A load estimation and early warning method for an aerial work platform

Abstract

This invention discloses a method for load estimation and early warning of a working platform, including collecting point cloud data of objects on the platform, constructing a three-dimensional model, then segmenting and identifying object categories, obtaining the density of objects by combining with a predefined density table, calculating the volume of each object based on the three-dimensional model, calculating the weight of each object, and summing them to obtain the total load. Simultaneously, it considers the weighing results of traditional weighing systems on the platform, taking the larger value as the final weighing result. This invention avoids the situation where the actual load is too high due to inaccurate weighing results from traditional weighing systems. It has real-time monitoring capabilities, enabling comparison and verification of detection results with traditional data, ensuring the reliability of load data, and providing effective protection for the safety of platform use.

Description

Technical Field

[0001] This invention relates to a method for load estimation and early warning of a working platform at height, belonging to the field of platform weighing technology. Background Technology

[0002] Traditional weighing systems for aerial platforms, based on deformation detection, primarily monitor the platform's load in real time by integrating load and pressure sensors. These sensors, installed in the platform's support structure, measure the weight on the platform and transmit the data to the control system. The system calculates the total load on the platform based on the real-time data and notifies the operator via a display screen or alarm system, ensuring that overloading does not occur during operation. Furthermore, traditional weighing systems also have overload protection functions; if the load exceeds a preset range, the system automatically issues an alarm or limits further platform elevation to prevent equipment damage or ensure operator safety.

[0003] Despite the many advanced features of traditional weighing systems, some shortcomings remain. For example, the accuracy and reliability of sensors can be affected by environmental factors such as temperature and humidity changes, which can cause fluctuations in weighing data. Secondly, if the system issues false alarms or fails to update load data in a timely manner, operators may overlook potential overload risks. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for estimating and warning the load of a working platform, which aims to ensure the reliability of platform load data and reduce the risk of platform overload.

[0005] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0006] This invention provides a method for load estimation and early warning of aerial work platforms, comprising the following steps:

[0007] Collect point cloud data of objects on the platform and construct a 3D model;

[0008] Based on the three-dimensional model, segmentation is performed to identify the category information corresponding to each object. Based on the category information, the density value corresponding to the object category is obtained from a predefined density table.

[0009] Based on the three-dimensional model, the volume of each object is calculated, and the weight of each object is calculated by combining the density value. The weights of all objects are added together to obtain the total load of the platform, and the first weighing result is obtained; the second weighing result is obtained by weighing the objects on the platform through the platform's preset weighing system.

[0010] The larger of the two weighing results is taken as the final weighing result. If the final weighing result exceeds the maximum weight that the platform can bear, the alarm device will be triggered.

[0011] Furthermore, the point cloud data of objects on the acquisition platform is used to construct a 3D model, specifically including:

[0012] Two 3D cameras are set up diagonally on the platform, and point cloud data from the two cameras are collected using a synchronous triggering mechanism.

[0013] The collected point cloud data undergoes preprocessing such as filtering, denoising, and downsampling to remove redundant noise points and background data, and extracts the 3D point cloud within the working platform area.

[0014] A pass-through filter is applied to the 3D point cloud to remove point clouds outside the working platform space, thereby obtaining a 3D model of the object on the platform.

[0015] Furthermore, the step of acquiring point cloud data from two cameras by diagonally positioning them on the platform and employing a synchronous triggering mechanism includes: calibrating the poses of the two cameras relative to the center of the workspace and obtaining the extrinsic parameter matrix. This includes a rotation matrix R and a translation vector t, which describe the position and orientation of each camera in the global coordinate system.

[0016] After calibration, the point cloud acquired by each camera is transformed from the camera coordinate system to a unified workspace coordinate system through the extrinsic parameter matrix; feature points are extracted and matched on the transformed point cloud to identify key points in the overlapping area. These key points are used to locate the overlapping area, initially align the point cloud, and accelerate the registration of the point cloud.

[0017] The iterative nearest point algorithm is used to accurately register the point cloud and minimize the error in the overlapping area;

[0018] Iterative optimization aligns the point clouds fused from the two cameras in the global coordinate system.

[0019] Furthermore, the transformation formula for converting the point cloud acquired by each camera from the camera coordinate system to a unified workspace coordinate system using the extrinsic parameter matrix is ​​as follows:

[0020] ;

[0021] In the formula, R is the extrinsic parameter matrix. The obtained rotation matrix, t is based on the extrinsic matrix The obtained translation vector, Let these be the coordinates of the point cloud in the workspace coordinate system. The coordinates of the point cloud in the camera coordinate system.

[0022] Furthermore, the iterative nearest-point algorithm is used to accurately register the point cloud and minimize the error in the overlapping region. The optimization formula is as follows:

[0023] ;

[0024] In the formula, R is the rotation matrix and t is the translation vector. Let the coordinates of the target point cloud be in the workspace coordinate system. Let the coordinates of the point cloud to be registered be in the workspace coordinate system. This indicates the number of valid matching point pairs between the target point cloud and the point cloud to be registered.

[0025] Further, the step of segmenting and identifying the category information corresponding to each object based on the 3D model, and obtaining the density value corresponding to the object category from a known density dataset based on the category information, includes:

[0026] The constructed 3D model is transformed into the input format required by the deep network model 3D-UNet; the deep network model 3D-UNet performs convolution and feature extraction on the 3D data to generate a 3D segmentation probability map of the same size as the input data, in which each voxel is assigned a class label;

[0027] After segmentation, the deep network model 3D-UNet assigns the category information corresponding to each object on the platform in three-dimensional space; combined with the object category, it looks up the density value of the object of that category from a predefined density table.

[0028] Further, the step of calculating the volume of each object based on the three-dimensional model and calculating the weight of each object by combining the density value includes:

[0029] Normal vector estimation is performed on the input point cloud. The normal vector of each point is calculated by fitting a local plane using the K-neighborhood, generating the normal vector field of the point cloud. An implicit surface model is constructed using the normal vector field of the point cloud, and the consistency of the normal vector direction is corrected by global optimization of the implicit surface model.

[0030] Point clouds are embedded into a multi-resolution octree structure, and the spatial distribution of the mesh structure is generated through hierarchical voxelization to ensure that the distribution of point clouds in the octree structure meets the closure requirement.

[0031] Based on the multi-resolution support, closure guarantee, and point cloud normal vectors provided by the mesh structure, the Poisson equation is solved. ,in It is an implicit function that defines the surface of an object. It is a vector field obtained by interpolating the point cloud normal vectors. Used to control the smoothness of curved surfaces This is used to guide the shape of the surface using the normal vector field, to obtain the zero isosurface of the implicit function, and then the Marching Cubes volume rendering algorithm is used to extract the zero isosurface from the implicit surface model to generate a triangular mesh.

[0032] Furthermore, the step of calculating the volume of each object based on the three-dimensional model and calculating the weight of each object by combining the density value also includes: performing mesh repair based on the obtained triangular mesh;

[0033] The volume of each object is calculated using the voxelization accumulation method after the repaired mesh is converted into a voxel mesh, and the number of occupied voxels is counted. The volume is ,in, It is the volume of the object. It is the volume of the voxel mesh. The weight is calculated by finding the object category density ρ of the segmentation results and combining it with the volume V. This completes the calculation of the volume and weight of each object.

[0034] Furthermore, the mesh repair includes:

[0035] Geometric methods were used to fit local surfaces and fill holes; a mesh smoothing algorithm was used to remove sharp areas; and self-intersecting regions were repaired through noise reduction and topology optimization.

[0036] Furthermore, based on the three-dimensional models of each object, the corresponding three-dimensional coordinate information is obtained, and the load distribution of the platform is obtained with the center position of the object as the center of gravity.

[0037] Formula used: Calculate the sum of the moments of each object's weight relative to the platform connection point between the work platform and the support platform; where: This is the sum of the torques generated by the gravity of the objects on the platform. For the weight of each object, This is the distance between the object's center of gravity and the point where it connects to the platform.

[0038] When the torque exceeds the preset limit threshold, the platform will output an alarm message cloud.

[0039] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0040] This invention provides a method for load estimation and early warning of aerial work platforms. It collects point cloud data of objects on the platform to construct a 3D model. Then, by segmenting and identifying object categories and combining this data with a predefined density table, the corresponding object density is obtained. Simultaneously, the volume of each object is calculated based on the 3D model, and the weight of each object is calculated and summed to obtain the total load. This invention can perform corresponding weight calculations for different objects on the platform, resulting in more accurate estimation results. Furthermore, this invention considers the weighing results of traditional weighing systems on the platform. By taking the larger value as the final weighing result, it effectively balances the error between the estimated result and traditional weighing, avoiding situations where the actual load is too high due to inaccurate weighing results from traditional systems. This effectively improves the reliability of the weighing results and provides effective protection for the safety of platform use. Attached Figure Description

[0041] Figure 1 This is a side view of the working platform of the load estimation and early warning method for elevated work platforms provided in an embodiment of the present invention;

[0042] Figure 2 This is a top view of the work platform for the load estimation and early warning method for high-altitude work platforms provided in this embodiment of the invention;

[0043] Figure 3 This is a flowchart of camera installation and 3D model construction for the load estimation and early warning method for elevated work platforms provided in this embodiment of the invention;

[0044] Figure 4 This is a schematic diagram of the 3D-UNet network structure of the load estimation and early warning method for elevated work platforms provided in this embodiment of the invention;

[0045] Figure 5 This is a deep learning recognition flowchart of the load estimation and early warning method for elevated work platforms provided in this embodiment of the invention;

[0046] Figure 6 This is a flowchart of the cumulative load calculation process for the load estimation and early warning method for elevated work platforms provided in this embodiment of the invention;

[0047] Figure 7 This is a flowchart of the platform load power failure restart process of the load estimation and early warning method for aerial work platforms provided in this embodiment of the invention;

[0048] Figure 8 This is a test flowchart of the RGB-D camera load estimation method for the load estimation and early warning method for elevated work platforms provided in this embodiment of the invention;

[0049] Figure 9 This is a flowchart of the real-time monitoring and alarm process for the load estimation and early warning method for elevated work platforms provided in this embodiment of the invention;

[0050] Figure 10 This is a schematic diagram of the weight dynamic calculation mechanism of the load estimation and early warning method for the aerial work platform provided in the embodiment of the present invention;

[0051] Figure 11 This is a schematic diagram of the calculation process of torque sum in the load estimation and early warning method for aerial work platforms provided in this embodiment of the invention. Detailed Implementation

[0052] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0053] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are used only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0054] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0055] Example 1

[0056] This embodiment introduces a method for load estimation and early warning of aerial work platforms, including the following steps:

[0057] S100: Collect point cloud data of objects on the platform and construct a 3D model.

[0058] Specifically, please refer to Figure 1 , Figure 2 and Figure 3Two 3D (RGB-D) cameras are set up diagonally on the platform. After the cameras are installed, a synchronous triggering mechanism is used to collect point cloud data from the two cameras to ensure the consistency of the point clouds in the time dimension.

[0059] For example, two RGB-D cameras are placed diagonally on the working platform, forming an intersecting viewpoint. This not only increases the overlapping area of ​​the point cloud but also avoids blind spots that would occur with a single camera. Taking a platform 2.4 meters long, 1.2 meters wide, and with a 1.4-meter-high fence as an example, the two RGB-D cameras should be installed at opposite corners of the platform to minimize the impact of the fence on the field of view. The cameras should be slightly tilted towards the center of the platform, with a recommended downward angle between 30° and 45°, so that the fields of view of the two cameras overlap in the middle area of ​​the platform. This design can completely cover the objects on the platform and provide sufficient overlap for subsequent point cloud fusion, facilitating accurate registration.

[0060] The dual-camera system requires precise intrinsic and extrinsic parameter calibration to ensure that the point clouds generated by the two cameras are aligned to the same coordinate system, ultimately achieving high-precision point cloud fusion. In this embodiment, a checkerboard calibration board is used to complete the intrinsic parameter calibration of each camera, obtaining the camera's intrinsic parameter matrix K, which includes the focal length. Principal point coordinates And distortion parameters; by calibrating the poses of the two cameras relative to the center of the workspace, the extrinsic parameter matrix is ​​obtained. This includes a rotation matrix R and a translation vector t, which describe the position and orientation of each camera in the global coordinate system.

[0061] After calibration, the point cloud of each camera is transformed from the camera coordinate system to a unified workspace coordinate system using an extrinsic parameter matrix. The transformation formula is as follows: In the formula, R is the extrinsic parameter matrix. The obtained rotation matrix, t is based on the extrinsic matrix The obtained translation vector, Let these be the coordinates of the point cloud in the workspace coordinate system. The coordinates of the point cloud in the camera coordinate system are given. The initially fused point cloud is then processed using Scale Invariant Feature Transform (SIFT) to extract and match feature points, identifying key points in overlapping regions. The role of these key points is as follows:

[0062] 1. Locating overlapping regions: By extracting SIFT feature points from point clouds from different viewpoints, the overlapping regions of point clouds collected by two cameras can be quickly identified, providing a spatial correlation basis for subsequent registration.

[0063] 2. Provide preliminary alignment: Matched SIFT feature points can establish a correspondence between two point clouds. By calculating the geometric transformation of the feature points (such as rotation and translation), the point clouds are initially aligned, reducing the initial error of subsequent fine registration.

[0064] 3. Accelerate the registration process: SIFT feature point matching provides high-quality initial pose estimation for the ICP (Iterative Closest Point) algorithm, avoiding the problem of ICP getting stuck in local optima or slow convergence due to excessive initial pose deviation.

[0065] After completing SIFT feature point extraction and matching, the ICP algorithm iteratively optimizes the initial alignment results provided by these key points to further refine the point cloud registration. The specific process is as follows:

[0066] Coarse registration stage: The matching results of SIFT feature points are used to calculate a rough rigid body transformation matrix (rotation matrix R and translation vector t) to initially align the two point clouds to the same coordinate system.

[0067] Fine registration stage: The ICP algorithm uses the coarse registration result as the initial value, and gradually optimizes the pose parameters by minimizing the distance between corresponding points in the point cloud, and finally achieves high-precision point cloud alignment.

[0068] Through this process, SIFT feature points play a crucial role throughout the entire point cloud registration process, providing key inputs for subsequent steps and significantly improving the registration efficiency and accuracy of the entire system.

[0069] Then, the Iterative Closest Point (ICP) algorithm is used to accurately register the point cloud, minimizing the error in the overlapping area to meet the accuracy requirements of subsequent volume and weight estimation. The optimized formula is as follows:

[0070] ;

[0071] In the formula, R is the rotation matrix and t is the translation vector. Let the coordinates of the target point cloud be in the workspace coordinate system. Let the coordinates of the point cloud to be registered be in the workspace coordinate system. This represents the number of effectively matched point pairs between the target point cloud and the point cloud to be registered. Through iterative optimization, the point clouds from the two cameras achieve a high degree of consistency in spatial position and attitude in the global coordinate system. The distance between corresponding points in the overlapping area is significantly reduced, and the rotation matrix R and translation vector t for point cloud quality inspection are precisely optimized, enabling seamless fusion of the object surface geometry from both perspectives and eliminating misalignment or ghosting. ICP iteratively optimizes and gradually reduces the average distance between corresponding points between point clouds, eventually converging to a local or global minimum, ensuring that the point clouds fused from the two cameras are aligned in the global coordinate system. There are no obvious misalignments or errors, thus providing complete and high-precision point cloud data for subsequent processing and analysis.

[0072] The collected point cloud data undergoes preprocessing such as filtering, denoising, and downsampling to remove redundant noise points and background data, and the 3D point cloud within the working platform area is extracted.

[0073] A pass-through filter is applied to the 3D point cloud to remove point clouds outside the working platform space, thereby obtaining a 3D model of the object on the platform.

[0074] S200. Based on the 3D model, segment and identify the category information corresponding to each object. Based on the category information, obtain the density value corresponding to the object category from the predefined density table.

[0075] To accurately identify the category attributes of objects on the work platform, please refer to [link / reference]. Figure 4 This invention uses the 3D-UNet deep network model to perform semantic segmentation on processed 3D point clouds to identify the categories of different objects. 3D-UNet is a deep learning model suitable for 3D data, a 3D extension of the classic 2D-UNet. It can effectively utilize 3D contextual information to accurately segment complex 3D structures. The 3D-UNet network architecture consists of an encoder (analysis path) and a decoder (synthesis path), each path containing four resolution levels. The encoder reduces image resolution through successive convolutional and pooling layers, while the decoder gradually restores image resolution through upsampling and convolutional layers. At each level of the decoder, there are shortcut connections from the corresponding encoder level; these connections help the network recover detailed information in the synthesis path. Finally, the network outputs the segmentation result through a 1×1×1 convolutional layer. This structure allows 3D-UNet to effectively learn from sparse annotations and generate accurate 3D image segments.

[0076] Specifically, in this embodiment, the constructed 3D model is first converted into the input format required by the deep network model 3D-UNet; the deep network model 3D-UNet generates a 3D segmentation probability map of the same size as the input data by performing convolution and feature extraction on the 3D data, in which each voxel is assigned a class label.

[0077] After segmentation, each object on the platform is assigned a corresponding category information in 3D space; based on the object category, the density value of that category is retrieved from a predefined density table. For example, for an object identified as a "counterweight," its density can be selected from known densities. Similarly, as long as the model contains conventional carrier densities, the density of real-time loads can be determined in this way.

[0078] It should be added that the deep network model 3D-UNet is trained using a 3D dataset with category labels, and its segmentation accuracy is optimized through supervised learning.

[0079] During the testing phase, the platform typically carries precisely weighed counterweights. Once operational, to facilitate high-altitude operations, personnel often carry demolition and rescue tools to the platform for work at height. Therefore, common object categories that need to be identified include loads, personnel, demolition tools (such as demolition hammers, fire axes, hydraulic shears, glass breakers, and rescue jacks), and rescue equipment (such as stretchers, rescue ropes, and rescue ladders). The table below shows the density values ​​for several common types of loads.

[0080] Table 1 Loading density of common aerial work platforms

[0081] Object categories describe Density value (kg / m³) personnel High-altitude workers 1000 counterweight Counterweight materials for aerial work platforms 7300 Aluminum alloy stretcher High-altitude rescue work 2700 steel pipe Steel pipes used for construction or support 7850

[0082] S300. Based on the 3D model, calculate the volume of each object, and combine it with the density value to calculate the weight of each object. Add up the weights of all objects to get the total load of the platform and obtain the weighing result.

[0083] Specifically, please refer to Figure 5 To calculate the volume and weight of each object separately, the segmentation results of the 3D point cloud need to be processed, and Poisson reconstruction needs to be performed separately for each segmented region.

[0084] An implicit surface model is constructed using the normal vector field of the point cloud, and a continuous closed mesh is generated.

[0085] First, normal vector estimation is performed on the input point cloud. The normal vector of each point is calculated by fitting a local plane using the K-neighborhood, generating the normal vector field of the point cloud. An implicit surface model is constructed using the normal vector field of the point cloud, and the consistency of the normal vector direction is corrected through global optimization of the implicit surface model.

[0086] Then, the point cloud data after denoising, downsampling and registration is embedded into a multi-resolution octree structure. The spatial distribution of the grid structure is generated by hierarchical voxelization to ensure that the distribution of the point cloud in the octree structure meets the closure requirement. The depth of the octree determines the resolution of the final grid.

[0087] Based on the multi-resolution support, closure guarantee, and point cloud normal vectors provided by the mesh structure, the Poisson equation is solved. ,in It is an implicit function that defines the surface of an object. It is a vector field obtained by interpolating the point cloud normal vectors. Used to control the smoothness of curved surfaces This is used to guide the shape of the surface using the normal vector field, and the implicit function zero isosurface is the generated mesh surface. Then, the Marching Cubes algorithm is used to extract the zero isosurface from the implicit surface model to generate a triangular mesh.

[0088] Based on the obtained triangular mesh, mesh repair is performed. To repair holes and irregular areas in the generated mesh, geometric methods can be used to fit local surfaces to fill holes, or topological methods can be used to detect non-closed boundaries and generate closed surfaces. At the same time, mesh smoothing algorithms (such as Laplacian smoothing) are used to remove sharp areas, and noise reduction and topology optimization are used to repair self-intersecting regions to ensure the closure and integrity of the mesh.

[0089] The volume of the repaired mesh was calculated using the voxelization accumulation method. After converting the mesh to a voxel mesh, the number of occupied voxels was counted. The volume is ,in, It is the volume of the object. It is the volume of the voxel mesh. Finally, the weight is calculated by finding the object category density ρ of the segmentation results and combining it with the volume V. This completes the calculation of the volume and weight of each object.

[0090] Each segmented object undergoes the above calculation independently to obtain its corresponding weight value. The sum of the weights of all objects represents the total load of the platform. This process combines the semantic segmentation capabilities of deep learning models with traditional physical calculation methods, seamlessly integrating geometric information and physical properties to achieve accurate load estimation.

[0091] Please see Figure 6 When loading objects or people onto the work platform, it is necessary to use the platform's entrances and exits, and loading is done one item at a time. The loading process is discrete, and the loading status of each object can be easily distinguished using image frame difference processing. Furthermore, after loading is complete, the multiple stacking of objects and the occlusion of personnel make it difficult for an RGB-D camera to fully capture the volume of each object. Therefore, this invention proposes a weight accumulation calculation mechanism. When it is determined that a single object has been stably loaded onto the platform or a single person has been stably positioned on the platform, the volume and weight of the newly added object are calculated through point cloud difference processing, thereby accumulating the total weight. The same principle applies when objects are moved off the platform or personnel are disembarking.

[0092] Furthermore, in traditional weighing systems, load cells are generally used to detect the load weight of the work platform. Therefore, please refer to... Figure 10 Based on the above content, this embodiment incorporates a dynamic weight calculation mechanism in conjunction with a weighing sensor. Specifically, when it is determined that a single object is stably loaded onto / unloaded from the platform or a single person stably stands on / leaves the platform, the volume and weight of the newly added / reduced object are calculated through point cloud subtraction processing, thereby dynamically calculating the real-time total weight.

[0093] It's important to note that, unlike ordinary loading and unloading operations, personnel on aerial work platforms are not stationary. They may move around the platform or transport objects from one location to another. During this process, there are often obstructions between people and objects, potentially leading to inaccurate weight calculations. However, the actual weight does not change under these circumstances. Since aerial work platforms are typically equipped with load cells, when these load cells are functioning normally, the RGB-D camera is activated and dynamic weight calculations are performed only when the load cell readings exceed a preset threshold.

[0094] In addition, please see Figure 7 To avoid data loss after a power outage, resulting in multiple objects being stacked but lacking previously accumulated data, this embodiment proposes a corresponding image data power-off memory method. When the device is powered on, the following process should be followed: when the platform is empty, the empty load result is directly output; when the platform is not empty, point cloud subtraction and deep learning recognition estimation are performed, and the point cloud file and weight value are updated and stored in real time.

[0095] It should be added that in traditional weighing systems, to accurately determine whether a load cell can weigh accurately, multiple precisely weighed counterweights are typically used for load weighing tests. The actual weight of the object is compared with the sensor's weighing value. Only if multiple weighings with different weights are performed accurately can the load cell's weighing value be determined to be accurate. Similarly, please refer to [link to relevant documentation]. Figure 8 To determine the accuracy of the deep network model-based estimation method in this embodiment, a load estimation test should also be performed. Unlike sensors, which only require counterweight weight testing, deep learning load estimation needs to be compared with the actual counterweight mass and the weighing value of a tested weighing sensor. Furthermore, a variety of common objects should be selected for testing to verify the accuracy of deep learning classification and segmentation.

[0096] S400: Combining the traditional weighing system with the weighing results of the object on the platform, the larger value is taken as the final weighing result. If the final weighing result exceeds the maximum weight that the platform can bear, an alarm device is triggered.

[0097] Please refer to Figure 9 This invention enhances safety by analyzing the output of a deep learning model in real time and sending the results to a monitoring platform. It employs a verification method with a traditional weighing system. The weights obtained from both systems are compared, and a pre-set threshold ensures the results are within a reasonable range. In cases where the traditional weighing system is known to be malfunctioning, the weight estimated by this invention is temporarily used. If the difference between the weight estimated by the traditional system and the method of this invention is within a reasonable and acceptable range, the slightly larger weight is used as the final weighing result to ensure safety. Furthermore, if the difference between weighing result one and weighing result two exceeds a preset threshold, a weighing alarm is output. If the final weighing result exceeds the maximum tolerable weight, an alarm device is triggered.

[0098] It should be noted that the warning rules and thresholds in this embodiment can be adjusted according to actual application scenarios to adapt to different monitoring needs. The warning mechanism of this invention can issue timely alarms when data fluctuations exceed the normal range, reminding operators to take appropriate measures.

[0099] Furthermore, in this embodiment, if the conventional weighing device is known to be malfunctioning, the first weighing result will be temporarily used as the final weighing result.

[0100] It's important to note that since the aerial work platform is installed at the top of the boom and connected only to one articulated boom, a platform connection point is formed on the work platform. Load limits are essentially designed to prevent excessive loads from exceeding the boom's connection strength and causing safety risks. However, in reality, besides excessive weight, uneven load distribution also poses safety risks. When the object's weight approaches its limit and is concentrated on one side, the risk is no less than that of a critical weight situation. To limit this, weight torque information is monitored.

[0101] Specifically, please refer to Figure 11 By segmenting the three-dimensional point cloud of the object as described above, in addition to volume information, the corresponding three-dimensional coordinate information is obtained based on the three-dimensional model of each object. Taking the center position of the object as the center of gravity, the load distribution of the platform is obtained.

[0102] Formula used: Calculate the sum of the moments of each object's weight relative to the platform connection point between the work platform and the support platform; where: This is the sum of the torques generated by the gravity of the objects on the platform. For the weight of each object, This is the distance between the object's center of gravity and the point where it connects to the platform.

[0103] When the torque exceeds the preset limit threshold, the platform will output an alarm message.

[0104] In summary, this invention employs multi-view point cloud data fusion and deep learning technology to achieve real-time detection and dynamic monitoring of the load on an aerial work platform. By deploying dual RGB-D cameras diagonally across the platform, RGB images and depth maps are acquired to generate multi-view point cloud data. Complementary perspectives effectively reduce blind spots and avoid object occlusion, ensuring the integrity of the point cloud data. After denoising and downsampling preprocessing, the point cloud data undergoes initial registration using SIFT feature point matching, followed by precise registration using the ICP algorithm. After filtering out non-platform point clouds using pass-through filtering, the U-Net deep learning model is used for semantic segmentation of the point cloud, identifying various objects on the platform and determining their density based on object category. Combining the segmentation results, a continuous mesh model is generated using Poisson reconstruction based on the point cloud normal vectors, and holes and irregular areas are repaired to ensure mesh closure. After obtaining the closed mesh, its volume is calculated using the projection method, and the weight is calculated by combining it with object density, and then superimposed to obtain the total load of the platform. This invention features real-time monitoring capabilities, enabling it to compare and verify detection results with traditional data to ensure the reliability of load data. It also triggers an alarm mechanism when the load exceeds a preset safety threshold, effectively preventing safety accidents.

[0105] Example 2: This example provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in Example 1.

[0106] Example 3: This example provides a computer device, including:

[0107] Memory, used to store computer programs / instructions;

[0108] A processor for executing the computer program / instructions to implement the steps of the method described in Embodiment 1.

[0109] Example 4: This example provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the methods described in Example 1 and above.

[0110] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

[0111] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0113] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0114] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure and not to limit its protection scope. Although this disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading this disclosure, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the invention, but these changes, modifications or equivalent substitutions are all within the protection scope of the pending claims.

Claims

1. A method for load estimation and early warning of an aerial work platform, characterized in that, Includes the following steps: Collect point cloud data of objects on the platform and construct a 3D model; Based on the three-dimensional model, segmentation is performed to identify the category information corresponding to each object. Based on the category information, the density value corresponding to the object category is obtained from a predefined density table. Based on the three-dimensional model, the volume of each object is calculated, and the weight of each object is calculated by combining the density value. The weights of all objects are added together to obtain the total load of the platform, and the first weighing result is obtained; the second weighing result is obtained by weighing the objects on the platform through the platform's preset weighing system. Combine the first and second weighing results and take the larger value as the final weighing result. If the final weighing result exceeds the maximum weight that the platform can bear, the alarm device will be triggered. The step of segmenting based on the 3D model to obtain category information corresponding to each object, and obtaining density values ​​corresponding to the object categories from a known density dataset based on the category information, includes: The constructed 3D model is transformed into the input format required by the deep network model 3D-UNet; the deep network model 3D-UNet performs convolution and feature extraction on the 3D data to generate a 3D segmentation probability map of the same size as the input data, in which each voxel is assigned a class label; After segmentation, the deep network model 3D-UNet assigns a category label to each object on the platform in three-dimensional space; based on the object category, it looks up the density value of that category of object from a predefined density table; The step of calculating the volume of each object based on the three-dimensional model and calculating the weight of each object by combining the density value includes: Normal vector estimation is performed on the input point cloud. The normal vector of each point is calculated by fitting a local plane using the K-neighborhood, generating the normal vector field of the point cloud. An implicit surface model is constructed using the normal vector field of the point cloud, and the consistency of the normal vector direction is corrected by global optimization of the implicit surface model. Point clouds are embedded into a multi-resolution octree structure, and the spatial distribution of the mesh structure is generated through hierarchical voxelization to ensure that the distribution of point clouds in the octree structure meets the closure requirements. Based on the multi-resolution support, closure guarantee, and point cloud normal vectors provided by the mesh structure, the Poisson equation is solved. ,in It is an implicit function that defines the surface of an object. It is a vector field obtained by interpolating the point cloud normal vectors. Used to control the smoothness of curved surfaces This is used to guide the shape of the surface using the normal vector field, obtain the zero isosurface of the implicit function, and use the Marching Cubes volume rendering algorithm to extract the zero isosurface from the implicit surface model to generate a triangular mesh; The step of calculating the volume of each object based on the three-dimensional model and calculating the weight of each object by combining the density value includes: performing mesh repair based on the obtained triangular mesh; The volume of each object is calculated using the voxel accumulation method after the repaired mesh is converted into a voxel mesh, and the number of occupied voxels is counted. The volume is ,in, It is the volume of the object. It is the volume of the voxel mesh; the weight is calculated by finding the object category density ρ of the segmentation results and combining it with the volume V. This completes the calculation of the volume and weight of each object.

2. The method for load estimation and early warning of a high-altitude work platform according to claim 1, characterized in that, The point cloud data of objects on the acquisition platform is used to construct a 3D model, specifically including: Two 3D cameras are set up diagonally on the platform, and point cloud data from the two cameras are collected using a synchronous triggering mechanism. The collected point cloud data undergoes preprocessing such as filtering, denoising, and downsampling to remove redundant noise points and background data, and extracts the 3D point cloud within the working platform area. A pass-through filter is applied to the 3D point cloud to remove point clouds outside the working platform space, thereby obtaining a 3D model of the object on the platform.

3. The method for load estimation and early warning of a high-altitude work platform according to claim 2, characterized in that, The method involves setting up two 3D cameras diagonally on the platform and acquiring point cloud data from both cameras using a synchronous triggering mechanism. This includes: calibrating the poses of the two cameras relative to the center of the workspace and obtaining the extrinsic parameter matrix. This includes a rotation matrix R and a translation vector t, which describe the position and orientation of each camera in the global coordinate system. After calibration, the point cloud collected by each camera is transformed from the camera coordinate system to a unified workspace coordinate system through the external parameter matrix. Feature points are extracted and matched on the transformed point cloud to identify key points in the overlapping area. The iterative nearest point algorithm is used to accurately register the point cloud and minimize the error in the overlapping area; Through iterative optimization, the point clouds fused from the two cameras are aligned in the global coordinate system.

4. The method for load estimation and early warning of a high-altitude work platform according to claim 3, characterized in that, The formula for transforming the point cloud data acquired by each camera from the camera coordinate system to a unified workspace coordinate system using the extrinsic parameter matrix is ​​as follows: ； In the formula, R is the extrinsic parameter matrix. The obtained rotation matrix, t is based on the extrinsic matrix The obtained translation vector, Let the coordinates of the point cloud be in the workspace coordinate system. The coordinates of the point cloud in the camera coordinate system.

5. The method for load estimation and early warning of a high-altitude work platform according to claim 3, characterized in that, The iterative nearest-point algorithm is used to accurately register the point cloud and minimize the error in the overlapping region. The optimization formula is as follows: ； In the formula, R is the rotation matrix and t is the translation vector. Let the coordinates of the target point cloud be in the workspace coordinate system. Let the coordinates of the point cloud to be registered be in the workspace coordinate system. This indicates the number of valid matching point pairs between the target point cloud and the point cloud to be registered.

6. The method for load estimation and early warning of a high-altitude work platform according to claim 1, characterized in that, The mesh repair process includes: Geometric methods were used to fit local surfaces and fill holes; a mesh smoothing algorithm was used to remove sharp areas; and self-intersecting regions were repaired through noise reduction and topology optimization.

7. The method for load estimation and early warning of a high-altitude work platform according to claim 1, characterized in that, Based on the three-dimensional models of each object, the corresponding three-dimensional coordinate information is obtained, and the load distribution of the platform is obtained with the center of gravity of the object as the center of gravity. Formula used: Calculate the sum of the moments of each object's weight relative to the platform connection point between the work platform and the support platform; where: This is the sum of the torques generated by the gravity of the objects on the platform. For the weight of each object, This is the distance between the object's center of gravity and the point where it connects to the platform. When the torque exceeds the preset limit threshold, the platform will output an alarm message.

Images