A method for real-time multi-target positioning and risk warning using AI-controlled spheres in gas construction scenarios

Abstract

This invention relates to the field of gas construction safety technology, and in particular to a method for real-time multi-target positioning and risk warning using an explosion-proof AI-controlled sphere in gas construction scenarios. The method includes the following steps: S1: multi-source data collaborative acquisition; S2: AI multi-target intelligent identification and real-time positioning; S3: scenario-based dynamic risk assessment; S4: graded early warning and linkage response; S5: data recording and model iterative optimization. This invention's method for real-time multi-target positioning and risk warning using an explosion-proof AI-controlled sphere in gas construction scenarios overcomes the shortcomings of traditional gas construction positioning methods, such as interference from obstructions, static lag in risk assessment, and weak early warning linkage. It is adaptable to various gas construction scenarios, possesses explosion-proof, low-power, and rapid deployment characteristics, and can significantly improve the intelligence and precision of gas construction safety protection. It can be widely applied to gas construction companies, emergency management agencies, etc., to ensure the safety of personnel, equipment, and gas facilities during construction, and to prevent safety accidents such as gas leaks and explosions.

Description

Technical Field

[0001] This invention relates to the field of gas construction safety technology, and in particular to a method for real-time multi-target positioning and risk warning using an explosion-proof AI-controlled sphere in gas construction scenarios. Background Technology

[0002] Gas pipeline construction is characterized by complex environments, numerous risks, and extremely high safety requirements. The construction process often faces challenges such as dense underground pipelines, variable geological conditions, and GNSS signal obstruction or absence. Traditional safety management methods are no longer sufficient to meet the needs of precise protection, and the main shortcomings are as follows: 1. Insufficient accuracy in multi-target positioning: Traditional positioning relies on a single GNSS or visual sensor, which is prone to positioning deviations in scenarios such as construction obstruction, electromagnetic interference, and underground operations. It is impossible to accurately obtain the relative positional relationship between construction personnel, equipment and gas pipelines, which can easily lead to risks such as collisions and accidental excavation.

[0003] 2. Static lag in risk assessment: Existing risk assessments are mostly based on fixed safety rules or manual inspections, without integrating dynamic factors such as real-time environmental data and target movement status. This makes it difficult to adapt to the safety requirements of different construction stages such as pipeline excavation and connection, and it is impossible to identify dynamically changing risks in a timely manner.

[0004] 3. Weak linkage between early warning and response: Early warning methods are mostly single sound and light prompts, lacking linkage mechanisms with construction equipment and gas pipeline control devices. In high-risk scenarios, it is impossible to quickly trigger emergency measures such as shutdown and gas cut-off, which can easily lead to the escalation of the situation.

[0005] 4. Limited Functionality of Explosion-Proof Equipment: Traditional explosion-proof monitoring equipment only has video acquisition capabilities, lacking AI intelligent analysis, multi-target tracking, and risk assessment capabilities, thus failing to achieve closed-loop management of the entire process from "perception to identification to assessment to early warning to response." Poor Scenario Adaptability: Safety thresholds and risk types vary significantly across different gas construction scenarios. Existing methods require reconstruction of algorithm models to adapt to new scenarios, resulting in low deployment efficiency and high costs.

[0006] To address the aforementioned issues, this invention proposes a method for real-time multi-target positioning and risk warning using an explosion-proof AI-controlled sphere in gas construction scenarios. Summary of the Invention

[0007] The main objective of this invention is to provide a method for real-time multi-target positioning and risk warning of explosion-proof AI-controlled spheres in gas construction scenarios, which can effectively solve the problems in the background technology.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for real-time multi-target positioning and risk warning using AI-controlled surveillance cameras in gas construction scenarios includes the following specific steps: S1: Multi-source data collaborative acquisition: The explosion-proof AI control ball integrates a visual acquisition module, a multimodal sensor group, and a high-precision positioning module to simultaneously collect image and video data, environmental perception data, and spatial location data of the construction area. S2: AI-powered multi-target intelligent recognition and real-time positioning: The collected data is processed based on a multimodal fusion algorithm to automatically identify target types such as construction workers, construction equipment, gas pipelines, fire sources, and dangerous area boundaries, and output the real-time three-dimensional coordinates and motion trajectories of each target. S3: Scenario-based dynamic risk assessment: By combining the gas construction safety rule base with real-time location data, a deep learning model is used to quantitatively assess the risk status of safety distance conflicts, unauthorized intrusions, and dangerous operations between targets, and output the risk level. S4: Tiered Early Warning and Coordinated Response: Based on the risk level, corresponding audible and visual warnings, terminal push notifications, and equipment linkage control commands are generated, while risk events and target trajectories are recorded to form a closed loop of safety management. S5: Model Iterative Optimization By collecting historical risk event data, positioning error data, and early warning feedback results, the multi-target identification model, positioning fusion algorithm, and risk assessment model are incrementally trained to continuously improve positioning accuracy and risk identification accuracy.

[0009] Preferably, in step S2, the AI ​​multi-target intelligent recognition and real-time localization adopts a fusion architecture of "YOLOv8 lightweight detection model + UWB localization module + Kalman filter", specifically including: The YOLOv8 model is used to extract visual features of multiple targets in images, enabling the classification and recognition of targets such as construction workers, excavators / welding machines, gas pipeline signs, and open flames / smoke. By integrating centimeter-level spatial location data from UWB positioning modules and combining it with Kalman filtering algorithms to correct target trajectory, positioning deviations caused by obstruction, electromagnetic interference, and missing GNSS signals in construction scenarios are resolved. It automatically associates target type with location data to generate structured location information that includes target ID, type, real-time coordinates, movement speed, and dwell time.

[0010] Preferably, in step S3, the scenario-based dynamic risk assessment adopts a fusion model of "Bayesian network + reinforcement learning", specifically including: Construct a gas construction safety risk factor library, covering scenario-based factors such as target safety distance thresholds, hazardous area access rules, equipment operation specifications, and gas leakage concentration thresholds; Using real-time multi-target positioning data and environmental perception data (combustible gas concentration, temperature and humidity, vibration value) as inputs, the probability of risk occurrence is quantified through a Bayesian network; By using reinforcement learning algorithms to dynamically update risk assessment rules, the system can adapt to the safety requirements of different construction stages such as pipeline excavation, connection, and maintenance, and achieve real-time iterative updates of risk levels.

[0011] Preferably, in step S4, the graded early warning and linkage response are divided into four levels according to risk level, specifically including: Level 1 Warning (Low Risk): Local audio and visual alerts are provided via explosion-proof AI surveillance cameras, and warning information is simultaneously pushed to the mobile terminals of on-site management personnel. Level 2 Warning (Medium Risk): Enhance the intensity of audible and visual warnings, coordinate with the construction area broadcasting system to provide voice reminders, and record the trajectory of the target violation; Level 3 warning (high risk): Triggers an emergency stop command for construction equipment, blocks access to the electronic fence in the danger zone, and sends real-time alarms and on-site images to the project management platform; Level 4 Warning (Extremely High Risk): In addition to the above response, the gas pipeline emergency shut-off valve control module will be automatically activated and the information will be simultaneously reported to the regulatory authority's emergency system.

[0012] Preferably, in step S1, the multimodal sensor group includes a combustible gas sensor, an infrared thermal imaging sensor, a vibration sensor, and a temperature and humidity sensor. After the data from each sensor is processed by explosion-proof encapsulation, it is preprocessed in real time by an edge computing module to filter out environmental noise interference.

[0013] Preferably, in step S2, the multi-target localization supports the simultaneous tracking of no less than 50 dynamic targets, the localization update frequency is ≥10Hz, the static target localization error is ≤5cm, the dynamic target localization error is ≤10cm, and the target type recognition accuracy is ≥98%.

[0014] Preferably, in step S3, the risk level is divided into four levels: low, medium, high, and extremely high. The triggering conditions include: the safety distance between multiple targets is less than a preset threshold, unauthorized personnel enter the dangerous area, construction equipment is too far from the gas pipeline, the concentration of combustible gas exceeds the safety threshold, and open flame / illegal welding operation is detected.

[0015] Preferably, it is compatible with various scenarios such as gas pipeline excavation, pipeline connection, equipment maintenance, and station construction. The target recognition parameters, safety thresholds, and risk assessment rules can be adjusted through scenario-based configuration interfaces, and it can be quickly deployed without reconstructing the algorithm model.

[0016] Preferably, the explosion-proof AI control ball adopts an Ex d IIC T6 level explosion-proof design, the algorithm module is adapted to low power consumption operation requirements, supports dual modes of battery power supply and solar charging, and works stably under ambient temperature of -40℃ to 85℃ and IP67 protection level.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. Accurate and stable positioning: The multi-modal fusion positioning architecture solves problems such as occlusion, electromagnetic interference, and missing GNSS signals, achieving centimeter-level positioning and continuous trajectory tracking of multiple targets, and significantly improving positioning reliability.

[0018] 2. Dynamic and Intelligent Risk Assessment: Integrating Bayesian networks and reinforcement learning, it enables the quantitative calculation of risk probabilities and the dynamic updating of assessment rules, adapting to the safety needs of different construction stages, and making risk identification more accurate; Efficient and Linked Early Warning and Response: The four-level hierarchical early warning mechanism covers the entire scenario from alert to emergency response, linking sound and light, terminals, equipment, and emergency systems to achieve early detection, early warning, and rapid response to risks, reducing accident losses.

[0019] 3. Strong environmental adaptability: The Ex d IIC T6 explosion-proof design and wide temperature range, high protection level adaptability, support dual power supply mode, and can operate stably in various complex gas construction scenarios; Flexible deployment and low cost: The scenario-based configuration interface can quickly adapt to new scenarios without reconstructing the model, and is compatible with existing construction management systems, reducing deployment and upgrade costs; Continuous performance optimization: The model iteration optimization mechanism can continuously improve system performance based on actual operation data, forming a safety management closed loop and ensuring long-term construction safety. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the overall method of the explosion-proof AI-controlled ball multi-target real-time positioning and risk warning method for gas construction scenarios according to the present invention. Figure 2 This is an AI multi-target positioning fusion architecture diagram of the explosion-proof AI deployment ball multi-target real-time positioning and risk warning method for gas construction scenarios according to the present invention. Figure 3 This is a structural diagram of a scenario-based dynamic risk assessment model for the real-time positioning and risk warning method of multi-target AI-controlled ball in gas construction scenarios according to the present invention. Figure 4 This is a logic diagram of hierarchical early warning and linkage response for the explosion-proof AI-controlled multi-target real-time positioning and risk early warning method in the gas construction scenario of the present invention. Detailed Implementation

[0021] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0022] like Figure 1-4 As shown, the method for real-time multi-target positioning and risk warning using an explosion-proof AI-controlled sphere in a gas construction scenario includes the following steps: S1: Multi-source data collaborative acquisition Data acquisition equipment: Based on an explosion-proof AI control ball, it integrates a visual acquisition module (high-definition camera, infrared thermal imaging lens), a multimodal sensor group, and a high-precision positioning module (UWB positioning module + GNSS / BeiDou assisted positioning).

[0023] Data collected: Simultaneously collect image and video data, environmental perception data (combustible gas concentration, temperature and humidity, vibration value, open flame / smoke signal), and spatial location data (position of the control ball itself and relative position of the target object) of the construction area.

[0024] Data preprocessing: After the data from each sensor is processed by Ex d IIC T6 level explosion-proof packaging, it is preprocessed in real time by the edge computing module built into the control ball to filter environmental noise interference, extract effective data features, and reduce the processing pressure of subsequent algorithms.

[0025] S2: AI Multi-Target Intelligent Recognition and Real-Time Positioning A fusion architecture combining "YOLOv8 lightweight detection model + UWB localization module + Kalman filter" is adopted to achieve accurate localization of multiple targets. Target recognition: The YOLOv8 lightweight model is used to process image and video data, automatically extracting visual features of targets such as construction workers, construction equipment such as excavators / welding machines, gas pipeline signs / exposed pipelines, open flames / smoke, and hazardous area boundaries, and completing target classification and recognition.

[0026] Location acquisition: The UWB positioning module measures the distance between the target and the control ball through wireless pulse signals, and outputs the target's centimeter-level three-dimensional coordinates by combining the control ball's own spatial position data; the GNSS / BeiDou module assists in calibration in open scenes to improve positioning stability.

[0027] Trajectory Correction: In construction scenarios, Kalman filtering algorithm is used to predict and correct the target's motion trajectory, eliminating positioning errors and outputting a smooth and continuous target motion trajectory.

[0028] Information output: Automatically associates target ID, type, real-time 3D coordinates, movement speed, dwell time and other information to generate structured positioning data, providing a basis for risk assessment.

[0029] S3: Scenario-based Dynamic Risk Assessment A hybrid model combining Bayesian networks and reinforcement learning is employed to achieve dynamic quantitative risk assessment. Construct a risk factor library: covering core safety factors in gas construction scenarios, including scenario-based factors such as multi-target safety distance thresholds, hazardous area access rules, equipment operation specifications, combustible gas concentration thresholds, and safety requirements during the construction phase.

[0030] Probability quantification: Using real-time positioning data of multiple targets and environmental perception data as input, Bayesian networks are used to characterize the causal relationship between risk factors and quantify the probability of occurrence of risks such as safety distance conflict, unauthorized entry, dangerous operation, and gas leakage.

[0031] Dynamic updates: Utilizing reinforcement learning algorithms, with the optimization goals of "risk identification accuracy + response timeliness", the risk assessment rules and thresholds are dynamically updated to adapt to the safety requirements of different construction stages such as pipeline excavation, connection, and maintenance, and to achieve real-time iterative updates of risk levels.

[0032] Risk classification: Outputs four risk levels: low, medium, high, and extremely high, clearly defining the risk triggers and the scope of impact.

[0033] S4: Tiered Early Warning and Coordinated Response Develop differentiated and coordinated response strategies based on risk levels to ensure timely early warning and efficient handling: Level 1 Warning (Low Risk): The explosion-proof AI surveillance camera activates local audio and visual alerts and simultaneously pushes the warning information (risk type, target location, and recommended measures) to the mobile terminal of on-site management personnel without interrupting construction.

[0034] Level 2 Warning (Medium Risk): Increase the intensity and frequency of audible and visual warnings, link the construction area broadcasting system to provide voice reminders, and automatically record the movement trajectory of illegal targets and details of risk events for easy follow-up.

[0035] Level 3 Early Warning (High Risk): Triggers an emergency shutdown command for construction equipment, blocks access to the electronic fence in the danger zone, and prohibits unauthorized personnel from entering; pushes alarm information and on-site images to the project management platform in real time, and notifies repair personnel to rush to the scene.

[0036] Level 4 Warning (Extremely High Risk): Based on the Level 3 warning, the gas pipeline emergency shut-off valve control module is automatically activated to block the leaking gas source; the emergency system of the regional emergency management department is reported simultaneously to initiate the cross-departmental collaborative response process.

[0037] S5: Data Recording and Model Iteration Optimization Explosion-proof adaptation and scenario-based deployment Explosion-proof design: The explosion-proof AI control ball adopts an Ex d IIC T6 level explosion-proof structure, and the shell is made of 304 stainless steel. The protection level reaches IP67, which is suitable for ambient temperature of -40℃~85℃. It can operate safely in flammable and explosive scenarios such as combustible gases and dust.

[0038] Power supply mode: Supports dual modes of battery power and solar charging. Low power consumption algorithm is adapted to outdoor scenarios without external power supply to ensure continuous working time meets construction needs.

[0039] Scenario adaptation: Through the scenario-based configuration interface, users can adjust the target recognition parameters, safety thresholds and risk assessment rules according to different scenarios such as pipeline excavation, connection and site construction, and can quickly deploy without reconstructing the algorithm model.

[0040] Model iterative optimization: Data collection: Continuously collect historical risk event data, location error data, early warning feedback results (such as false alarm / missed alarm records), construction scenario change data, etc.

[0041] Incremental training: Using collected historical data, the multi-target recognition model, localization fusion algorithm, and risk assessment model are incrementally trained to optimize model parameters and continuously improve target recognition accuracy, localization precision, and risk assessment accuracy.

[0042] Each step forms a closed-loop collaboration to ensure intelligent protection throughout the entire process: The multi-source preprocessed data of S1 provides high-quality input for the localization of S2 and the risk assessment of S3; The structured location data of S2 and the environmental data of S1 together drive the risk level calculation of S3; The risk level of S3 directly determines the early warning method and linkage measures of S4, so as to achieve precise matching of "risk-response"; The early warning feedback data and historical operation data of S4 provide a basis for model iteration and optimization, and continuously improve system performance; Explosion-proof design and scenario-based configuration are integrated throughout the entire process to ensure the stable operation of the system in complex construction environments.

[0043] Example: Application of urban gas pipeline excavation construction scenario 1. Application Scenarios Overview The gas pipeline upgrade and renovation project on a main road in a city involves a dense network of underground pipelines (including power and water supply pipelines) in the construction area. The geological structure is soft soil, posing a risk of foundation pit collapse. During the construction period, it is necessary to ensure the normal gas supply to surrounding residents. The project adopts an uninterrupted excavation method, involving equipment such as excavators and welding machines, and 15 construction workers. It is necessary to closely monitor the safe distance between personnel, equipment and gas pipelines to prevent risks such as gas leaks and mechanical collisions with pipelines.

[0044] 2. Implementation Steps Details S1, Multi-source data collaborative acquisition deployment Three explosion-proof AI monitoring balls were deployed around the construction area, using a tripod and magnetic combination for installation, covering the entire excavation face and the exposed gas pipeline area; the monitoring balls were set up in both battery and solar charging modes to ensure continuous operation.

[0045] The multimodal sensor array collects data in real time: combustible gas concentration (monitoring methane leaks), infrared thermal imaging (identifying open flames / welding machine high-temperature points), vibration values ​​(monitoring whether pipelines are impacted), and temperature and humidity (adapting algorithm parameters to the environment); the vision module collects 1080P high-definition video, and the UWB positioning module deploys 4 anchor points to achieve full coverage of the construction area.

[0046] The edge computing module filters noise from the sensor data and transmits the valid data to the AI ​​processing unit.

[0047] S2, AI multi-target intelligent recognition and real-time positioning The YOLOv8 model automatically identifies targets such as construction workers (including whether they are wearing safety helmets), excavators, welding machines, exposed gas pipelines, and foundation pit boundaries, with an accuracy rate of over 98%.

[0048] The UWB positioning module outputs centimeter-level three-dimensional coordinates of each target. Combined with GNSS-assisted calibration, it obtains the real-time distance between the excavator bucket and the gas pipeline, and the relative position of construction personnel and the foundation pit boundary.

[0049] When the excavator's operation obstructs the UWB signal, the Kalman filter algorithm predicts the current position based on the historical motion trajectory, corrects the positioning deviation, and ensures the trajectory is continuous and stable.

[0050] Output structured data: target ID (personnel 01-15, equipment 01-02), type, real-time coordinates, movement speed, dwell time, updated every 100ms.

[0051] S3, Scenario-based Dynamic Risk Assessment Risk factor database configuration: safety distance threshold between gas pipelines and equipment ≥ 1.5m, safety distance between personnel and foundation pit boundary ≥ 0.8m, safety threshold for combustible gas concentration ≤ 0.5% VOL, welding machine operation must be ≥ 3m away from pipelines.

[0052] The Bayesian network inputs real-time data: the distance between the excavator and the pipeline is 1.2m (below the threshold), the concentration of combustible gas is 0.3% VOL (normal), and one construction worker is 0.5m away from the edge of the foundation pit (below the threshold). Quantitative calculations show that "the probability of equipment colliding with the pipeline is 65% and the probability of personnel falling is 70%", which is comprehensively judged as a level two warning.

[0053] The reinforcement learning algorithm dynamically increases the risk weight of "equipment-pipe distance" based on the current construction stage (exposed pipeline work) to ensure that the assessment is adapted to the needs of the scenario.

[0054] S4. Tiered Early Warning and Coordinated Response The system triggered a level-two warning: three surveillance cameras simultaneously activated high-frequency audio and visual alerts, and a voice reminder was broadcast in the construction area: "The excavator is approaching the gas pipeline. Construction workers, please stay away from the edge of the pit."

[0055] The on-site management personnel received an early warning message on their mobile terminals, which included the risk type, the ID of the target at risk, and a screenshot of the real-time location. The management personnel rushed to the scene to guide the adjustment of the excavator's working position and reminded the personnel at risk to stay away from the danger zone.

[0056] The system automatically records the movement trajectory of the at-risk target, the warning time, and the handling results, forming a risk event file.

[0057] S5, Data Recording and Model Iteration Optimization Data was continuously collected during construction: positioning deviation records, early warning response time, false alarms / missed alarms (no false alarms this time), and feedback on the handling effect.

[0058] After the project was completed, the YOLOv8 model was incrementally trained using historical data to optimize the target recognition algorithm in soft soil environments; the conditional probability table of the Bayesian network was adjusted to improve the quantitative accuracy of the risk of "equipment-pipeline distance".

[0059] 3. Application Effects This embodiment achieves intelligent safety protection throughout the entire construction process through this method, successfully avoiding the risks of equipment approaching pipelines at close range three times and personnel unauthorizedly approaching the foundation pit twice; the positioning deviation is controlled within 10cm, the early warning response time is ≤1s, and the risk disposal can be completed without interrupting construction, which not only ensures construction safety but also does not affect the project progress. Compared with the traditional manual inspection method, the safety management efficiency is improved by more than 60%.

[0060] In summary, the present invention provides a method for real-time multi-target positioning and risk warning in gas construction scenarios using an explosion-proof AI-controlled sphere. This method integrates multi-source sensing devices through an explosion-proof AI-controlled sphere to simultaneously collect visual, environmental, and spatial location data of the construction area. It employs a fusion architecture of "YOLOv8 lightweight detection model + UWB positioning module + Kalman filter" to achieve accurate identification and real-time positioning of multiple targets, including construction personnel, equipment, and gas pipelines. A dynamic risk assessment model is constructed based on "Bayesian network + reinforcement learning," and risk levels are quantified using a gas construction safety rule base. Four-level graded early warnings are triggered according to risk levels, linking audible and visual prompts, equipment control, and emergency system reporting to form a closed-loop safety management system. This invention addresses the shortcomings of traditional gas construction positioning methods, such as interference from obstructions, static lag in risk assessment, and weak early warning linkage. It is adaptable to various gas construction scenarios, possesses explosion-proof, low-power, and rapid deployment characteristics, and can significantly improve the intelligence and precision of gas construction safety protection.

[0061] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for real-time multi-target positioning and risk warning using an explosion-proof AI-controlled sphere in gas construction scenarios, characterized in that, The specific steps include the following: S1: Multi-source data collaborative acquisition: The explosion-proof AI control ball integrates a visual acquisition module, a multimodal sensor group, and a high-precision positioning module to simultaneously collect image and video data, environmental perception data, and spatial location data of the construction area. S2: AI-powered multi-target intelligent recognition and real-time positioning: The collected data is processed based on a multimodal fusion algorithm to automatically identify target types such as construction workers, construction equipment, gas pipelines, fire sources, and dangerous area boundaries, and output the real-time three-dimensional coordinates and motion trajectories of each target. S3: Scenario-based dynamic risk assessment: By combining the gas construction safety rule base with real-time location data, a deep learning model is used to quantitatively assess the risk status of safety distance conflicts, unauthorized intrusions, and dangerous operations between targets, and output the risk level. S4: Tiered Early Warning and Coordinated Response: Based on the risk level, corresponding audible and visual warnings, terminal push notifications, and equipment linkage control commands are generated, while risk events and target trajectories are recorded to form a closed loop of safety management. S5: Data Recording and Model Iteration Optimization By collecting historical risk event data, positioning error data, and early warning feedback results, the multi-target identification model, positioning fusion algorithm, and risk assessment model are incrementally trained to continuously improve positioning accuracy and risk identification accuracy.

2. The method for real-time positioning and risk warning of multiple targets using an explosion-proof AI-controlled sphere in a gas construction scenario according to claim 1, wherein in step S2, the AI ​​multi-target intelligent recognition and real-time positioning adopts a fusion architecture of "YOLOv8 lightweight detection model + UWB positioning module + Kalman filter", specifically including: The YOLOv8 model is used to extract visual features of multiple targets in images, enabling the classification and recognition of targets such as construction workers, excavators / welding machines, gas pipeline signs, and open flames / smoke. By integrating centimeter-level spatial location data from UWB positioning modules and combining it with Kalman filtering algorithms to correct target trajectory, positioning deviations caused by obstruction, electromagnetic interference, and missing GNSS signals in construction scenarios are resolved. It automatically associates target type with location data to generate structured location information that includes target ID, type, real-time coordinates, movement speed, and dwell time.

3. The method for real-time multi-target positioning and risk warning of explosion-proof AI-controlled spheres in gas construction scenarios according to claim 1, characterized in that: In S3, the scenario-based dynamic risk assessment adopts a fusion model of "Bayesian network + reinforcement learning", specifically including: Construct a gas construction safety risk factor library, covering scenario-based factors such as target safety distance thresholds, hazardous area access rules, equipment operation specifications, and gas leakage concentration thresholds; Using real-time multi-target positioning data and environmental perception data (combustible gas concentration, temperature and humidity, vibration value) as inputs, the probability of risk occurrence is quantified through a Bayesian network; By using reinforcement learning algorithms to dynamically update risk assessment rules, the system can adapt to the safety requirements of different construction stages such as pipeline excavation, connection, and maintenance, and achieve real-time iterative updates of risk levels.

4. The method for real-time multi-target positioning and risk warning of explosion-proof AI-controlled spheres in gas construction scenarios according to claim 1, characterized in that: In S4, the tiered early warning and coordinated response are divided into four levels according to risk level, specifically including: Level 1 Warning (Low Risk): Local audio and visual alerts are provided via explosion-proof AI surveillance cameras, and warning information is simultaneously pushed to the mobile terminals of on-site management personnel. Level 2 Warning (Medium Risk): Enhance the intensity of audible and visual warnings, coordinate with the construction area broadcasting system to provide voice reminders, and record the trajectory of the target violation; Level 3 warning (high risk): Triggers an emergency stop command for construction equipment, blocks access to the electronic fence in the danger zone, and sends real-time alarms and on-site images to the project management platform; Level 4 Warning (Extremely High Risk): In addition to the above response, the gas pipeline emergency shut-off valve control module will be automatically activated and the information will be simultaneously reported to the regulatory authority's emergency system.

5. The method for real-time multi-target positioning and risk warning of explosion-proof AI-controlled spheres in gas construction scenarios according to claim 1, characterized in that: In step S1, the multimodal sensor group includes a combustible gas sensor, an infrared thermal imaging sensor, a vibration sensor, and a temperature and humidity sensor. After the data from each sensor is processed by explosion-proof encapsulation, it is preprocessed in real time by an edge computing module to filter out environmental noise interference.

6. The method for real-time multi-target positioning and risk warning of explosion-proof AI-controlled spheres in gas construction scenarios according to claim 1, characterized in that: In S2, multi-target localization supports the simultaneous tracking of no less than 50 dynamic targets, with a localization update frequency of ≥10Hz, a static target localization error of ≤5cm, a dynamic target localization error of ≤10cm, and a target type recognition accuracy of ≥98%.

7. The method for real-time multi-target positioning and risk warning of explosion-proof AI-controlled spheres in gas construction scenarios according to claim 1, characterized in that: In step S3, the risk level is divided into four levels: low, medium, high, and extremely high. The triggering conditions include: the safe distance between multiple targets is less than a preset threshold, unauthorized personnel enter the dangerous area, construction equipment is too far from the gas pipeline, the concentration of combustible gas exceeds the safety threshold, and open flame / illegal welding operation is detected.

8. The method for real-time multi-target positioning and risk warning of explosion-proof AI-controlled spheres in gas construction scenarios according to claim 1, characterized in that: It is adaptable to various scenarios such as gas pipeline excavation, pipeline connection, equipment maintenance, and station construction. Target recognition parameters, safety thresholds, and risk assessment rules can be adjusted through scenario-based configuration interfaces, allowing for rapid deployment without reconstructing the algorithm model.

9. The method for real-time multi-target positioning and risk warning of explosion-proof AI-controlled spheres in gas construction scenarios according to claim 1, characterized in that: The explosion-proof AI control ball adopts an Ex d IIC T6 level explosion-proof design. The algorithm module is adapted to low power consumption operation requirements and supports dual modes of battery power supply and solar charging. It can work stably under ambient temperature conditions of -40℃ to 85℃ and IP67 protection level.

Images