An oil field electronic fence intelligent entity out-of-bound behavior identification and graded early warning method and system based on visual perception

Abstract

This invention relates to a vision-based method and system for identifying and classifying the boundary crossing behavior of intelligent agents in oilfield electronic fences. Belonging to the fields of oilfield perimeter security and computer vision technology, it addresses the problems of existing electronic fences, such as inability to identify target types, inability to quantify boundary crossing risks, single warning levels, and low system integration. The method involves target detection and continuous tracking of oilfield perimeter image data to obtain target type and movement trajectory. Boundary crossing verification is performed on the target's center position based on a polygonal electronic fence, and the boundary crossing distance from the target to the fence boundary is calculated. This distance is then fused with target type risk weights to obtain an adjusted boundary crossing distance. Classified warnings are implemented based on preset thresholds. Simultaneously, the warning information is visualized and logged, achieving accurate identification and differentiated warnings for boundary crossing behavior. This invention is applicable to perimeter security monitoring and intelligent risk warning in complex environments such as oilfield mining areas, well sites, and oil and gas storage areas.

Description

Technical Field

[0001] It involves the fields of computer vision, geospatial positioning, deep learning, and multi-target tracking technology, specifically the recognition and hierarchical early warning of boundary crossing behavior of intelligent agents in oilfield electronic fences based on visual perception. Background Technology

[0002] Perimeter security technology for oilfields is a crucial infrastructure for ensuring the safety of oil and gas production. It has evolved from early, single-physical protection to intelligent, multi-modal sensing technologies. Currently, mainstream electronic fence systems primarily utilize physical sensing methods such as infrared beam detectors, vibrating fiber optic cables, and microwave radar for perimeter intrusion detection. These technologies trigger boundary crossing alarms by detecting signal interruptions or abnormal changes. For example, infrared beam detectors identify intrusion behavior through beam obstruction, while vibrating fiber optic cables identify intrusion events through fence vibration signals. While these solutions are simple to implement and have fast response times, they generally lack the ability to identify intrusion targets, failing to distinguish between different types of targets such as people, vehicles, or animals, and are prone to false alarms or missed alarms in complex environments.

[0003] With the development of computer vision and deep learning technologies, some research has begun to introduce video surveillance and target detection technologies into the perimeter security field. Convolutional neural networks are used to identify targets in video frames, enabling the detection of people or vehicles. For example, target detection methods based on algorithms such as the YOLO series or Faster R-CNN have been applied in scenarios such as mining areas and industrial parks to improve the intelligence level of security systems. Meanwhile, multi-target tracking algorithms such as DeepSORT and ByteTrack are also used to achieve continuous tracking of target trajectories, thereby assisting in judging the movement behavior of targets. However, existing visual security systems often focus on target detection or behavior recognition, frequently lacking deep integration with geospatial fencing, and failing to form a complete closed loop from target recognition to boundary violation determination.

[0004] Regarding the integration of electronic fences and geographic information systems, some existing technologies use GPS or RTK positioning to delineate electronic fence areas and combine target positioning data for boundary crossing judgment. For example, in vehicle monitoring or drone no-fly zone management, boundary crossing alarms are triggered by comparing geographic coordinates with preset area boundaries. However, such solutions mostly rely on high-precision positioning equipment and are mainly aimed at single target objects, lacking direct support for visual target detection, making them difficult to apply to complex scenarios in oil fields where personnel, vehicles, and unidentified objects are mixed.

[0005] Regarding early warning mechanisms, existing electronic fence systems typically use fixed thresholds to trigger alarms, lacking quantitative analysis and risk classification of boundary-crossing behaviors. For example, most systems only trigger a single alarm signal when a target touches or enters the fence, without considering factors such as the distance of the boundary crossing and the type of target for differentiated evaluation, resulting in insufficient precision in early warning information. Furthermore, at the data level, oilfield security scenarios are limited by acquisition conditions, and real boundary-crossing sample data is relatively scarce. Existing model training largely relies on limited datasets, leading to weak model generalization ability and significant performance degradation in complex environments such as sandstorms, low light, and multi-target interference.

[0006] In terms of system integration, existing technologies often disperse data processing, target detection, fence verification, and early warning functions across different modules or systems, lacking a unified integration platform. This results in complex system deployment and high operation and maintenance costs. Furthermore, fence configuration largely relies on fixed coordinates or backend parameter settings, lacking intuitive visual interaction methods and making it difficult to quickly adjust according to actual oilfield production conditions.

[0007] In summary, existing technologies suffer from several drawbacks, including limited sensing methods, inability to identify target types, lack of quantitative and hierarchical assessment of boundary crossing risks, lack of integrated collaboration between target detection and tracking, insufficient security data samples leading to poor model generalization ability, inadequate fusion of electronic fences and visual perception, low system integration, and poor configuration and maintenance flexibility. Summary of the Invention

[0008] To address the shortcomings of existing technologies, such as limited sensing methods, inability to identify target types, lack of quantitative and hierarchical assessment of boundary crossing risks, lack of integrated coordination between target detection and tracking, insufficient security data samples leading to poor model generalization ability, inadequate fusion of electronic fences and visual perception, low system integration, and poor configuration and maintenance flexibility, the technical solution provided by this invention is as follows: A method for identifying and hierarchically warning of boundary crossing behavior of intelligent agents in oilfield electronic fences based on visual perception, comprising: The steps to configure geofence parameters for the oilfield perimeter, set target type risk weights and early warning level thresholds, and output fence boundary data and risk parameters. Based on the fence boundary data and risk parameters, the steps include acquiring oilfield perimeter image data and performing target detection and multi-target tracking, identifying target types and generating unique target identifiers, target center positions and movement trajectories; The steps include determining the spatial relationship between the target's center location and the fence boundary data, identifying the target's boundary crossing status, and filtering out boundary-crossing targets. The steps are: calculating the shortest distance from the target to the fence boundary based on the center position of the target and the fence boundary data, obtaining the boundary crossing distance, and outputting the boundary crossing distance data; The steps are as follows: Based on the fusion processing of the cross-boundary distance data and the target type risk weight, the adjusted cross-boundary distance is obtained and compared with the warning level threshold to determine the target warning level and generate corresponding warning information. Based on the aforementioned warning information, differentiated prompts are provided to targets, and the target type, target location, boundary crossing distance, and warning level are recorded, thereby realizing the steps of visual display and log storage of boundary crossing behavior in the oilfield.

[0009] Furthermore, in a preferred embodiment, the geographic electronic fence adopts a polygon boundary representation method, and the coordinates of the polygon vertices can be interactively modified in a visual interface. The modification results are synchronized in real time for boundary crossing determination.

[0010] Furthermore, in a preferred embodiment, target detection is used to identify four types of targets: people, vehicles, unidentified objects, and wild animals, and the detection results are filtered by confidence to eliminate targets with low confidence.

[0011] Furthermore, in a preferred embodiment, multi-target tracking associates and matches targets based on their appearance features and motion trajectories, and assigns a unique identifier to each target to maintain continuity across frames.

[0012] Furthermore, in a preferred embodiment, the boundary crossing determination is achieved by judging the spatial relationship between the target center point and the boundary of the polygonal fence. When the target center point is located outside the fence, it is determined to be a boundary crossing.

[0013] Furthermore, in a preferred embodiment, error correction processing is performed on the target center location data, including filtering out abnormal fluctuation positions and smoothing continuous positions, in order to improve the accuracy of the cross-boundary distance calculation.

[0014] Based on the same inventive concept, this invention also provides a vision-based oilfield electronic fence intelligent agent boundary crossing behavior recognition and hierarchical early warning system, comprising: A module for configuring geofence parameters for oilfield perimeters, setting target type risk weights and early warning level thresholds, and outputting fence boundary data and risk parameters; Based on the fence boundary data and risk parameters, a module is used to acquire oilfield perimeter image data, perform target detection and multi-target tracking, identify target types, and generate unique target identifiers, target center positions, and movement trajectories. A module that determines the target's boundary crossing status and filters out targets crossing the boundary based on the spatial relationship between the target's center location and the fence boundary data; This module calculates the shortest distance from the target to the fence boundary based on the center position of the target and the fence boundary data, obtains the boundary crossing distance, and outputs the boundary crossing distance data. A module that integrates the cross-boundary distance data with the target type risk weight to obtain the adjusted cross-boundary distance and compares it with the warning level threshold to determine the target warning level and generate corresponding warning information. Based on the aforementioned warning information, the module provides differentiated prompts to targets and records the target type, target location, boundary crossing distance, and warning level, thereby enabling the visualization and log storage of boundary crossing behavior in the oilfield.

[0015] Based on the same inventive concept, the present invention also provides a computer storage medium for storing a computer program, wherein when the computer program is read by a computer, the computer executes the method described thereon.

[0016] Based on the same inventive concept, the present invention also provides a computer, including a processor and a storage medium, wherein when the processor reads a computer program stored in the storage medium, the computer executes the method described thereon.

[0017] Based on the same inventive concept, the present invention also provides a computer program product, which, when executed, implements the method described.

[0018] Compared with the prior art, the advantages of the technical solution provided by the present invention are as follows: By introducing a dual-source data processing mechanism, the system leverages the synergistic effect of synthetic data generation based on GAN networks and incremental learning of real data in the data processing layer. This enables the system to construct training datasets covering multiple target types and various boundary crossing scenarios in oilfield scenarios where samples are scarce. Compared with existing methods that rely solely on real data, this effectively improves the model's adaptability to dust storms, low light conditions, and complex interference environments, thereby significantly enhancing the generalization ability and stability of the early warning model.

[0019] By integrating deep learning object detection and multi-object tracking technologies into the object detection and tracking layer, the system enables the identification of types of people, vehicles, unidentified objects, and wild animals and the tracking of their continuous trajectories. This transforms boundary crossing judgment from single-frame detection to continuous judgment based on time-series trajectories. Compared with existing technologies that only use object detection or simple inter-frame comparison, this effectively reduces the problems of target loss, ID switching, and misjudgment, and improves the continuity and accuracy of boundary crossing behavior recognition.

[0020] By constructing a geofence verification mechanism based on polygon spatial positioning, and realizing flexible configuration and real-time synchronization of fence coordinates in the parameter configuration layer and visualization layer, the system can quickly adapt to different oilfield perimeter structures without restarting. Compared with the traditional fixed coordinate or hard-coded fence method, it significantly improves the flexibility of fence configuration and the efficiency of engineering application adaptation, while improving the spatial accuracy of boundary crossing judgment.

[0021] By introducing the shortest distance calculation method from the target center point to the fence boundary into the boundary crossing distance quantification calculation module, and combining it with coordinate precision control and Gaussian filtering error processing mechanism, the boundary crossing behavior is transformed from a simple "whether it has crossed the boundary" to a quantifiable continuous distance index. Compared with the existing technical solutions that only perform boundary trigger judgment, it can more accurately reflect the degree of target boundary crossing and provide a reliable quantitative basis for subsequent risk assessment.

[0022] By introducing a fusion calculation mechanism of target type weight and boundary crossing distance into the dynamic hierarchical early warning judgment module, different types of targets exhibit differentiated risk levels under the same boundary crossing distance, realizing multi-level early warning classification based on the adjusted boundary crossing distance. Compared with the traditional single-level alarm method triggered by a unified threshold, this method is more in line with the actual security risk distribution of oilfields and improves the pertinence and decision reference value of early warning information.

[0023] By setting up a weight adaptive update mechanism based on historical security data and expert experience, the target type weights can be dynamically adjusted according to the actual operation of the oilfield. Compared with fixed weights or manual static configuration, this can continuously optimize the risk matching capability of the early warning model and ensure that the system has high adaptability and accuracy in different time periods and different scenarios.

[0024] By constructing a multi-level closed-loop architecture consisting of a parameter configuration layer, a data processing layer, a target detection and tracking layer, a geofence verification layer, a hierarchical early warning layer, and a visualization layer, data connectivity and result feedback are achieved between modules. Compared with existing distributed security systems, this effectively reduces data fragmentation and redundant processing between systems, and improves overall processing efficiency and system synergy.

[0025] By integrating fence configuration, target detection, trajectory display, boundary crossing marking, and early warning information display into the visualization monitoring module, maintenance personnel can intuitively obtain target behavior and early warning status. Compared with traditional systems that only provide alarm signals or simple logs, this significantly improves the system's human-computer interaction efficiency and on-site decision-making capabilities.

[0026] By automatically recording all information such as target ID, type, location, boundary crossing distance, and warning level in the early warning information push and log recording modules, and supporting local storage and tracing, the system has a complete security data closed loop. Compared with existing technologies that lack recording or only record alarm results, it can provide reliable data support for subsequent event analysis, model optimization, and security management.

[0027] By unifying and dynamically adjusting the target type weight, early warning threshold, and fence boundary during system initialization and parameter configuration, the system can quickly complete deployment and strategy optimization according to different oilfield scenarios. Compared with traditional systems that require complex backend configuration or redevelopment, this significantly reduces deployment costs and maintenance complexity.

[0028] It is suitable for perimeter security monitoring and intelligent boundary risk classification and early warning in complex environments such as oilfields, well sites, and oil and gas storage areas. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall system architecture; Figure 2 This is the core processing flowchart. Detailed Implementation

[0030] To make the advantages and benefits of the technical solution provided by the present invention clearer, the technical solution provided by the present invention will now be described in further detail with reference to the accompanying drawings, specifically: Implementation Method 1: This implementation method provides a vision-based method for identifying and hierarchically warning of boundary crossing behavior of intelligent agents in oilfield electronic fences, including: The steps to configure geofence parameters for the oilfield perimeter, set target type risk weights and early warning level thresholds, and output fence boundary data and risk parameters. Based on the fence boundary data and risk parameters, the steps include acquiring oilfield perimeter image data and performing target detection and multi-target tracking, identifying target types and generating unique target identifiers, target center positions and movement trajectories; The steps include determining the spatial relationship between the target's center location and the fence boundary data, identifying the target's boundary crossing status, and filtering out boundary-crossing targets. The steps are: calculating the shortest distance from the target to the fence boundary based on the center position of the target and the fence boundary data, obtaining the boundary crossing distance, and outputting the boundary crossing distance data; The steps are as follows: Based on the fusion processing of the cross-boundary distance data and the target type risk weight, the adjusted cross-boundary distance is obtained and compared with the warning level threshold to determine the target warning level and generate corresponding warning information. Based on the aforementioned warning information, differentiated prompts are provided to targets, and the target type, target location, boundary crossing distance, and warning level are recorded, thereby realizing the steps of visual display and log storage of boundary crossing behavior in the oilfield.

[0031] The geographic electronic fence uses a polygon boundary representation, and the coordinates of the polygon vertices can be interactively modified in a visual interface. The modification results are synchronized in real time for boundary violation determination.

[0032] Target detection is used to identify four types of targets: people, vehicles, unidentified objects, and wild animals. The detection results are then filtered based on confidence to remove targets with low confidence.

[0033] Multi-target tracking associates and matches targets based on their appearance features and motion trajectories, and assigns a unique identifier to each target to maintain continuity across frames.

[0034] Boundary crossing is determined by judging the spatial relationship between the target center point and the boundary of the polygonal fence. When the target center point is located outside the fence, it is judged as crossing the boundary.

[0035] Error correction processing is performed on the target center location data, including filtering out abnormal fluctuations and smoothing continuous locations, to improve the accuracy of boundary crossing distance calculation.

[0036] Based on the same inventive concept, this invention also provides a vision-based oilfield electronic fence intelligent agent boundary crossing behavior recognition and hierarchical early warning system, comprising: A module for configuring geofence parameters for oilfield perimeters, setting target type risk weights and early warning level thresholds, and outputting fence boundary data and risk parameters; Based on the fence boundary data and risk parameters, a module is used to acquire oilfield perimeter image data, perform target detection and multi-target tracking, identify target types, and generate unique target identifiers, target center positions, and movement trajectories. A module that determines the target's boundary crossing status and filters out targets crossing the boundary based on the spatial relationship between the target's center location and the fence boundary data; This module calculates the shortest distance from the target to the fence boundary based on the center position of the target and the fence boundary data, obtains the boundary crossing distance, and outputs the boundary crossing distance data. A module that integrates the cross-boundary distance data with the target type risk weight to obtain the adjusted cross-boundary distance and compares it with the warning level threshold to determine the target warning level and generate corresponding warning information. Based on the aforementioned warning information, the module provides differentiated prompts to targets and records the target type, target location, boundary crossing distance, and warning level, thereby enabling the visualization and log storage of boundary crossing behavior in the oilfield.

[0037] A computer storage medium is also provided for storing a computer program, which, when read by the computer, executes the method.

[0038] A computer is also provided, including a processor and a storage medium, wherein the computer executes the method when the processor reads a computer program stored in the storage medium.

[0039] A computer program product is also provided, which, when executed, implements the method described.

[0040] Implementation Method Two: This implementation method is a further detailed description of the technical solution provided in Implementation Method One, specifically: This embodiment provides a method and system for identifying and classifying the boundary crossing behavior of intelligent agents in oilfield electronic fences based on visual perception. The overall concept does not revolve solely around the single judgment of "whether the boundary has been crossed," but rather designs a system around the complete business chain in oilfield perimeter security scenarios, addressing "what the target is, whether the target has crossed the boundary, to what extent the boundary has been crossed, at what level of warning should be issued, and how to achieve full-process visualization and traceability management." This technical solution is applicable to scenarios with complex perimeter environments, significant changes in lighting, strong dust interference, mixed traffic of personnel and vehicles, and the risk of unidentified object intrusion, such as oilfield mining areas, well sites, and oil and gas storage areas. By integrating visual target recognition, multi-target continuous tracking, geographic electronic fence spatial verification, quantitative calculation of boundary crossing distance, dynamic judgment of risk level, and visualized management of warning information, the system can continuously, accurately, and differentiatedly identify and classify the boundary crossing behavior in complex oilfield environments.

[0041] In practical implementation, the first step is to construct the overall system operating environment. This system includes parameter configuration, data processing, target detection and multi-target tracking, geofence verification, boundary crossing distance quantification, dynamic hierarchical early warning judgment, and visualization monitoring and log management. These components are not isolated; rather, data is transferred in the following order: configuration, training, detection, verification, calculation, early warning, display, and recording. The parameter configuration section outputs fence boundaries, risk weights, and threshold parameters, which serve as the basis for subsequent data generation, target boundary crossing judgment, and early warning level determination. The data processing section outputs standardized training data and model update results, which serve as model support for target recognition and early warning decision-making. The target detection and multi-target tracking section outputs target type, target identifier, target center location, and movement trajectory, which serve as input for fence verification and boundary crossing distance calculation. The geographic electronic fence verification section outputs the determination result of whether the target has crossed the boundary, which serves as a prerequisite for distance quantification and early warning screening. The boundary crossing distance quantification calculation section outputs the actual degree of target boundary crossing, which serves as an important basis for early warning level determination. The dynamic hierarchical early warning determination section outputs early warning level and early warning information. Finally, the visualization monitoring and log management section completes real-time display, differentiated reminders, and full recording, thus forming a complete closed loop.

[0042] First, the system is initialized and its parameters are configured. After system startup, the visual acquisition equipment, computing and processing equipment, and display interface are initialized, making the camera, video stream access module, image processing module, target detection model, multi-target tracking model, and early warning judgment model available for use. To ensure system adaptability to different deployment environments, the running hardware is identified during initialization, automatically matching the central processing unit or graphics processing unit operating mode according to the site configuration, enabling real-time visual perception and model calculation. After the basic environment is established, a geographic electronic fence is configured in the visualization interface. This fence uses a polygonal region representation, and multiple vertex coordinates can be set according to the actual boundary shape of the oilfield perimeter. By connecting these vertices sequentially, a closed fence area is formed. The fence coordinates are not hard-coded into the program but are saved in a parameterized manner. Maintenance personnel can directly modify the fence vertex positions or add and delete fence boundary points in the interface, allowing the fence to flexibly adapt to different areas such as well sites, oil depot perimeters, passageways, and loading / unloading areas. After the fence coordinates are configured, they are synchronously stored and sent to the fence verification and visualization parts in real time, ensuring that all subsequent target location judgments are based on the fence boundary.

[0043] While configuring the fence, it is also necessary to set risk weights for different target types. These weights reflect the risk differences of different target types in oilfield perimeter security. For example, unauthorized personnel entry typically poses a higher security risk, vehicle intrusion may affect production operations and safety passages, unidentified object intrusion poses a potential threat, and while wild animal entry may trigger alarms, it generally carries a relatively lower risk. Therefore, the system presets different weight values ​​for different targets, allowing for different warning levels triggered by different targets at the same boundary crossing distance. In addition to target weights, warning level thresholds also need to be set, dividing warnings into multiple levels and assigning different alert methods to different levels. The aforementioned fence coordinates, target type weights, and warning level thresholds are uniformly saved as system operating parameters and serve as the basic input for subsequent data generation, real-time identification, boundary crossing distance correction, and warning level determination.

[0044] After parameter configuration, the system enters the dual-source data preparation and loading stage. Due to the characteristics of oilfield perimeter security scenarios—few real boundary-crossing samples, high collection costs, and insufficient scene coverage—training solely with real-world data can easily lead to insufficient model recognition capabilities in environments such as sandstorms, low-light nighttime conditions, target occlusion, and multi-target cross-movement. Therefore, this implementation adopts a dual-source data processing mechanism combining synthetic and real data. For the synthetic data portion, the system automatically generates simulated target samples including personnel, vehicles, unidentified objects, and wild animals based on preset oilfield scene backgrounds, fence boundary areas, and target type rules. During generation, not only are the appearance and movement states of the targets in the images generated, but also the target trajectory, target location sequence, target category label, boundary crossing label, boundary crossing distance label, and corresponding warning level label are generated simultaneously. By setting different movement directions, speeds, entry angles, dwell times, and boundary crossing depths, a large number of oilfield perimeter boundary-crossing samples can be constructed to compensate for the scarcity of real data.

[0045] For the real data portion, the system supports loading on-site collected data from video frame files, trajectory files, or structured annotation files. After loading, the system automatically parses the target type, target location, timestamp, trajectory information, etc., and converts them into a standard format recognizable by the system. For data from different sources and in different formats, standardization processing is required to ensure consistency in coordinate representation, category labels, time order, and annotation fields, facilitating unified use for model training and subsequent inference. During real data processing, newly added on-site samples can be continuously supplemented, introducing newly collected out-of-bounds behavior samples into the model training process, thereby achieving continuous model optimization based on real business feedback. After synthetic data generation and real data standardization processing, a dataset suitable for model training is obtained and input into the early warning model training section and the target detection model optimization section, providing stable and sufficient data support for subsequent real-time detection.

[0046] After obtaining the training data, the model is trained and updated. During training, the processed data can be divided into a training set and a validation set. The training set is used to optimize the target detection model, tracking and association strategy, and early warning judgment logic, while the validation set is used to verify the model's recognition accuracy, trajectory continuity, and the rationality of hierarchical early warning in different scenarios. After training is completed, the optimal model parameters are saved and loaded as a real-time inference model during system runtime. Through this step, the system already possesses the capability for target recognition, motion tracking, and early warning judgment adapted to oilfield scenarios before formal operation.

[0047] After model preparation is complete, the target detection and multi-target tracking stage begins. This stage involves continuously acquiring image sequences of the oilfield perimeter via cameras or video streams, and performing target detection on each frame. During detection, the images are first preprocessed to ensure that image size, color channels, and brightness distribution meet the model input requirements. Then, deep features are extracted from the images, and target regions are located and classified, outputting the detection box position and target category for each target. Target categories include at least personnel, vehicles, unidentified objects, and wild animals. To improve detection reliability, a confidence level filtering rule is set for the detection results, discarding results below a set confidence threshold to reduce interference from environmental noise, shadows, reflections, and non-target backgrounds.

[0048] After single-frame detection, multi-target tracking is further performed to prevent target identity loss or duplicate counting between consecutive frames. During tracking, the system uses the target's appearance features, spatial position changes, and motion trends in consecutive frames to perform association matching on detected targets and assign a unique identifier to each target. This unique identifier remains unchanged throughout the target's existence, allowing the system to determine whether an object crossing the boundary at a certain moment is the same object as at the previous moment. As the video stream continues to input, the system continuously updates the center position, historical trajectory, and current state of each target, forming a continuous motion description of the target in the time dimension. In this way, subsequent boundary crossing judgments are no longer static judgments based on isolated single-frame positions, but rather dynamic identification based on the target's temporal information with continuous identity, thereby improving the stability and accuracy of boundary crossing judgments.

[0049] After obtaining the target's unique identifier, category information, and center location, the system enters the geofence boundary verification stage. The core task of this stage is to determine whether the target is currently outside the fence. To do this, the target's center location in the image coordinate system is converted into planar position coordinates corresponding to the fence boundary through camera calibration and coordinate mapping. Since the fence is represented by a polygonal boundary, it is necessary to determine the spatial relationship between the target's center point and this polygonal region. When the target's center point is located outside the polygon, it is determined to be outside the boundary; when the target's center point is located inside the polygon or falls on the polygon's boundary, it is determined to be within the boundary. Through this method of judgment based on the relationship between the target's center location and the polygonal region, the system can achieve high-precision boundary verification under complex perimeter shapes, rather than performing a coarse judgment only for simple rectangular areas or fixed single-line boundaries.

[0050] This fence verification step also features dynamic adaptation capabilities. When maintenance personnel modify the fence boundary coordinates in the visual interface, the updated fence information can be immediately synchronized to the fence verification section, allowing the system to perform boundary violation judgments based on the new perimeter boundaries without system downtime. This is highly practical for oilfield sites where the monitored area frequently needs to be changed due to construction adjustments, temporary work isolation, or changes in access routes. Through this step, the system outputs the current boundary violation status of each target and filters out targets determined to be boundary violations, sending them to the subsequent boundary violation distance quantification calculation step.

[0051] For targets already determined to have crossed the boundary, a further quantitative calculation of the boundary crossing distance is performed. Many existing electronic fence systems can only provide a binary conclusion of "crossed the boundary" or "not crossed the boundary," failing to distinguish the degree of boundary crossing. This implementation further calculates the actual distance the target exceeds the fence boundary, making the warning results more precise. Specifically, using the target's center position as a reference, the shortest spatial distance from that point to the nearest boundary of the fence is calculated. When the target is outside the fence, this shortest distance is taken as the target's actual boundary crossing distance. To make this distance more reliable, the system first performs precision control and error correction on the target position coordinates obtained from visual acquisition. The target position originates from the visual perception module, combined with the planar coordinates obtained from camera calibration. The fence boundary originates from a high-precision coordinate acquisition method and is converted into calculable planar projection coordinates. Due to the complex on-site environment, the coordinate sequence may be subject to noise interference. Therefore, the target coordinates are filtered to remove abnormal fluctuation points, and the neighborhood mean is used to correct positions with excessively large single-frame errors. After several consecutive frames of stable position, they are then used for distance calculation. This avoids drastic fluctuations in the boundary crossing distance due to instantaneous detection jitter, thereby improving the stability of subsequent warning level determination.

[0052] After obtaining the actual boundary crossing distance, the system enters the dynamic hierarchical early warning determination stage. This stage is a crucial technical part of this implementation method. Its core lies not in simply comparing whether the boundary crossing distance exceeds a single threshold, but in combining the target type risk weight with the actual boundary crossing distance to obtain a risk assessment result that better aligns with oilfield security logic. Specifically, the system reads the corresponding weight according to the target category and uses this weight to correct the actual boundary crossing distance, obtaining an adjusted boundary crossing distance. The adjusted boundary crossing distance can be understood as a comprehensive risk quantification result based on the actual spatial boundary crossing degree, superimposed with target type risk factors. For high-risk targets, even if their actual boundary crossing distance is small, they may reach a higher warning level after weight correction; for low-risk targets, the comprehensive risk may be relatively low at the same boundary crossing distance. In this way, the system no longer applies the same set of early warning rules to all targets, but instead makes differentiated judgments based on the nature of the target.

[0053] In the specific judgment process, the adjusted boundary crossing distance is compared with multiple pre-set threshold levels. If the adjusted boundary crossing distance does not reach the warning threshold, it is considered as no warning; if it reaches a lower range, it is judged as a Level 1 warning; if it enters the middle range, it is judged as a Level 2 warning; and if it reaches a higher range, it is judged as a Level 3 warning. Different levels correspond to different warning presentation methods. For example, a Level 1 warning is a normal interface prompt, used to remind on-duty personnel to pay attention; a Level 2 warning is a pop-up reminder, used to indicate obvious risks; and a Level 3 warning is triggered by both a pop-up and an audible alarm, used to indicate high-risk events and require prompt handling. Through this dynamic hierarchical approach, the system can elevate boundary crossing behavior from a simple trigger-based alarm to an intelligent warning process that is explainable, quantifiable, and allows for differentiated handling.

[0054] To avoid a disconnect between the fixed target type weights and the actual risk distribution, this implementation also includes a dynamic weight adjustment mechanism. Based on statistical results of historical oilfield security events over a period of time, the system can adaptively adjust the risk weights for different target types within a small range. For example, if there is a significant increase in vehicle intrusion incidents in a certain oilfield during a certain period, the system can appropriately increase the risk weight of vehicle targets; if a certain type of animal frequently triggers inefficient warnings, its weight can be adjusted to improve risk matching. This adjustment process does not rely entirely on manual reprogramming but is completed based on a preset cycle and actual statistical results, ensuring that the system remains consistent with the actual risk structure even after long-term operation.

[0055] After determining the warning level, the system enters the real-time warning information push and log recording phase. The main task of this phase is to promptly present the generated warning results to operations and maintenance personnel and retain all key process data. For real-time pushes, the system displays corresponding information differentiated according to the warning level, marking the location, type, distance of the intruding target, and warning level on the visual interface, and triggering different alert formats based on the level. This allows on-duty personnel to intuitively see on the monitoring screen which target has crossed the boundary, to what extent it has crossed, and the level of risk, enabling them to respond quickly.

[0056] Simultaneously, the system fully records each detection and warning event. The record includes at least the detection time, unique target identifier, target type, target center location, boundary crossing status, actual boundary crossing distance, adjusted boundary crossing distance, and final warning level. This data is stored in a local database or designated storage unit, supporting subsequent queries based on time, target type, warning level, and other criteria. Log recording not only serves post-event traceability and event review but also provides a data foundation for subsequent model optimization, risk statistical analysis, and fence strategy adjustments. For example, by statistically analyzing the time distribution of high-level warnings in a certain area, it can be determined whether the area requires further strengthened control; or by analyzing long-term logs to discover frequent false alarms for a certain type of target, the model and weight parameters can be optimized in reverse.

[0057] Finally, the entire process results are visualized and interactively managed. The visualization interface, serving as the human-computer interaction platform for the entire system, not only displays results but also handles parameter configuration, fence editing, data management, model training triggering, and log querying. The main monitoring interface draws the current electronic fence boundary, allowing maintenance personnel to intuitively see the system's monitoring range. Target detection boxes, target center positions, and historical movement trajectories are overlaid on the video feed, making the target movement process clearly visible. For targets that cross the boundary, the crossing distance and warning level are additionally marked next to them, enabling personnel to quickly distinguish different risk targets. For objects currently triggering alarms, their status can be highlighted in a more prominent manner. The interface also allows for setting up fence parameter modification entry points, enabling personnel to directly drag vertices and adjust boundaries on the graphical interface; setting up entry points for modifying target type weights and warning thresholds to adjust strategies based on on-site conditions; and setting up log query and data saving entry points for managing historical records and exporting analytical data. Through this design, the entire system is no longer a simple patchwork of multiple independent functional modules but rather an integrated operation of configuration, identification, judgment, alarm, and traceability, all carried out through a unified visualization platform.

[0058] In summary, the technical solution provided in this implementation method is a holistic solution that starts with visual perception, uses geographic electronic fences as spatial constraints, employs continuous target tracking as a dynamic basis, quantifies boundary crossing distances as a risk characterization method, uses target type weight correction as a differentiated judgment mechanism, provides multi-level early warning outputs as business results, and integrates visualization and log recording as a management closed loop. It does not simply rely on a single detection algorithm or fence judgment rule, but rather coordinates the design at the system level for data sources, target identification, trajectory maintenance, spatial verification, risk calculation, early warning handling, and operation and maintenance management in oilfield perimeter security. This enables continuous identification, precise quantification, and dynamic hierarchical early warning of boundary crossing behavior by perimeter agents in complex oilfield environments.

[0059] Implementation Method 3: This implementation method is described in detail with reference to the accompanying drawings. Specific embodiments are provided to further illustrate the technical solutions offered above. Specifically: To address the core shortcomings of existing oilfield electronic fence boundary crossing early warning technologies, this implementation proposes a vision-based method and system for identifying and hierarchically warning of boundary crossing behavior by intelligent agents in oilfield electronic fences, aiming to achieve the following objectives: By integrating visual perception and geographic electronic fence technology, it can accurately identify the types of targets that cross the boundary, and set differentiated weights for different types of targets to solve the industry pain point of undifferentiated early warning; A dynamic early warning level determination method based on quantitative calculation of boundary crossing distance and target type weighting is designed to realize the graded assessment of boundary crossing risk in oilfields, replacing the traditional single early warning mode; Integrating target detection and multi-target tracking technologies improves the accuracy of target detection and the continuity of tracking in complex oilfield scenarios, and reduces the false alarm and false negative rates of boundary crossing judgments; A dual-source processing mechanism of synthetic data generation + real data loading is constructed to make up for the lack of real data in oilfield perimeter security and improve the generalization ability of early warning models. Develop the function of flexibly configuring geofence coordinates to adapt to the scene adjustment needs of different oilfield perimeters and improve the practicality of the system; Build an integrated visual interactive system that integrates core functions across the entire process to achieve real-time display, automatic recording, and traceability of early warning information, thereby reducing the operational costs for maintenance personnel; It achieves an accuracy rate of ≥95% in recognizing boundary crossing behavior of perimeter intelligent agents in complex oilfield environments, and the early warning level judgment is highly matched with the actual security risks in oilfields, meeting the needs of harsh environment adaptation and actual production for perimeter security in oilfields.

[0060] Technical solution The technical solution of this implementation is based on a closed-loop process of dual-source data processing → target detection and multi-target tracking → geographic electronic fence verification → boundary crossing distance quantification → dynamic hierarchical early warning → visual monitoring and recording. It covers six core functional modules, and all modules have been improved to address the shortcomings of existing technologies. The core formulas and judgment rules are derived from the core logic of target type weight design, boundary crossing distance calculation, and early warning level classification, realizing the integrated design of visual perception and oilfield electronic fence early warning.

[0061] System Overall Architecture The system architecture of this implementation is centered on the fusion of visual perception and geographic electronic fences, supported by deep learning, multi-target tracking, and geospatial positioning technologies, and uses full-process visualization as the operational platform. It constructs a six-layer closed-loop architecture, enabling data exchange and result feedback between each layer. The specific architecture layers are as follows: Parameter configuration layer: Enables flexible setting and updating of oilfield geofence coordinates, target type weight configuration, early warning level threshold adjustment, and output of personalized security parameters; Data processing layer: Enables adaptive generation of synthetic security data and standardized loading of real security data, providing data support for early warning model training and target detection; Target detection and tracking layer: Based on deep learning target detection and multi-target tracking technology, it realizes accurate detection and continuous tracking of targets at the oilfield perimeter, and outputs information such as target type, location, and trajectory; Geofencing verification layer: Based on the spatial positioning principle of polygon geofencing, it completes the boundary judgment of the target location and outputs the verification result of whether the target has crossed the boundary; Tiered early warning layer: By quantifying the distance of crossing the boundary and integrating it with the target type weight, the early warning level is dynamically determined, and the early warning level and early warning information are output. Visualization layer: Enables integrated display of fence configuration, target detection and tracking, boundary crossing verification, and tiered early warning results, while automatically recording all operation and early warning logs to achieve data traceability.

[0062] Core module design The six core modules of this implementation method are: dual-source data processing module, geographic electronic fence verification module, target detection and multi-target tracking module, boundary crossing distance quantification calculation module, dynamic hierarchical early warning judgment module, and integrated visualization monitoring module. The design ideas, core improvements, and core formulas / judgment rules of each module are as follows: Dual-source data processing module Design Concept: A dual-data source supply mechanism is constructed based on a GAN network for generating synthetic data and an incremental learning mechanism using real samples. The GAN network simulates scenarios such as dust storms, low light levels, and multi-target interference at the oilfield perimeter, generating labeled boundary-crossing trajectory data. Real samples are used in an incremental learning mode to continuously iterate and optimize the early warning model. Labels include target type, boundary-crossing status, and early warning level. Real data supports standardized format loading and parsing, enabling seamless switching between the two data sources and providing sufficient data for training the early warning model.

[0063] Core improvements: Addressing the scarcity of real-world data for oilfield perimeter security; synthetic data can be used to simulate oilfield security scenarios with different target types and varying boundary crossing distances in batches; real-world data undergoes automated parsing and feature extraction, improving data processing efficiency by ≥50%. Key Design: Synthetic Data: The system presets the background of the oilfield perimeter and the coordinates of the geofence, randomly generates trajectory data for four types of targets: personnel, vehicles, unidentified objects, and wild animals, automatically calculates the boundary crossing status, boundary crossing distance, and warning level, and outputs a synthetic dataset with complete annotations. Real Data: Supports loading real trajectory / video frame labeled data in mainstream formats such as CSV, automatically parses target type, location and other feature information, completes data standardization processing, and adapts to the input requirements of early warning models and detection and tracking modules.

[0064] Geographic electronic fence verification module Design concept: Based on the principle of polygon geospatial positioning, a flexibly configurable electronic fence is constructed. The target center coordinates obtained by target detection and tracking are spatially compared with the fence coordinates to accurately verify the target's boundary crossing behavior.

[0065] Key improvements: Replacing the traditional fixed-coordinate electronic fence design, it supports real-time modification of the fence polygon boundary coordinates via a visual interface. After modification, it automatically synchronizes to the target detection and tracking module within 1 second, without requiring a system restart, and quickly adapts to different oilfield perimeter scenarios; polygon point detection based on target center coordinates improves the accuracy of boundary crossing verification, reducing the false judgment rate by ≥80%. Core determination rule: Let the geographic electronic fence be an n-sided polygon with vertex coordinates P1(x1,y1), P2(x2,y2),...,Pn(xn,yn). The target center coordinates are P(x,y). The boundary crossing state is determined by the positional relationship between the spatial point and the polygon. If P(x,y) is outside the polygon, it is considered to be out of bounds; If P(x,y) is inside the polygon / on its boundary, it is considered not to have crossed the boundary.

[0066] Target detection and multi-target tracking module Design Concept: This design integrates deep learning-based object detection and multi-object tracking technologies. First, it detects objects in video frames / images of the oilfield perimeter, identifies object types, and outputs bounding boxes. Then, it continuously tracks the detected targets using multi-object tracking technology, outputting unique identifiers and motion trajectories to ensure the continuity of boundary crossing detection. Core Improvement: This design replaces the traditional single-object detection mode, achieving an integrated design of detection and tracking. It solves the problems of missed target detection and tracking interruptions in complex oilfield scenarios, improving the continuity of target tracking by ≥90%. Key Design Features: Target detection: Extract depth features from oilfield perimeter images / video frames to accurately identify four types of targets: personnel, vehicles, unidentified objects, and wild animals, and output the coordinates of the target detection box and the target type; Multi-target tracking: Based on the appearance features and movement trajectory of the target, a unique identifier is assigned to each target to achieve continuous tracking of the target and update the center coordinates of the target in real time, providing dynamic position data for fence verification.

[0067] Cross-boundary distance quantization calculation module Coordinate acquisition accuracy requirements: The target location coordinates are acquired through a visual perception module, and the planar coordinate accuracy is ±0.05m after camera calibration. The geofence boundary coordinates are acquired using RTK positioning, with latitude and longitude coordinate accuracy of ±0.1m and planar projection coordinate accuracy of ±0.08m.

[0068] Error handling method: Gaussian filtering algorithm is used to remove coordinate acquisition noise. Outliers with single-frame coordinate errors exceeding ±0.1m are corrected by neighborhood mean. After three consecutive frames of coordinate stability, they are used in distance calculation.

[0069] Mathematical model and parameter range: The shortest distance algorithm of planar polygons is used to calculate d (P,F). Parameter limits: the angle between the target point and the fence edge is ≥15°, the distance calculation iteration is ≤5 times, the calculation result is retained to 2 decimal places, and the effective value range of the out-of-bounds distance D is 0~500m.

[0070] Core formula: Let the center coordinates of the out-of-bounds target be P(x,y), the geographic electronic fence be a polygon F, and the vertical distance from the target to the fence boundary be D. Then the out-of-bounds distance is:

[0071] Where d(P,F) is the spatial perpendicular distance from the target point P to the polygon F, D≥0, and the larger the value of D, the higher the degree of boundary violation.

[0072] Dynamic hierarchical early warning judgment module Design Concept: As the core innovative module of this implementation method, differentiated risk weights are first set for different types of outbound targets. Then, the outbound distance is fused with the target type weight to obtain the adjusted outbound distance. Finally, warning levels are divided based on the threshold of the adjusted outbound distance, realizing dynamic hierarchical warnings. Core Improvement: Replacing the traditional single warning level mode, this is the first time that the target type weight and outbound distance are fused as the basis for determining the warning level, realizing differentiated classification of outbound risk and significantly improving the reference value of warning information. Core formulas and judgment rules: Weighting coefficient determination method: Based on the historical incident data of perimeter security in the oil field over the past 3 years (personnel intrusion accounted for 68%, vehicle illegal crossing accounted for 22%, unidentified object intrusion accounted for 9%, and wild animal interference accounted for 1%), combined with the scoring method of petroleum security industry experts (weight ratio 7:3), the basic risk weights of the four types of targets are determined.

[0073] Mapping relationship between weight and warning level: The target type weight directly participates in the calculation of the adjusted out-of-bounds distance. The higher the weight, the faster the warning level increases under the same out-of-bounds distance, and the higher the risk matching degree.

[0074] Dynamic adjustment mechanism: The system supports automatic weight updates based on quarterly security data of the oilfield. Every 90 days, based on the statistical results of boundary crossing incidents in the oilfield, the weight coefficients are adaptively corrected within a range of ±0.05, without the need for manual adjustment. Target type weight assignment: Fixed weights Wt are assigned to four types of targets according to their oilfield security risk levels.

[0075] Adjusted out-of-bounds distance calculation: The actual out-of-bounds distance is combined with the target type weight to obtain the adjusted out-of-bounds distance. Achieve quantitative adjustment of risk:

[0076] Warning Level Determination: Based on the adjusted threshold of the out-of-bounds distance, the warning level is divided into 4 levels (Level 0 is no warning, and Levels 1-3 are progressively higher warning levels):

[0077] Integrated Visual Monitoring Module Design concept: Build a visual interactive interface that integrates full-process functions such as parameter configuration, data management, model training, real-time detection and tracking, boundary crossing verification, hierarchical early warning, and log recording. This enables one-click operation and real-time display of results for oilfield electronic fence early warning, while automatically recording all early warning information and operation logs, and supporting local data storage and traceability.

[0078] Key improvements: Solving the problem of fragmented traditional early warning processes and achieving integrated functionality across the entire process; visually displaying target trajectories, boundary crossing locations, and early warning levels, making operation intuitive and reducing the operational costs for maintenance personnel by ≥60%; and automating the recording of early warning logs to solve the problem of untraceable data. Key Design: Interactive configuration: Supports real-time visual modification of geofence coordinates, one-click saving and synchronization to the target detection module, and supports visual modification and updating of target type weight and warning level threshold. Real-time display: Draw electronic fences, target detection boxes, and target trajectories on the interface, and label the target type, boundary crossing distance, and warning level; Log recording: Automatically records information such as detection time, target type, boundary crossing status, boundary crossing distance, and warning level. Supports querying, clearing, and local saving of logs; Model Management: Integrates a visual operation for synthetic data generation, real data loading, and early warning model training, allowing model training to be completed without the need for professional coding.

[0079] Core Implementation Steps The vision-based oilfield electronic fence intelligent agent boundary crossing behavior recognition and hierarchical early warning method of this embodiment is specifically implemented in steps S1-S7, with each step executed sequentially and data exchanged to form a complete closed loop of security detection-boundary crossing verification-hierarchical early warning: S1: System Initialization and Parameter Configuration Start the visualization system, complete the initialization of devices and modules, and adapt to the CPU / GPU hardware environment; Configure the initial polygon coordinates of the geofence in the visual interface, support real-time modification of the fence boundary coordinates, and directly synchronize the modification command to the target detection module, and synchronously set the target type weight and warning level distance threshold. Initialize the target detection and tracking module and the early warning model to prepare for subsequent processes.

[0080] S2: Preparation and Loading of Dual-Source Data Select the data source through the visual interface: synthetic data or real data; If you choose to use synthetic data: set the number of samples, the system will generate synthetic data of oilfield perimeter security with complete annotations in batches and output the synthetic dataset; If you choose real data: Select the real data file path, the system will automatically load and parse the data, complete the standardization process, and output standardized real data; The processed dataset is allocated to the early warning model training module to provide data support for the training of the early warning level determination model. Train the early warning level determination model, and complete the model optimization and saving.

[0081] S3: Object Detection and Multi-Object Tracking Initiate real-time visual acquisition (camera / video stream) of the oilfield perimeter to acquire continuous image / video frames; Perform target detection on each frame of the image, identify the target type and output the coordinates of the target detection box, and discard low-confidence detection results; The system performs multi-target tracking on detected targets, assigns a unique identifier to each target, calculates the target center coordinates in real time, and outputs the target's unique ID, type, center coordinates, and motion trajectory.

[0082] S4: Geographic Fence Boundary Verification Input the target center coordinates obtained from target detection and tracking into the geographic electronic fence verification module; Based on the principle of polygon spatial positioning, the boundary status (out of bounds / not out of bounds) of each target is determined. Output the target's out-of-bounds status verification results to provide a filtering basis for calculating the out-of-bounds distance.

[0083] S5: Boundary Crossing Distance Calculation and Dynamic Hierarchical Early Warning Judgment For targets that are detected as crossing the boundary, the actual crossing distance D from the target to the electronic fence boundary is calculated using the boundary crossing distance quantification calculation module. Based on the target type, match the corresponding risk weight Wt and calculate the adjusted out-of-bounds distance Dadjust. Based on the adjusted threshold for the distance of intrusion, the warning level L for each intruding target is determined; Corresponding warning information is generated for targets with different warning levels. Level 1 warning is a general prompt, Level 2 warning is a pop-up reminder, and Level 3 warning is a pop-up and sound alarm.

[0084] S6: Real-time push notifications and log recording of early warning information Different warning information is pushed to the visual interface according to the different warning levels; The system automatically records all information for this warning, including detection time, target ID, target type, center coordinates, boundary crossing distance, adjusted boundary crossing distance, and warning level. The warning logs are stored in a local database for subsequent querying and tracing.

[0085] S7: Visualization and operation of the entire process results In the monitoring canvas of the visualization interface, draw the geographic electronic fence, target detection box, target center coordinates and movement trajectory; The target is labeled, displaying the target ID and target type. For targets that cross the boundary, the crossing distance and warning level are additionally labeled. Operations and maintenance personnel can perform operations such as modifying fence coordinates, querying early warning logs, and saving data in the interface, realizing full-process visual interaction.

[0086] The core innovations of this implementation method are reflected in six aspects. All innovations address the core shortcomings of existing oilfield electronic fence boundary crossing early warning technologies, possessing significant technical originality and practical application value. The specific innovations are as follows: A boundary crossing identification method that integrates visual perception and geographic electronic fence: This method is the first to integrate computer vision target type recognition technology into oilfield electronic fence early warning, replacing the traditional single inductive boundary crossing judgment, achieving accurate identification of boundary crossing targets and differentiated risk assessment, and solving the industry pain point of traditional technology lacking target type recognition. Dynamic early warning level determination mechanism based on target type weight and boundary crossing distance: Differentiated risk weights are designed for target types, and the adjusted boundary crossing distance is obtained by integrating the quantitative calculation of boundary crossing distance. Dynamic graded early warning is realized based on threshold, replacing the traditional single early warning mode, so that the early warning level is highly matched with the actual security risks of the oilfield. A dual-data source adaptive processing mechanism based on GAN network and incremental learning from real samples: This mechanism constructs a dual-data source model that combines batch generation of synthetic data using GAN network and incremental learning from real samples, effectively addressing the scarcity of real data for oilfield perimeter security. This provides ample data support for training early warning models and significantly improves the model's generalization ability. Integrated design of target detection and multi-target tracking: It integrates deep learning target detection and multi-target tracking technologies to achieve accurate detection and continuous tracking of targets in complex oilfield scenarios, providing dynamic and continuous target location data for boundary crossing judgment, reducing the rate of missed detection and false alarm, and improving the continuity and accuracy of boundary crossing identification; Flexible configurable polygon geofence design: Breaking through the limitations of traditional fixed coordinate geofences, it supports flexible modification of the geofence polygon coordinates through a visual interface, and can quickly adapt to the scene adjustment needs of different oilfield perimeters, improving the system's practicality and adaptability. The oilfield electronic fence early warning system is an integrated visualization system that integrates parameter configuration, data processing, detection and tracking, fence verification, hierarchical early warning, and log recording. It features an intuitive visual interface that enables real-time display and automated recording of early warning information. This system solves the problem of fragmented traditional processes, reduces the operating costs for maintenance personnel, and ensures the traceability of early warning data.

[0087] Combination Figure 1 It displays the visualization layer, hierarchical early warning layer, geofence verification layer, target detection and tracking layer, data processing layer, and parameter configuration layer; the data flow between each layer is indicated by arrows.

[0088] Combination Figure 2 The process demonstrates the following steps: S1 System initialization and parameter configuration → S2 Preparation and loading of dual-source data → S3 Target detection and multi-target tracking → S4 Geographic electronic fence boundary crossing verification → S5 Boundary crossing distance calculation and dynamic hierarchical early warning judgment → S6 Real-time push of early warning information and log recording → S7 Visualization and operation of the entire process results.

[0089] The above description of several specific embodiments further details the technical solution provided by the present invention in order to highlight the advantages and benefits of the technical solution provided by the present invention. However, the above-described specific embodiments are not intended to limit the present invention. Any reasonable modifications and improvements to the present invention, combinations of embodiments, and equivalent substitutions based on the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying and hierarchically warning of boundary crossing behavior of intelligent agents in oilfield electronic fences based on visual perception, characterized in that, include: The steps to configure geofence parameters for the oilfield perimeter, set target type risk weights and early warning level thresholds, and output fence boundary data and risk parameters. Based on the fence boundary data and risk parameters, the steps include acquiring oilfield perimeter image data and performing target detection and multi-target tracking, identifying target types and generating unique target identifiers, target center positions and movement trajectories; The steps include determining the spatial relationship between the target's center location and the fence boundary data, identifying the target's boundary crossing status, and filtering out boundary-crossing targets. The steps are: calculating the shortest distance from the target to the fence boundary based on the center position of the target and the fence boundary data, obtaining the boundary crossing distance, and outputting the boundary crossing distance data; The steps are as follows: Based on the fusion processing of the cross-boundary distance data and the target type risk weight, the adjusted cross-boundary distance is obtained and compared with the warning level threshold to determine the target warning level and generate corresponding warning information. Based on the aforementioned warning information, differentiated prompts are provided to targets, and the target type, target location, boundary crossing distance, and warning level are recorded, thereby realizing the steps of visual display and log storage of boundary crossing behavior in the oilfield.

2. The method for identifying and hierarchically warning of boundary crossing behavior of intelligent agents in oilfield electronic fences based on visual perception, as described in claim 1, is characterized in that... The geographic electronic fence uses a polygon boundary representation, and the coordinates of the polygon vertices can be interactively modified in a visual interface. The modification results are synchronized in real time for boundary violation determination.

3. The method for identifying and hierarchically warning of boundary crossing behavior of intelligent agents in oilfield electronic fences based on visual perception, as described in claim 1, is characterized in that... Target detection is used to identify four types of targets: people, vehicles, unidentified objects, and wild animals. The detection results are then filtered based on confidence to remove targets with low confidence.

4. The method for identifying and hierarchically warning of boundary crossing behavior of intelligent agents in oilfield electronic fences based on visual perception, as described in claim 1, is characterized in that... Multi-target tracking associates and matches targets based on their appearance features and motion trajectories, and assigns a unique identifier to each target to maintain continuity across frames.

5. A method for identifying and hierarchically warning of boundary crossing behavior of an intelligent agent in an oilfield electronic fence based on visual perception, as described in claim 1, is characterized in that... Boundary crossing is determined by judging the spatial relationship between the target center point and the boundary of the polygonal fence. When the target center point is located outside the fence, it is judged as crossing the boundary.

6. The method for identifying and hierarchically warning of boundary crossing behavior of intelligent agents in oilfield electronic fences based on visual perception, as described in claim 1, is characterized in that... Error correction processing is performed on the target center location data, including filtering out abnormal fluctuations and smoothing continuous locations, to improve the accuracy of boundary crossing distance calculation.

7. A vision-perception-based intelligent agent boundary crossing behavior recognition and hierarchical early warning system for oilfield electronic fences, characterized in that, include: A module for configuring geofence parameters for oilfield perimeters, setting target type risk weights and early warning level thresholds, and outputting fence boundary data and risk parameters; Based on the fence boundary data and risk parameters, a module is used to acquire oilfield perimeter image data, perform target detection and multi-target tracking, identify target types, and generate unique target identifiers, target center positions, and movement trajectories. A module that determines the target's boundary crossing status and filters out targets crossing the boundary based on the spatial relationship between the target's center location and the fence boundary data; This module calculates the shortest distance from the target to the fence boundary based on the center position of the target and the fence boundary data, obtains the boundary crossing distance, and outputs the boundary crossing distance data. A module that integrates the boundary crossing distance data with the target type risk weight to obtain the adjusted boundary crossing distance and compares it with the warning level threshold to determine the target warning level and generate corresponding warning information. Based on the aforementioned warning information, the module provides differentiated prompts for targets and records the target type, target location, boundary crossing distance, and warning level, thereby enabling the visualization and log storage of boundary crossing behavior in the oilfield.

8. A computer storage medium for storing computer programs, characterized in that, When the computer program is read by the computer, the computer executes the method of claim 1.

9. A computer, comprising a processor and a storage medium, characterized in that, When the processor reads the computer program stored in the storage medium, the computer executes the method of claim 1.

10. A computer program product, as a computer program, is characterized by: When the computer program is executed, it implements the method of claim 1.

Images