Highway construction safety early warning system and method based on multi-source perception

Abstract

The present application relates to the field of road traffic control, specifically to a highway construction safety early warning system and method based on multi-source perception. The specific implementation process includes: collecting construction site data, completing space-time registration and data fusion under GIS; deducing mechanical crossing operation area and vehicle driving blind area, generating virtual construction fence; executing target attribute difference, identifying construction personnel and locking site intrusion target; constructing a swinging dynamics model of the suspended object, predicting the time-varying trajectory envelope and performing dynamic intersection detection; identifying space-time trajectory interference nodes and solving man-machine interaction conflict items, triggering a targeted adaptive warning mechanism to output differentiated warning information. The present application realizes the active identification and accurate warning of the interaction risk of man, machine and vehicle in the highway construction scene through multi-source perception fusion and risk warning, effectively improves the warning lag problem caused by fragmented perception, and improves the construction safety guarantee capability and the traffic operation safety level.

Description

Technical Field

[0001] This invention relates to the field of road traffic control technology, specifically to a highway construction safety early warning system and method based on multi-source sensing. Background Technology

[0002] Highway construction is a crucial component of transportation infrastructure development. In highway construction scenarios, such as bridge and interchange construction, the construction area is linearly distributed and highly open, with construction activities overlapping with personnel and vehicle traffic in both time and space. Current highway construction safety management generally employs a combination of manual inspections, video surveillance, and several automated safety auxiliary systems to monitor and manage the construction site. Some systems collect data on the operational status of construction machinery, others record personnel entry / exit or location information, and still others perform environmental monitoring or intrusion detection in specific areas. These technologies have improved construction safety to some extent and have been applied in traffic safety management practices.

[0003] However, existing technologies have inherent limitations in practical applications. They are often designed around a single monitoring object or function, with different systems typically operating independently, lacking a unified data organization and correlation mechanism. This makes it difficult to comprehensively reflect the overall changes in the status of personnel, equipment, and the environment at the construction site. In scenarios with complex construction activities and variable traffic conditions, this decentralized information acquisition method is not conducive to forming a complete and continuous understanding of the safety situation. Existing safety monitoring and early warning methods mostly rely on pre-set fixed rules or static threshold conditions, and their applicability is limited when construction stages change, work methods are adjusted, or external traffic conditions change. When the construction site exhibits dynamic changes, existing technologies struggle to reflect potential safety risks in a timely and accurate manner, resulting in insufficient flexibility and foresight in safety management.

[0004] In summary, due to the significant uncertainty of the highway construction site environment, the distribution of personnel, vehicles, and equipment changes continuously over time. Faced with complex environments and multiple targets, it is often difficult to balance complete perception with consistent information. The dispersed nature of safety information increases the difficulty of comprehensively assessing and issuing early warnings regarding the construction traffic safety situation.

[0005] To address this, a highway construction safety early warning system and method based on multi-source perception were proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a highway construction safety early warning system and method based on multi-source sensing, which realizes safety early warning for highway construction through multi-source sensing.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A highway construction safety early warning system based on multi-source sensing includes: The data processing module collects real-time data on mechanical motion vectors, wind load parameters, personnel identification beacons, and radar point cloud clusters at the construction site, and delineates virtual construction fences that include mechanical cross-operation areas and vehicle blind spots based on GIS data. The intrusion identification module uses a target attribute difference algorithm within a virtual construction fence to map the intersection of the physical bounding box generated by the radar point cloud clusters with the personnel identity beacon, and then reversely locks onto the on-site intrusion target without beacon features. The conflict resolution module constructs a dynamic model of the suspended load's swing. Based on the mechanical motion vector and wind load parameters, it calculates the time-varying trajectory envelope of the suspended load within a future time window under inertial action. It then performs dynamic intersection detection between the time-varying trajectory envelope and the spatial positions of adjacent machinery and traffic flow in the mechanical cross-operation area. Based on the detection results, it identifies spatiotemporal trajectory interference nodes and resolves human-machine interaction conflict terms by combining vehicle blind spots. The targeted early warning module triggers a targeted adaptive early warning mechanism based on intruding targets and human-computer interaction conflicts. It projects visual assistance and collision countdown to construction equipment, activates location-based tactile vibration reminders for construction personnel, and triggers directional sound and light alarms for intruding targets.

[0008] Preferably, the specific implementation process of real-time acquisition of mechanical motion vectors, wind load parameters, personnel identification beacons, and radar point cloud clusters at the construction site, and delineation of virtual construction fences including mechanical cross-operation areas and vehicle blind spots based on GIS data, includes: Encoders and tilt sensors deployed on construction machinery read rotation angle and amplitude data in real time, constructing a mechanical motion vector; wind speed and direction sensors installed on the top of the construction machinery boom acquire wind load parameters; ultra-wideband positioning base stations are used to calculate the personnel identification beacons of construction workers; lidar and millimeter-wave radar are driven to acquire radar point cloud clusters; spatiotemporal registration is performed based on a unified GIS, and rigid structures in the static building information of the construction site are imported as feature anchor points to perform rigid body transformation correction on the radar point cloud clusters. Spatial drift caused by metal fence reflection and transient occlusion is suppressed by calculating the Euclidean distance field variance of continuous frame point clouds; mechanical cross-operation areas are deduced using mechanical motion vectors based on the three-dimensional voxels of transport vehicles; a driver's viewpoint is simulated using light projection, and a dynamic view cone is generated by combining vehicle geometry, on-site material stacking height, and road longitudinal slope curvature, and the area not covered by the dynamic view cone is defined as the vehicle driving blind spot; according to the construction stage markers and traffic organization status, the mechanical cross-operation areas and the vehicle driving blind spots are spatially mapped and superimposed to generate a virtual construction fence.

[0009] Preferably, in the virtual construction fence, the specific implementation process of using the target attribute difference algorithm to map the intersection of the physical bounding box generated by the radar point cloud cluster with the personnel identity beacon, and then reversely locking the on-site intruder without beacon features includes: The received radar point cloud clusters are filtered, denoised, and segmented using Euclidean clustering to extract obstacle point cloud sets with independent spatial features and construct corresponding 3D physical bounding boxes. Simultaneously, real-time 3D coordinate data of all personnel identification beacons at the construction site are acquired and mapped to a geographic information system coordinate system unified with the radar data. Spatial location matching is performed using the target attribute difference algorithm to determine whether a personnel identification beacon exists within each physical bounding box. Bounding boxes containing personnel identification beacons are marked as valid targets. The remaining physical bounding boxes that do not match any beacon data are filtered out using set difference logic. After filtering out static environmental interference based on object motion morphology characteristics, the remaining physical bounding boxes are defined as intruding targets on site.

[0010] Preferably, the specific implementation process of constructing a dynamic model of the suspended load's swing, and calculating the time-varying trajectory envelope of the suspended load within a future time window under inertial action based on the mechanical motion vector and wind load parameters, includes: The hoisting machinery control bus reads the wire rope lowering length and the mass of the suspended object, establishing a hoisting oscillation dynamic model based on the Lagrange method. The mechanical motion vector is decomposed into tangential acceleration generated by rotational reversal and normal acceleration generated by amplitude change, which are then input into the hoisting oscillation dynamic model as excitation sources. Based on the wind speed and direction data in the wind load parameters, and combined with the drag coefficient of the hoisting object's windward surface, aerodynamic interference terms are calculated and coupled to the hoisting oscillation dynamic model as non-conservative generalized forces. The coupled hoisting oscillation dynamic model is then discretely iterated to predict the instantaneous displacement sequence of the hoisting object relative to the lifting point. The instantaneous displacement sequence is kinematically superimposed with the real-time geodetic coordinates at the end of the hoisting machinery boom to synthesize the absolute spatial motion path of the hoisting object. Based on the geometric dimensions of the hoisting object, the absolute spatial motion path is subjected to three-dimensional spatial expansion processing to generate a time-varying trajectory envelope defining the potential sweeping area of ​​the hoisting object.

[0011] Preferably, the specific implementation process for dynamically intersecting the time-varying trajectory envelope with the spatial positions of adjacent machinery and traffic flow in the machinery intersection area includes: The time-varying trajectory envelope is uniformly mapped to the global coordinate system of the construction scene based on GIS, and time-series aligned with the adjacent machinery and traffic flow. Dynamic spatial bounding volumes updated over time are generated for the boom structure, slewing range, and operating posture of the adjacent machinery. Continuous motion bounding volumes with time attributes are generated for the traffic flow based on vehicle type, speed, and road topology. Under a unified time axis, continuous spatial overlap determination and nearest-distance evolution analysis are performed on the time-varying trajectory envelope and various bounding volumes to form an intersection criterion describing the change in spatial proximity over time. A conflict information flow containing potential contact time, spatial location, and target category is output based on the intersection criterion.

[0012] Preferably, the specific implementation process of identifying spatiotemporal trajectory interference nodes based on the detection results and calculating human-computer interaction conflict terms in conjunction with vehicle blind spots includes: Based on the conflict information flow output by dynamic intersection detection, the set of spatial positions corresponding to the contact moment is analyzed under a unified time reference framework. Based on the set of spatial positions, a spatiotemporal correlation matrix describing the relative motion relationship between the suspended object, machinery, vehicle, and personnel within the same time slice is constructed. From this matrix, the spatial distance convergence trend and motion direction coupling features are extracted, and spatiotemporal trajectory interference nodes are determined. The spatiotemporal trajectory interference nodes are mapped to the vehicle driving blind zone defined by the vehicle geometry, driver's line of sight occlusion relationship, and road structure constraints. It is determined whether the interference node falls into the effective perception missing area of ​​the corresponding vehicle. If the determination is successful, a human-machine interaction conflict term representing the interaction risk between personnel, machinery, and vehicle is generated by combining the personnel identity beacon and the spatial attributes of the intruding target on site.

[0013] Preferably, the specific implementation process of triggering the targeted adaptive early warning mechanism based on the intruding target and human-computer interaction conflict includes: A unified risk situation coding is applied to the intruding target and human-machine interaction conflict items at the site. The target category, spatial location, approach rate, and associated equipment type are fused to generate risk trigger description information. Based on the risk trigger description information, a targeted adaptive early warning mechanism is triggered to determine the type of early warning object and the corresponding early warning output channel. For construction equipment, the risk trigger description information is converted into visual prompts associated with the equipment's operating status and collision countdown parameters associated with time, and projected onto the equipment's operating interface. For construction personnel, a tactile vibration mode matching the risk level is activated through the personnel identification beacon. For the intruding target at the site, sound and light devices deployed in the construction area are driven to generate directional warning signals. The early warning process is dynamically adjusted as the risk situation is updated to maintain real-time consistency between the safety early warning output and the construction site status.

[0014] A multi-source sensing-based method for early warning of highway construction safety includes: Real-time collection of mechanical motion vectors, wind load parameters, personnel identification beacons, and radar point cloud clusters at the construction site; and delineation of virtual construction fences including mechanical cross-operation areas and vehicle blind spots based on GIS data. In the virtual construction fence, the target attribute difference algorithm is used to map the intersection of the physical bounding box generated by the radar point cloud cluster with the personnel identity beacon, and then reversely lock the on-site intruder without beacon characteristics. A dynamic model of the swing of the suspended object is constructed. Based on the mechanical motion vector and wind load parameters, the time-varying trajectory envelope of the suspended object within the future time window under inertial action is calculated. The time-varying trajectory envelope is dynamically intersected with the spatial positions of adjacent machinery and traffic flow in the mechanical cross-operation area. Based on the detection results, spatiotemporal trajectory interference nodes are identified, and human-machine interaction conflict terms are calculated in combination with vehicle driving blind spots. Based on the intrusion of targets and human-computer interaction conflicts, a targeted adaptive early warning mechanism is triggered, which projects visual assistance and collision countdown to construction equipment, activates positioning tactile vibration reminders for construction personnel, and triggers directional sound and light alarms for intrusion of targets.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention integrates multi-source heterogeneous sensing information such as mechanical motion vectors, wind load parameters, personnel identification beacons, and radar point cloud clusters, and constructs a virtual construction fence that includes mechanical cross-operation areas and vehicle blind spots under a unified GIS spatial benchmark. This enables holistic and continuous perception of the status of personnel, equipment, vehicles, and environment at highway construction sites, improving the problems of fragmented systems, isolated information, and difficulty in forming a complete safety situation awareness.

[0016] 2. This invention introduces a dynamic model of suspended object swing, which couples the mechanical motion state with environmental wind load factors to generate a time-varying trajectory envelope of the suspended object within a future time window. Through dynamic intersection detection and spatiotemporal trajectory interference node calculation, it realizes early warning of potential collision and human-machine interaction risks, breaking through the limitations of relying on static rules or post-event alarms, and improving the initiative and accuracy of construction safety early warning.

[0017] 3. Based on the conflict between the intruder and the human-computer interaction, this invention constructs a targeted adaptive early warning mechanism. It can output differentiated and multimodal early warning information to construction equipment, construction personnel and intruders according to different risk objects and risk forms. This enables the early warning content to be accurately matched with the source of risk, reduces interference from invalid alarms, and improves the timeliness and effectiveness of safety intervention at the construction site. It is suitable for complex application scenarios such as highway construction with strong openness and dynamic changes in operation status. Attached Figure Description

[0018] Figure 1This is a structural diagram of the highway construction safety early warning system based on multi-source sensing proposed in this invention; Figure 2 This is a flowchart of the highway construction safety early warning method based on multi-source sensing proposed in this invention; Figure 3 This is a schematic diagram of the highway construction safety early warning process proposed in this invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It must be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to constitute any limitation on the scope of protection of this invention. Therefore, all equivalent changes or modifications conceived by those skilled in the art based on the content disclosed in this invention without inventive effort should fall within the scope of protection claimed by this invention.

[0020] Reference Figures 1 to 3 This invention provides a highway construction safety early warning system and method based on multi-source sensing, the technical solution of which is as follows: Example 1: This embodiment provides a highway construction safety early warning system based on multi-source sensing, referring to... Figure 1 The system includes a data processing module, an intrusion detection module, a conflict resolution module, and a targeted early warning module.

[0021] The data processing module collects real-time data on mechanical motion vectors, wind load parameters, personnel identification beacons, and radar point cloud clusters at the construction site, and delineates virtual construction fences that include mechanical cross-operation areas and vehicle blind spots based on GIS data. The intrusion identification module uses a target attribute difference algorithm within a virtual construction fence to map the intersection of the physical bounding box generated by the radar point cloud clusters with the personnel identity beacon, and then reversely locks onto the on-site intrusion target without beacon features. The conflict resolution module constructs a dynamic model of the suspended load's swing. Based on the mechanical motion vector and wind load parameters, it calculates the time-varying trajectory envelope of the suspended load within a future time window under inertial action. It then performs dynamic intersection detection between the time-varying trajectory envelope and the spatial positions of adjacent machinery and traffic flow in the mechanical cross-operation area. Based on the detection results, it identifies spatiotemporal trajectory interference nodes and resolves human-machine interaction conflict terms by combining vehicle blind spots. The targeted early warning module triggers a targeted adaptive early warning mechanism based on intruding targets and human-computer interaction conflicts. It projects visual assistance and collision countdown to construction equipment, activates location-based tactile vibration reminders for construction personnel, and triggers directional sound and light alarms for intruding targets.

[0022] Furthermore, the specific implementation process of real-time acquisition of mechanical motion vectors, wind load parameters, personnel identification beacons, and radar point cloud clusters at the construction site, and delineating virtual construction fences including mechanical cross-operation zones and vehicle blind spots based on GIS data, includes: Encoders and tilt sensors deployed on construction machinery read rotation angle and amplitude data in real time, constructing a mechanical motion vector; wind speed and direction sensors installed on the top of the construction machinery boom acquire wind load parameters; ultra-wideband positioning base stations are used to calculate the personnel identification beacons of construction workers; lidar and millimeter-wave radar are driven to acquire radar point cloud clusters; spatiotemporal registration is performed based on a unified GIS, and rigid structures in the static building information of the construction site are imported as feature anchor points to perform rigid body transformation correction on the radar point cloud clusters. Spatial drift caused by metal fence reflection and transient occlusion is suppressed by calculating the Euclidean distance field variance of continuous frame point clouds; mechanical cross-operation areas are deduced using mechanical motion vectors based on the three-dimensional voxels of transport vehicles; a driver's viewpoint is simulated using light projection, and a dynamic view cone is generated by combining vehicle geometry, on-site material stacking height, and road longitudinal slope curvature, and the area not covered by the dynamic view cone is defined as the vehicle driving blind spot; according to the construction stage markers and traffic organization status, the mechanical cross-operation areas and the vehicle driving blind spots are spatially mapped and superimposed to generate a virtual construction fence.

[0023] Specifically, by deploying angle encoders and tilt sensors on the slewing and luffing mechanisms of construction machinery such as tower cranes and crawler cranes, real-time information on the slewing angle, luffing amplitude, and attitude changes of the machinery is collected. The sampling period of the angle encoder is no more than 100 milliseconds, and the angular resolution of the tilt sensor is no less than 0.1 degrees, thereby constructing a continuously updated mechanical motion vector to describe the actual motion state of the construction machinery in three-dimensional space. This mechanical motion vector is aligned with other sensing data in the system using timestamps, providing an accurate motion basis for subsequent work area simulations. Wind speed and direction sensors are installed at the top of the construction machinery boom, with the installation height consistent with the main lifting operation height, to acquire real-time wind speed and direction changes within the construction area. In actual engineering tests, the wind speed and direction sensors can stably collect wind speed changes within the range of 0 to 10 meters per second, maintaining a data update delay of no more than 200 milliseconds even under gust conditions, thus forming wind load parameters reflecting environmental disturbance characteristics and providing environmental input for the dynamic evolution of construction risk areas.

[0024] For personnel detection, several ultra-wideband positioning base stations are deployed within the construction area to perform three-dimensional positioning calculations on the personnel identification beacons worn by construction workers. These beacons output the real-time spatial location of personnel under a unified coordinate reference, with a positioning accuracy controllable within 30 centimeters, and are updated at a frequency not exceeding 200 milliseconds, enabling the system to continuously monitor the movement trajectory of construction workers within the construction area. The personnel identification beacon data, serving as a crucial basis for distinguishing authorized construction personnel from abnormal intruders, is synchronously input into subsequent spatial analysis processes.

[0025] For target environment perception, lidar and millimeter-wave radar are deployed at construction area entrances and exits, vehicle intersections, and locations where visibility is easily obstructed. These lidar and millimeter-wave radars work together to acquire radar point cloud clusters covering the construction area. The radar point cloud clusters contain spatial scattered information about targets such as construction machinery, transport vehicles, construction personnel, and temporary obstacles. Their refresh rate is maintained above 10 Hz in practical applications, meeting the real-time perception requirements of dynamic construction traffic scenarios. To address issues such as metal fencing, multipath reflections, and short-term occlusion at the construction site, the system performs consistency analysis on consecutive frame radar point clouds during the point cloud processing stage.

[0026] In the multi-source data fusion stage, spatiotemporal registration processing is performed based on a unified geographic information system, mapping mechanical motion vectors, wind load parameters, personnel identification beacons, and radar point cloud clusters to the same spatial coordinate system. Static building information from the construction site, such as road structures, bridge piers, supports, and temporary protective facilities, is imported. Rigid structures with long-term stability are used as feature anchor points to perform rigid body transformation correction on the radar point cloud clusters, eliminating the impact of sensor installation errors and attitude drift on spatial consistency. Furthermore, by calculating the Euclidean distance field variance between consecutive frame point clouds, spatial drift of the point cloud caused by metal enclosure reflections or transient occlusion is suppressed, ensuring that the corrected point cloud remains stable and convergent within the static structure region, thereby improving the reliability of the spatial extrapolation results. Based on this, the spatial range that construction machinery may cover under operating conditions such as rotation and amplitude variation is extrapolated using the three-dimensional voxels of transport vehicles and the aforementioned mechanical motion vectors. This generates a mechanical cross-operation area reflecting the spatial overlap under conditions of multiple machines operating simultaneously, which can be updated in real time with changes in machine attitude.

[0027] By simulating the driver's viewpoint using light projection, and combining this with vehicle geometry, material stacking height at the construction site, and road slope and curvature information, a dynamic visual cone is generated that changes with the vehicle's driving state. This dynamic visual cone characterizes the driver's effective field of vision in their current posture, and areas not covered by the dynamic visual cone are defined as blind spots. In actual testing, for large vehicles such as concrete trucks, a stable blind spot forms in the near-front area of ​​the vehicle under the presence of construction barriers. This blind spot dynamically adjusts with changes in road slope and vehicle posture. Finally, based on construction stage markers and real-time traffic organization status, the mechanical intersection area and the vehicle blind spot are spatially mapped and overlaid in a unified spatial coordinate system to generate a virtual construction fence describing the boundary of high-risk construction areas. This virtual construction fence dynamically updates with construction stage transitions, changes in mechanical operation status, and traffic flow adjustments, enabling the system to express the potential traffic safety risk range within the construction area in a structured manner.

[0028] This embodiment can unify and spatially integrate dispersed mechanical, personnel, vehicle, and environmental perception information in the complex traffic scenario of highway construction. It improves the problem that highway construction risk areas rely on manual experience for delineation and are difficult to adjust in real time according to the operation status. This enables dynamic modeling of construction traffic risk areas, providing a reliable foundation for subsequent intrusion identification and safety warning, and improving the foresight and accuracy of traffic safety management at the construction site.

[0029] Furthermore, in the virtual construction fence, the specific implementation process of using the target attribute difference algorithm to map the intersection of the physical bounding boxes generated by the radar point cloud clusters with personnel identity beacons, and then reversely locking onto the on-site intruders without beacon features, includes: The received radar point cloud clusters are filtered, denoised, and segmented using Euclidean clustering to extract obstacle point cloud sets with independent spatial features and construct corresponding 3D physical bounding boxes. Simultaneously, real-time 3D coordinate data of all personnel identification beacons at the construction site are acquired and mapped to a geographic information system coordinate system unified with the radar data. Spatial location matching is performed using the target attribute difference algorithm to determine whether a personnel identification beacon exists within each physical bounding box. Bounding boxes containing personnel identification beacons are marked as valid targets. The remaining physical bounding boxes that do not match any beacon data are filtered out using set difference logic. After filtering out static environmental interference based on object motion morphology characteristics, the remaining physical bounding boxes are defined as intruding targets on site.

[0030] Specifically, the radar point cloud cluster contains spatial scattered information of all perceptible targets within the construction site. Its data density and refresh rate are configured according to the scale of the construction site. In practical engineering applications, the point cloud refresh rate can be stably maintained above 10 Hz. To address common noise points, suspended dust, and distant weakly reflective targets in the construction environment, the received radar point cloud cluster undergoes filtering and denoising processing to remove outliers and low-confidence point clouds, thereby improving the stability of subsequent spatial analysis. After filtering, Euclidean clustering is further performed on the point cloud data to aggregate spatially adjacent point clouds with continuous distribution characteristics into independent obstacle point cloud sets.

[0031] After point cloud clustering, a corresponding 3D physical bounding box is constructed for each obstacle point cloud set. This 3D physical bounding box describes the minimum circumscribed range of the target in space, and its size parameters are updated in real time according to the target's shape, reflecting the spatial occupancy of construction personnel, vehicles, mechanical parts, or other temporary obstacles. In actual tests at bridge construction sites, the height of the 3D physical bounding box constructed for a single adult target is typically between 1.5 and 1.8 meters, while the length and width of the bounding box for vehicle targets are significantly larger than those for personnel targets. Real-time 3D coordinate data of all personnel identification beacons within the construction site are simultaneously acquired. These beacons are calculated using ultra-wideband positioning technology, and their coordinate information is updated in the system with a delay of no more than 200 milliseconds and mapped to a geographic information system coordinate system consistent with the radar point cloud data. Through this mapping process, the personnel identification beacon data and the 3D physical bounding boxes generated by the radar point cloud can be directly compared under the same spatial reference, thus avoiding matching errors caused by inconsistencies in coordinate systems.

[0032] After data preparation, a target attribute differential algorithm is introduced to perform spatial location matching operations on the 3D physical bounding boxes generated from the radar point cloud and the personnel identification beacon data. Specifically, when performing spatial location matching, the target attribute differential algorithm first sets a 3D spatial tolerance buffer area for each generated radar 3D physical bounding box based on the radar point cloud positioning accuracy and the system error of the ultra-wideband base station. For example, the bounding box is extended outward by 20 centimeters in each of its length, width, and height directions. The system detects whether the 3D coordinates of the personnel identification beacon fall within this extended tolerance buffer area. If they do, the Euclidean distance between the beacon coordinates and the geometric center point of the bounding box is further calculated, and the bounding box with the smallest distance that meets the tolerance requirement is uniquely bound to the beacon, thereby resolving ownership conflicts when multiple bounding boxes overlap. After completing the traversal and binding of all beacons, set differential logic is used to mark all remaining bounding boxes that are not bound to any beacons as primary suspicious targets. For these initially suspected targets, the system utilizes motion morphology features to filter out static environmental interference. Specifically, it calculates the centroid displacement variance and average velocity vector of the target over ten consecutive time slices. When the average moving speed of a target is consistently below the starting walking speed of 0.3 meters per second, it is identified as a stationary stockpile of materials or fixed facility and eliminated. In actual construction scenarios, this method can effectively filter out bounding boxes corresponding to static structures such as bridge piers, supports, or fences, avoiding misjudgments as intruders. After eliminating static targets, the remaining unmatched 3D physical bounding boxes with dynamic motion characteristics are defined as intruders on site. These intruders may originate from temporary workers not wearing identification beacons, or from vehicles or unrelated pedestrians who have mistakenly entered the construction area.

[0033] This embodiment introduces a target attribute difference algorithm, which can automatically distinguish between valid and invalid targets in highway construction scenarios with multiple targets and complex environments, and accurately lock onto abnormal intrusion behavior without identity beacons. This effectively improves the shortcomings of relying solely on a single positioning system or video recognition, which is difficult to cover non-cooperative targets, thereby enhancing the integrity and intelligence level of traffic safety management at construction sites.

[0034] Furthermore, a dynamic model of the suspended load's swing is constructed. Based on the mechanical motion vector and wind load parameters, the specific implementation process of calculating the time-varying trajectory envelope of the suspended load within a future time window under inertial action includes: The hoisting machinery control bus reads the wire rope lowering length and the mass of the suspended object, establishing a hoisting oscillation dynamic model based on the Lagrange method. The mechanical motion vector is decomposed into tangential acceleration generated by rotational reversal and normal acceleration generated by amplitude change, which are then input into the hoisting oscillation dynamic model as excitation sources. Based on the wind speed and direction data in the wind load parameters, and combined with the drag coefficient of the hoisting object's windward surface, aerodynamic interference terms are calculated and coupled to the hoisting oscillation dynamic model as non-conservative generalized forces. The coupled hoisting oscillation dynamic model is then discretely iterated to predict the instantaneous displacement sequence of the hoisting object relative to the lifting point. The instantaneous displacement sequence is kinematically superimposed with the real-time geodetic coordinates at the end of the hoisting machinery boom to synthesize the absolute spatial motion path of the hoisting object. Based on the geometric dimensions of the hoisting object, the absolute spatial motion path is subjected to three-dimensional spatial expansion processing to generate a time-varying trajectory envelope defining the potential sweeping area of ​​the hoisting object.

[0035] Specifically, the basic operating parameters directly related to the lifting operation are read in real time through the control bus interface inside the lifting machinery. These basic operating parameters include the real-time lowering length of the wire rope, the nominal mass of the load, and the current operating status indicator. The lowering length of the wire rope is used to determine the geometric constraint relationship between the load and the lifting point, and the load mass is used to characterize its inertial characteristics during motion. In practical engineering applications, these parameters are typically refreshed by the lifting machinery control system at a cycle of no more than 100 milliseconds, which meets the accuracy requirements for continuous modeling of the lifting process. Based on the parameters of the lowering length of the wire rope and the mass of the load, a swing dynamics model of the load is established to describe its dynamic behavior. Specifically, the construction process involves selecting two orthogonal swing angles of the wire rope relative to the vertical line of gravity as generalized coordinates, and constructing a Lagrangian function that includes the kinetic energy and gravitational potential energy of the load. The motion at the end of the robotic boom is decomposed, where the tangential acceleration is obtained by multiplying the rotational angular acceleration by the current amplitude, and the normal acceleration is the time derivative of the variable amplitude velocity. These two are substituted into the equation of motion as nonholonomic constraints of the system. The fourth-order Runge-Kutta method was used to numerically solve the second-order nonlinear differential equations, thereby iteratively calculating the swing angle and angular velocity of the suspended load at the next moment. After the dynamic model was constructed, the motion state of the construction machinery was introduced as an external excitation input. The mechanical motion state originated from the mechanical motion vector, which contained the spatial motion information generated by the lifting machinery during operations such as slewing and luffing. The mechanical motion vector was decomposed into tangential motion components caused by slewing and normal motion components caused by luffing, and these motion components were transformed into equivalent excitation sources that affect the suspended load system. The dynamic model can not only reflect the natural swing behavior of the suspended load under static lifting point conditions, but also realistically describe the superimposed influence of the lifting machinery operation on the trajectory of the suspended load.

[0036] Meanwhile, wind load parameters from the construction site are incorporated into the dynamic model of the suspended load's sway. These wind load parameters are collected in real time by wind speed and direction sensors deployed within the construction area, with the sampling height matched to the operating height of the suspended load to ensure the representativeness of the environmental data. During model calculations, the aerodynamic disturbance to the load's sway is estimated based on wind speed, wind direction, and the geometric characteristics of the load's windward side, and this disturbance is treated as a non-conservative external force and incorporated into the dynamic model. Through this approach, the model can respond to uncertainties such as gusts and crosswinds in the actual construction environment, avoiding errors caused by trajectory prediction under ideal environmental assumptions alone.

[0037] After coupling mechanical excitation and wind load, discrete-time step calculations are performed on the swing dynamics model of the suspended object to predict its motion state within a preset future time window. This time window can be configured according to construction safety management requirements; in practical engineering cases, it is typically set to 3 to 10 seconds to balance predictive foresight and real-time computation. Through continuous iterative calculations, the system obtains a sequence of instantaneous displacement changes of the suspended object relative to the lifting point. This displacement sequence describes the possible swing amplitude and direction changes of the suspended object within a future timeframe. The relative displacement sequence information of the suspended object is kinematically superimposed with the real-time geodetic coordinates of the end of the crane boom. These end-boom coordinates are provided by the crane's attitude sensing system, and their spatial accuracy meets the requirements of construction-level safety analysis. Through this superposition process, the system converts the relative motion result, originally referenced to the lifting point, into the absolute spatial motion path of the suspended object in the global coordinate system of the construction site, thereby achieving spatial reconstruction of the actual trajectory of the suspended object.

[0038] After obtaining the absolute spatial motion path of the suspended object, a three-dimensional spatial dilation process is performed on the motion path based on the geometric dimensions of the suspended object. The geometric dimensions include the length, width, and height parameters of the suspended object, reflecting the area occupied by the suspended object in space. By dilating the motion path, the system can generate a spatial region covering all potential motion positions of the suspended object within the prediction time window. This spatial region constitutes the time-varying trajectory envelope defining the potential sweep range of the suspended object.

[0039] This embodiment, by introducing dynamic modeling of the swinging load and time-varying trajectory envelope, overcomes the limitations of relying solely on static operating ranges or empirical thresholds for safety assessment, enabling predictive characterization of lifting operation risks. This transforms safety warnings from reactive responses to proactive interventions, improving the accuracy and timeliness of risk identification and response in highway construction scenarios.

[0040] Furthermore, the specific implementation process of dynamically intersecting the time-varying trajectory envelope with the spatial positions of adjacent machinery and traffic flow in the machinery intersection area includes: The time-varying trajectory envelope is uniformly mapped to the global coordinate system of the construction scene based on GIS, and time-series aligned with the adjacent machinery and traffic flow. Dynamic spatial bounding volumes updated over time are generated for the boom structure, slewing range, and operating posture of the adjacent machinery. Continuous motion bounding volumes with time attributes are generated for the traffic flow based on vehicle type, speed, and road topology. Under a unified time axis, continuous spatial overlap determination and nearest-distance evolution analysis are performed on the time-varying trajectory envelope and various bounding volumes to form an intersection criterion describing the change in spatial proximity over time. A conflict information flow containing potential contact time, spatial location, and target category is output based on the intersection criterion.

[0041] Specifically, the global coordinate system is based on high-precision surveying data from the construction site, encompassing spatial information such as road centerlines, access roads, bridge structures, and temporary facilities. This ensures that the trajectory envelope of the hoisted object can be aligned with other dynamic and static elements on site under the same spatial reference. The method uses a timestamp synchronization mechanism to align the time-varying trajectory envelope with the status data of adjacent machinery and traffic flow on the construction site, ensuring that data from different sources participate in subsequent analysis within the same time reference frame, thereby avoiding temporal misjudgments caused by differences in sampling periods.

[0042] After unifying space and time, a dynamic spatial enclosure is constructed for adjacent construction machinery located within the machinery intersection area to describe their operational status. This dynamic spatial enclosure is determined by the boom structure parameters, real-time slewing angle, luffing range, and current operating posture of the construction machinery. Its spatial form is continuously updated over time, reflecting the spatial range that the machinery may occupy during actual operation. In practical engineering applications, this enclosure accurately covers the movement area formed by the tower crane boom, hook, and related structures during slewing and luffing, thus providing a reliable foundation for subsequent spatial intersection analysis. Simultaneously, a continuous motion enclosure with temporal attributes is constructed for the traffic flow within the construction area. This traffic flow is modeled based on vehicle type, real-time speed, and road topology. For example, for large vehicles such as concrete trucks and dump trucks, the enclosure considers not only the vehicle's own geometry but also its direction of travel and braking characteristics to predict its possible location in the near future. In this way, the movement of vehicles within the construction area is no longer considered as discrete location points but is described as a continuous spatially occupied area evolving over time, more realistically reflecting the potential conflict between vehicles and lifting operations.

[0043] Under a unified timeline, continuous spatial overlap determination and nearest-distance evolution analysis are performed on the time-varying trajectory envelope and various enclosing bodies of adjacent machinery and traffic flow. The spatial overlap determination is used to identify whether different spatial bodies overlap within any time slice, while the nearest-distance evolution analysis is used to characterize the trend of the minimum spatial distance between the suspended object's trajectory envelope and other targets over time. Through continuous analysis of the distance change trend, the system can identify the risk state of continuous spatial distance convergence before actual contact occurs, thereby forming an intersection criterion describing the change of spatial proximity over time.

[0044] Based on the aforementioned intersection criteria, a conflict information stream is output, containing the potential contact time, spatial location, and target category. This conflict information stream not only indicates the existence of a potential risk but also explicitly provides the specific time window, spatial coordinates, and target type involved, such as adjacent construction machinery, transport vehicles, or other dynamic targets. In a preferred embodiment, when a tower crane is hoisting a steel component, the system identifies that its hoisting trajectory envelope will converge to a position less than 1 meter in spatial distance with the dynamic enclosure of another tower crane's boom in approximately 4 seconds, and simultaneously detects a concrete transport truck entering the area along a construction access road. The system then generates a conflict information stream containing the aforementioned time and spatial information, providing a direct basis for subsequent risk assessment and early warning triggering.

[0045] This embodiment, by introducing an intersection detection mechanism of time-varying trajectory envelopes and multiple types of dynamic bounding bodies, enables continuous and forward-looking identification of potential conflicts between construction machinery and traffic flow. This effectively improves the accuracy and timeliness of judgment regarding multi-target interaction risks in complex construction scenarios, providing reliable data support and a decision-making basis for safety early warning in highway construction environments.

[0046] Furthermore, the specific implementation process of identifying spatiotemporal trajectory interference nodes based on the detection results and calculating human-computer interaction conflict terms in conjunction with vehicle blind spots includes: Based on the conflict information flow output by dynamic intersection detection, the set of spatial positions corresponding to the contact moment is analyzed under a unified time reference framework. Based on the set of spatial positions, a spatiotemporal correlation matrix describing the relative motion relationship between the suspended object, machinery, vehicle, and personnel within the same time slice is constructed. From this matrix, the spatial distance convergence trend and motion direction coupling features are extracted, and spatiotemporal trajectory interference nodes are determined. The spatiotemporal trajectory interference nodes are mapped to the vehicle driving blind zone defined by the vehicle geometry, driver's line of sight occlusion relationship, and road structure constraints. It is determined whether the interference node falls into the effective perception missing area of ​​the corresponding vehicle. If the determination is successful, a human-machine interaction conflict term representing the interaction risk between personnel, machinery, and vehicle is generated by combining the personnel identity beacon and the spatial attributes of the intruding target on site.

[0047] Specifically, based on the conflict information flow, potential contact events are analyzed within a unified temporal reference framework. The conflict information flow includes the potential contact time, corresponding spatial location, and category information of the targets involved. The system extracts the set of spatial locations corresponding to each potential contact time by performing temporal backtracking and alignment of this information. This set of spatial locations describes the relative spatial distribution of suspended objects, construction machinery structures, passing vehicles, and personnel within the same time slice, thus providing foundational data for subsequent motion relationship analysis.

[0048] After obtaining the set of spatial locations, a spatiotemporal correlation matrix is ​​constructed to characterize the relative motion relationships based on the spatial coordinates and motion state information of multiple targets within a unified time slice. Specifically, the construction process involves first counting the total number N of all identified targets (lifted objects, construction machinery, vehicles, and personnel) within the current time slice, and then establishing an N x N matrix. Each element in this matrix represents the real-time minimum Euclidean distance between the physical bounding boxes of the corresponding two targets. To extract the spatial distance convergence trend, the system maintains a cache of the matrix sequence containing the past five consecutive time slices in memory. The system performs a time-dimension differentiation operation on the element value at that position. If the result is negative and the absolute value is consistently greater than zero, a distance convergence trend is determined. Simultaneously, for motion direction coupling characteristics, the system calculates the dot product of the unit vector of the line connecting the centroids of two targets and their respective current velocity vectors. When the dot product result indicates that the velocity components of both targets are moving towards each other along the connecting line and their relative proximity exceeds 0.5 meters per second, motion direction coupling is confirmed.

[0049] The spatiotemporal correlation matrix comprehensively reflects the spatial distance change trends and movement direction relationships among suspended objects, machinery, vehicles, and personnel. By analyzing the spatial distance evolution within continuous time slices, the system can identify key location points where spatial distances continuously converge and movement directions exhibit significant coupling characteristics. These key location points are defined as spatiotemporal trajectory interference nodes, representing the spatial locations where risk interactions are most likely to occur in the near future. These spatiotemporal trajectory interference nodes are mapped to the vehicle's blind spot, defined by vehicle geometry, driver's line-of-sight occlusion, and road structure constraints. The vehicle blind spot takes into account differences in vehicle type; for example, concrete trucks, dump trucks, or large pump trucks have varying degrees of perception loss due to factors such as cab height, vehicle length, and cargo box structure. Simultaneously, road structure information, such as the width of construction access roads, roadside barriers, and temporary protective facilities, is also incorporated into the blind spot modeling process to reflect the actual visual occlusion situation in the construction scenario. Through this mapping process, the system can determine whether the spatiotemporal trajectory interference node falls within the corresponding vehicle's effective perception loss area.

[0050] If the spatiotemporal trajectory interference node is determined to be located within the vehicle's blind spot, the risk is further refined by combining personnel identification beacon data or the spatial attributes of the intruding target. The personnel identification beacon is used to distinguish between authorized construction personnel and invalid targets, and its real-time spatial location can be correlated with the interference node. When a personnel identification beacon or a dynamic object identified as an intruding target is present near the interference node, the system generates a human-machine interaction conflict term representing the risk of interaction between personnel, machinery, and vehicles. This conflict term not only describes the fact that a risk exists but also includes the type of object involved, its spatial location, and temporal characteristics, which is used to support the triggering of subsequent targeted safety warnings. In a preferred embodiment, when a concrete transport truck enters the work area along the construction access road, the system detects through dynamic intersection detection that the spatial distance between the envelope of the hoisted object's trajectory and the vehicle's moving enclosure tends to converge rapidly after about 3 seconds. Further analysis shows that the spatiotemporal trajectory interference node corresponding to this convergence position is located within the blind spot of the driver's right front side, and radar and identification beacon data confirm that there is a construction worker who has not evacuated in time near this position. Based on this, the system generates corresponding human-computer interaction conflict items and triggers differentiated safety reminders for vehicle drivers and construction workers in subsequent warnings, thereby completing risk intervention before an accident occurs.

[0051] This embodiment, through the spatiotemporal trajectory interference node identification and vehicle driving blind spot coupling analysis mechanism, can accurately identify key interaction scenarios that are imperceptible to drivers but pose significant risks in complex construction traffic environments. It effectively improves the shortcomings of judging solely based on spatial distance thresholds or single target locations, thereby enhancing the pertinence and reliability of human-computer interaction risk identification and providing a technical means with practical engineering value for traffic safety early warning in highway construction scenarios.

[0052] Furthermore, the specific implementation process of triggering the targeted adaptive early warning mechanism based on the intruding target and human-computer interaction conflict includes: A unified risk situation coding is applied to the intruding target and human-machine interaction conflict items at the site. The target category, spatial location, approach rate, and associated equipment type are fused to generate risk trigger description information. Based on the risk trigger description information, a targeted adaptive early warning mechanism is triggered to determine the type of early warning object and the corresponding early warning output channel. For construction equipment, the risk trigger description information is converted into visual prompts associated with the equipment's operating status and collision countdown parameters associated with time, and projected onto the equipment's operating interface. For construction personnel, a tactile vibration mode matching the risk level is activated through the personnel identification beacon. For the intruding target at the site, sound and light devices deployed in the construction area are driven to generate directional warning signals. The early warning process is dynamically adjusted as the risk situation is updated to maintain real-time consistency between the safety early warning output and the construction site status.

[0053] Specifically, a unified risk situation coding is applied to the identified intrusion targets and human-machine interaction conflicts. This risk situation coding uses target category, spatial location, approach rate, and associated equipment type as core elements to fuse and express multi-source risk information. The risk situation coding employs a 64-bit binary data frame structure. The high eight bits are defined as the target category segment, distinguishing object attributes such as construction personnel, transport vehicles, and unidentified intruders. The last sixteen bits are defined as the spatial index segment, using spatial mapping logic to convert the three-dimensional grid coordinates of spatiotemporal interference nodes in the virtual construction fence into a unique linear address code. The middle eight bits are defined as the risk level segment, writing the corresponding urgency code based on the quantification result of the convergence trend between the approach rate value and spatial distance. The low thirty-two bits are defined as the equipment association segment, used to record the unique hardware identification number and current operation status code of the machinery involved in the conflict. During the generation process, the above segmented data are sequentially concatenated through displacement and logical OR operations, ultimately fused into a unique hexadecimal risk feature string. This string is the risk trigger description information, providing a unified input for subsequent targeted early warning decisions.

[0054] Upon obtaining the risk trigger description information, a targeted adaptive early warning mechanism is triggered, automatically determining the corresponding early warning object type and early warning output channel. This mechanism calculates the estimated collision time based on the approach rate and spatial distance, and classifies the risk level into three levels according to this time parameter: Level 1 (low risk) is determined when the estimated collision time is between three and five seconds; Level 2 (medium risk) is determined when it is between 1.5 and three seconds; and Level 3 (high risk) is determined when it is less than 1.5 seconds. For different risk levels, the system calls preset differentiated response parameters: for Level 1 risk, the personnel identification beacon emits intermittent weak vibrations at a frequency of one hertz; for Level 2 risk, the vibration frequency is increased to four hertz and the intensity is increased; for Level 3 risk, a continuous strong vibration mode is activated to instruct emergency avoidance. In multi-object concurrent conflict scenarios, the system executes a priority early warning principle based on vulnerable groups, that is, it prioritizes sending trigger commands to the construction personnel's identification beacon, and then sends braking prompts to the mechanical equipment operating terminal, ensuring that personnel have the longest possible reaction window. The warning is downgraded or lifted only when the real-time calculated risk level decreases and remains at a low threshold for more than one second. When the risk level increases, it is upgraded immediately without delay, thus preventing frequent signal jumps in critical states.

[0055] During the early warning process for construction workers, a tactile vibration pattern matching the risk level is triggered by a personnel identification beacon. This tactile vibration pattern has different vibration frequencies and durations set according to the urgency of the risk, enabling construction workers to effectively perceive risk information even in noisy environments or with limited visibility. In a preferred embodiment, when a construction worker enters a vehicle's blind spot and highly overlaps with a spatiotemporal trajectory interference node, their identification beacon triggers a high-frequency, short-period vibration pattern, prompting them to quickly evacuate the current area, thereby reducing the risk of human-vehicle conflict.

[0056] During the early warning process for intruders, audible and visual devices deployed within the construction area generate directional warning signals. These devices automatically adjust the direction of the sound source and the orientation of the lights based on the location of the intruder, concentrating the warning signal on the area where the intruder is located and avoiding interference with other normal work areas. In a preferred embodiment, when the system detects an unauthorized person entering the hoisting work area without an identification tag, nearby audible and visual devices immediately emit directional sound and flashing lights towards that area, guiding the person away from the danger zone quickly.

[0057] Throughout the early warning process, the targeted adaptive early warning mechanism dynamically adjusts the content and intensity of the early warning output according to the real-time changes in the risk situation. When the risk approaches lessen or the target moves away from the danger zone, the system automatically lowers the early warning level or terminates the early warning output, thereby maintaining real-time consistency between the safety early warning behavior and the actual state of the construction site, and avoiding unnecessary impact on work efficiency caused by continuous high-intensity early warnings.

[0058] This embodiment introduces a targeted adaptive early warning mechanism based on risk situation coding, which effectively transforms complex human-computer interaction risk identification results into executable and perceptible safety intervention measures. This not only improves the pertinence and timeliness of early warning information but also reduces false alarm rates and vigilance fatigue at construction sites. Thus, a closed-loop linkage between risk identification, decision output, and safety intervention is achieved in highway construction scenarios, providing practically valuable early warnings for road traffic construction safety.

[0059] Example 2: Reference Figure 2 This embodiment proposes a highway construction safety early warning method based on multi-source sensing, including: Real-time collection of mechanical motion vectors, wind load parameters, personnel identification beacons, and radar point cloud clusters at the construction site; and delineation of virtual construction fences including mechanical cross-operation areas and vehicle blind spots based on GIS data. In the virtual construction fence, the target attribute difference algorithm is used to map the intersection of the physical bounding box generated by the radar point cloud cluster with the personnel identity beacon, and then reversely lock the on-site intruder without beacon characteristics. A dynamic model of the swing of the suspended object is constructed. Based on the mechanical motion vector and wind load parameters, the time-varying trajectory envelope of the suspended object within the future time window under inertial action is calculated. The time-varying trajectory envelope is dynamically intersected with the spatial positions of adjacent machinery and traffic flow in the mechanical cross-operation area. Based on the detection results, spatiotemporal trajectory interference nodes are identified, and human-machine interaction conflict terms are calculated in combination with vehicle driving blind spots. Based on the intrusion of targets and human-computer interaction conflicts, a targeted adaptive early warning mechanism is triggered, which projects visual assistance and collision countdown to construction equipment, activates positioning tactile vibration reminders for construction personnel, and triggers directional sound and light alarms for intrusion of targets.

[0060] Furthermore, the real-time acquisition, fusion, and virtual construction fence construction of multi-source sensing data specifically involves: Multi-source sensing devices are deployed along the existing main line of the highway and the construction work area at the construction site. Angle encoders, tilt sensors, and mechanical control bus interfaces installed on the lifting machinery acquire real-time data on the machinery's rotation angle, amplitude variation, and operating posture, constructing a continuously updated mechanical motion vector. Simultaneously, ultra-wideband positioning base stations are deployed within the construction area to perform 3D spatial positioning calculations on personnel identification beacons worn by construction workers. LiDAR and millimeter-wave radars are deployed at construction area entrances and exits, vehicle intersections, and locations with potential visual obstructions to continuously collect radar point cloud clusters covering vehicles, personnel, and large components. Wind speed and direction sensors are simultaneously integrated to obtain wind load parameters such as wind speed and direction. After time synchronization, anomaly removal, and noise suppression of the multi-source data, spatiotemporal registration is performed within a unified geographic information system coordinate system. The system imports architectural information such as road structures, bridge components, and temporary protective facilities at the construction site as a static environmental base. Based on the mechanical motion vector, it deduces the mechanical cross-operation area and calculates the vehicle blind spot by combining vehicle geometry, driving direction, and line-of-sight occlusion relationship. Finally, the above areas are spatially superimposed to generate a virtual construction fence that is dynamically updated with the construction stage, thereby realizing a structured expression of the construction risk area.

[0061] Furthermore, within the virtual construction fence, the attributes of on-site targets are differentiated, and intruding targets are locked in reverse. Specifically: After the virtual construction fence is constructed, filtering and spatial clustering are performed on the received radar point cloud clusters to extract target point cloud sets with independent spatial continuity, and a corresponding 3D physical bounding box is constructed for each set. Simultaneously, the real-time 3D coordinates of all personnel identification beacons at the construction site are acquired and mapped to a geographic information system coordinate system consistent with the radar point cloud. By introducing a target attribute difference algorithm, spatial inclusion relationship judgment is performed on each physical bounding box to determine whether personnel identification beacons exist within it. When a beacon mapping point exists inside the bounding box, it is marked as a construction personnel target; when the bounding box does not match any beacons and exhibits continuous displacement characteristics, static environmental interference is further filtered out by combining the movement direction and velocity changes, and the remaining bounding boxes are defined as intruding targets on site.

[0062] Furthermore, the prediction of the time-varying trajectory envelope of the suspended object based on mechanical motion and wind load, and the subsequent spatial conflict detection, are as follows: After target identification, the hoisting machinery control bus reads the wire rope lowering length, load mass, and current operational status parameters to construct a swing dynamics model describing the load's dynamic behavior, using the mechanical motion vector as excitation input. Simultaneously, aerodynamic disturbances are introduced into the model calculation process based on real-time collected wind speed and direction parameters. Through discrete iteration of the model, the instantaneous displacement of the load relative to the lifting point within a future time window is predicted. This displacement sequence is then superimposed with the real-time geodetic coordinates of the hoisting machinery boom end to synthesize the absolute spatial motion path of the load in the construction scenario. Combining the load's geometric dimensions, a three-dimensional spatial dilation process is performed on the motion path to generate a time-varying trajectory envelope describing the potential sweep area of ​​the load. This trajectory envelope is mapped to the global coordinate system of the construction scenario and temporally aligned and dynamically intersected with the dynamic spatial bounding volumes of adjacent machinery and the continuous motion bounding volumes of traffic flow, thus outputting a conflict information flow containing the potential contact time and spatial location.

[0063] Furthermore, the specific steps for resolving human-computer interaction conflicts and triggering targeted adaptive early warning based on spatiotemporal trajectory interference nodes are as follows: Within a unified time reference framework, a set of spatial locations corresponding to potential contact moments is extracted, and a spatiotemporal correlation structure describing the relative motion relationships of the suspended object, machinery, vehicles, and personnel is constructed. From this structure, spatiotemporal trajectory interference nodes with continuously converging spatial distances and coupling trends in motion directions are identified. These interference nodes are further mapped to the vehicle's blind spot, defined by vehicle geometry, driver line-of-sight occlusion, and road structure constraints. When an interference node is determined to be located in an area lacking effective perception, a corresponding human-machine interaction conflict term is generated by combining the spatial attributes of personnel identification beacons or intruding targets. Risk status encoding is applied to the conflict term, triggering a targeted adaptive early warning mechanism that outputs differentiated early warning information to construction equipment, construction personnel, and intruding targets.

[0064] This embodiment effectively improves the problems of delayed risk identification, serious generalization of early warning, and untimely safety intervention under complex traffic conditions in existing highway construction scenarios by using multi-source perception, risk prediction, conflict resolution and targeted early warning processes, thus ensuring both construction safety and traffic efficiency.

[0065] Example 3: This embodiment deploys the aforementioned highway construction safety early warning system and method based on multi-source sensing entirely in the safety early warning center of a construction site, referring to... Figure 3 This enables safety early warning for highway construction.

[0066] In this embodiment, the main line of the highway is closed for construction in sections at night, with three independent but overlapping construction sections arranged longitudinally. Each construction section is equipped with cranes, transport vehicles, and on-site workers, and there is a risk of traffic flow intersections and machinery interference between adjacent sections. Multi-source sensing devices are deployed in each construction section, and cross-section data aggregation and collaborative processing are achieved through a central edge computing node. The multi-source sensing devices include angle encoders and tilt sensors installed on the cranes to continuously collect information on machinery rotation, amplitude changes, and attitude changes; ultra-wideband positioning base stations deployed within the construction area for three-dimensional positioning of construction workers wearing identification beacons; lidar and millimeter-wave radars located at the entrances and exits of the construction area and in the central median strip to stably acquire point cloud data of vehicles, personnel, and components under nighttime conditions; and wind speed and direction sensors to collect information on wind speed, wind direction, and visibility changes. All types of data are collected synchronously using a unified time base and uploaded to the edge computing node in real time.

[0067] During the data fusion phase, the system first performs spatiotemporal registration and consistency constraint processing on multi-source data based on a unified geographic information system coordinate system. By denoising and clustering radar point cloud clusters, multiple target point cloud sets with independent spatial characteristics are extracted, and corresponding 3D physical bounding boxes are constructed. Simultaneously, the 3D coordinates obtained from personnel identification beacons are mapped to the same coordinate system. The system uses a target attribute difference algorithm to determine the spatial correspondence between each physical bounding box and the identification beacon, thereby identifying targets carrying beacons as construction workers and targets without matching beacons but exhibiting continuous movement characteristics as intruders. During a nighttime construction operation, the system detected a private vehicle not included in the construction schedule slowly entering the construction fence area in the second construction section and completed the intruder determination within approximately one second.

[0068] The system acquires data from the crane control bus, including the wire rope lowering length, load mass, and current operating status. It then combines this data with real-time collected mechanical motion vectors and wind load parameters to construct a dynamic model of the load's sway. Discrete iterative calculations are performed on the model to predict the instantaneous displacement of the load within a future time window. This displacement result is then superimposed with the real-time geodetic coordinates of the crane boom end to generate the absolute spatial motion path of the load. Further, considering the load's geometry, the path is spatially expanded to form a time-varying trajectory envelope describing the load's potential sweep range. In practical applications, when nighttime gusts increase to approximately five meters per second, the system predicts a significant increase in the load's lateral sway and generates an expanded trajectory envelope in advance for subsequent risk assessment.

[0069] The time-varying trajectory envelope is mapped to the global coordinate system of the construction scene and dynamically intersected with the dynamic enclosing bodies of machinery and the continuous moving enclosing bodies of traffic flow in adjacent construction sections. For traffic flow, the system generates moving enclosing bodies with time attributes based on vehicle type, average speed, and road topology; for adjacent machinery, it generates spatial enclosing bodies that change over time based on their operating posture. By continuously analyzing the spatial proximity relationships between enclosing bodies under a unified time axis, the system outputs a conflict information flow containing potential contact moments, spatial locations, and target categories. Based on this, the system extracts the set of spatial locations corresponding to key contact moments under a unified time reference framework and constructs a spatiotemporal association structure describing the relative motion relationships of suspended objects, machinery, vehicles, and personnel. By analyzing the spatial distance convergence trend and motion direction coupling characteristics, the system identifies spatiotemporal trajectory interference nodes with significant risk characteristics and maps these nodes to the vehicle driving blind spots defined by vehicle geometry, driver line-of-sight occlusion, and temporary facilities in the construction area. When it is determined that the interference node is located in an area where the vehicle's effective perception is missing and is close to construction personnel or intruders in the target space, the system generates a corresponding human-machine interaction conflict item. During the early warning output phase, the system uniformly codes the risk status of intruding targets and human-machine interaction conflicts on site, and triggers a targeted adaptive early warning mechanism. For construction equipment, the system overlays and displays risk areas and remaining safety time prompts on the equipment operation terminals; for construction personnel, high-intensity tactile vibrations are triggered through identity beacons to enhance nighttime perception; for intruding targets, directional warning signals are emitted through sound and light devices along the construction area to guide them to leave the danger zone as soon as possible.

[0070] This embodiment, under conditions of multiple parallel construction sections and low visibility, effectively improves the problems of discontinuous risk identification and delayed early warning response under nighttime construction and complex traffic organization conditions through multi-source perception, risk prediction, conflict identification, and targeted early warning. Its adaptability and reliability in complex road traffic construction environments provide stable and scalable technical support for highway construction safety management.

[0071] It should be clarified that the embodiments described above are merely exemplary and are intended to aid in understanding the present invention, not to limit it. Those skilled in the art can make various changes and modifications after grasping the core ideas of the present invention. Therefore, the scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A highway construction safety early warning system based on multi-source sensing, characterized in that, include: The data processing module collects real-time data on mechanical motion vectors, wind load parameters, personnel identification beacons, and radar point cloud clusters at the construction site, and delineates virtual construction fences that include mechanical cross-operation areas and vehicle blind spots based on GIS data. The intrusion identification module uses a target attribute difference algorithm within a virtual construction fence to map the intersection of the physical bounding box generated by the radar point cloud clusters with the personnel identity beacon, and then reversely locks onto the on-site intrusion target without beacon features. The conflict resolution module constructs a dynamic model of the suspended load's swing. Based on the mechanical motion vector and wind load parameters, it calculates the time-varying trajectory envelope of the suspended load within a future time window under inertial action. It then performs dynamic intersection detection between the time-varying trajectory envelope and the spatial positions of adjacent machinery and traffic flow in the mechanical cross-operation area. Based on the detection results, it identifies spatiotemporal trajectory interference nodes and resolves human-machine interaction conflict terms by combining vehicle blind spots. The targeted early warning module triggers a targeted adaptive early warning mechanism based on intruding targets and human-computer interaction conflicts. It projects visual assistance and collision countdown to construction equipment, activates location-based tactile vibration reminders for construction personnel, and triggers directional sound and light alarms for intruding targets.

2. The highway construction safety early warning system based on multi-source sensing according to claim 1, characterized in that, The specific implementation process of real-time acquisition of mechanical motion vectors, wind load parameters, personnel identification beacons, and radar point cloud clusters at the construction site, and delineation of virtual construction fences including mechanical cross-operation zones and vehicle blind spots based on GIS data, includes: Encoders and tilt sensors deployed on construction machinery read rotation angle and amplitude data in real time, constructing a mechanical motion vector; wind speed and direction sensors installed on the top of the construction machinery boom acquire wind load parameters; ultra-wideband positioning base stations are used to calculate the personnel identification beacons of construction workers; lidar and millimeter-wave radar are driven to acquire radar point cloud clusters; spatiotemporal registration is performed based on a unified GIS, and rigid structures in the static building information of the construction site are imported as feature anchor points to perform rigid body transformation correction on the radar point cloud clusters. Spatial drift caused by metal fence reflection and transient occlusion is suppressed by calculating the Euclidean distance field variance of continuous frame point clouds; mechanical cross-operation areas are deduced using mechanical motion vectors based on the three-dimensional voxels of transport vehicles; a driver's viewpoint is simulated using light projection, and a dynamic view cone is generated by combining vehicle geometry, on-site material stacking height, and road longitudinal slope curvature, and the area not covered by the dynamic view cone is defined as the vehicle driving blind spot; according to the construction stage markers and traffic organization status, the mechanical cross-operation areas and the vehicle driving blind spots are spatially mapped and superimposed to generate a virtual construction fence.

3. The highway construction safety early warning system based on multi-source sensing according to claim 1, characterized in that, In virtual construction fencing, the specific implementation process of using a target attribute difference algorithm to map the intersection of the physical bounding boxes generated by radar point cloud clusters with personnel identity beacons, and then reversely locking onto unidentified intruders on-site, includes: The received radar point cloud clusters are filtered, denoised, and segmented using Euclidean clustering to extract obstacle point cloud sets with independent spatial features and construct corresponding 3D physical bounding boxes. Simultaneously, real-time 3D coordinate data of all personnel identification beacons at the construction site are acquired and mapped to a geographic information system coordinate system unified with the radar data. Spatial location matching is performed using the target attribute difference algorithm to determine whether a personnel identification beacon exists within each physical bounding box. Bounding boxes containing personnel identification beacons are marked as valid targets. The remaining physical bounding boxes that do not match any beacon data are filtered out using set difference logic. After filtering out static environmental interference based on object motion morphology characteristics, the remaining physical bounding boxes are defined as intruding targets on site.

4. The highway construction safety early warning system based on multi-source sensing according to claim 1, characterized in that, The specific implementation process of constructing a dynamic model of the suspended load's swing, and calculating the time-varying trajectory envelope of the suspended load within a future time window under inertial action based on the mechanical motion vector and wind load parameters, includes: The hoisting machinery control bus reads the wire rope lowering length and the mass of the suspended object, establishing a hoisting oscillation dynamic model based on the Lagrange method. The mechanical motion vector is decomposed into tangential acceleration generated by rotational reversal and normal acceleration generated by amplitude change, which are then input into the hoisting oscillation dynamic model as excitation sources. Based on the wind speed and direction data in the wind load parameters, and combined with the drag coefficient of the hoisting object's windward surface, aerodynamic interference terms are calculated and coupled to the hoisting oscillation dynamic model as non-conservative generalized forces. The coupled hoisting oscillation dynamic model is then discretely iterated to predict the instantaneous displacement sequence of the hoisting object relative to the lifting point. The instantaneous displacement sequence is kinematically superimposed with the real-time geodetic coordinates at the end of the hoisting machinery boom to synthesize the absolute spatial motion path of the hoisting object. Based on the geometric dimensions of the hoisting object, the absolute spatial motion path is subjected to three-dimensional spatial expansion processing to generate a time-varying trajectory envelope defining the potential sweeping area of ​​the hoisting object.

5. The highway construction safety early warning system based on multi-source sensing according to claim 1, characterized in that, The specific implementation process of dynamically intersecting the time-varying trajectory envelope with the spatial positions of adjacent machinery and traffic flow in the machinery intersection area includes: The time-varying trajectory envelope is uniformly mapped to the global coordinate system of the construction scene based on GIS, and time-series aligned with the adjacent machinery and traffic flow. Dynamic spatial bounding volumes updated over time are generated for the boom structure, slewing range, and operating posture of the adjacent machinery. Continuous motion bounding volumes with time attributes are generated for the traffic flow based on vehicle type, speed, and road topology. Under a unified time axis, continuous spatial overlap determination and nearest-distance evolution analysis are performed on the time-varying trajectory envelope and various bounding volumes to form an intersection criterion describing the change in spatial proximity over time. A conflict information flow containing potential contact time, spatial location, and target category is output based on the intersection criterion.

6. The highway construction safety early warning system based on multi-source sensing according to claim 5, characterized in that, The specific implementation process of identifying spatiotemporal trajectory interference nodes based on the detection results and calculating human-computer interaction conflict terms in conjunction with vehicle blind spots includes: Based on the conflict information flow output by dynamic intersection detection, the set of spatial positions corresponding to the contact moment is analyzed under a unified time reference framework. Based on the set of spatial positions, a spatiotemporal correlation matrix describing the relative motion relationship between the suspended object, machinery, vehicle, and personnel within the same time slice is constructed. From this matrix, the spatial distance convergence trend and motion direction coupling features are extracted, and spatiotemporal trajectory interference nodes are determined. The spatiotemporal trajectory interference nodes are mapped to the vehicle driving blind zone defined by the vehicle geometry, driver's line of sight occlusion relationship, and road structure constraints. It is determined whether the interference node falls into the effective perception missing area of ​​the corresponding vehicle. If the determination is successful, a human-machine interaction conflict term representing the interaction risk between personnel, machinery, and vehicle is generated by combining the personnel identity beacon and the spatial attributes of the intruding target on site.

7. The highway construction safety early warning system based on multi-source sensing according to claim 1, characterized in that, The specific implementation process of triggering the targeted adaptive early warning mechanism based on intruding targets and human-computer interaction conflicts includes: A unified risk situation coding is applied to the intruding target and human-machine interaction conflict items at the site. The target category, spatial location, approach rate, and associated equipment type are fused to generate risk trigger description information. Based on the risk trigger description information, a targeted adaptive early warning mechanism is triggered to determine the type of early warning object and the corresponding early warning output channel. For construction equipment, the risk trigger description information is converted into visual prompts associated with the equipment's operating status and collision countdown parameters associated with time, and projected onto the equipment's operating interface. For construction personnel, a tactile vibration mode matching the risk level is activated through the personnel identification beacon. For the intruding target at the site, sound and light devices deployed in the construction area are driven to generate directional warning signals. The early warning process is dynamically adjusted as the risk situation is updated to maintain real-time consistency between the safety early warning output and the construction site status.

8. A method for early warning of highway construction safety based on multi-source sensing, characterized in that, include: Real-time collection of mechanical motion vectors, wind load parameters, personnel identification beacons, and radar point cloud clusters at the construction site; and delineation of virtual construction fences including mechanical cross-operation areas and vehicle blind spots based on GIS data. In the virtual construction fence, the target attribute difference algorithm is used to map the intersection of the physical bounding box generated by the radar point cloud cluster with the personnel identity beacon, and then reversely lock the on-site intruder without beacon characteristics. A dynamic model of the swing of the suspended object is constructed. Based on the mechanical motion vector and wind load parameters, the time-varying trajectory envelope of the suspended object within the future time window under inertial action is calculated. The time-varying trajectory envelope is dynamically intersected with the spatial positions of adjacent machinery and traffic flow in the mechanical cross-operation area. Based on the detection results, spatiotemporal trajectory interference nodes are identified, and human-machine interaction conflict terms are calculated in combination with vehicle driving blind spots. Based on the intrusion of targets and human-computer interaction conflicts, a targeted adaptive early warning mechanism is triggered, which projects visual assistance and collision countdown to construction equipment, activates positioning tactile vibration reminders for construction personnel, and triggers directional sound and light alarms for intrusion of targets.

Images