Method and system for detecting intrusion of personnel in dangerous area of operation of a bridge crane

Abstract

The present application relates to a kind of bridge crane operation dangerous area personnel intrusion detection method and system, comprising: by sensor, obtain hoisting object real-time position and speed data, adopt deep learning algorithm analysis hoisting object movement trajectory, obtain hoisting object movement trajectory prediction result;According to hoisting object movement trajectory prediction result, in combination with load mass and rope length data, adopt dynamics model to calculate dangerous area boundary, determine dynamic dangerous area range;By image acquisition device, obtain operation area image data, identify the position of personnel in image, obtain personnel position data;According to personnel position data, when personnel has entered dynamic dangerous area range, generate early warning signal, using three-dimensional visualization technology early warning signal is superimposed to real-time image on site, obtain visual warning information, and visual warning information is presented by display device.The present application realizes the dynamic accurate prevention and control of dangerous area by multidimensional technology fusion.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bridge cranes, in particular to a bridge crane operation dangerous area personnel intrusion detection method and system. BACKGROUND

[0002] As an important material handling equipment in modern industrial production, the bridge crane is widely used in workshops, warehouses, wharfs and other places. Its bridge structure spanning above the working area and the characteristic of running along the track enable it to realize the hoisting operation of heavy objects in three-dimensional space. However, this operation mode also brings significant safety hazards: the swing characteristics of the hoisted object, the inertial effect of the load mass and the openness of the operation area make the area below and around the crane become a high-risk area.

[0003] Traditional methods mostly rely on fixed area division or simple sensor detection, but due to the failure to consider the dynamic influence of hoisted object motion state and load parameters, the pre-warning area often does not match the actual dangerous area, resulting in high false alarm rate or increased risk of missing reports. SUMMARY

[0004] Therefore, the technical problem to be solved by the present application is to overcome the problem that the pre-warning area of the bridge crane in the prior art does not match the actual dangerous area, resulting in high false alarm rate or increased risk of missing reports, and to provide a bridge crane operation dangerous area personnel intrusion detection method and system, which realizes dynamic and accurate prevention and control of the dangerous area through multi-dimensional technology integration.

[0005] To solve the above technical problems, the present application provides a bridge crane operation dangerous area personnel intrusion detection method, comprising the following steps:

[0006] Obtain real-time position and speed data of the hoisted object through a sensor, analyze the motion trajectory of the hoisted object using a deep learning algorithm, and obtain a hoisted object motion trajectory prediction result;

[0007] According to the hoisted object motion trajectory prediction result, combine the load mass and rope length data, and use a dynamics model to calculate the dangerous area boundary to determine the dynamic dangerous area range;

[0008] Obtain operation area image data through an image acquisition device, identify the position of personnel in the image, and obtain personnel position data;

[0009] When the personnel has entered the dynamic dangerous area range according to the personnel position data, a warning signal is generated, the warning signal is superimposed on the real-time image on site using three-dimensional visualization technology, and visual warning information is obtained;

[0010] The visual warning information is presented through a display device, data is transmitted using a real-time streaming protocol, and low-delay warning information display is obtained.

[0011] In an embodiment of the present application, real-time position and speed data of the hoisted object are acquired by sensors, a deep learning algorithm is used to analyze the motion trajectory of the hoisted object, and a motion trajectory prediction result of the hoisted object is obtained, including:

[0012] Real-time position and speed data of the hoisted object are collected by sensors and stored as a time series data set;

[0013] The time series data set is preprocessed using data processing technology to obtain standardized data;

[0014] The standardized data is trained using a long short-term memory network to obtain a trajectory prediction model;

[0015] Real-time position and speed data are analyzed according to the trajectory prediction model to obtain predicted trajectory data;

[0016] If the deviation between the predicted trajectory data and the real-time monitoring data exceeds a preset threshold, the predicted trajectory data is corrected using a Kalman filter algorithm to obtain optimized trajectory data;

[0017] The motion trend of the hoisted object is calculated according to the optimized trajectory data to obtain motion trajectory change characteristics;

[0018] The real-time monitoring system is updated by the motion trajectory change characteristics to obtain a hoisted object trajectory prediction result.

[0019] In an embodiment of the present application, according to the hoisted object motion trajectory prediction result, combined with load mass and rope length data, a dynamic dangerous area range is determined by calculating the dangerous area boundary using a dynamics model, including:

[0020] The hoisted object motion trajectory prediction data, load mass data, and rope length data are acquired, and a preprocessing algorithm is used to clean the data to obtain standardized motion parameters;

[0021] According to the standardized motion parameters, a dynamics model is used to calculate the speed distribution characteristics and acceleration changes of the hoisted object at different time points to obtain a dynamic motion state;

[0022] The speed distribution characteristics and acceleration changes are extracted from the dynamic motion state, combined with the rope length data and the rope tension, and the offset range of the hoisted object position is calculated to determine a potential impact area;

[0023] If the potential impact area exceeds a preset area safety factor, a finite element analysis method is used to calculate the dangerous area boundary to obtain a preliminary boundary range;

[0024] According to the preliminary boundary range, the boundary range is adjusted in combination with environmental wind speed data to obtain a final range of the dynamic dangerous area;

[0025] Boundary point coordinates are extracted from the final range of the dynamic dangerous area, a region grid is generated by using a geometric algorithm, and visualized data of the dangerous area is obtained;

[0026] It is judged whether the dynamic dangerous area overlaps with a preset safety area through the visualized data, and if the dynamic dangerous area overlaps with the preset safety area, boundary adjustment parameters of the dynamic dangerous area are generated.

[0027] In an embodiment of the present application, if the range of the dynamic dangerous area changes, the boundary data is reorganized at a high frequency through an edge computing device to obtain a real-time updated dangerous area boundary, including:

[0028] The boundary data of the dangerous area is obtained from the sensor through the edge computing device, data preprocessing technology is used to clean the collected data to obtain a standardized boundary data set;

[0029] If the characteristic value of the standardized boundary data set deviates from a preset threshold value, the data is partitioned through an incremental K-means clustering algorithm to determine a dynamically changing boundary subset;

[0030] According to the dynamically changing boundary subset, a sliding window technology is used to segment the high-frequency data stream to obtain a time series boundary segment;

[0031] The time series boundary segment is reorganized at a high frequency through the edge computing device, and a weighted average method is used to fuse multiple segments of data to generate continuous boundary update data;

[0032] If the continuity of the boundary update data meets a preset condition, the boundary points are smoothed through a geometric interpolation algorithm to obtain a smoothed dangerous area boundary;

[0033] According to the smoothed dangerous area boundary, a rasterization technology is used to map the boundary to a two-dimensional plane to generate a real-time updated dangerous area boundary;

[0034] The generated dangerous area boundary is continuously verified through a real-time monitoring system, an anomaly detection algorithm is used to judge the stability of the boundary data, and a final boundary output is obtained.

[0035] In an embodiment of the present application, image data of a work area is obtained through an image acquisition device, a position of a person in the image is recognized to obtain person position data, including:

[0036] Video stream is obtained from a field camera, and image data is extracted through an image processing module;

[0037] YOLO algorithm is used to detect the target of the image data to obtain preliminary person position data;

[0038] The data analysis module performs coordinate correction on the preliminary personnel location data to determine the precise location coordinates.

[0039] If the location coordinates do not match the preset threshold range, the image data will be re-extracted through the image processing module.

[0040] Based on precise location coordinates, a Kalman filter algorithm is used to track personnel positions and obtain continuous location data.

[0041] The continuous location data is updated through a real-time processing module to obtain dynamic location information;

[0042] Data fusion technology is used to integrate dynamic location information to determine personnel location data.

[0043] In one embodiment of the present invention, when there is insufficient light or interference from obstructions, a multimodal fusion algorithm is used to combine infrared sensor data to obtain highly reliable personnel location data, including:

[0044] Multi-source data input is obtained from personnel location data and infrared sensor data. Data preprocessing techniques are used to clean and format the data to obtain a standardized dataset.

[0045] If the standardized dataset contains noise or missing values, the personnel location data and infrared sensor data are smoothed using the Kalman filter algorithm to obtain a smoothed location dataset.

[0046] Based on the smoothed location dataset, a multimodal fusion algorithm is used to weight and integrate personnel location data and infrared sensor data to obtain a fused location dataset.

[0047] If the deviation of the fused location dataset exceeds a preset threshold in a low-light environment, the fused location dataset is optimized by adjusting the weights of the infrared sensor data to obtain an optimized location dataset.

[0048] Based on the optimized location dataset, the particle filter algorithm is used to dynamically correct the interference from obstructions, and the corrected location dataset is obtained.

[0049] If there is an inconsistency between the corrected location dataset and the historical location data, time series analysis techniques are used to smooth and predict the location data to obtain the predicted location dataset.

[0050] Based on the predicted location dataset, the data is converted into three-dimensional location information through spatial coordinate mapping technology to obtain highly reliable personnel location data.

[0051] In one embodiment of the present invention, when it is determined from personnel location data that a person has entered a dynamic danger zone, an early warning signal is generated. Three-dimensional visualization technology is used to overlay the early warning signal onto a real-time on-site image to obtain visualized early warning information, including:

[0052] Real-time personnel location data is acquired and compared with preset danger zone boundary thresholds to determine whether the personnel have entered the danger zone and obtain the boundary entry status.

[0053] If the boundary entry state is "entering", then an early warning signal is generated through the early warning signal generation algorithm to obtain early warning signal data;

[0054] Acquire real-time images of the scene, and use image preprocessing algorithms to denoise and enhance the real-time images to obtain processed scene images;

[0055] Using 3D visualization technology, the early warning signal data and the processed on-site images are mapped to coordinates to obtain a 3D signal overlay model;

[0056] By using an image data fusion algorithm, the three-dimensional signal superposition model is fused with the processed on-site image to obtain preliminary visual early warning information;

[0057] Based on the initial visual warning information, the lighting and transparency are adjusted through rendering optimization algorithms to obtain the final visual warning information.

[0058] In one embodiment of the present invention, visual warning information is presented through a display device, and data is transmitted using a real-time streaming protocol to obtain low-latency warning information display, including:

[0059] Raw early warning data is obtained through a real-time streaming protocol, and the data stream is processed using the WebRTC protocol to obtain a low-latency data transmission stream.

[0060] Key early warning information is extracted from low-latency data transmission streams, and a timestamp synchronization mechanism is used to ensure the temporal consistency of the early warning information.

[0061] If the temporal consistency of the early warning information meets the preset threshold, the information is divided into blocks using data sharding technology to obtain the block-based early warning data.

[0062] Based on the segmented early warning data, the data is published to the on-site display devices using the MQTT protocol, resulting in rapidly distributed early warning information.

[0063] For rapidly distributed alert information, the data content is parsed using JSON format to determine whether the information meets the compatibility requirements of the display device;

[0064] If the information meets the compatibility requirements of the display device, it will be presented on the on-site display device through streaming rendering technology to obtain real-time visualized early warning information display;

[0065] Based on the real-time visualized early warning information, a feedback loop mechanism is used to collect and display latency data to determine the stability optimization requirements of the transmission protocol.

[0066] In one embodiment of the present invention, based on the display of low-latency warning information, a warning signal is pushed to the operation terminal to obtain real-time operation command feedback, including:

[0067] Acquire data from external sensors and generate early warning signals;

[0068] The warning signal is pushed to the operation terminal through a preset communication protocol;

[0069] The warning signal is analyzed on the operating terminal, and an operating interface is generated for display.

[0070] If the parsed signal strength exceeds the preset threshold, the instruction generation mechanism is triggered to obtain the operation instruction;

[0071] Operation commands are transmitted to the feedback processing module via an encrypted channel to obtain command confirmation;

[0072] Machine learning classification algorithms are used to determine the validity of instructions and generate execution signals.

[0073] Update the system response status based on the execution signal to achieve real-time feedback.

[0074] To address the aforementioned technical problems, this invention also provides a personnel intrusion detection system for hazardous areas of bridge crane operations, comprising:

[0075] The trajectory prediction module acquires real-time position and speed data of the hoisted object through sensors, and uses deep learning algorithms to analyze the motion trajectory of the hoisted object to obtain the predicted motion trajectory result of the hoisted object;

[0076] The dynamic hazard area calculation module calculates the boundary of the hazard area based on the predicted trajectory of the hoisted object, combined with the load mass and rope length data, and determines the range of the dynamic hazard area using a dynamic model.

[0077] The personnel identification module acquires image data of the work area through an image acquisition device, identifies the location of personnel in the image, and obtains personnel location data.

[0078] The early warning generation module generates an early warning signal when it determines that a person has entered a dynamic danger zone based on the personnel location data. It then uses 3D visualization technology to overlay the early warning signal onto the real-time on-site image to obtain visualized early warning information.

[0079] The visualization transmission module presents visual warning information through a display device, and uses a real-time streaming protocol to transmit data, resulting in low-latency warning information display.

[0080] The technical solution of the present invention has the following advantages compared with the prior art:

[0081] This invention discloses a method for detecting personnel intrusion into hazardous areas of bridge crane operations. It acquires real-time data of the hoisted object using sensors, predicts its trajectory using deep learning, and calculates the dynamic hazardous area by combining load and rope information. Simultaneously, a multimodal fusion algorithm is employed to accurately locate personnel on-site, and their positions are compared with the real-time updated hazardous area boundaries. When personnel are detected entering the hazardous area, this invention generates an early warning signal, which is then overlaid onto a real-time on-site image using 3D visualization technology. The warning information is then presented on a display device with low latency and pushed to the operating terminal. This method enables real-time early warning of dynamic hazardous areas during hoisting operations, effectively improving construction safety and reducing the probability of accidents. Attached Figure Description

[0082] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein:

[0083] Figure 1 This is a flowchart of the steps of the method for detecting personnel intrusion into the hazardous area of ​​a bridge crane operation according to the present invention;

[0084] Figure 2 This is a flowchart of the steps in the method for predicting the motion trajectory of a hoisted object according to the present invention;

[0085] Figure 3 This is a flowchart of the steps of the dynamic hazardous area accurate calculation method based on multi-factor fusion of the present invention;

[0086] Figure 4 This is a flowchart of the steps of the real-time dynamic update method for dangerous area boundaries based on edge computing of the present invention;

[0087] Figure 5 This is a flowchart of the steps of the real-time accurate positioning and tracking method for personnel based on multimodal data fusion of the present invention;

[0088] Figure 6 This is a flowchart of the robust personnel positioning method based on multimodal sensor fusion of the present invention;

[0089] Figure 7 This is a flowchart of the steps of the intelligent visualization early warning method based on three-dimensional spatial mapping of the present invention;

[0090] Figure 8This is a flowchart of the steps of the low-latency early warning information display method based on real-time streaming transmission of the present invention;

[0091] Figure 9 This is a flowchart of the steps of the early warning response closed-loop control method based on intelligent decision-making of the present invention;

[0092] Figure 10 This is a structural framework diagram of the personnel intrusion detection system for hazardous areas of bridge crane operations according to the present invention. Detailed Implementation

[0093] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0094] Reference Figure 1 As shown, this invention discloses a method for detecting personnel intrusion in hazardous areas of bridge crane operations, comprising the following steps:

[0095] The real-time position and speed data of the hoisted object are acquired by sensors, and the motion trajectory of the hoisted object is analyzed by deep learning algorithms to obtain the motion trajectory prediction results of the hoisted object;

[0096] By using sensors to collect real-time motion parameters of the hoisted object and performing time-series analysis on historical motion data through deep learning algorithms, the trajectory of the hoisted object in the next few seconds can be accurately predicted. This solves the problem of insufficient prediction of the motion trend of objects by traditional methods.

[0097] Based on the predicted trajectory of the hoisted object, combined with the load mass and rope length data, a dynamic model is used to calculate the boundary of the danger zone and determine the range of the dynamic danger zone.

[0098] By further combining key parameters such as load mass and rope length, and using a dynamic model to calculate the swing envelope area, the delineation of the danger boundary is no longer a simple geometric projection, but an intelligent calculation result that comprehensively considers dynamic factors such as inertial force and centrifugal force, fundamentally changing the crude mode of fixed threshold warning.

[0099] Image data of the work area is acquired by an image acquisition device, and the positions of personnel in the images are identified to obtain personnel position data;

[0100] In terms of personnel detection, machine vision technology is used to locate the coordinates of personnel in the work area in real time. By comparing the spatial relationship with the dynamically calculated boundary of the danger zone, intrusion judgment with millimeter-level accuracy is achieved.

[0101] When it is determined that a person has entered a dynamic danger zone based on the personnel location data, an early warning signal is generated. The early warning signal is then superimposed onto the real-time on-site image using 3D visualization technology to obtain visualized early warning information.

[0102] When intrusion is detected, the system not only generates regular warning signals, but also uses 3D visualization technology to integrate the warning information with real-world images of the scene in augmented reality, enabling operators to intuitively understand the three-dimensional spatial distribution of the danger zone.

[0103] Visual warning information is presented through display devices, and data is transmitted using real-time streaming protocols to achieve low-latency warning information display.

[0104] Using real-time streaming protocols to transmit data can significantly reduce data transmission and processing latency, ensuring that warning information is delivered to the display device within milliseconds. This allows operators or on-site workers to see the danger warning immediately and take quick safety actions such as avoidance or stopping the machine, effectively reducing the probability of collisions or crushing accidents.

[0105] The method of this invention breaks through the mindset of static protection in principle. Through the synergy of motion prediction, dynamic modeling and augmented reality, the safety protection range always changes in sync with the real risks. This avoids the loss of work efficiency caused by over-vigilance and eliminates the safety hazards caused by blind spots in protection. Ultimately, it realizes the technological leap from "passive response" to "active prevention" in crane operation safety protection. It can realize real-time early warning of dynamic dangerous areas in hoisting operations, effectively improve construction safety and reduce the probability of accidents.

[0106] In this embodiment, the prediction of the hoisted object's trajectory is one of the key technical aspects, and its accuracy directly affects the reliability of subsequent hazardous area calculations and early warnings. (Referring to...) Figure 2 As shown, this invention further discloses a method for predicting the motion trajectory of a hoisted object, which features a refined design for the trajectory prediction process and constructs a complete dynamic prediction and correction mechanism, including:

[0107] Data acquisition and storage: The three-dimensional coordinates and velocity vectors of the hoisted object are acquired through high-precision sensors (such as GNSS, UWB or vision sensors) to form a time-stamped sequence dataset. This time-series storage method provides complete historical motion characteristics for subsequent analysis.

[0108] Data preprocessing: Sliding window normalization and outlier removal techniques are used to standardize the raw data, effectively eliminating sensor noise and transient interference, and improving data quality.

[0109] Deep learning model training: Long Short-Term Memory (LSTM) network is introduced as the core prediction model. The unique forget gate and memory unit design of this network structure enable it to learn the short-term fluctuation pattern and long-term trend characteristics of the hoisting object's motion at the same time. Compared with ordinary neural networks or traditional physical models, it is more adaptable to the complex pattern of periodic oscillation and sudden motion commonly seen in crane operations.

[0110] Real-time trajectory prediction: By dynamically adjusting network parameters through the backpropagation algorithm and analyzing real-time position and velocity data, the predicted output continuously approaches the actual motion trajectory, providing a preliminary basis for calculating dangerous areas.

[0111] Prediction deviation detection and correction: The predicted trajectory output by the LSTM model is compared with the actual data monitored by the sensors in real time. When the deviation between the two exceeds the preset threshold (usually set according to the crane's working level and safety standards), the Kalman filter correction mechanism is triggered. This dual verification design has important technical value: Kalman filtering makes the optimal estimate of the predicted trajectory through the state space model. Its recursive calculation characteristics can effectively suppress the error accumulation that may occur in the long-term prediction of the LSTM model. It is particularly suitable for handling motion changes caused by sudden braking or load sway.

[0112] Motion trend analysis: The corrected and optimized trajectory data not only includes the current position information, but also extracts higher-order motion features such as acceleration and jerk through differential operations. These feature vectors are input into the motion trend analysis module to calculate the spatial position envelope range that the hoisted object may reach in the next few seconds.

[0113] System updates and feedback: These dynamically updated prediction results are pushed to the monitoring system to ensure that the calculation and early warning of dangerous areas are always based on the latest and most accurate motion data, providing high-confidence input parameters for the calculation of dangerous areas.

[0114] This multi-level trajectory prediction architecture has produced significant technical effects: the LSTM network solves the nonlinear motion problem that is difficult to model using traditional methods, and can autonomously learn the motion law of the hoisted object from historical data without relying on precise physical equations; while the introduction of Kalman filtering forms a closed-loop system for error correction, ensuring that prediction stability can still be maintained when the sensor fails briefly or the motion state changes abruptly; the synergy between the two enables trajectory prediction to have both the ability of deep learning to recognize complex patterns and the fast response characteristics of filtering algorithms to real-time data.

[0115] From an application perspective, this prediction method can adapt to changes in motion characteristics of cranes of different tonnages and under different lifting conditions. When the load undergoes large-amplitude swings or rapid translation, it can provide reliable trajectory prediction, laying the foundation for the accurate delineation of subsequent dynamic hazard areas. Furthermore, by continuously updating motion trajectory characteristics, the system achieves adaptive optimization of the prediction model. As operating time accumulates, the model gradually adapts to the specific crane's operating habits and load characteristics, forming personalized prediction capabilities, thereby continuously improving the accuracy of the entire safety protection system.

[0116] In actual working conditions, the motion of the hoisted object is affected by various dynamic factors, including changes in load mass, rope length adjustments, and sudden changes in wind speed. These factors can all cause nonlinear changes in the boundary of the danger zone. To address this issue, [the following is a reference to...] Figure 3 As shown, the present invention further discloses a method for accurate calculation of dynamic hazardous areas based on multi-factor fusion, including:

[0117] Data acquisition and preprocessing: The input trajectory prediction data, load mass, and rope length are cleaned and standardized to eliminate measurement errors and outliers, ensuring the reliability of the input parameters.

[0118] Dynamic state calculation: Based on the standardized motion parameters, a multibody dynamics model is used for accurate calculation. This model can simultaneously consider the translational motion, swinging motion of the hoisted object, and the elastic deformation of the rope. By solving the differential equations of motion, the instantaneous velocity and acceleration distribution of the hoisted object at each time point are obtained. Compared with the simple geometric projection method, this dynamic analysis based on physical laws can more accurately reflect the true motion state of the hoisted object during variable speed motion or sudden braking.

[0119] Offset Range Analysis: After obtaining the accurate dynamic motion state, the influence of velocity distribution characteristics and acceleration changes on the position of the hoisted object is further analyzed. By establishing the functional relationship between rope tension and swing angle, the maximum possible offset range of the hoisted object in each direction is calculated. This calculation process takes into account the dynamic effects caused by changes in rope length—longer ropes amplify the swing amplitude, while shorter ropes increase the swing frequency. The calculation parameters are dynamically adjusted to adapt to these changes.

[0120] Finite element boundary calculation: When the calculation results show that the potential impact area exceeds the preset safety factor, a more accurate finite element analysis method is initiated. The hoisting object, rope and its connecting mechanism are discretized into a finite element model. The precise motion trajectory under complex stress conditions is solved by numerical calculation, thereby obtaining a more accurate boundary of the danger zone.

[0121] Environmental factor correction: To further improve the reliability of boundary calculation, the key parameter of ambient wind speed is also introduced. The influence of wind load on the movement of the hoisted object is analyzed through fluid dynamics principles. The wind speed data is converted into equivalent force and input into the dynamic model to realize dynamic adjustment of the boundary of the dangerous area. This multi-factor fusion calculation method can adapt to the operational needs under different weather conditions and ensure that the dangerous area range can be updated in a timely manner when the wind changes suddenly.

[0122] Visual data processing: After completing the boundary calculation, the discrete boundary points are transformed into a continuous regional grid using computational geometry algorithms. This process not only facilitates subsequent visualization but also provides an accurate geometric basis for overlap detection.

[0123] Safety Zone Overlap Detection: When an overlap between a dynamic hazardous area and a preset safe area (such as a fixed equipment area or personnel passageway) is detected, boundary adjustment parameters are automatically generated, and the safety boundary is recalculated through an optimization algorithm to ensure the safe distance between each area.

[0124] This refined method for calculating hazardous areas has yielded significant technical benefits: the introduction of a dynamic model enables the system to accurately capture the physical nature of the hoisting motion, avoiding errors caused by simple geometric projection; the addition of finite element analysis improves the calculation accuracy under complex working conditions, especially when dealing with large mass loads or long rope operations; and the real-time fusion of ambient wind speed enables the system to adapt to changes in the external environment.

[0125] Overall, this method has achieved a leap from static estimation to dynamic and accurate calculation of hazardous areas, enabling the safety protection range to be adjusted in real time according to changes in the operation status. It avoids the loss of operation efficiency caused by overly conservative calculations and prevents safety hazards caused by insufficient calculations, providing a reliable technical guarantee for the intelligent safety protection of bridge cranes.

[0126] In actual operating environments, the motion state of the hoisted object may change abruptly, such as emergency braking, load swaying, or sudden external interference. These situations can cause rapid changes in the boundary of the danger zone. If a traditional centralized calculation method is used, it is difficult to respond to these rapid changes in a timely manner due to data transmission delays and computing resource limitations. This may lead to delayed updates of the danger zone, resulting in a lack of timeliness in safety protection. In addition, drastic fluctuations in boundary data may also trigger frequent false alarms in the early warning system, affecting normal operating efficiency.

[0127] To solve the above problems, refer to Figure 4 As shown, the present invention also provides a method for real-time dynamic updating of dangerous area boundaries based on edge computing, including:

[0128] Boundary data acquisition and preprocessing: Deploy a data preprocessing module on an edge computing device close to the data source to clean and standardize the raw boundary data acquired by the sensors in real time, eliminate measurement noise and outliers, and ensure the quality of the input data.

[0129] Dynamic Change Detection and Partitioning: When the characteristic values ​​of standardized boundary data (such as the rate of curvature change, displacement velocity, etc.) exceed the preset safety threshold, the incremental K-means clustering algorithm is immediately activated. This algorithm can perform dynamic clustering analysis only on newly arrived data without recalculating all the data, quickly identifying boundary subsets that have undergone significant changes, greatly improving the response speed to abrupt changes. Compared with traditional batch clustering, this incremental processing method significantly reduces computational latency, enabling the system to capture local changes in dangerous areas in a timely manner.

[0130] Time-series data segmentation: For high-frequency changing boundary data streams, a sliding window technique is used for time-series segmentation, ensuring that the data segments within each time window maintain temporal and spatial continuity.

[0131] Multi-source data fusion and updating: Edge computing devices process these data fragments in parallel, fusing the calculation results from multiple windows using a weighted average algorithm. Newer data fragments are assigned higher weights to ensure that boundary updates reflect the latest dynamic states in a timely manner. This processing method preserves the trend characteristics of historical data while highlighting the contribution of the latest changes, effectively balancing the needs of data continuity and real-time performance.

[0132] Boundary smoothing optimization: When the continuity of boundary update data meets the preset conditions, a geometric interpolation algorithm is further applied to smooth the discrete boundary points, eliminating the jagged boundaries caused by sensor jitter or calculation fluctuations, and generating a smooth curve that conforms to the laws of physical motion. This smoothing is not a simple mathematical filter, but an intelligent interpolation that combines the kinematic characteristics of the crane, which can maintain the physical rationality of the boundary shape.

[0133] Two-dimensional plane mapping transformation: To facilitate subsequent visualization and spatial analysis, rasterization technology is used to convert the smoothed vector boundaries into grid data on a two-dimensional plane; this transformation not only reduces the complexity of data processing, but also facilitates the overlay analysis of dangerous areas and other safe areas.

[0134] Boundary stability verification: The generated boundary data is continuously verified through a real-time monitoring module. An anomaly detection algorithm based on statistical process control is used to determine the stability of the boundary data, eliminate abnormal fluctuations caused by instantaneous interference, and ensure that the output boundary data is both timely and maintains sufficient stability.

[0135] The entire process forms a closed-loop control. Any abnormal change in the boundary data will trigger a recalculation and verification process, ensuring that a reliable protection boundary can be output under various operating conditions.

[0136] This edge computing-based dynamic boundary reorganization method offers significant technical advantages: the edge computing deployment model solves the latency problem caused by traditional cloud processing, enabling most computing tasks to be completed near the data generation location; the combination of incremental clustering and sliding window technology achieves real-time response to rapidly changing boundary data; the application of weighted averaging and geometric interpolation ensures a smooth transition of boundary updates; and the final anomaly detection mechanism effectively improves the robustness of the system. Overall, this method enables the boundary of dangerous areas to adjust at millisecond speeds to follow changes in the movement state of the hoisted object, avoiding safety hazards caused by update delays and preventing false alarms caused by boundary jitter, providing real-time and stable safety protection for bridge crane operations.

[0137] In this embodiment, identifying the location of personnel through an image acquisition device is one of the key technical aspects. Therefore, this invention further discloses a method for real-time accurate positioning and tracking of personnel based on multimodal data fusion, referring to... Figure 5 As shown, it includes:

[0138] Video data acquisition and preprocessing: Real-time video streams are acquired from multi-angle cameras deployed at the work site. The image processing module performs preliminary frame extraction and enhancement processing, including automatic white balance adjustment and dynamic range optimization, to address the differences in imaging quality under different lighting conditions.

[0139] Deep learning object detection: The YOLO (You Only Look Once) object detection algorithm is used to analyze the preprocessed image. This single-stage detection algorithm maintains a high detection speed while improving the recognition ability of small-sized targets through multi-scale feature fusion, making it particularly suitable for large-scale monitoring scenarios such as crane operation areas.

[0140] Precise spatial coordinate correction: The initial personnel position data output by the algorithm will be processed by a dedicated coordinate correction module. This module combines the camera's calibration parameters and the on-site coordinate system transformation relationship to convert the image coordinates into real-world coordinates, eliminating measurement errors caused by lens distortion and viewing angle tilt.

[0141] Abnormal data self-checking mechanism: To ensure the accuracy of location data, a strict verification mechanism is set up: when the coordinates of a person are detected to be outside the preset reasonable threshold range (such as abnormal height or sudden speed), the image re-acquisition and processing process will be automatically triggered. This self-checking mechanism effectively avoids abnormal data output caused by instantaneous interference or detection errors.

[0142] Intelligent motion state tracking: For the verified accurate position coordinates, the Kalman filter algorithm is further applied to establish a state space model of the personnel's motion. The continuous tracking of the personnel's position is achieved through a prediction-correction closed-loop mechanism. The Kalman filter is particularly suitable for processing noisy observation data. It can effectively smooth position jumps caused by detection fluctuations or partial occlusion. At the same time, it can predict the short-term movement trend of the personnel through the kinematic model and maintain the continuity of tracking during the detection interval.

[0143] Real-time data processing and updates: These tracking data are continuously received and updated, and a multi-threaded architecture is used to ensure that processing timeliness is maintained even with high frame rate video input.

[0144] Multi-source data fusion optimization: Detection results from different cameras are spatiotemporally aligned and weighted by confidence to solve the single-view occlusion problem and improve positioning accuracy. This multi-source information fusion not only expands the system's monitoring coverage but also significantly reduces false detection and false negative rates through mutual verification of multi-angle observation data. The final output personnel location data includes not only the current precise coordinates but also derived information such as movement speed and direction, providing comprehensive and reliable input for subsequent intrusion assessment in dangerous areas.

[0145] This personnel detection and tracking solution boasts significant technical advantages: the YOLO algorithm ensures a high detection rate even in complex environments; coordinate correction and self-checking mechanisms effectively improve the accuracy of location data; the introduction of Kalman filtering provides tolerance for temporary occlusion and detection fluctuations; and multi-source data fusion enhances overall reliability. In terms of application results, this method can adapt to various complex situations at crane operation sites. Even in challenging scenarios such as changing lighting, partial occlusion, or rapid personnel movement, it can continuously output stable and accurate personnel location information, providing a solid data foundation for intrusion detection in hazardous areas.

[0146] In actual operation scenarios, factors such as nighttime operation, equipment shadows, and stacked goods often lead to a decline in the image quality captured by visible light cameras. Meanwhile, obstructions such as mobile devices and building structures can cause interruptions in personnel detection. These environmental interferences make the detection system that relies solely on vision have obvious shortcomings in terms of reliability, which may lead to missed personnel detection or false location reports, seriously affecting the accuracy of safety warnings.

[0147] To solve the above problems, refer to Figure 6 As shown, this invention proposes a robust personnel localization method based on multimodal sensor fusion, comprising:

[0148] Multi-source data acquisition and preprocessing: Construct a multi-source data input channel to simultaneously acquire personnel location data from visible light cameras and thermal imaging data from infrared sensors. Through a specially designed data preprocessing process, the two types of data are spatiotemporally aligned and standardized to ensure that data from different modalities can be fused under a unified coordinate system and time series reference.

[0149] Adaptive data smoothing: To address the unavoidable sensor noise problem in industrial environments, an adaptive Kalman filter algorithm is used to smooth the raw data. This algorithm can dynamically adjust the filtering parameters according to the data quality, effectively suppressing random noise while preserving valid motion information.

[0150] Intelligent weighted data fusion: After data preprocessing, the core multimodal fusion stage is entered. The weighted fusion algorithm designed here fully considers the performance characteristics of different sensors in different environments: when the light is sufficient, visual data is given higher weight to take advantage of its high resolution, while under low light conditions, the contribution of infrared data is automatically increased. This dynamic weight adjustment mechanism is triggered by a preset ambient brightness threshold to ensure that the most reliable data source can be selected adaptively.

[0151] Dynamic correction in occlusion scenarios: To further improve positioning accuracy under occlusion conditions, a particle filter algorithm is introduced to establish a probabilistic model of personnel movement. By sampling and evaluating a large number of possible positions using particles, the problem of loss of position information caused by brief occlusion is effectively overcome. Another advantage of particle filtering is that it can integrate prior motion patterns. When a person is completely occluded, their possible position can still be predicted based on historical trajectories.

[0152] Temporal consistency guarantee: To ensure the temporal consistency of the output data, a time series analysis module is also integrated. This module establishes an autoregressive prediction model by analyzing the historical change patterns of location data. When a significant deviation between the current data and the predicted value is detected, a data smoothing and compensation mechanism is automatically activated. This time series processing not only improves the accuracy of single-point location, but also outputs a smooth trajectory that conforms to the laws of human movement.

[0153] 3D spatial coordinate mapping: Through precise spatial coordinate mapping, 2D sensor data is converted into position information in the 3D work space, providing a complete location description for subsequent hazardous area assessment.

[0154] This multimodal fusion scheme brings significant technological improvements: the introduction of infrared sensors fundamentally solves the detection problem in low-light environments; the dynamic weighted fusion algorithm fully utilizes the complementary advantages of different sensors; particle filtering technology effectively addresses occlusion issues; and time-series analysis ensures the continuity of motion trajectories. In practical terms, this method enables the personnel detection system to maintain stable performance under various harsh environmental conditions, reducing the false alarm rate by an order of magnitude and significantly decreasing false alarms caused by environmental interference. This highly reliable positional awareness capability provides all-weather, all-condition reliable protection for the safety of bridge crane operations, significantly enhancing the practical value of the entire early warning system.

[0155] In this embodiment, the use of 3D visualization technology to generate early warning information is another key technology of the present invention, as described below. Figure 7 As shown, this invention proposes an intelligent visualization early warning method based on three-dimensional spatial mapping, comprising:

[0156] Real-time intrusion status detection: Establish a precise spatial location comparison mechanism to determine the spatial relationship between the real-time acquired three-dimensional coordinates of personnel and the dynamically calculated boundary of the danger zone with millimeter-level precision, ensuring the accuracy of intrusion detection.

[0157] Tiered warning signal generation: When it is confirmed that personnel have entered a dangerous area, differentiated warning signal data will be generated based on the preset warning level. This data not only includes conventional warning signs, but also integrates key information such as the degree of danger and the depth of intrusion.

[0158] Real-world image optimization processing: Professional-grade preprocessing is performed on real-time images from the scene. Adaptive noise reduction algorithms are used to eliminate noise from the image sensor, and dynamic range extension technology is used to improve the detail representation in areas with uneven lighting, laying the foundation for subsequent visualization fusion.

[0159] Precise 3D spatial mapping: By establishing the camera parameter model and the on-site coordinate system transformation relationship, the abstract warning signal data is accurately projected onto the corresponding position in the real scene image. This mapping is not a simple two-dimensional superposition, but an intelligent fusion that takes into account three-dimensional spatial characteristics such as perspective transformation and occlusion relationship, ensuring that the warning signs can maintain the correct spatial correspondence with the equipment and structures in the actual scene.

[0160] Intelligent environment adaptive fusion: After initial fusion, the brightness, transparency and edge effects of the warning signs are automatically adjusted according to the current ambient light conditions and background complexity, so that the warning information can maintain the best recognition in different visual environments.

[0161] Multi-dimensional warning effect optimization: Activate the enhanced display mode to ensure warning effect through high-contrast coloring and dynamic flashing prompts.

[0162] The technological advantages of this visual early warning solution are reflected in several aspects: precise spatial mapping ensures a perfect match between the warning area and the actual danger zone, eliminating the positioning deviation present in traditional methods; intelligent light and shadow adjustment makes the warning information clearly distinguishable under various lighting conditions; the three-dimensional display allows operators to intuitively understand the spatial distribution of the danger zone, especially the risk location in the vertical direction; and differentiated warning design helps to quickly determine the level of danger. From practical application results, this method significantly improves the efficiency of warning information transmission, enabling operators to accurately identify dangerous situations in a short time. This efficient and intuitive visual early warning greatly enhances the safety response capability of bridge crane operation sites and provides reliable technical protection for preventing collision accidents.

[0163] In real-world industrial environments, the transmission delay of early warning information directly impacts the timeliness of safety protection. When hoisted objects move rapidly or personnel suddenly enter a dangerous area, even with highly accurate detection algorithms, serious safety accidents can still occur if the early warning information cannot be delivered to the display device in a timely manner. Therefore, this invention also provides a low-latency early warning information display method based on real-time streaming transmission, referring to... Figure 8 As shown, it includes:

[0164] Real-time data stream acquisition and transmission: The raw warning data stream is processed using WebRTC (Web Real-Time Communication), a protocol designed specifically for real-time communication. WebRTC is an open-source technology that supports real-time audio and video communication and data transmission for browsers and mobile applications. The congestion control algorithm and adaptive bit rate adjustment mechanism built into WebRTC can dynamically optimize transmission parameters according to network conditions, ensuring data integrity while controlling transmission latency to the millisecond level. The protocol also integrates forward error correction (FEC) and intelligent balancing mechanism for packet loss retransmission to ensure that a stable data stream can still be maintained when the network fluctuates.

[0165] Timing synchronization and consistency verification: During transmission, a precise timestamp synchronization mechanism is used to align all warning information in time. This synchronization not only takes into account network transmission delays, but also compensates for clock deviations between different display devices, ensuring strict synchronization of multi-terminal displays. This is especially important for multi-screen collaborative warnings in large-scale work sites.

[0166] Intelligent data chunking: To further improve transmission efficiency, the verified warning data is intelligently chunked. Differentiated chunking strategies are adopted based on information priority and content characteristics: key location and warning data are transmitted quickly in small chunks, while auxiliary background information is appropriately merged to reduce protocol overhead.

[0167] Efficient information distribution: The segmented data is published to various display terminals on site through the lightweight MQTT (Message Queuing Telemetry Transport) protocol. MQTT is a lightweight Internet of Things (IoT) communication protocol based on the publish / subscribe model, which is particularly suitable for one-to-many industrial scenarios. Its extremely low protocol overhead and hierarchical mechanism can maximize the saving of bandwidth resources while ensuring the reliable transmission of critical information.

[0168] Data format compatibility verification: After receiving the data, the display device first extracts the content through a JSON (JavaScript Object Notation) parser and performs a strict format compatibility check. This check not only verifies the integrity of the data, but also automatically adjusts the rendering parameters according to the device performance to ensure the best display effect from high-end monitoring screens to portable terminals.

[0169] Streaming Parallel Rendering: The final streaming rendering stage adopts a decoupled pipeline architecture, which processes data parsing, content preparation and image rendering in parallel, making full use of the rendering capabilities of modern GPUs.

[0170] Dynamic performance optimization: Real-time collection of latency data from each stage, and continuous optimization of transmission parameters and rendering strategies through feedback loop. This adaptive mechanism can automatically maintain the best performance state as the operating environment changes (such as network load fluctuations, increase or decrease of display devices, etc.).

[0171] It is particularly worth noting that the entire transmission chain adopts end-to-end latency monitoring. Any latency anomalies in any link will be immediately identified and optimized in a targeted manner to ensure low latency throughout the entire process from the generation to the display of the warning information.

[0172] The technological advantages of this low-latency early warning and display system are reflected in multiple aspects: the combination of WebRTC and MQTT achieves the best balance between transmission efficiency and reliability; timestamp synchronization ensures consistency of display across multiple terminals; intelligent data fragmentation optimizes network resource utilization; and the streaming rendering architecture fully leverages the potential of hardware acceleration. In practical terms, this method enables operators to obtain clear and accurate early warning information at the first sign of danger, significantly improving the timeliness and reliability of crane operation safety protection. This real-time guarantee ensures timely warnings even in the face of rapidly moving loads or sudden intrusions, buying valuable reaction time for on-site personnel and effectively preventing safety accidents.

[0173] Finally, based on the low-latency warning information display, it is also necessary to push the warning signal to the operation terminal to obtain real-time operation command feedback, referring to... Figure 9As shown, the present invention also provides a closed-loop control method for early warning response based on intelligent decision-making, comprising:

[0174] Multi-source sensor data fusion: Integrates real-time data streams from multiple sources, including intrusion status of hazardous areas, motion parameters of hoisted objects, environmental monitoring indicators, etc., and generates structured early warning signals through multi-dimensional feature fusion.

[0175] Priority communication transmission: These signals are transmitted to the operating terminal via an optimized communication protocol stack that employs a priority queue management mechanism to ensure that high-level warnings can be delivered in a timely manner despite network congestion.

[0176] Dynamic interface adaptation generation: After receiving the signal, the terminal device is decoded in real time by the embedded parsing engine and dynamically generates an adapted operation interface according to the warning type. For example, different visual codes are used for different levels of danger (red flashing indicates emergency stop, yellow warning indicates caution and avoidance). This ergonomic design greatly reduces the cognitive load of operators.

[0177] Intelligent command decision generation: When the strength of the parsed signal exceeds the preset threshold (such as continuous intrusion into a dangerous area or the hoisted object rapidly approaching personnel), the command generation mechanism will be automatically triggered. This mechanism is not a simple matching of preset rules, but an intelligent decision-making mechanism that combines the context information of the current operation scenario. For example, when hoisting precision equipment, it will prioritize deceleration rather than emergency stop to avoid damage to the goods.

[0178] Secure encrypted transmission verification: The generated instructions are transmitted to the feedback processing module through an encrypted channel based on the national cryptographic algorithm. The encryption process is implemented in a lightweight manner to avoid introducing additional delays, while meeting the communication security requirements of industrial control systems. After receiving the instructions, the feedback processing module will confirm the validity of the instructions through a multi-factor authentication mechanism, including operator identity verification and analysis of the logical rationality of the instructions.

[0179] Machine learning effectiveness evaluation: To ensure the scientific nature of the instructions, the system introduces an online learning classification model to evaluate the effectiveness of the instructions. This model is trained on historical operation data and can identify potential erroneous instructions (such as erroneous instructions to continue operation under high wind conditions) and provide correction suggestions. The model adopts an incremental learning architecture, which can continuously optimize the decision logic as the system runs and gradually adapt to the operating habits under specific working conditions.

[0180] Real-time system status updates: The final verified execution signal will update the system status in real time, including adjusting the crane's operating parameters, triggering the auxiliary braking device, or notifying surrounding personnel to avoid the situation, forming a complete "perception-decision-execution" closed loop.

[0181] This early warning and response method achieves significant technological advancements: the intelligent command generation mechanism reduces operator error; encrypted communication and multi-factor authentication ensure the security of control commands; and the machine learning classifier effectively prevents risks caused by human error. This method not only improves the security chain for personnel intrusion detection but also, by introducing intelligent decision-making elements, enables crane systems to evolve from passive early warning to proactive protection, providing a new generation of technological paradigms for safety management in industrial settings. Particularly in complex operating conditions, its rapid response and accurate decision-making capabilities effectively prevent sudden risks that traditional methods struggle to address, significantly improving the safety level of heavy equipment operations.

[0182] Reference Figure 10 As shown, in order to implement the above method, the present invention also provides a personnel intrusion detection system for hazardous areas of bridge crane operations, comprising:

[0183] The trajectory prediction module acquires real-time position and speed data of the hoisted object through sensors, and uses deep learning algorithms to analyze the motion trajectory of the hoisted object to obtain the predicted motion trajectory result of the hoisted object;

[0184] The dynamic hazard area calculation module calculates the boundary of the hazard area based on the predicted trajectory of the hoisted object, combined with the load mass and rope length data, and determines the range of the dynamic hazard area using a dynamic model.

[0185] The personnel identification module acquires image data of the work area through an image acquisition device, identifies the location of personnel in the image, and obtains personnel location data.

[0186] The early warning generation module generates an early warning signal when it determines that a person has entered a dynamic danger zone based on the personnel location data. It then uses 3D visualization technology to overlay the early warning signal onto the real-time on-site image to obtain visualized early warning information.

[0187] The visualization transmission module presents visual warning information through a display device, and uses a real-time streaming protocol to transmit data, resulting in low-latency warning information display.

[0188] Specifically, the system further includes a boundary reorganization module, which is used to reorganize the boundary data frequently through an edge computing device to obtain the real-time updated boundary of the dangerous area when the range of the dynamic dangerous area changes.

[0189] Specifically, the system also includes a multimodal fusion module, which, when there is insufficient light or interference from obstructions, uses a multimodal fusion algorithm to combine infrared sensor data to obtain highly reliable personnel location data.

[0190] Specifically, the system also includes an instruction feedback module, which is used to push the warning signal to the operation terminal based on the low-latency warning information display, and obtain real-time operation instruction feedback.

[0191] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for detecting intrusion of a person into a dangerous area of operation of a bridge crane, characterized in that: The method comprises the following steps: Obtain real-time position and speed data of the hoisting object through sensors, analyze the motion trajectory of the hoisting object using a deep learning algorithm, and obtain a motion trajectory prediction result of the hoisting object; According to the motion trajectory prediction result of the hoisting object, combined with the load mass and rope length data, a dynamic dangerous area range is determined by using a dynamics model to calculate the boundary of the dangerous area, including the following steps: obtaining hoisting object motion trajectory prediction data, load mass data and rope length data, using a preprocessing algorithm to clean the data to obtain standardized motion parameters; according to the standardized motion parameters, the dynamics model is used to calculate the speed distribution characteristics and acceleration changes of the hoisting object at different time points to obtain a dynamic motion state; from the dynamic motion state, the speed distribution characteristics and acceleration changes are extracted, combined with the rope length data and rope tension, the offset range of the hoisting object position is calculated, and the potential impact area is determined; if the potential impact area exceeds the preset regional safety factor, the finite element analysis method is used to calculate the boundary of the dangerous area to obtain a preliminary boundary range; according to the preliminary boundary range, combined with the environmental wind speed data, the boundary range is adjusted to obtain the final range of the dynamic dangerous area; from the final range of the dynamic dangerous area, the boundary point coordinates are extracted, and a region grid is generated using a geometric algorithm to obtain the visualization data of the dangerous area; through the visualization data, it is judged whether the dynamic dangerous area overlaps with the preset safety area, and if it overlaps, the boundary adjustment parameters of the dynamic dangerous area are generated; Obtain image data of the work area through an image acquisition device, identify the position of personnel in the image, and obtain personnel position data; When it is determined that the personnel have entered the range of the dynamic dangerous area according to the personnel position data, a warning signal is generated, and a three-dimensional visualization technology is used to superimpose the warning signal on the real-time image on site to obtain visual warning information; The visual warning information is presented through a display device, and real-time streaming protocol transmission data is used to obtain low-delay warning information display.

2. The method of detecting intrusion of personnel into a hazardous area of a bridge crane operation according to claim 1, characterized in that: Obtain real-time position and speed data of the hoisting object through sensors, analyze the motion trajectory of the hoisting object using a deep learning algorithm, and obtain a motion trajectory prediction result of the hoisting object, including: Collect real-time position and speed data of the hoisting object through sensors and store them as time series data sets; Preprocess the time series data sets using data processing technology to obtain standardized data; Train the standardized data through a long short-term memory network to obtain a trajectory prediction model; Analyze the real-time position and speed data according to the trajectory prediction model to obtain prediction trajectory data; If the deviation between the prediction trajectory data and the real-time monitoring data exceeds a preset threshold, the prediction trajectory data is corrected through a Kalman filter algorithm to obtain optimized trajectory data; Calculate the motion trend of the hoisting object according to the optimized trajectory data to obtain motion trajectory change characteristics; Update the real-time monitoring system through the motion trajectory change characteristics to obtain the hoisting object trajectory prediction result.

3. The method of detecting intrusion of personnel into a hazardous area of a bridge crane operation of claim 1, wherein: If the dynamic dangerous area range changes, the boundary data is reorganized at a high frequency through an edge computing device to obtain real-time updated dangerous area boundaries, including: The edge computing device obtains boundary data of the dangerous area from the sensor, and adopts a data preprocessing technology to clean the collected data to obtain a standardized boundary data set; If the characteristic value of the standardized boundary data set deviates from a preset threshold value, the data is partitioned through an incremental K-means clustering algorithm to determine a dynamically changing boundary subset; According to the dynamically changing boundary subset, a sliding window technology is used to segment the high-frequency data stream to obtain a time series boundary segment; The edge computing device recombines the time series boundary segment at high frequency, fuses multiple segments of data by using a weighted average method, and generates continuous boundary update data; If the continuity of the boundary update data meets a preset condition, the boundary points are smoothed through a geometric interpolation algorithm to obtain a smoothed dangerous area boundary; According to the smoothed dangerous area boundary, a rasterization technology is used to map the boundary to a two-dimensional plane to generate a real-time updated dangerous area boundary; The generated dangerous area boundary is continuously verified through a real-time monitoring system, the stability of the boundary data is judged by using an anomaly detection algorithm, and the final boundary output is obtained.

4. The method of detecting intrusion of personnel into a hazardous area of a bridge crane operation of claim 1, wherein: An image acquisition device is used to obtain image data of a work area, and the position of personnel in the image is identified to obtain personnel position data, including: A video stream is obtained from a field camera, and image data is extracted by an image processing module; A YOLO algorithm is used to detect targets in the image data to obtain preliminary personnel position data; The preliminary personnel position data is corrected in coordinates by a data analysis module to determine accurate position coordinates; If the position coordinates do not match a preset threshold range, the image data is reextracted by the image processing module; According to the accurate position coordinates, a Kalman filtering algorithm is used to track the position of personnel to obtain continuous position data; The continuous position data is updated by a real-time processing module to obtain dynamic position information; Data fusion technology is used to integrate the dynamic position information to determine the personnel position data.

5. The method of detecting intrusion of personnel into a hazardous area of a bridge crane operation of claim 1, wherein: In the case of insufficient light or obstruction interference, a multi-modal fusion algorithm is used in combination with infrared sensor data to obtain high-reliability personnel position data, including: Multi-source data input is obtained from personnel position data and infrared sensor data, and data preprocessing technology is used to clean and format the data to obtain a standardized data set; If the standardized data set has noise or missing values, the personnel position data and the infrared sensor data are smoothed by a Kalman filtering algorithm to obtain a smoothed position data set; According to the smoothed position data set, a multi-modal fusion algorithm is used to weight and integrate the personnel position data and the infrared sensor data to obtain a fused position data set; If the deviation of the fused position data set exceeds a preset threshold value in an insufficient light environment, the weight of the infrared sensor data is adjusted to optimize the fused position data set to obtain an optimized position data set; According to the optimized position data set, a particle filtering algorithm is used to dynamically correct the obstruction interference to obtain a corrected position data set; If the corrected position data set is inconsistent with historical position data, the position data is smoothed and predicted by time series analysis technology to obtain a predicted position data set; According to the predicted position data set, the data is converted into three-dimensional position information through a spatial coordinate mapping technique to obtain high-reliability personnel position data.

6. The method of detecting intrusion of personnel into a hazardous area of a bridge crane operation of claim 1, wherein: According to the personnel position data, when it is determined that the personnel has entered the range of the dynamic danger area, a warning signal is generated, and a three-dimensional visualization technique is used to superimpose the warning signal on the real-time image on site to obtain visual warning information, including: Real-time personnel position data is obtained, compared with the preset danger area boundary threshold, and it is determined whether the personnel position enters the danger area to obtain a boundary entry state; If the boundary entry state is entering, a warning signal is generated through a warning signal generation algorithm to obtain warning signal data; A real-time image on site is obtained, and an image preprocessing algorithm is used to denoise and enhance the real-time image to obtain a processed on-site image; A three-dimensional signal superposition model is obtained by using a three-dimensional visualization technique to perform coordinate mapping on the warning signal data and the processed on-site image; Through an image data fusion algorithm, the three-dimensional signal superposition model and the processed on-site image are fused to obtain preliminary visual warning information; For the preliminary visual warning information, a rendering optimization algorithm is used to adjust the light and transparency to obtain the final visual warning information.

7. The method of detecting intrusion of personnel into a hazardous area of a bridge crane operation of claim 1, wherein: The visual warning information is presented through a display device, and real-time streaming protocol is used to transmit data to obtain low-latency warning information display, including: Raw warning data is obtained through the real-time streaming protocol, and the data stream is processed using the WebRTC protocol to obtain a low-latency data transmission stream; Key warning information is extracted from the low-latency data transmission stream, and a timestamp synchronization mechanism is used to determine the timing consistency of the warning information; If the timing consistency of the warning information meets the preset threshold, the information is processed in blocks through a data slicing technique to obtain blocked warning data; According to the blocked warning data, the MQTT protocol is used to publish to the display device on site to obtain quickly distributed warning information; For the quickly distributed warning information, the data content is parsed through the JSON format to determine whether the information meets the display device compatibility requirements; If the information meets the display device compatibility requirements, the information is presented on the display device on site through a streaming rendering technique to obtain real-time visual warning information display; According to the real-time visual warning information display, a feedback loop mechanism is used to collect display delay data to determine the stability optimization requirements of the transmission protocol.

8. The method of detecting intrusion of personnel into a hazardous area of a bridge crane operation of claim 1, wherein: According to the low-latency warning information display, the warning signal is pushed to the operation terminal to obtain real-time operation instruction feedback, including: Obtain external sensor data to generate a warning signal; The warning signal is pushed to the operation terminal through a preset communication protocol; The warning signal is parsed on the operation terminal to generate an operation interface display; If the signal strength after parsing exceeds the preset threshold, an instruction generation mechanism is triggered to obtain an operation instruction; The operation instruction is transmitted to the feedback processing module through an encrypted channel to obtain instruction confirmation; A machine learning classification algorithm is used to determine the validity of the instruction to generate an execution signal; According to the execution signal, the system response state is updated to complete real-time feedback.

9. A system for detecting intrusion of a person into a dangerous area of a bridge crane operation for implementing the method according to any one of claims 1 to 8, characterized in that including: The trajectory prediction module obtains real-time position and speed data of the hoisted object through sensors, analyzes the motion trajectory of the hoisted object by using a deep learning algorithm, and obtains a motion trajectory prediction result of the hoisted object. The dynamic dangerous area calculation module calculates the boundary of the dangerous area by using a dynamics model according to the motion trajectory prediction result of the hoisted object, in combination with load mass and rope length data, and determines the range of the dynamic dangerous area. The personnel identification module obtains image data of the work area by using an image acquisition device, identifies the position of personnel in the image, and obtains personnel position data. The early warning generation module generates an early warning signal when it is determined that the personnel have entered the range of the dynamic dangerous area according to the personnel position data, superimposes the early warning signal onto real-time images on the scene by using three-dimensional visualization technology, and obtains visualized early warning information. The visualized transmission module presents the visualized early warning information by using a display device, transmits data by using a real-time streaming protocol, and obtains low-delay early warning information display.

Images