Bridge tunnel fire identification and accurate positioning method and system based on multi-source fusion
By using a collaborative architecture between edge computing gateways and cloud-based main control platforms, the problem of difficult fire source location caused by complex multi-source interference in bridge and tunnel fire identification is solved, achieving highly robust and high-precision three-dimensional fire source location and ensuring the reliable execution of fire-fighting equipment.
Patent Information
- Application Number
- CN202611276861.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-21
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-08-21
AI Technical Summary
In confined spaces such as bridges and tunnels, fire identification and location face interference from multiple sources and complex working conditions, including vision system failure, sensor malfunction, misalignment of coordinate systems between data sources, and difficulty in data fusion, resulting in extremely low robustness and accuracy of fire source location.
By adopting a hardware-software co-engineering architecture of edge computing gateway and cloud main control platform, and through spatial dimensionality reduction and alignment of multi-source sensing data, steady-state background reconstruction, multimodal fire confidence early warning, pose update with slip rate penalty, and swarm intelligence algorithm, a space-thermodynamic joint cost function is constructed to achieve three-dimensional high robust absolute physical positioning of fire source.
Achieve three-dimensional, highly robust, absolute physical positioning of fire sources under complex operating conditions, reduce false alarm rates, ensure precise linkage of fire-fighting equipment, control positioning errors to the centimeter level, and shorten response time to the second level.
Smart Images

Figure CN122799558B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of Internet of Things data processing and intelligent monitoring technology, and more specifically, to a method and system for identifying and accurately locating bridge and tunnel fires based on multi-source fusion. Background Technology
[0002] Currently, fire identification and location in confined spaces such as bridges and tunnels face extremely challenging multi-source and complex interference: On the one hand, the dense smoke and drastic changes in lighting accompanying fires can easily cause conventional optical vision systems to fail, and transient thermal noise caused by heavy vehicle exhaust can seriously interfere with environmental temperature measurement. Furthermore, harsh conditions can easily cause sensor physical layer failures and signal jumps, thus inducing false alarms. On the other hand, when introducing mobile inspection vehicles to assist in positioning, water accumulation in tunnels can easily cause wheel slippage and lead to irreversible integral drift in traditional odometers. In addition, ultra-wideband (UWB) non-line-of-sight multipath pulse noise caused by reflectors such as metal pipe walls makes the spatial six-degree-of-freedom pose estimation of the vehicle itself extremely unstable. Moreover, when faced with heterogeneous data such as one-dimensional fiber optic temperature measurement, two-dimensional infrared images, and three-dimensional radar / base stations, existing technologies struggle to overcome the problems of coordinate system misalignment and physical quantity hierarchy mismatch between data sources. Forced fusion can easily lead to dimensional collapse, ultimately resulting in the inability of existing systems to achieve highly robust and accurate three-dimensional absolute physical positioning of fire sources under the superposition of the above-mentioned multiple harsh environments.
[0003] Therefore, the present invention provides a method and system for identifying and accurately locating bridge and tunnel fires based on multi-source fusion, thereby improving the above-mentioned technical problems. Summary of the Invention
[0004] This disclosure aims to address the shortcomings of existing technologies by providing a method and system for bridge and tunnel fire identification and precise location based on multi-source fusion. The invention adopts a hardware-software collaborative architecture of edge computing gateway and cloud main control platform. The edge hardware completes the natural feature space dimensionality reduction self-calibration, steady-state background reconstruction, and multimodal fire confidence warning triggering. The cloud cluster reconstructs the carrier's absolutely smooth six-degree-of-freedom pose based on slip rate penalty error state Kalman and inverse smoothing filter. Finally, a self-consistent space-thermodynamic joint cost function of physical quantity hierarchy is constructed, and a swarm intelligence algorithm is introduced to perform global extremum optimization. Thus, under complex and constrained working conditions, the three-dimensional highly robust absolute physical location of the fire source and the precise linkage of disaster prevention equipment are achieved.
[0005] To achieve the above objectives, the present disclosure proposes the following technical solutions:
[0006] In a first aspect, embodiments of this disclosure propose a method for identifying and accurately locating bridge and tunnel fires based on multi-source fusion, comprising the following steps:
[0007] S1. Acquisition of multi-source sensing data: Acquiring multi-source sensing data within the confined space of bridges and tunnels; the multi-source sensing data includes one-dimensional fiber optic temperature measurement data, two-dimensional infrared images, three-dimensional radar point clouds, and the underlying kinematic signals of the mobile inspection vehicle;
[0008] S2, Spatial Dimensionality Reduction Alignment and Steady-State Background Reconstruction: Perform natural feature-guided spatial dimension reduction alignment on the multi-source sensing data, and reconstruct the large-scale thermodynamic steady-state background field to obtain the true temperature rise.
[0009] S3. Fault-tolerant diagnosis and multimodal solution: Perform fault-tolerant diagnosis of the sensor physical layer based on the discrete variance of the signal samples, and perform multimodal solution of the fire comprehensive confidence; when the fire comprehensive confidence is greater than the safety judgment threshold, reverse solve the spatial direction vector and latch the absolute timestamp of the trigger warning;
[0010] S4. Forward tightly coupled pose state update: Using the absolute timestamp as the alignment origin, extract the mechanical deviation fitting wheel slip ratio scale factor, and perform forward tightly coupled six-degree-of-freedom pose state update.
[0011] S5. Reverse smoothing pose reconstruction: Call the wheel slip ratio scale factor as the covariance penalty term to perform reverse smoothing and reconstruct the state space global absolute smooth six-degree-of-freedom pose.
[0012] S6. Joint Cost Function Construction and Optimization Linkage: Following the confidence weight optimization based on confidence modulation, a space-thermodynamic joint cost function to eliminate dimensional collapse is constructed based on the global absolute smooth six-degree-of-freedom pose, the spatial direction vector, and the actual temperature rise. The swarm intelligence optimization algorithm is called to perform spatial iterative optimization, and the actual three-dimensional absolute coordinate vector of the fire source is output to link with disaster prevention equipment.
[0013] As a preferred technical solution of the present invention, the multi-source sensing data is acquired by a distributed fiber optic temperature measurement system, an ultra-wideband positioning base station, an infrared thermal imager, a mechanically rotating 3D lidar, an electrochemical carbon monoxide sensor, a micromechanical inertial measurement unit, and a digital wheel encoder.
[0014] The process of reconstructing the large-scale thermodynamic steady-state background field to obtain the true temperature rise specifically includes:
[0015] Extract the preset spatial filtering resolution window and time sliding window length, perform dual low-pass integral filtering in both spatial and temporal dimensions on the transient discrete temperature pulse value, and reconstruct the thermodynamic steady-state background field.
[0016] Extract the highest temperature connected region of local infrared radiation, and calculate the actual temperature rise by subtracting its temperature from the temperature of the thermodynamic steady-state background field.
[0017] As a preferred embodiment of the present invention, the step of performing sensor physical layer fault-tolerant diagnosis and performing fire comprehensive confidence multimodal solution specifically includes:
[0018] The sample discrete variance of each sensor modal signal is calculated in real time. The sigmoid activation function is used to smoothly map the sample discrete variance to the preset environmental baseline discrete variance, and the health reliability factor of each modality is calculated and output.
[0019] A semi-S-shaped function is used to map the actual temperature rise and gas concentration into physical membership probability values, respectively; the frequency probability of samples falling into discrete state intervals within a time sliding window is statistically analyzed, and the signal uncertainty is quantified by Shannon information entropy.
[0020] By combining the health reliability factors of each modality, the signal uncertainty, and the physical membership probability value, the overall fire confidence level is calculated and output.
[0021] As a preferred embodiment of the present invention, the fitting of the wheel slip ratio scaling factor and the use of the wheel slip ratio scaling factor as a covariance penalty term to perform inverse smoothing specifically include:
[0022] Extract the inertial displacement gradient and subtract it from the desired wheel displacement gradient, then fit and output the wheel slip ratio scaling factor;
[0023] The wheel slip ratio scaling factor is substituted into the error state Kalman filter engine to perform observation updates that include position and attitude states;
[0024] The wheel slip ratio scaling factor is combined with a preset slip penalty amplification coefficient and introduced as a penalty term into the calculation of the inverse smoothing gain matrix. The forward filtering estimate at the end of the time window is used as the initial condition, and the inverse smoothing gain matrix is used to eliminate high-frequency discrete oscillations in reverse, thereby reconstructing the global absolute smooth six-degree-of-freedom pose.
[0025] As a preferred embodiment of the present invention, the construction of the space-thermodynamic joint cost function to eliminate dimensional collapse specifically includes:
[0026] The fiber temperature rise is obtained by subtracting the steady-state background temperature, and then precisely matched with the one-dimensional fire source exponential temperature rise decay physical model to construct the space-thermodynamic joint cost function. :
[0027] ;
[0028] in, Let represent the three-dimensional absolute coordinate vector of the real fire source to be solved, and let have one-dimensional coordinate components. ; and The confidence weight coefficient representing the dynamic fusion of spatial geometric constraints and thermodynamic constraints; and These represent the smoothed position coordinate vector and the global three-dimensional attitude rotation matrix in the global absolutely smooth six-degree-of-freedom pose, respectively. and Represents the extrinsic parameter matrix and translation vector of a rigid transformation; and These represent the average physical depth distance and the spatial direction vector, respectively; and These represent the real-time measured temperature value and the thermodynamic steady-state background field temperature value, respectively. and These represent the initial maximum temperature parameter of the fire source and the longitudinal physical index attenuation coefficient, respectively. Indicates the total number of fiber optic temperature measurement nodes; The standard vector 2-norm operator.
[0029] Secondly, this disclosure proposes a bridge and tunnel fire identification and precise positioning system based on multi-source fusion, the system including on-site sensing hardware, edge computing gateway and cloud main control platform;
[0030] The on-site sensing hardware is used to acquire multi-source sensing data within the confined space of bridges and tunnels; the multi-source sensing data includes environmental thermodynamic field data, spatial geometric ranging data, and the underlying kinematic signals of the mobile inspection vehicle;
[0031] The edge computing gateway is used to perform natural feature-guided spatial dimensionality reduction and alignment on the multi-source sensing data and reconstruct a large-scale thermodynamic steady-state background field; perform sensor physical layer fault tolerance diagnosis and multimodal solution of fire comprehensive confidence based on the discrete variance of signal samples; when the fire comprehensive confidence is greater than the safety judgment threshold, reversely solve the spatial direction vector and latch the absolute timestamp of the early warning and report it.
[0032] The cloud-based main control platform is used to extract mechanical deviation fitting wheel slip rate scaling factors and perform forward tightly coupled six-degree-of-freedom pose state updates, using the absolute timestamp as the alignment origin; it calls the wheel slip rate scaling factor as a covariance penalty term to perform reverse smoothing reconstruction of the global absolute smooth six-degree-of-freedom pose; it takes confidence weights, constructs a space-thermodynamic joint cost function to eliminate dimensional collapse and performs swarm intelligence space iterative optimization, and outputs the real three-dimensional absolute coordinate vector of the fire source to link disaster prevention equipment.
[0033] As a preferred embodiment of the present invention, the edge computing gateway specifically includes:
[0034] Tensor processing unit is used to perform targetless spatial dimensionality reduction alignment and homogeneous projection mapping; perform dual integral low-pass filtering to reconstruct the thermodynamic steady-state background field; and extract the local infrared radiation highest temperature connected domain and its mapped average physical depth distance.
[0035] The arithmetic logic unit is used to calculate the actual temperature rise based on the thermodynamic steady-state background field; perform the sensor physical layer fault tolerance diagnosis and the fire comprehensive confidence multimodal solution; and when the confidence exceeds the limit, solve the spatial direction vector and perform data packet reporting.
[0036] As a preferred embodiment of the present invention, the cloud-based main control platform specifically includes:
[0037] A graphics processing unit cluster is used to extract the inertial displacement gradient and fit the wheel slip ratio scale factor by subtracting the wheel slip ratio from the absolute timestamp as the alignment origin, and to perform the forward tightly coupled six-degree-of-freedom pose state update.
[0038] The vector extension instruction set of the central processing unit is used to call the wheel slip ratio scale factor to calculate the inverse smoothing gain matrix and perform the inverse smoothing to reconstruct the global absolute smoothing six-degree-of-freedom pose;
[0039] The kernel computing engine is used to subtract the steady-state background temperature to obtain the true fiber temperature rise; it undertakes confidence modulation optimization of confidence weights; it constructs the space-thermodynamic joint cost function and calls the multi-golden-eagle optimization algorithm to execute the swarm intelligence space iterative optimization.
[0040] Thirdly, embodiments of this disclosure provide an electronic device, including: a processor and a memory;
[0041] The memory is used to store computer-executable instructions;
[0042] The processor is used to execute the computer-executable instructions to realize a method for identifying and accurately locating bridge and tunnel fires based on multi-source fusion.
[0043] Fourthly, this disclosure proposes a computer-readable storage medium storing computer-executable instructions. When these computer-executable instructions are executed by a processor, a method for identifying and accurately locating bridge and tunnel fires based on multi-source fusion is implemented.
[0044] In summary, the present invention has the following beneficial effects:
[0045] Firstly, this invention uses the tensor processing unit of the edge computing gateway to perform spatial dimensionality reduction alignment and homogeneous projection mapping without a target object based on the natural line-plane coplanar constraint. On the cloud main control platform, a space-thermodynamic joint cost function is constructed based on the coordinate system rigid mapping and orthogonal dimensionality reduction strategy. This solves the alignment problem in visual failure scenarios and the dimensionality collapse problem caused by the forced fusion of mismatched physical quantity levels.
[0046] Secondly, this invention reconstructs the large-scale thermodynamic steady-state background field through dual integral low-pass filtering to extract the true temperature rise, eliminating the interference of transient high-frequency thermal noise such as heavy vehicle exhaust; at the same time, it performs sensor physical layer fault-tolerant diagnosis based on the discrete variance of signal samples and integrates Shannon information entropy to perform multimodal calculation, avoiding signal jumps caused by hard thresholds and effectively suppressing system false alarms caused by dust and oil.
[0047] Third, this invention extracts the mechanical deviation fitting wheel slip ratio scaling factor and substitutes it into the error state Kalman filter (ESKF) engine to perform forward tightly coupled six-degree-of-freedom pose state update; at the same time, it calls the slip ratio scaling factor as the covariance penalty term to perform Rauch-Tung-Striebel (RTS) reverse smoothing, thereby deeply compensating for the odometer integral drift induced by tunnel water accumulation, and inversely eliminating ultra-wideband (UWB) non-line-of-sight multipath impulse noise and high-frequency discrete oscillations, reconstructing a global absolutely smooth six-degree-of-freedom pose.
[0048] Fourth, this invention is based on a deep hardware-software collaborative architecture and introduces a multi-golden eagle optimization algorithm with an embedded Levy flight mechanism to perform swarm intelligent spatial iterative optimization. It uses a long-tail distribution and large-step-size jump mechanism to avoid the algorithm getting trapped in local optima. In environments without calibration plates and with dense smoke, this solution can control the three-dimensional absolute positioning error of the real fire source to the centimeter level, shorten the overall fire confirmation response time of the system to the second level, and almost completely suppress false alarms caused by working condition interference, ultimately ensuring the reliable execution of the three-dimensional precise directional spraying operation of the fire monitor. Attached Figure Description
[0049] Figure 1 A flowchart illustrating a method for identifying and accurately locating bridge and tunnel fires based on multi-source fusion, provided in an embodiment of the present invention.
[0050] Figure 2 This is a framework diagram of a bridge and tunnel fire identification and precise positioning system based on multi-source fusion, provided in an embodiment of the present invention. Detailed Implementation
[0051] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.
[0052] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0053] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.
[0054] Furthermore, the technical features involved in the various embodiments of this application described below can be combined with each other as long as they do not conflict with each other.
[0055] This disclosure aims to address the challenges of fusing multi-source heterogeneous data in confined spaces like bridges and tunnels under harsh conditions (such as visual failure, transient thermal noise, slippage, drift, and non-line-of-sight multipath interference), resulting in extremely low robustness and accuracy in fire source localization. Therefore, this disclosure proposes a multi-source fusion-based method and system for bridge and tunnel fire identification and precise localization to achieve low false alarm identification and accurate three-dimensional absolute localization of fire sources in complex and confined environments. The method and system employ a hardware-software co-engineering architecture of an edge computing gateway and a cloud-based main control platform. Through edge-side multimodal confidence screening, cloud-based six-degree-of-freedom pose smoothing reconstruction combined with slip rate penalty (ESKF+RTS), and the construction of a space-thermodynamic joint cost function for swarm intelligence extreme value optimization, it eliminates the paradox of dimensional collapse in multi-source complex interference and heterogeneous data, thereby achieving highly robust three-dimensional absolute physical localization of fire sources and precise directional spraying of firefighting equipment.
[0056] Please refer to Figure 1 , Figure 1 A framework diagram of a bridge and tunnel fire identification and precise positioning system based on multi-source fusion according to an embodiment of the present disclosure is shown. The system includes: on-site sensing hardware, edge computing gateway and cloud main control platform;
[0057] The on-site sensing hardware is used to collect multi-source sensing data in the confined space of bridges and tunnels; specifically, it includes: a distributed fiber optic temperature measurement system fixedly laid on the roof of the bridge and tunnel, an ultra-wideband positioning base station deployed on the side wall; and infrared thermal imagers, mechanical rotating 3D lidar, electrochemical carbon monoxide sensors, micromechanical inertial measurement units and digital wheel encoders mounted on the mobile inspection vehicle.
[0058] An edge computing gateway is used to acquire multi-source sensing data and execute multi-modal feature preprocessing and early warning triggering logic at the edge. The edge computing gateway internally includes:
[0059] The tensor processing unit is used to perform objectless spatial dimensionality reduction alignment and homogeneous projection mapping; perform double integral low-pass filtering to reconstruct the thermodynamic steady-state background field; and extract the local infrared radiation highest temperature connected domain and its mapped average physical depth distance.
[0060] The arithmetic logic unit is used to calculate the actual temperature rise based on the steady-state background field; perform sensor physical layer fault tolerance diagnosis based on the discrete variance of signal samples; perform multimodal solution of fire comprehensive confidence by fusing Shannon information entropy; and when the confidence exceeds the safety judgment threshold, reversely solve the spatial direction vector, encapsulate the extracted multidimensional local features and trigger absolute timestamp into target early warning data packets for reporting.
[0061] The cloud-based main control platform is used to receive target early warning data packets and perform full-domain six-degree-of-freedom pose reconstruction of the carrier and extreme value optimization of the fire source space. The cloud-based main control platform specifically includes:
[0062] The graphics processing unit cluster is used to extract the inertial displacement gradient and the wheel slip ratio scale factor by subtracting the wheel slip ratio from the warning timestamp as the alignment origin, and then substitute it into the error state Kalman filter engine to perform forward tightly coupled six-DOF pose state update.
[0063] The central processing unit's vector extension instruction set is used to call the slip rate scaling factor as a covariance penalty term to calculate the inverse smoothing gain matrix and perform Rauch-Tung-Striebel (RTS) inverse smoothing to reconstruct the global absolute smooth six-DOF pose.
[0064] The kernel computing engine is used to subtract the steady-state background temperature to obtain the true fiber temperature rise; it undertakes confidence modulation to optimize the confidence weight; it constructs a space-thermodynamic joint cost function to eliminate dimensional collapse, and calls the multi-golden eagle optimization algorithm embedded with the Levy flight mechanism to perform swarm intelligent spatial iterative optimization, and outputs the true three-dimensional absolute coordinate vector of the fire source to link the fire monitor.
[0065] Please refer to Figure 2 , Figure 2The flowchart of the bridge and tunnel fire identification and precise location method based on multi-source fusion according to an embodiment of this disclosure is shown. The main implementers of the method are an edge computing gateway and a cloud-based main control platform. The overall process mainly includes the following six steps:
[0066] Step 1: System initialization, definition of space physical baseline and loading of global prior parameters.
[0067] To address the technical shortcomings of missing coordinate references and unclear parameter sources during the computation of multi-source heterogeneous data, the distributed control system prioritizes reading operations from the underlying non-volatile memory during the startup phase to establish a global parameter mapping table.
[0068] Specifically, the edge computing gateway and the cloud control platform load the following spatial baselines and global static parameters:
[0069] Space physical reference definitions: The origin of the global geodetic coordinate system is defined as the center point of the bridge and tunnel entrance, with the positive X-axis strictly parallel to the fiber optic laying path (the longitudinal central axis of the bridge and tunnel), the Y-axis perpendicular to the sidewall, and the Z-axis perpendicular to the horizontal plane and upwards; the local coordinate system of the infrared camera is defined (the origin is located at the optical center, the Z-axis is forward along the optical axis, the X-axis is to the right, and the Y-axis is downwards); the coordinate system of the carrier's centroid is defined (the origin is located at the carrier's geometric centroid, the X-axis is along the forward direction, the Y-axis is to the left, and the Z-axis is upwards).
[0070] Hardware and spatial prior parameter loading: Reading the third-order intrinsic parameter matrix of the factory-preset infrared camera. Read the pre-calibrated third-order rotational extrinsic parameter matrix from the local coordinate system of the infrared camera to the coordinate system of the carrier's centroid. With the three-dimensional translation extrinsic vector Read the third-order rotational extrinsic parameter matrix from the inertial measurement unit coordinate system to the carrier's center-of-mass coordinate system. Read the zero-bias error compensation vector of the inertial measurement unit. Read the three-dimensional deployment absolute coordinate set of ultra-wideband positioning base stations. Read the coordinate set of the total longitudinal physical length and cross-sectional boundary dimensions of the bridge and tunnel;
[0071] Model and Algorithm Hyperparameter Loading: Reading the Spatial Filter Resolution Window of Fiber Optic Thermometry With time sliding window length Read the modes of each sensor Environmental reference signal discrete variance (Specifically includes the discrete variance of the environmental reference signal in the infrared mode) Discrete variance of environmental reference signal with gas mode ), sensor fault activation steepness adjustment coefficient κ; read the infrared temperature threshold extracted from the connected components. With minimum area threshold Read the transition interval parameters (slope control coefficient) of the semi-S-shaped temperature membership function. temperature difference at the center ) and gas membership distribution parameters (slope control coefficient) Central concentration ); Read the fire confidence level safety judgment threshold Read the number of frames in the sliding window of information entropy statistics. Total number of discrete state intervals Read the constant vector of gravitational acceleration. Discrete-time state transition matrix including attitude and position Basic process noise covariance matrix With the noise covariance matrix of ultra-wideband ranging observation Read the number of sliding window periods and the slip penalty amplification factor for the fitted wheel slip ratio. Read the confidence level nonlinear gain coefficient Adjustment coefficient with basic weight normalization Read the longitudinal physical index attenuation coefficient of spatial thermal radiation. Initial maximum temperature parameters of the fire source and the total number of fiber optic temperature measurement nodes Read the iterative convergence threshold of the coplanar constraint self-calibration objective function. Population size of the multi-gold eagle optimization algorithm Convergence tolerance Maximum number of iterations 3D step size scaling control vector The characteristic index parameter β of the long-tailed Levy distribution.
[0072] After completing the above baseline definition and parameter loading, the system bus releases its suspended state and activates the underlying perception task in step 2.
[0073] Step 2: The edge computing gateway performs natural feature-guided spatial dimensionality reduction and alignment, and reconstructs the large-scale thermodynamic steady-state background field.
[0074] S2.1 Perform natural feature-guided spatial dimensionality reduction alignment;
[0075] To address the visual failure problem caused by drastic changes in ambient lighting within tunnels, an edge computing gateway acquires the original 3D point cloud output from a 3D LiDAR and a 2D infrared image output from an infrared camera. The tensor processing unit of the edge computing gateway extracts the 3D line set of the maintenance track edge from the point cloud and the high-temperature boundary from the infrared image. Based on the core mechanism of natural line-surface features satisfying geometric coplanar constraints, the tensor processing unit uses the minimum sum of squared distances from the projection of point cloud edge points to the thermal boundary lines in the infrared image as the objective cost function. It then employs the Levenberg-Marquardt algorithm to perform nonlinear optimization to obtain the optimal solution. This completes the object-free self-calibration and homogeneous projection mapping:
[0076] ;
[0077] ;
[0078] in, The system represents the self-calibrated optimization objective cost function. It solves for the optimal extrinsic parameters by minimizing this function, and the optimal extrinsic parameters are determined when the extrinsic parameter falls to the convergence threshold loaded in step 1. The optimization process ends when the time is right. This represents the total number of point cloud feature points extracted and participating in the coplanar constraint optimization; This represents the coordinate vector of the i-th feature point in the original 3D point cloud. This indicates the extraction of the same data from the infrared image. The corresponding two-dimensional high-heat boundary line equation; Represents the physical projection function based on the intrinsic parameter matrix; Operator for calculating the geometric Euclidean distance from a point to a line in a two-dimensional plane; Represents the local coordinate vector of the original 3D point cloud; and This represents the third-order orthogonal rotation matrix and three-dimensional translation vector from the local coordinate system of the lidar to the local coordinate system of the infrared camera, as output by the optimization calculation of the tensor processing unit. This represents the physical depth distance value of the laser point cloud projection onto the optical axis of the infrared camera; and This represents the x and y coordinates of the merged pixel generated by projection alignment; This represents the infrared camera intrinsic parameter matrix loaded in step 1.
[0079] S2.2, Reconstructing the large-scale thermodynamic steady-state background field;
[0080] Simultaneously, in response to the transient thermal noise caused by heavy vehicle exhaust, the edge computing gateway continuously reads fiber optic temperature pulses and extracts the data loaded in step 1. and Configure the spatiotemporal window and perform a double integral low-pass filter to reconstruct the thermodynamic steady-state background field:
[0081] ;
[0082] in, Indicates the time of the fiber optic system In longitudinal distance The transient discrete temperature pulse value output at the location; This represents the k-th temperature measurement node in the calculated output. In absolute moment Thermodynamic steady-state background temperature value; Represents the time integral variable, Represents the sliding variable of time integration; and Let these represent the spatial distance integral element and the time integral element, respectively.
[0083] After the calculation is completed, the edge computing gateway's direct memory access controller writes the fused pixel coordinates into static random access memory; simultaneously, it writes the steady-state background temperature array... The data is pushed to the in-memory database for persistent caching, and then a hardware interrupt is triggered to wake up the multimodal feature extraction thread, which then proceeds to step 3.
[0084] Step 3: The edge computing gateway extracts the temperature rise-related multimodal features, performs fault tolerance diagnosis and confidence calculation, and reverse-engineers the spatial direction vector.
[0085] S3.1 Extract local physical features and the actual temperature rise;
[0086] The edge computing gateway's tensor processing unit, based on fused pixel coordinates, calls the temperature threshold loaded in step 1. With area threshold Extract the connected region with the highest local infrared radiation temperature, and map it to obtain the average physical depth distance of the corresponding region. To enable upstream and downstream parameter linkage, the edge computing gateway obtains the current coarse one-dimensional vertical coordinates through the system bus. The corresponding steady-state background temperature is retrieved from the database, and the actual temperature rise is calculated by the arithmetic logic unit.
[0087] ;
[0088] in, This represents the highest absolute temperature value extracted from the local infrared radiation. This represents the steady-state background temperature value obtained from step S2.2; This represents the calculated actual temperature rise.
[0089] S3.2 Perform sensor physical layer fault tolerance diagnosis;
[0090] The arithmetic logic unit of the edge computing gateway calculates the discrete variance of the sensor signal samples in real time. Based on the fault activation steepness parameter, the arithmetic logic unit uses a sigmoid activation function to achieve a soft assessment of the sensor's health, avoiding signal jumps caused by hard thresholds.
[0091] ;
[0092] in, This represents the sensor mode index, including infrared modes. With gas mode ; This represents the calculated discrete variance of the signal samples; and This represents the discrete variance of the environmental baseline and the steepness adjustment coefficient loaded in step 1; This represents the sensor fault activation steepness adjustment coefficient loaded in step 1; Represents the modality of the calculated output Health reliability factor (including and ).
[0093] S3.3 Perform multimodal fire confidence calculation;
[0094] The temperature rise is positively correlated with the probability of a fire. The arithmetic logic unit uses a semi-S-shaped function to achieve a smooth mapping from the temperature rise and gas concentration to the physical membership probability. Subsequently, the Shannon information entropy is used to calculate the dynamic quantification of signal uncertainty through statistical sample frequency, and the overall fire confidence level is fused and output.
[0095] ;
[0096] ;
[0097] ;
[0098]
[0099] in: and These represent the calculated physical membership probability values for temperature and gas concentration, respectively. This represents the concentration sequence obtained from the carbon monoxide sensor output; , , , , This represents the total number of discrete intervals of information entropy and the membership distribution parameters loaded in step 1; This indicates the number of frames loaded in step 1. Within a limited time sliding window, modality The sample falls into the first Frequency probability of a discrete state interval; This represents the spatial information entropy calculated using Shannon's formula (including the entropy for the temperature rise sequence). Compared with gas sequences ); This represents the overall fire confidence level output by the arithmetic logic unit.
[0100] S3.4, Reverse solution of spatial direction vector and triggering of data transmission;
[0101] If the judgment Greater than the safety judgment threshold The edge computing gateway immediately reads the underlying hardware clock and latches the absolute timestamp of the current alert. Simultaneously, the arithmetic logic unit calls the intrinsic parameter inverse matrix to strip away the two-dimensional image dimension, generating a spatial ray vector pointing to the physical center of the fire source:
[0102] ;
[0103] in: This represents the three-dimensional absolute direction unit vector pointing towards the center of the fire source in the local coordinate system of the infrared camera, generated by the inverse projection. Represents the standard vector 2-norm operator; Representing the intrinsic parameter matrix The inverse matrix; This indicates that the edge computing gateway reads the absolute timestamp generated in real time by the hardware clock the instant the fire confidence level exceeds the limit.
[0104] Subsequently, the edge computing gateway will average the depth distance Directional unit vector Infrared modal health factors Gas mode health factors Confidence and timestamp The data is packaged into a target warning data packet and pushed to the cloud main control platform to wake up the cloud graphics processor cluster.
[0105] Step 4: The cloud-based main control platform extracts the mechanical deviation fitting slip compensation scale and performs forward tightly coupled six-degree-of-freedom pose estimation based on error state Kalman filtering.
[0106] S4.1 Extract the mechanical deviation and fit the slip compensation scale;
[0107] To address the issue of odometer integral drift caused by water accumulation in tunnels, a cloud-based graphics processing unit (GPU) cluster parses data packets to... To align the origin with the real-time fiber optic data stream for time synchronization; based on the physical mechanism that real inertial displacement is unaffected by slippage, the GPU extracts the acceleration data of the inertial measurement unit, and combines it with the extrinsic parameter matrix loaded in step 1. With zero bias vector The wheel slip ratio is precisely rotated and projected onto the coordinate system of the carrier's center of mass; then, the difference between its corresponding inertial displacement gradient and the desired wheel displacement gradient is used to fit the wheel slip ratio scaling factor.
[0108] ;
[0109] in, This represents the wheel slip ratio scaling factor generated by dynamic fitting. This represents the desired mechanical linear velocity measured and output by the digital wheel encoder. This represents the three-dimensional physical acceleration vector output by the micromechanical inertial measurement unit; This represents the constant vector of gravitational acceleration in the corresponding coordinate system for the loading in step 1; This represents the time differential element.
[0110] S4.2, Perform forward tightly coupled six-DOF pose state update;
[0111] Meanwhile, the GPU constructs a multi-point ranging observation equation based on Euclidean geometry for the j-th effective ultra-wideband base station:
[0112] ;
[0113] Subsequently, the GPU will slide the scale factor. Substitute the error state Kalman filter (ESKF) engine into the data, and process the data including position state. Perform forward inference and tightly coupled observation updates with the six-DOF full state vector of the attitude state:
[0114] ;
[0115] ;
[0116] ;
[0117] ;
[0118] in, and These represent the loading steps in step 1. The three-dimensional absolute coordinates and measured distance values of each base station; It indicates that it follows the observation noise covariance matrix. The corresponding channel noise; This represents the forward prediction six-degree-of-freedom state vector of the carrier at step k; ⊕ represents the three-dimensional absolute position coordinate subvector extracted from the six-degree-of-freedom state; ⊕ represents the pose addition update operator on the state manifold space of the SE(3) Lie group; This represents the Kalman gain matrix of the calculated output; This represents the Jacobian observation matrix of the ultra-wideband observation equation with respect to the state variables; Represents the Jacobian observation matrix The transpose of the matrix; This indicates the matrix operation of finding the inverse of a square matrix; and These represent the actual ranging observation vector and the theoretical ranging observation vector based on state prediction for the ultra-wideband base station group, respectively. This represents the forward six-degree-of-freedom state vector of the carrier at step k after the observation update. and Let represent the prediction and update error covariance matrices at step k, respectively; Represents the identity matrix; This represents the global parameters loaded in step 1.
[0119] The graphics processing unit cluster will use the state vector and covariance matrix Pushed into the high-speed circular buffer. After the tail pointer out-of-bounds flag is set, the system kernel wakes up the RTS inverse smoothing filter daemon.
[0120] Step 5: The cloud-based main control platform calls the slip rate factor to perform inverse gain penalty calculation and reconstructs the state space to smooth the six-degree-of-freedom pose in reverse.
[0121] S5.1, Call the slip ratio factor to perform inverse gain penalty;
[0122] To address non-line-of-sight multipath impulse noise in ultra-wideband radio frequency signals, the cloud-based main control platform invokes the Rauch-Tung-Striebel (RTS) inverse smoothing algorithm based on the covariance matrix. Based on the physical mechanism of the significant decrease in the reliability of odometry forward observations during slippage periods, the system incorporates the slip rate scaling factor generated in step S4.1 as a covariance penalty term into the inverse smoothing gain matrix, thereby reducing the weight of forward predictions.
[0123] ;
[0124] in: This represents the inverse smoothing gain matrix (6×6 dimensions) obtained after slip ratio penalty correction calculation. Indicates the retrieved first Step state error covariance matrix (6×6 dimensions); This represents the state transition matrix loaded in step 1. The transpose of the matrix; This represents the wheel slip ratio scaling factor. , This represents the state transition matrix, slip penalty amplification factor, and basic process noise covariance matrix loaded in step 1.
[0125] S5.2 Reconstructing the state space and smoothing the six-DOF pose in reverse;
[0126] Inverse smoothing uses the forward filter estimate at the end of the time window as the initial condition for inverse inference. The carrier position and attitude state are updated synchronously with full-state inverse smoothing to eliminate high-frequency discrete oscillations.
[0127] ;
[0128] in, and This represents the high-precision, absolutely smooth, six-DOF state vectors of the carrier at the k-th and (k+1)-th steps of the solution output; This represents the six-degree-of-freedom state vector output from the forward system state prediction from step k to step k+1.
[0129] Reverse engineering back to the timestamp Termination; the system terminates from its six-degree-of-freedom state vector. Extract smooth position coordinates and synchronously corrected global three-dimensional attitude rotation matrix of the carrier It is then transferred to the global optimization module.
[0130] Step 6: The cloud-based main control platform undertakes the confidence modulation optimization weight, constructs a joint cost function based on the coordinate system rigid mapping and orthogonal dimensionality reduction strategy, and executes swarm intelligence spatial iterative optimization and device linkage.
[0131] S6.1, Optimize confidence weights by modulating confidence levels;
[0132] The kernel computing engine of the cloud-based main control platform extracts health factors and confidence levels from the early warning data packets, and dynamically generates confidence weights for subsequent cost functions.
[0133] ;
[0134] ;
[0135] in: and This represents the dynamic fusion confidence weight coefficient of the calculated spatial geometric constraint term and thermodynamic constraint term; , , ξ represents the infrared modal health factor, gas modal health factor, and confidence level extracted from the data packet in step 3; ξ and ω represent the nonlinear gain coefficient and adjustment coefficient loaded in step 1.
[0136] S6.2 Construct a joint cost function to eliminate dimensionality collapse;
[0137] To address the technical shortcomings of misaligned coordinate systems and mismatched physical quantity hierarchies in heterogeneous data sources, the kernel computing engine calls rigid body transformation extrinsic parameters to transform the ray to the carrier's centroid coordinate system, explicitly extracting the orthogonal projection component of the X-axis of the coordinates to be determined. Simultaneously, by subtracting the steady-state background temperature to obtain the true fiber optic temperature rise, a joint cost function is constructed by accurately matching the one-dimensional fire source exponential temperature rise decay physical model.
[0138] ;
[0139] in: The joint optimization objective cost function representing the self-consistent spatial geometry and thermodynamics of the constructed system is denoted as . Let represent the final three-dimensional absolute coordinate vector of the actual fire source to be solved, and ; , This represents the smoothed position coordinate vector extracted in step S5.2 and the global three-dimensional attitude rotation matrix; , This represents the local average depth distance and direction unit vector generated in steps S3.1 and S3.4; and These represent the real-time measured temperature value of the k-th node and the steady-state background temperature value cached in the database, respectively. The difference between the two represents the actual fiber temperature rise under thermodynamic constraints. Indicates the vector to be extracted. One-dimensional coordinate components of orthogonal projection on the X-axis of the global geodetic coordinate system; , , , , This indicates the corresponding global static parameter loaded in step 1.
[0140] S6.3, Introducing Levy flight execution swarm intelligent spatial iterative optimization;
[0141] The main control platform calls the bridge and tunnel physical boundary parameters loaded in step 1 as environmental constraints, and initializes the scale to... To optimize the population, a multi-golden eagle optimization algorithm is introduced to perform iterative optimization in the execution space. Based on the core mechanism of long-tail distribution leading to large-step jumps, the algorithm incorporates the Levy flight mechanism to expand the population search space and avoid getting trapped in local optima.
[0142] ;
[0143] in, and This represents the absolute coordinate vector of the virtual fire source prediction generated in the (i+1)th and i-th iterations; This represents the globally optimal position prediction vector in the current iteration population that minimizes the objective cost function; and Represents an independently generated three-dimensional random number vector that follows a Gaussian normal distribution; This represents the Hadamard product operator at the matrix element level. and This indicates the corresponding algorithm hyperparameters loaded in step 1.
[0144] S6.4, Iteration Termination and Precise Equipment Linkage;
[0145] When the objective cost function The difference between two adjacent generations falls to the convergence tolerance applied in step 1. The following occurs, or the number of iterations reaches the maximum number of iterations loaded in step 1. At that time, the cloud-based main control platform terminated its iterative optimization process.
[0146] This embodiment performs in-depth hardware and software collaborative compensation for non-line-of-sight errors, heterogeneous multi-source interference, and slip drift through the aforementioned steps. Compared with a single infrared or point cloud detection scheme, in environments without calibration plates and with dense smoke, it can control the three-dimensional absolute positioning error of the fire source to the centimeter level, shorten the overall fire confirmation response time of the system to the second level, and almost completely suppress false alarms caused by operating condition interference.
[0147] Finally, the cloud-based main control platform extracts the globally optimal solution that has converged through optimization as the three-dimensional absolute coordinate vector of the real fire source. It is converted into standard industrial control messages and transmitted through the fiber optic backbone to the automatic homing fire monitor controller in the disaster prevention zone, which drives the water cannon servo motor to reverse the spatial yaw and pitch movements and perform three-dimensional precise directional spraying operations.
[0148] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.
Claims
1. A method for identifying and accurately locating bridge and tunnel fires based on multi-source fusion, characterized in that, The method includes the following steps: S1. Acquisition of multi-source sensing data: Acquiring multi-source sensing data within the confined space of bridges and tunnels; the multi-source sensing data includes one-dimensional fiber optic temperature measurement data, two-dimensional infrared images, three-dimensional radar point clouds, and the underlying kinematic signals of the mobile inspection vehicle; S2, Spatial Dimensionality Reduction Alignment and Steady-State Background Reconstruction: Perform natural feature-guided spatial dimension reduction alignment on the multi-source sensing data, and reconstruct the large-scale thermodynamic steady-state background field to obtain the true temperature rise. S3. Fault-tolerant diagnosis and multimodal solution: Perform fault-tolerant diagnosis of the sensor physical layer based on the discrete variance of the signal samples, and perform multimodal solution of the fire comprehensive confidence; when the fire comprehensive confidence is greater than the safety judgment threshold, reverse solve the spatial direction vector and latch the absolute timestamp of the trigger warning; S4. Forward tightly coupled pose state update: Using the absolute timestamp as the alignment origin, extract the mechanical deviation fitting wheel slip ratio scale factor, and perform forward tightly coupled six-degree-of-freedom pose state update. S5. Reverse smoothing pose reconstruction: Call the wheel slip ratio scale factor as the covariance penalty term to perform reverse smoothing and reconstruct the state space global absolute smooth six-degree-of-freedom pose. S6. Joint Cost Function Construction and Optimization Linkage: Following the confidence modulation optimization confidence weight, based on the global absolute smooth six-degree-of-freedom pose, the spatial direction vector and the real temperature rise, a space-thermodynamic joint cost function to eliminate dimensional collapse is constructed. The swarm intelligence optimization algorithm is called to perform spatial iterative optimization and output the real three-dimensional absolute coordinate vector of the fire source to link with disaster prevention equipment. The multi-source sensing data is acquired by a distributed fiber optic temperature measurement system, an ultra-wideband positioning base station, an infrared thermal imager, a mechanically rotating 3D lidar, an electrochemical carbon monoxide sensor, a micromechanical inertial measurement unit, and a digital wheel encoder. The process of reconstructing the large-scale thermodynamic steady-state background field to obtain the true temperature rise specifically includes: Extract the preset spatial filtering resolution window and time sliding window length, perform dual low-pass integral filtering in both spatial and temporal dimensions on the transient discrete temperature pulse value, and reconstruct the thermodynamic steady-state background field. Extract the highest temperature connected region of local infrared radiation, and calculate the actual temperature rise by subtracting its temperature from the temperature of the thermodynamic steady-state background field. The on-site sensing hardware for collecting the multi-source sensing data specifically includes: a distributed fiber optic temperature measurement system fixedly laid on the roof of the bridge and tunnel, an ultra-wideband positioning base station deployed on the side wall; and an infrared thermal imager, a mechanically rotating 3D lidar, an electrochemical carbon monoxide sensor, a micromechanical inertial measurement unit, and a digital wheel encoder mounted on the mobile inspection vehicle. The process of performing fault-tolerant diagnostics at the sensor physical layer and performing multimodal calculations of fire comprehensive confidence specifically includes: The sample discrete variance of each sensor modal signal is calculated in real time. The sigmoid activation function is used to smoothly map the sample discrete variance to the preset environmental baseline discrete variance, and the health reliability factor of each modality is calculated and output. A semi-S-shaped function is used to map the actual temperature rise and gas concentration into physical membership probability values, respectively; the frequency probability of samples falling into discrete state intervals within a time sliding window is statistically analyzed, and the signal uncertainty is quantified by Shannon information entropy. By combining the health reliability factors of each modality, the signal uncertainty, and the physical membership probability value, the overall fire confidence level is calculated and output.
2. The method for bridge and tunnel fire identification and precise location based on multi-source fusion according to claim 1, characterized in that, The fitting of the wheel slip ratio scaling factor, and the use of the wheel slip ratio scaling factor as a covariance penalty term to perform inverse smoothing, specifically include: Extract the inertial displacement gradient and subtract it from the desired wheel displacement gradient, then fit and output the wheel slip ratio scaling factor; The wheel slip ratio scaling factor is substituted into the error state Kalman filter engine to perform observation updates that include position and attitude states; The wheel slip ratio scaling factor is combined with a preset slip penalty amplification coefficient and introduced as a penalty term into the calculation of the inverse smoothing gain matrix. The forward filtering estimate at the end of the time window is used as the initial condition, and the inverse smoothing gain matrix is used to eliminate high-frequency discrete oscillations in reverse, thereby reconstructing the global absolute smooth six-degree-of-freedom pose.
3. The method for bridge and tunnel fire identification and precise location based on multi-source fusion according to claim 1, characterized in that, The construction of the joint space-thermodynamic cost function to eliminate dimensional collapse specifically includes: The fiber temperature rise is obtained by subtracting the steady-state background temperature, and then precisely matched with the one-dimensional fire source exponential temperature rise decay physical model to construct the space-thermodynamic joint cost function. : ; in, Let represent the three-dimensional absolute coordinate vector of the real fire source to be solved, and let have one-dimensional coordinate components. and The confidence weight coefficient representing the dynamic fusion of spatial geometric constraints and thermodynamic constraints; and These represent the smoothed position coordinate vector and the global three-dimensional attitude rotation matrix in the global absolutely smooth six-degree-of-freedom pose, respectively. and Represents the extrinsic parameter matrix and translation vector of a rigid transformation; and These represent the average physical depth distance and the spatial direction vector, respectively; and These represent the real-time measured temperature value and the thermodynamic steady-state background field temperature value, respectively. and These represent the initial maximum temperature parameter of the fire source and the longitudinal physical index attenuation coefficient, respectively. Indicates the total number of fiber optic temperature measurement nodes; The standard vector 2-norm operator.
4. A bridge and tunnel fire identification and precise location system based on multi-source fusion, characterized in that, The system is used to implement the bridge and tunnel fire identification and precise location method based on multi-source fusion as described in any one of claims 1 to 3. The system includes on-site sensing hardware, edge computing gateway and cloud main control platform. The on-site sensing hardware is used to acquire multi-source sensing data within the confined space of bridges and tunnels; the multi-source sensing data includes environmental thermodynamic field data, spatial geometric ranging data, and the underlying kinematic signals of the mobile inspection vehicle; The edge computing gateway is used to perform natural feature-guided spatial dimensionality reduction and alignment on the multi-source sensing data and reconstruct a large-scale thermodynamic steady-state background field; perform sensor physical layer fault tolerance diagnosis and multimodal solution of fire comprehensive confidence based on the discrete variance of signal samples; when the fire comprehensive confidence is greater than the safety judgment threshold, reversely solve the spatial direction vector and latch the absolute timestamp of the early warning and report it. The cloud-based main control platform is used to extract mechanical deviation fitting wheel slip rate scaling factors and perform forward tightly coupled six-degree-of-freedom pose state updates, using the absolute timestamp as the alignment origin; it calls the wheel slip rate scaling factor as a covariance penalty term to perform reverse smoothing reconstruction of the global absolute smooth six-degree-of-freedom pose; it takes confidence weights, constructs a space-thermodynamic joint cost function to eliminate dimensional collapse and performs swarm intelligence space iterative optimization, and outputs the real three-dimensional absolute coordinate vector of the fire source to link disaster prevention equipment.
5. The bridge and tunnel fire identification and precise positioning system based on multi-source fusion according to claim 4, characterized in that, The edge computing gateway specifically includes: Tensor processing unit is used to perform targetless spatial dimensionality reduction alignment and homogeneous projection mapping; perform dual integral low-pass filtering to reconstruct the thermodynamic steady-state background field; and extract the local infrared radiation highest temperature connected domain and its mapped average physical depth distance. The arithmetic logic unit is used to calculate the actual temperature rise based on the thermodynamic steady-state background field; perform the sensor physical layer fault tolerance diagnosis and the fire comprehensive confidence multimodal solution; and when the confidence exceeds the limit, solve the spatial direction vector and perform data packet reporting.
6. The bridge and tunnel fire identification and precise positioning system based on multi-source fusion according to claim 4, characterized in that, The cloud-based main control platform specifically includes: A graphics processing unit cluster is used to extract the inertial displacement gradient and fit the wheel slip ratio scale factor by subtracting the wheel slip ratio from the absolute timestamp as the alignment origin, and to perform the forward tightly coupled six-degree-of-freedom pose state update. The vector extension instruction set of the central processing unit is used to call the wheel slip ratio scale factor to calculate the inverse smoothing gain matrix and perform the inverse smoothing to reconstruct the global absolute smoothing six-degree-of-freedom pose; The kernel computing engine is used to subtract the steady-state background temperature to obtain the true fiber temperature rise; it undertakes confidence weight optimization based on confidence modulation; it constructs the space-thermodynamic joint cost function and calls the multi-golden-eagle optimization algorithm to execute the swarm intelligence space iterative optimization.
7. An electronic device, characterized in that, include: Processor and memory; The memory is used to store computer-executable instructions; The processor is configured to execute the computer-executable instructions to implement the bridge and tunnel fire identification and precise location method based on multi-source fusion as described in any one of claims 1 to 3.
8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the bridge and tunnel fire identification and precise location method based on multi-source fusion as described in any one of claims 1 to 3.
Citation Information
Patent Citations
Highway tunnel fire monitoring system and method based on image fusion
CN113596335A
Visual SLAM system construction method suitable for urban underground rescue
CN118243083A
