Three-dimensional uncertain risk perception decision method for construction site crane operation
By combining a three-dimensional probabilistic occupancy field model with a digital twin platform, the optimal collision-free path is generated and risks are perceived in real time, which solves the problem of dynamic uncertainty risk in crane operations at construction sites and improves construction safety and system adaptability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUAIYIN INSTITUTE OF TECHNOLOGY
- Filing Date
- 2026-05-09
- Publication Date
- 2026-07-28
AI Technical Summary
Existing crane operation safety protection and path planning technologies cannot effectively handle the dynamic uncertainty risks at construction sites, leading to frequent high-risk incidents that seriously threaten personnel safety and cause economic losses.
By employing a three-dimensional probabilistic occupancy field model combined with a digital twin platform, the three-dimensional environment is reconstructed using multi-source sensor data. The optimal collision-free path is planned using conditional risk value, and real-time risk perception and decision-making are performed through an AR helmet, forming a closed-loop control.
It enables proactive and adaptive control of dynamic and uncertain risks at the construction site, significantly reducing the incidence of high-risk accidents and improving operational safety and system adaptability.
Smart Images
Figure CN122472525A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent safety decision-making and automated crane operation technology at construction sites, and particularly to a three-dimensional uncertainty risk perception and decision-making method for crane operations at construction sites. Background Technology
[0002] Currently, in construction sites such as building engineering, bridge construction, and large equipment installation, cranes are core heavy machinery, and their operational safety is of paramount importance. However, the on-site environment is complex and ever-changing, with two prominent sources of uncertainty and risk: first, the unpredictable swaying of the suspended load due to factors such as wind and inertia; and second, the potential intrusion caused by the dynamic movement of construction personnel and machinery. These uncertainties make the crane operating area a dynamically changing, high-risk zone.
[0003] Existing crane safety protection and path planning technologies are mostly based on deterministic environmental assumptions. For example, traditional obstacle avoidance algorithms typically divide the environment into static safe zones and obstacle zones, failing to effectively handle and represent the aforementioned dynamic and uncertain risks. Their limitations are mainly reflected in:
[0004] (1) Relying on two-dimensional lasers or simple geometric models, it is difficult to reconstruct three-dimensional spatial information in a realistic and comprehensive manner, and it is even more impossible to quantify the possibility of the existence of obstacles;
[0005] (2) The planning algorithm aims to avoid known static obstacles, but lacks the ability to actively avoid tail risks with low probability and high consequences. When an accident occurs, the system reacts slowly.
[0006] The aforementioned technological limitations make it difficult for existing risk perception systems to cope with the uncertainties in real construction scenarios. High-risk events occur frequently, seriously threatening personnel safety and easily causing huge economic losses and project delays. With the continuous improvement of safety requirements and the in-depth promotion of the intelligent construction strategy, the development of a new generation of intelligent safety systems that can proactively perceive, quantify, and make decisions on uncertain risks has become an urgent need for the industry. Summary of the Invention
[0007] Purpose of the invention: To address the above problems, the purpose of this invention is to provide a three-dimensional uncertainty risk perception and decision-making method for crane operations at construction sites.
[0008] Technical solution: The present invention provides a three-dimensional uncertainty risk perception and decision-making method for crane operations at construction sites, comprising the following steps:
[0009] Step 1: Collect multi-source sensor data from the construction site using cameras and radar. Utilize a three-dimensional probabilistic occupancy field model to perform three-dimensional reconstruction of the suspended objects, personnel, and environment at the construction site, obtaining the probability value of the presence of obstacles at each spatial point and its corresponding confidence level.
[0010] Step 2: The digital twin platform receives actual operation data and performs online simulation, dynamically optimizes and updates the three-dimensional probability occupancy field model, and uses the updated three-dimensional probability occupancy field model to generate a three-dimensional probability map.
[0011] Step 3: Based on the 3D probabilistic map, conditional risk value is introduced into the sampling and expansion process of the fast-expanding random tree. The optimization objective is to minimize the tail risk and path length of the space traversed by the path. The optimal collision-free path at the end of the crane load is generated through differentiable optimization.
[0012] Step 4: Transmit the 3D probability map, risk indicators, and optimal collision-free path to the AR helmet, and use the AR helmet to overlay and display the risk heat map, obstacle probability distribution, and recommended path in real time, supporting construction personnel to interactively perceive and make decisions on risk information.
[0013] Step 5: Continuously monitor environmental changes during the operation. When the 3D probabilistic map is updated, trigger path replanning to form a closed-loop decision.
[0014] Preferably, in step 1, two-dimensional image data of the construction site is acquired using a camera, and three-dimensional point cloud data is acquired using radar. The two-dimensional image data and the three-dimensional point cloud data are then time-stamped and spatially registered to form a unified spatiotemporal data stream.
[0015] The registered multi-sensor data are fused and then input into the 3D probabilistic map model, as shown below:
[0016] ,
[0017] in, For spatial coordinates, As the perspective direction, For color, For density, The probability of existence. Confidence level;
[0018] The confidence upper bound for each voxel unit is calculated using the following formula:
[0019] ,
[0020] In the formula, The probability prediction mean exists. For the predicted standard deviation;
[0021] Obtain the mean of the predicted existence probability of each voxel. With confidence upper bound A three-dimensional probability map.
[0022] Preferably, step 2 includes:
[0023] The digital twin platform receives a 3D probabilistic map as the initial virtual environment, receives real-time operation data from the actual crane, performs dynamic scene simulation in the initial virtual environment, and predicts potential risk areas.
[0024] The simulation results are compared with the actual perceived risk events to calculate the prediction bias of the model;
[0025] Based on the prediction deviation, the parameters in the three-dimensional probabilistic occupancy field model are automatically adjusted.
[0026] Preferably, step 3 includes:
[0027] A random tree is constructed from the crane's starting point. When performing random sampling in three-dimensional space, regions with low probability in the three-dimensional probability map are prioritized.
[0028] For each newly generated path node, calculate its own risk, and calculate the cumulative conditional risk value of the entire path back from that node to the root node. Only if the accumulated risk is below the risk constraint threshold Only when the current node is included in the feasible path tree will an initial path from the starting point to the target point that satisfies the risk constraints be obtained;
[0029] The planned initial path is then optimized to be differentiable, with the objective function being:
[0030] ,
[0031] In the formula, The path risk integral. For path length, For the weighting factor;
[0032] By adjusting the risk confidence level and trade-off coefficients It generates paths that adapt to different operational safety standards and efficiency requirements, and uses these paths as the optimal collision-free paths.
[0033] Preferably, step 4 includes:
[0034] The optimal collision-free path and real-time risk heat map are visualized and rendered using an AR headset, while simultaneously receiving risk preference parameters input by the operator through an interactive interface. Adjust the risk constraint threshold.
[0035] Beneficial effects: Compared with the prior art, the significant advantages of this invention are:
[0036] 1. This invention elevates environmental perception from a deterministic description to a probabilistic spatial uncertainty quantification through a three-dimensional probabilistic occupancy field model, providing a formal basis for systems to handle fuzzy problems such as sensor noise, dynamic obstacles, and partial observability in the real world.
[0037] 2. This invention relies on a digital twin platform to realize a continuous learning mechanism based on online simulation and comparison with real data, enabling the system to continuously improve the fidelity of scene representation and the applicability of decision-making models through parameter adaptive optimization, and possessing lifelong learning characteristics.
[0038] 3. This invention adopts a path planning algorithm based on the conditional risk value criterion, realizing a robust decision-making paradigm for tail risk avoidance, shifting from optimizing expected performance to ensuring the safety boundary under worst-case conditions; through an augmented reality interface, situation sharing and interpretable interaction are realized, allowing operators to dynamically adjust risk preference parameters based on experience, forming a collaborative decision-making closed loop of hybrid augmented intelligence.
[0039] 4. This invention enables proactive, adaptive, and evolutionary control of dynamic and uncertain risks in hoisting operations, which fundamentally surpasses traditional passive safety systems based on fixed rules or deterministic maps, and is expected to significantly reduce the incidence of high-risk accidents. Attached Figure Description
[0040] Figure 1 This is a flowchart of the present invention;
[0041] Figure 2 A flowchart for risk perception and decision-making in crane operations at construction sites;
[0042] Figure 3 This is a schematic diagram of digital twins and model updates. Detailed Implementation
[0043] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the embodiments of the present invention, and not all structures.
[0044] In the following description, specific details such as target system architecture and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.
[0045] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.
[0046] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.
[0047] Furthermore, in the description of this application and the appended claims, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0048] References to "one embodiment" or "some embodiments" in this specification mean that one or more embodiments of this application include the target features, structures, or characteristics described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.
[0049] Combination Figure 1 and Figure 2 As shown in this embodiment, the three-dimensional uncertainty risk perception and decision-making method for crane operations at construction sites includes the following steps:
[0050] Step 1: Collect multi-source sensor data from the construction site using cameras and radar. Utilize a three-dimensional probabilistic occupancy field model to perform three-dimensional reconstruction of the suspended objects, personnel, and environment at the construction site, obtaining the probability value of the presence of obstacles at each spatial point and its corresponding confidence level.
[0051] Furthermore, in step 1, two-dimensional image data of the construction site is collected using a camera, and three-dimensional point cloud data is collected using radar. The two-dimensional image data and the three-dimensional point cloud data are time-stamp aligned and spatially registered to form a unified spatiotemporal data stream.
[0052] The registered multi-sensor data are fused and then input into the 3D probabilistic map model, as shown below:
[0053] ,
[0054] in, For spatial coordinates, As the perspective direction, For color, For density, The probability of existence. Confidence level;
[0055] The confidence upper bound for each voxel unit is calculated using the following formula:
[0056] ,
[0057] In the formula, The probability prediction mean exists. For the predicted standard deviation;
[0058] Obtain the mean of the predicted existence probability of each voxel. With confidence upper bound A three-dimensional probability map.
[0059] In one example, at least one fisheye camera and at least one millimeter-wave radar are installed on the crane boom and slewing platform. The fisheye camera is used to acquire hemispherical panoramic optical images of the crane's operating area, while the millimeter-wave radar is used to detect the three-dimensional coordinates and radial velocity of the suspended load, personnel, and equipment. The fisheye camera should be an industrial-grade model with high dynamic range to ensure clear images even under complex lighting conditions such as strong light and shadow. The millimeter-wave radar should be a frequency-modulated continuous wave radar with high point cloud density to accurately detect weakly reflective targets. During installation, the fisheye camera should be placed at the highest point of the crane boom to achieve complete coverage of the operating area; the millimeter-wave radar should be installed in a location with minimal vibration, ensuring its scanning plane covers the hook's range of motion and areas with frequent personnel movement on the ground. Lens distortion correction is performed on the fisheye camera images. Motion distortion compensation and noise filtering are performed on the millimeter-wave radar point cloud. The most crucial aspects are timestamp alignment and spatial registration. Through joint calibration technology, the camera and radar coordinate systems are aligned to the crane's base coordinate system, and image and point cloud data are uploaded to the central processing unit via the vehicle's Ethernet. This transmission method ensures high bandwidth, low latency, and strong anti-interference capabilities, providing synchronous and complementary multi-source perception data for subsequent fusion processing. Operators must wear AR helmets with see-through capabilities. These helmets should maintain a high-speed data connection to the onboard server via WiFi 6 or 5G technology to ensure the real-time nature of the visualized information. This wireless connection allows for free head movement for the operator, avoiding the constraints of cables and making it suitable for large-scale operational scenarios.
[0060] In the 3D probabilistic map model, neural radiation field technology is used to reconstruct and render 3D scenes using images as supervision signals and point clouds as spatial constraints. This 3D probabilistic map model can query the occupancy probability and color of any point in space and generate a dense 3D voxel map with probabilistic significance. The final output is the occupancy probability value of each voxel unit and its uncertainty estimate.
[0061] Step 2: The digital twin platform receives actual operation data and performs online simulation, dynamically optimizes and updates the three-dimensional probability occupancy field model, and uses the updated three-dimensional probability occupancy field model to generate a three-dimensional probability map.
[0062] like Figure 3 As shown, step 2 further includes:
[0063] The digital twin platform receives a 3D probabilistic map as the initial virtual environment and receives real-time operational data from the actual crane (including but not limited to structural posture data, motion control data, basic load parameters, dynamic data of the hoisted object, meteorological data, task instruction data, and operator input data). It then performs dynamic scene simulations (such as simulating hoisted object swinging, personnel movement, and other dynamic scenarios) in the initial virtual environment and predicts potential risk areas.
[0064] The simulation results are compared with the actual perceived risk events to calculate the prediction bias of the model;
[0065] Based on prediction bias, the parameters in the 3D probabilistic occupancy field model are automatically adjusted. Specifically, a continuous, data-driven model performance evaluation and optimization closed loop is established within the digital twin platform. It utilizes perceptual defects exposed by simulation, automatically transforming these defects into targeted training signals for localized, incremental fine-tuning of the massive 3D-POF model. Learning from actual operational errors and experience, it continuously evolves the accuracy and reliability of its environmental understanding.
[0066] A digital twin platform is a deeply customized system designed for self-evolution. It is not a general-purpose commercial simulation software, but a specialized platform that integrates high-fidelity crane and environmental models, possesses online simulation and risk prediction capabilities, and embeds automatic model parameter optimization algorithms. Its core value lies not in its resemblance to a real crane, but in its ability to safely and cost-effectively expose the defects of the perception model in virtual space and drive it to continuously improve itself. This enables the system to become increasingly adaptable to specific construction site environments as operating time increases.
[0067] Step 3: Based on the 3D probabilistic map, conditional risk value is introduced into the sampling and expansion process of the fast-expanding random tree. The optimization objective is to minimize the tail risk and path length of the space traversed by the path. The optimal collision-free path at the end of the crane load is generated through differentiable optimization.
[0068] Furthermore, step 3 includes:
[0069] A random tree is constructed from the crane's starting point. When performing random sampling in three-dimensional space, regions with low probability in the three-dimensional probability map are prioritized.
[0070] For each newly generated path node, calculate its own risk, and calculate the cumulative conditional risk value of the entire path back from that node to the root node. Only if the accumulated risk is below the risk constraint threshold Only when the current node is selected is it included in the feasible path tree, resulting in an initial path from the starting point to the target point that satisfies the risk constraints. Specifically, the risk of a node itself is a probability distribution, which can be queried from a probability map for the node coordinates. By querying its attributes in the three-dimensional probability occupancy field, the predicted mean of the existence probability can be obtained. and the predicted standard deviation and belief in the upper realm The distribution of node risk is entirely determined by Decide;
[0071] The planned initial path is then optimized to be differentiable, with the objective function being:
[0072] ,
[0073] In the formula, The path risk integral. For path length, For the weighting factor;
[0074] By adjusting the risk confidence level and trade-off coefficients It generates paths that adapt to different operational safety standards and efficiency requirements, and uses these paths as the optimal collision-free paths.
[0075] objective function The goal is to unify the sometimes contradictory objectives of minimizing tail risk and minimizing path length into a rigorous mathematical optimization framework. This is achieved by adjusting the risk confidence level of the parameters. Risk-length tradeoff coefficient The system can flexibly adapt to different operational safety standards and efficiency requirements.
[0076] Step 4: Combine the 3D probability map and risk indicators. The optimal collision-free path is transmitted to the AR helmet, and the AR helmet is used to overlay and display the risk heat map, obstacle probability distribution and recommended path in real time, supporting construction personnel to interactively perceive and make decisions on risk information.
[0077] Furthermore, step 4 includes:
[0078] The optimal collision-free path and real-time risk heat map are visualized and rendered using an AR headset, while simultaneously receiving risk preference parameters input by the operator through an interactive interface. Adjust the risk constraint threshold.
[0079] Specifically, the risk information contained in the front-end three-dimensional probability occupancy field is processed through a series of steps, including spatial mapping, risk quantification, and visualization encoding. Then, using computer graphics methods, it is transformed into an intuitive, colorful risk warning overlaid on the real world, serving as a real-time risk heat map.
[0080] Step 5: Continuously monitor environmental changes during the operation. When the 3D probabilistic map is updated, trigger path replanning to form a closed-loop decision.
Claims
1. A three-dimensional uncertainty risk perception and decision-making method for crane operations at construction sites, characterized in that, Includes the following steps: Step 1: Collect multi-source sensor data from the construction site using cameras and radar. Utilize a three-dimensional probabilistic occupancy field model to perform three-dimensional reconstruction of the suspended objects, personnel, and environment at the construction site, obtaining the probability value of the presence of obstacles at each spatial point and its corresponding confidence level. Step 2: The digital twin platform receives actual operation data and performs online simulation, dynamically optimizes and updates the three-dimensional probability occupancy field model, and uses the updated three-dimensional probability occupancy field model to generate a three-dimensional probability map. Step 3: Based on the 3D probabilistic map, conditional risk value is introduced into the sampling and expansion process of the fast-expanding random tree. The optimization objective is to minimize the tail risk and path length of the space traversed by the path. The optimal collision-free path at the end of the crane load is generated through differentiable optimization. Step 4: Transmit the 3D probability map, risk indicators, and optimal collision-free path to the AR helmet, and use the AR helmet to overlay and display the risk heat map, obstacle probability distribution, and recommended path in real time, supporting construction personnel to interactively perceive and make decisions on risk information. Step 5: Continuously monitor environmental changes during the operation. When the 3D probabilistic map is updated, trigger path replanning to form a closed-loop decision.
2. The three-dimensional uncertainty risk perception and decision-making method for crane operations at construction sites according to claim 1, characterized in that, In step 1, two-dimensional image data of the construction site is collected using a camera, and three-dimensional point cloud data is collected using radar. The two-dimensional image data and three-dimensional point cloud data are time-stamp aligned and spatially registered to form a unified spatiotemporal data stream. The registered multi-sensor data are fused and then input into the 3D probabilistic map model, as shown below: ， in, For spatial coordinates, As the perspective direction, For color, For density, The probability of existence. Confidence level; The confidence upper bound for each voxel unit is calculated using the following formula: ， In the formula, The probability prediction mean exists. For the predicted standard deviation; Obtain the mean of the predicted existence probability of each voxel. upper confidence level A three-dimensional probability map.
3. The three-dimensional uncertainty risk perception and decision-making method for crane operations at construction sites according to claim 1, characterized in that, Step 2 includes: The digital twin platform receives a 3D probabilistic map as the initial virtual environment, receives real-time operation data from the actual crane, performs dynamic scene simulation in the initial virtual environment, and predicts potential risk areas. The simulation results are compared with the actual perceived risk events to calculate the prediction bias of the model; Based on the prediction deviation, the parameters in the three-dimensional probabilistic occupancy field model are automatically adjusted.
4. The three-dimensional uncertainty risk perception and decision-making method for crane operations at construction sites according to any one of claims 1 to 3, characterized in that, Step 3 includes: A random tree is constructed from the crane's starting point. When performing random sampling in three-dimensional space, regions with low probability in the three-dimensional probability map are prioritized. For each newly generated path node, calculate its own risk, and calculate the cumulative conditional risk value of the entire path back from that node to the root node. Only if the accumulated risk is below the risk constraint threshold Only when the current node is included in the feasible path tree will an initial path from the starting point to the target point that satisfies the risk constraints be obtained; The planned initial path is then optimized to be differentiable, with the objective function being: ， In the formula, The path risk integral. For path length, For the weighting factor; By adjusting the risk confidence level and trade-off coefficients It generates paths that adapt to different operational safety standards and efficiency requirements, and uses these paths as the optimal collision-free paths.
5. The three-dimensional uncertainty risk perception and decision-making method for crane operations at construction sites according to claim 4, characterized in that, Step 4 includes: The optimal collision-free path and real-time risk heat map are visualized and rendered using an AR headset, while simultaneously receiving risk preference parameters input by the operator through an interactive interface. Adjust the risk constraint threshold.