained in multiple consecutive frames, temporal consistency is determined based on spatial location, scale, and number association. Candidate target point cloud clusters that appear in all consecutive preset frames and meet the conditions of spatial location and scale consistency are identified as stable target point cloud clusters.

6. The anti-collision method for gantry crane boom based on millimeter-wave radar according to claim 1, characterized in that, For each acquisition time, based on linear feature models, area feature models, and robust feature models, multiple candidate distance values ​​between the target point cloud cluster and the boom axis are calculated. The minimum value among these candidate distance values ​​is selected as the minimum distance value between the target point cloud cluster and the boom axis, forming a sequence of minimum distance values ​​arranged in chronological order, including: Based on the geometric features of the target point cloud cluster and the three-dimensional geometric model of the boom, a linear feature model, a planar feature model, and a robust feature model are constructed respectively. For each acquisition time, the linear feature model, the area feature model, and the robust feature model are applied respectively to calculate the first candidate distance value, the second candidate distance value, and the third candidate distance value at that acquisition time. The minimum value among the first candidate distance value, the second candidate distance value, and the third candidate distance value at the acquisition time shall be taken as the minimum distance value between the target point cloud cluster and the boom axis at the acquisition time. The minimum distance values ​​between the target point cloud cluster and the boom axis at each sampling time are arranged in chronological order to form the minimum distance value sequence.

7. The anti-collision method for gantry crane boom based on millimeter-wave radar according to claim 6, characterized in that, For each acquisition time, the linear feature model, the area feature model, and the robust feature model are applied respectively to calculate the first candidate distance value, the second candidate distance value, and the third candidate distance value at that acquisition time, including: The linear feature model is applied to perform principal component analysis on the target point cloud cluster to determine the main direction, and a robust straight line fitting is performed using a random sampling consensus algorithm to obtain the feature line and the set of interior points of the feature line; the nearest distance value from each point in the set of interior points to the boom axis is calculated, and the minimum value is taken as the first candidate distance value at the acquisition time; The geometric outer envelope corresponding to the target point cloud cluster after noise and outliers are constructed by applying the surface feature model, and the nearest distance from each vertex or sampling point on the surface of the geometric outer envelope to the boom axis is calculated, and the minimum value among them is taken as the second candidate distance value at the acquisition time. The robust feature model is applied to divide the target point cloud cluster into multiple sectors according to the azimuth angle; core points that meet the local point density conditions are selected in each sector, and the nearest distance value between each core point and the boom axis is calculated; quantile statistics or median statistics are performed on the nearest distance values ​​between each core point and the boom axis in each sector to obtain the representative distance value of each sector; the minimum value among the representative distance values ​​of each sector is taken as the third candidate distance value at the acquisition time.

8. The anti-collision method for gantry crane boom based on millimeter-wave radar according to claim 1, characterized in that, The step of performing multi-frame temporal filtering and robust estimation on the minimum distance value sequence to obtain the nearest distance assessment value between the target object and the boom includes: The minimum distance value sequence is time-series filtered based on a time sliding window to obtain the distance estimate at the current time. Determine whether the target point cloud cluster is in a low-speed stable state; If the target point cloud cluster is in a low-speed stable state, the nearest distance evaluation value of the previous moment will be used as the nearest distance evaluation value of the current moment. Otherwise, the distance estimate at the current moment is used as the nearest distance assessment at the current moment.

9. The anti-collision method for the boom of a gantry crane based on millimeter-wave radar according to any one of claims 2-8, characterized in that, The step of determining the current safety alarm zone of the target object based on the nearest distance assessment value, and performing anti-collision control based on the safety alarm zone of the target object, includes: If the nearest distance assessment value is greater than the first equivalent radius If the target object is located in a safe zone, no warning signal or control command will be output. If the nearest distance assessment value is less than or equal to the first equivalent radius And greater than the second equivalent radius If the target object is located in the attention zone, a warning signal is output. If the nearest distance assessment value is less than or equal to the second equivalent radius And greater than the third equivalent radius If the target object is located in the deceleration zone, a deceleration control command is output to cause the gantry crane to decelerate. If the nearest distance assessment value is less than or equal to the third equivalent radius If the target object is located in the stopping area, a stopping control command is output to cause the gantry crane to perform an emergency stop.

10. A gantry crane boom anti-collision device based on millimeter-wave radar, characterized in that, The device includes: The construction module is used to construct a dynamic three-dimensional safety protection model of the boom of a gantry crane. The dynamic three-dimensional safety protection model includes a three-dimensional capsule that is dynamically updated based on the real-time posture of the boom, and a multi-level safety alarm area divided by the spatial region of the three-dimensional capsule. The suppression module is used to perform coordinate transformation on the point cloud data acquired by the millimeter-wave radar to obtain initial point cloud data, and based on the dynamic three-dimensional security protection model, to filter out irrelevant points and suppress noise points on the initial point cloud data to obtain effective target point cloud data; wherein, the filtering out irrelevant points and suppressing noise points includes self-echo suppression based on black-and-white lists and evidence accumulation mechanisms. The clustering module is used to perform clustering analysis on the effective target point cloud data using an adaptive density clustering algorithm, and to perform multi-frame temporal consistency verification to determine the target point cloud clusters that exist continuously and stably. The calculation module is used to calculate multiple candidate distance values ​​between the target point cloud cluster and the boom axis based on the linear feature model, the area feature model and the robust feature model for each acquisition time, and select the minimum value from the multiple candidate distance values ​​as the minimum distance value between the target point cloud cluster and the boom axis, forming a sequence of minimum distance values ​​arranged in chronological order; The evaluation module is used to perform multi-frame temporal filtering and robust estimation on the minimum distance value sequence to obtain the nearest distance evaluation value between the target object and the boom, wherein the target object is the object corresponding to the target point cloud cluster; The control module is used to determine the current safety alarm zone of the target object based on the nearest distance assessment value, and to perform anti-collision control based on the safety alarm zone of the target object.

Images