Cable-type heat-sensing fire detection method and system with intelligent positioning function

Abstract

This invention provides a cable-based temperature-sensing fire detection method and system with intelligent positioning function, relating to the field of artificial intelligence technology. It includes deploying fiber optic temperature-sensing cables along a preset path, acquiring real-time three-dimensional coordinates of temperature measurement points using reflective markers and multi-view 3D reconstruction technology, correlating temperature data with coordinates to form a temperature field dataset, generating a three-dimensional temperature field grid data using a distance-attenuation-based weight optimization method, and providing temperature warnings and location information accordingly. This invention achieves accurate location and real-time warning of early-stage temperature anomalies in fires, improving the accuracy and response speed of fire monitoring.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a cable-type heat-sensing fire detection method and system with intelligent positioning function. Background Technology

[0002] In large and complex spaces such as underground utility tunnels, high-rise warehouses, and other similar locations, timely and accurate detection and location of fire sources are crucial for ensuring personnel safety and minimizing property damage. Cable-based heat-sensing fire detection technology, as a linear distributed temperature measurement system, can continuously monitor temperature changes along a spatial path and has been widely applied in various fire early warning systems. Traditional cable-based heat-sensing fire detection systems typically employ fiber optic distributed temperature measurement technology or special temperature-sensitive materials, achieving temperature monitoring by measuring light signal attenuation or resistance changes.

[0003] Traditional cable-based temperature sensing systems lack real-time spatial positioning capabilities, making it impossible to accurately obtain the real-time deformation position of the cable in three-dimensional space. This results in difficulty in accurately locating the spatial coordinates of the fire source in complex environments. Conventional temperature field reconstruction methods do not consider the optimization of measurement point weights. When the distribution of measurement points is uneven or the local temperature gradient changes significantly, the interpolation accuracy is insufficient, and it cannot truly reflect the characteristics of hotspot diffusion. Existing systems generally adopt fixed threshold alarm mechanisms and lack intelligent early warning strategies that consider spatial location factors, which easily leads to false alarms or missed alarms, especially in the early stages of a fire when the temperature distribution shows local anomalies. Summary of the Invention

[0004] This invention provides a cable-type heat-sensing fire detection method and system with intelligent positioning function, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a cable-type heat-sensing fire detection method with intelligent positioning function, comprising:

[0006] Multiple fiber optic temperature sensing cables are laid along multiple preset longitudinal paths within the space to be measured. Reflective markings are set on the surface of the fiber optic temperature sensing cables at preset intervals, and multiple temperature measuring points are set on each fiber optic temperature sensing cable at preset intervals.

[0007] Images of reflective markers are acquired, and the two-dimensional coordinate information of the reflective markers in the images is extracted. The real-time three-dimensional spatial position of the reflective markers is calculated using a multi-view three-dimensional reconstruction algorithm. The real-time deformation profile of the fiber optic temperature sensing cable is obtained using a cubic spline curve fitting method. Based on the relative distance between the real-time three-dimensional spatial position of the reflective markers and the temperature measuring point, the real-time three-dimensional coordinate information of the temperature measuring point is calculated.

[0008] By associating temperature data with the real-time three-dimensional coordinate information of the corresponding temperature measurement points, a temperature field dataset is obtained.

[0009] A distance-attenuation-based weight initialization method is used to assign initial weights to the temperature measurement points of each temperature field dataset. The weight coefficients are dynamically adjusted according to the magnitude of the local temperature gradient. The optimal weight combination is determined through iterative optimization to generate three-dimensional temperature field grid data.

[0010] Based on the three-dimensional temperature field grid data and combined with the preset warning temperature threshold, temperature warnings and abnormal temperature location information are provided.

[0011] Extracting the two-dimensional coordinate information of reflective markers in the image and calculating the real-time three-dimensional spatial position of the reflective markers using a multi-view three-dimensional reconstruction algorithm includes:

[0012] Using the reflective marker area as the seed point, the region growing algorithm is used to extract complete candidate reflective areas, calculate the centroid coordinates of the candidate reflective areas, and obtain the two-dimensional coordinates and confidence scores of the reflective markers.

[0013] The camera intrinsic parameter matrix and distortion coefficients are obtained, and the relative pose between cameras is calculated based on epipolar geometry constraints to establish a unified spatial coordinate system. The two-dimensional coordinates of the reflective markers are projected onto the unified spatial coordinate system to establish the correspondence between the reflective markers under different viewpoints. Combining triangulation and observation iterative optimization methods, the real-time three-dimensional spatial position of the reflective markers is output.

[0014] Projecting the two-dimensional coordinates of the reflective markers onto a unified spatial coordinate system, establishing the correspondence between the reflective markers from different viewpoints, and combining triangulation and observational iterative optimization methods, the real-time three-dimensional spatial position of the reflective markers is output, including:

[0015] The two-dimensional coordinates of the reflective markers are projected onto a unified spatial coordinate system, and the projection relationship equation between the two-dimensional image coordinates and the three-dimensional spatial coordinates is established. The initial solution is obtained using the linear least squares method, and a spatial position continuity constraint term is introduced. The three-dimensional coordinates are then iteratively optimized by combining the observation weights.

[0016] The covariance matrix of the three-dimensional coordinates is calculated based on the Jacobian matrix of the observation equation, and the standard deviation of the coordinate components is obtained; the reprojection error of the reflective marker point at each viewpoint is calculated based on the covariance matrix and the standard deviation of the coordinate components.

[0017] Observational data with reprojection errors greater than a preset error threshold are discarded. The optimization solution is then re-executed based on the filtered observational data until the preset convergence condition is met, and the final three-dimensional spatial position of the reflective marker points is output.

[0018] The real-time deformation profile of the fiber optic temperature sensing cable is obtained using a cubic spline curve fitting method. Based on the real-time three-dimensional spatial position of the reflective marker point and its relative distance to the temperature measuring point, the real-time three-dimensional coordinate information of the temperature measuring point is calculated, including:

[0019] The three-dimensional spatial positions of the reflective markers are sorted according to the laying path of the fiber optic temperature sensing cable to construct a marker sequence;

[0020] The node vector distribution is calculated based on the distance and curvature change between adjacent marker points in the marker point sequence. The smoothness of the curve shape is controlled by introducing an adjustable tension coefficient. The spatial distribution of control points is determined by combining the geometric constraints at the marker points. The control point weight coefficient is optimized by using the least squares method to generate a continuous and smooth deformation curve of the fiber optic temperature sensing cable.

[0021] Based on the initial distance ratio between the temperature measuring point and adjacent reflective marker points, arc length parameterized interpolation positioning is performed on the continuous smooth deformation curve. The local arc length change caused by fiber deformation is calculated, the position of the temperature measuring point is compensated in real time, and the real-time three-dimensional coordinate information of the temperature measuring point is output.

[0022] Based on the initial distance ratio between the temperature measuring point and adjacent reflective marker points, arc length parameterized interpolation is performed on a continuous smooth deformation curve to locate the point. The local arc length change caused by fiber deformation is calculated, and real-time compensation for the temperature measuring point position is performed, including:

[0023] Based on the initial distance ratio between the temperature measuring point and the adjacent reflective marker point, a continuous smooth deformation curve is fitted using the non-uniform rational B-spline method. The deformation curve is then represented by arc length parameterization. The curvature value at the curve node is calculated. Based on the curvature value, the parameter interval is divided using an adaptive step size, and the correspondence between the parameter and the arc length is established.

[0024] Calculate the local deformation characteristics of the optical fiber, establish a mapping relationship between the local deformation characteristics and the position offset based on the correspondence, calculate the compensation amount of the temperature measuring point position using the piecewise cubic Hermit interpolation method, iteratively optimize the temperature measuring point position in the parameter space until the preset accuracy requirements are met, and realize real-time compensation of the temperature measuring point position.

[0025] A distance-attenuation-based weight initialization method is used to assign initial weights to the temperature measurement points in each temperature field dataset. The weight coefficients are dynamically adjusted according to the magnitude of the local temperature gradient, and the optimal weight combination is determined through iterative optimization to generate three-dimensional temperature field mesh data, including:

[0026] The local feature scale is determined based on the spatial density distribution of temperature measurement points. The local feature scale is then substituted into the attenuation function to calculate the distance attenuation coefficient. The distance attenuation coefficient is corrected by combining the measurement uncertainty information, and initial weights are assigned to the temperature measurement points of each temperature field dataset.

[0027] The weighted average temperature is calculated using the distance matrix between grid points and temperature measurement points to construct the initial value of the temperature field; the initial value of the temperature field is then fitted by least squares to generate a local temperature distribution function.

[0028] The temperature gradient vector is solved based on the local temperature distribution function, and the weighting coefficients are dynamically adjusted for the temperature gradient vector.

[0029] An iterative optimization method is used to alternately update the weight coefficients and the local temperature distribution function until the temperature field changes in adjacent iterations meet the convergence condition. Based on the final determined weight coefficients and the temperature data in the temperature field dataset, a three-dimensional temperature field mesh is generated.

[0030] A second aspect of the present invention provides a cable-type heat-sensing fire detection system with intelligent positioning function, comprising:

[0031] The first unit is used to lay multiple fiber optic temperature sensing cables along multiple preset longitudinal paths in the space to be measured. The surface of the fiber optic temperature sensing cables is set with reflective marking points at preset intervals, and multiple temperature measuring points are set at preset intervals on each fiber optic temperature sensing cable.

[0032] The second unit is used to acquire images of reflective markers, extract the two-dimensional coordinate information of reflective markers in the images, calculate the real-time three-dimensional spatial position of reflective markers using a multi-view three-dimensional reconstruction algorithm, obtain the real-time deformation profile of the fiber optic temperature sensing cable using a cubic spline curve fitting method, and calculate the real-time three-dimensional coordinate information of the temperature measuring point based on the relative distance between the real-time three-dimensional spatial position of the reflective markers and the temperature measuring point.

[0033] The third unit is used to associate temperature data with the real-time three-dimensional coordinate information of the corresponding temperature measurement point to obtain a temperature field dataset.

[0034] The fourth unit is used to assign initial weights to the temperature measurement points of each temperature field dataset using a distance decay-based weight initialization method. The weight coefficients are dynamically adjusted according to the magnitude of the local temperature gradient, and the optimal weight combination is determined through iterative optimization to generate three-dimensional temperature field mesh data.

[0035] The fifth unit is used to provide temperature warnings and abnormal temperature location information based on the three-dimensional temperature field grid data and a preset warning temperature threshold.

[0036] A third aspect of the present invention,

[0037] An electronic device is provided, comprising:

[0038] processor;

[0039] Memory used to store processor-executable instructions;

[0040] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0041] Fourth aspect of the present invention,

[0042] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0043] The beneficial effects of this application are as follows:

[0044] This invention achieves precise fire detection by deploying multiple fiber optic temperature sensing cables along a preset path within the measured space, and combining reflective markers and multi-view 3D reconstruction algorithms. This solves the problem that traditional cable-type temperature sensing fire detectors cannot accurately locate the fire source, and enables rapid and accurate location of abnormal fire temperatures.

[0045] This invention employs a distance-attenuation-based weight initialization method and a dynamic adjustment of weight coefficients based on local temperature gradients to construct high-precision three-dimensional temperature field grid data. Compared with traditional methods, this method more accurately reflects the spatial temperature distribution and improves the accuracy of fire monitoring and early warning.

[0046] This invention correlates temperature data with the real-time three-dimensional coordinate information of temperature measurement points to form a complete temperature field dataset, enabling comprehensive intelligent monitoring of large-space environments. It is applicable to various complex spatial environments, improves the ability to provide early fire warnings and predict fire development trends, and has broad application prospects. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating the cable-type heat-sensing fire detection method with intelligent positioning function according to an embodiment of the present invention.

[0048] Figure 2 This is a schematic diagram of the process for generating a three-dimensional temperature field mesh based on distance decay weights. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0051] Figure 1 This is a flowchart illustrating the method for dynamic identification and prediction of building fire loads using artificial intelligence, as described in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0052] Multiple fiber optic temperature sensing cables are laid along multiple preset longitudinal paths within the space to be measured. Reflective markings are set on the surface of the fiber optic temperature sensing cables at preset intervals, and multiple temperature measuring points are set on each fiber optic temperature sensing cable at preset intervals.

[0053] Images of reflective markers are acquired, and the two-dimensional coordinate information of the reflective markers in the images is extracted. The real-time three-dimensional spatial position of the reflective markers is calculated using a multi-view three-dimensional reconstruction algorithm. The real-time deformation profile of the fiber optic temperature sensing cable is obtained using a cubic spline curve fitting method. Based on the relative distance between the real-time three-dimensional spatial position of the reflective markers and the temperature measuring point, the real-time three-dimensional coordinate information of the temperature measuring point is calculated.

[0054] By associating temperature data with the real-time three-dimensional coordinate information of the corresponding temperature measurement points, a temperature field dataset is obtained.

[0055] A distance-attenuation-based weight initialization method is used to assign initial weights to the temperature measurement points of each temperature field dataset. The weight coefficients are dynamically adjusted according to the magnitude of the local temperature gradient. The optimal weight combination is determined through iterative optimization to generate three-dimensional temperature field grid data.

[0056] Based on the three-dimensional temperature field grid data and combined with the preset warning temperature threshold, temperature warnings and abnormal temperature location information are provided.

[0057] In one optional implementation, extracting the two-dimensional coordinate information of the reflective markers in the image and calculating the real-time three-dimensional spatial position of the reflective markers using a multi-view three-dimensional reconstruction algorithm includes:

[0058] Using the reflective marker area as the seed point, the region growing algorithm is used to extract complete candidate reflective areas, calculate the centroid coordinates of the candidate reflective areas, and obtain the two-dimensional coordinates and confidence scores of the reflective markers.

[0059] The camera intrinsic parameter matrix and distortion coefficients are obtained, and the relative pose between cameras is calculated based on epipolar geometry constraints to establish a unified spatial coordinate system. The two-dimensional coordinates of the reflective markers are projected onto the unified spatial coordinate system to establish the correspondence between the reflective markers under different viewpoints. Combining triangulation and observation iterative optimization methods, the real-time three-dimensional spatial position of the reflective markers is output.

[0060] Multiple camera devices are used, each aimed at the same scene to capture images. Image frames containing reflective markers are obtained from each camera. Reflective markers are extracted from the images, and binarization processing with a grayscale threshold of 220 is used to convert the images into binary images to highlight the features of reflective areas. In the binary image, connected regions with brightness values ​​higher than the threshold are identified as candidate reflective marker regions.

[0061] For each candidate reflective region, a region growing algorithm is implemented using the reflective marker region as the seed point. The initial seed point has a gray value of 235, and the gray value tolerance for region growing is set to 15. Starting from the seed point, its 8 neighboring pixels are checked. If the gray value of the neighboring pixels is within the tolerance range, the pixel is included in the current region and added to the seed point set. This process is repeated until no new pixels are added, thereby extracting the complete candidate reflective region.

[0062] To ensure the accuracy of the extracted area, an area constraint is set, that is, the area of ​​the reflective area should be between 25 and 400 square pixels. If the area is too small, it will result in noise; if it is too large, it will result in interference caused by camera overexposure or other reflective objects.

[0063] For candidate reflective regions that meet the area constraints, their centroid coordinates are calculated. The centroid coordinates are calculated using a weighted average of the coordinates of all pixels within the region, with the weights being the pixel grayscale values. For example, for a candidate region containing 100 pixels, the weighted average of the x and y coordinates of all pixels is calculated, assuming the resulting centroid coordinates are (253.75, 142.38) pixels.

[0064] A confidence score is assigned to each reflective marker. The confidence score is calculated based on the region's morphological characteristics and grayscale distribution, specifically using the ratio of the region's mean grayscale value to its standard deviation. For example, if a reflective region has a mean grayscale value of 240 and a standard deviation of 8, its confidence score is 30. Reflective markers with a confidence score greater than 20 are considered valid markers.

[0065] During the camera calibration phase, at least 20 images containing a standard checkerboard pattern are used for camera intrinsic parameter calibration. The camera's intrinsic parameter matrix and distortion coefficients are obtained through checkerboard corner detection and optimization algorithms. For example, the intrinsic parameter matrix of a 1080p resolution camera is as follows: focal length fx=1024.5, fy=1026.8, principal point coordinates cx=639.5, cy=359.7, distortion coefficients k1=-0.1528, k2=0.0362, p1=0.0007, p2=-0.0004.

[0066] The coordinate system of one camera in a multi-camera system is chosen as the world coordinate system. The extrinsic parameters of the other cameras relative to this coordinate system are calculated. The extrinsic parameter calculation is based on epipolar geometry constraints, using at least eight pairs of matching points to solve for the fundamental matrix. The fundamental matrix is ​​further decomposed to obtain the rotation matrix and translation vector between the cameras, establishing a unified spatial coordinate system. For example, the rotation angle of the second camera relative to the first camera is (0.05, -0.12, 0.03) radians, and the translation vector is (348.5, -52.3, 75.6) millimeters.

[0067] The two-dimensional coordinates of the reflective markers are projected onto a unified spatial coordinate system to establish the correspondence between the reflective markers from different perspectives. Combining triangulation and observation iterative optimization methods, the real-time three-dimensional spatial position of the reflective markers is output.

[0068] The above technical solutions enable accurate two-dimensional extraction and stable three-dimensional spatial position reconstruction of reflective markers, which can be widely applied in fields such as motion capture, surgical navigation, and industrial measurement.

[0069] In one optional implementation, the two-dimensional coordinates of the reflective markers are projected onto a unified spatial coordinate system to establish the correspondence between the reflective markers from different viewpoints. Combining triangulation and observation-iterative optimization methods, the real-time three-dimensional spatial position of the reflective markers is output, including:

[0070] The two-dimensional coordinates of the reflective markers are projected onto a unified spatial coordinate system, and the projection relationship equation between the two-dimensional image coordinates and the three-dimensional spatial coordinates is established. The initial solution is obtained using the linear least squares method, and a spatial position continuity constraint term is introduced. The three-dimensional coordinates are then iteratively optimized by combining the observation weights.

[0071] The covariance matrix of the three-dimensional coordinates is calculated based on the Jacobian matrix of the observation equation, and the standard deviation of the coordinate components is obtained; the reprojection error of the reflective marker point at each viewpoint is calculated based on the covariance matrix and the standard deviation of the coordinate components.

[0072] Observational data with reprojection errors greater than a preset error threshold are discarded. The optimization solution is then re-executed based on the filtered observational data until the preset convergence condition is met, and the final three-dimensional spatial position of the reflective marker points is output.

[0073] A standard checkerboard calibration board was used to jointly calibrate the multi-camera system, obtaining the intrinsic and extrinsic parameter matrices for each camera. The intrinsic parameter matrix includes focal length, principal point coordinates, and distortion parameters, while the extrinsic parameter matrix describes the camera's position and orientation in the world coordinate system. Taking an eight-camera system as an example, the focal length parameters of camera 1 obtained through calibration are fx = 800 pixels and fy = 805 pixels; the principal point coordinates are cx = 640 pixels and cy = 480 pixels; and the radial distortion coefficients are k1 = -0.12 and k2 = 0.08. The rotation angles of camera 1 relative to the world coordinate system are 15 degrees around the x-axis, -8 degrees around the y-axis, and 2 degrees around the z-axis; the translation vectors are 1.2 meters in the x-direction, 0.8 meters in the y-direction, and 2.5 meters in the z-direction.

[0074] The process of establishing the projection relationship equation between 2D image coordinates and 3D spatial coordinates includes two stages: coordinate system transformation and perspective projection. 3D points in the world coordinate system are transformed to the camera coordinate system using a rotation matrix and a translation vector. The rotation matrix is ​​calculated using Euler angles. 3D points in the camera coordinate system are then mapped to the 2D image plane using a perspective projection model. Pixel coordinate u equals the focal length fx multiplied by the camera coordinate system x component divided by the camera coordinate system z component plus the principal point coordinate cx. Pixel coordinate v equals the focal length fy multiplied by the camera coordinate system y component divided by the camera coordinate system z component plus the principal point coordinate cy. For a reflective marker point with spatial coordinates of (10.5, 8.3, 15.2) meters, after projection transformation by camera 1, its pixel coordinates on the image plane are (512, 384).

[0075] Each camera's observation of a reflective marker provides two constraints, corresponding to the projection relationships of the u and v coordinates, respectively. When n cameras observe the same reflective marker, 2n constraint equations can be established to solve for the three unknown 3D coordinate components. The nonlinear projection equations are linearized and expanded near the initial estimation point, ignoring higher-order terms to obtain a system of linear equations. Taking eight cameras simultaneously observing a reflective marker as an example, 16 linear constraint equations are established to solve for the three unknowns, forming an overdetermined system of equations. By calculating the transpose of the constraint matrix multiplied by its inverse, and then multiplied by the transpose of the constraint matrix and the observation vector, the least-squares solution for the 3D coordinates is obtained. In actual calculations, the initial 3D coordinate estimation result for a certain reflective marker is (10.48, 8.35, 15.18) meters.

[0076] A motion model is established to predict the position of the reflective marker in the current frame. Commonly used models include the constant velocity model and the constant acceleration model. The constant velocity model assumes that the marker maintains a constant velocity for a short period of time, and the current position is equal to the previous frame position plus the velocity multiplied by the time interval. The constant acceleration model adds an acceleration term to the constant velocity model, and the current position is equal to the previous frame position plus the velocity multiplied by the time interval plus half the acceleration multiplied by the square of the time interval. Taking the constant velocity model as an example, if the position of the reflective marker in the previous frame is (10.45, 8.28, 15.17) meters, and the estimated velocity is (0.05, 0.02, 0.03) meters per frame, then the predicted position of the current frame is (10.5, 8.3, 15.2) meters. The continuity constraint term is incorporated into the optimization objective function in a weighted manner. The weight coefficients are dynamically adjusted according to the smoothness of the motion; the smoother the motion, the greater the weight.

[0077] Weights are assigned to each camera's observations. The weight calculation comprehensively considers factors such as the distance between the camera and the marker point, the observation angle, image sharpness, and occlusion. Distance weights use an inverse proportional function; the closer the distance, the greater the weight. Angle weights are determined by the angle between the line of sight and the marker point's normal vector; the smaller the angle, the greater the weight. Sharpness weights are evaluated using the image gradient magnitude; a larger gradient indicates a sharper image and a higher weight. In one measurement, a camera 2 meters from the marker point with a good viewing angle had a weight of 0.95, a camera 5 meters away with a slightly worse viewing angle had a weight of 0.75, and a camera with slight occlusion had a weight reduced to 0.4. Iterative optimization uses the Levenburg-Marquardt algorithm. In each iteration, the reprojection error of each camera is estimated based on the current 3D coordinates, the weights are updated, and the 3D coordinates are recalculated. The iteration process continues until the coordinate change is less than a preset threshold of 0.001 meters.

[0078] Each row of the Jacobian matrix corresponds to an observation equation, and each column corresponds to a parameter to be estimated. For the observation equation at pixel coordinate u, its partial derivative with respect to 3D coordinate X is equal to the focal length fx divided by the camera coordinate z component; its partial derivative with respect to Y is zero; and its partial derivative with respect to Z is equal to the negative focal length fx multiplied by the camera coordinate x component divided by the square of the camera coordinate z component. Similarly, the partial derivative for pixel coordinate v is calculated. Taking an 8-camera system as an example, the Jacobian matrix is ​​16 rows and 3 columns, with every two rows corresponding to the u and v observation equations for one camera. The Jacobian matrix is ​​calculated at the current 3D coordinate estimation point and is continuously updated during the iteration process.

[0079] The covariance matrix is ​​equal to the transpose of the Jacobian matrix multiplied by the weight matrix multiplied by the inverse of the Jacobian matrix. The weight matrix is ​​a diagonal matrix, with diagonal elements representing the weights of each observation. The diagonal elements of the covariance matrix represent the variances of the X, Y, and Z coordinate components, while the off-diagonal elements represent the covariances between coordinate components. The standard deviation of each coordinate component is the square root of its corresponding variance. In a certain calculation, the standard deviations were 0.024 meters in the X direction, 0.031 meters in the Y direction, and 0.045 meters in the Z direction, indicating that the positioning uncertainty is greatest in the Z direction. The covariance matrix can also be used to calculate the confidence ellipsoid and assess the reliability of 3D positioning.

[0080] The reprojection process uses the same geometric model and parameters as the forward projection to convert 3D coordinates into pixel coordinates. The Euclidean distance between the reprojected pixel coordinates and the actual observed pixel coordinates is the reprojection error. For camera i, the reprojection error is equal to the square root of the sum of the squared differences between the reprojected u coordinates and the observed u coordinates, plus the square root of the sum of the squared differences between the reprojected v coordinates and the observed v coordinates. In the actual calculation, the reprojected coordinates of camera 1 are (513, 385) pixels, the observed coordinates are (512, 384) pixels, and the reprojection error is 1.414 pixels. The reprojection error of each camera reflects the degree of consistency between the current 3D coordinate estimation and the observed data.

[0081] A preset error threshold is set, typically between 2 and 5 pixels. The specific value is determined based on image resolution and accuracy requirements. Observations with reprojection errors exceeding the threshold are considered outliers, caused by factors such as occluded markers, excessive image noise, or inaccurate calibration parameters. Outlier observations are discarded during subsequent optimization and do not participate in the calculation of 3D coordinates. For example, with an error threshold of 3 pixels, if the reprojection error of camera 5 is 4.2 pixels, it will be discarded as it exceeds the threshold; the reprojection error of camera 3 is 2.1 pixels, which is less than the threshold and is retained. The filtered observation data is used to reconstruct the optimization problem, updating the Jacobian matrix and weight matrix.

[0082] The iterative optimization process continuously updates the 3D coordinate estimates until convergence conditions are met. These convergence conditions include indices such as the coordinate change between two consecutive iterations, the change in the objective function value, and the gradient norm. The coordinate change threshold is typically set to 0.001 meters, indicating that convergence is considered achieved when the changes in the X, Y, and Z coordinate components are all less than this value. The objective function value change rate threshold is set to 0.01%, and iteration stops when the relative change in the function value is less than this threshold. The gradient norm threshold, set to 0.001, is used to determine whether the area near the optimum has been reached.

[0083] In one optional implementation, a cubic spline curve fitting method is used to obtain the real-time deformation profile of the fiber optic temperature sensing cable. Based on the real-time three-dimensional spatial position of the reflective marker point and the relative distance between the temperature measuring point and the real-time three-dimensional coordinate information of the temperature measuring point is calculated, including:

[0084] The three-dimensional spatial positions of the reflective markers are sorted according to the laying path of the fiber optic temperature sensing cable to construct a marker sequence;

[0085] The node vector distribution is calculated based on the distance and curvature change between adjacent marker points in the marker point sequence. The smoothness of the curve shape is controlled by introducing an adjustable tension coefficient. The spatial distribution of control points is determined by combining the geometric constraints at the marker points. The control point weight coefficient is optimized by using the least squares method to generate a continuous and smooth deformation curve of the fiber optic temperature sensing cable.

[0086] Based on the initial distance ratio between the temperature measuring point and adjacent reflective marker points, arc length parameterized interpolation positioning is performed on the continuous smooth deformation curve. The local arc length change caused by fiber deformation is calculated, the position of the temperature measuring point is compensated in real time, and the real-time three-dimensional coordinate information of the temperature measuring point is output.

[0087] Based on the actual laying path of the fiber optic temperature sensing cable, the captured three-dimensional spatial location data of the reflective markers are sorted. Assuming that the three-dimensional coordinates of 10 reflective markers are obtained, namely P1(10.5, 20.3, 5.2), P2(15.7, 22.1, 5.5), P3(22.3, 25.6, 6.0), etc., these points are sorted according to the physical continuity of the cable laying to form an ordered sequence of markers. The nearest neighbor search algorithm can be used in the sorting process, that is, starting from the starting marker, the nearest unselected point is searched as the next point until all points are included in the sequence.

[0088] After sorting, the distance and curvature change between adjacent marker points are calculated to determine the node vector distribution of the cubic spline curve. For each pair of adjacent marker points Pi and Pi+1, their Euclidean distance di,i+1 is calculated. In practical applications, the distance values ​​can be d1,2=6.3 cm and d2,3=7.8 cm. Based on these distance values ​​and the distribution characteristics of the points, a non-uniform node vector sequence is constructed. In determining the node vector distribution, node values ​​can be assigned according to the proportional relationship between the distances between adjacent points. For example, the node vector can be set as [0, 0, 0, 0, 0.15, 0.32, 0.48, 0.63, 0.79, 0.91, 1, 1, 1, 1], where the values ​​in the middle part reflect the distance ratio between marker points.

[0089] To control the smoothness of the curve, a tension coefficient α is introduced. This coefficient typically ranges from 0.1 to 0.9. In this embodiment, α is set to 0.3. A smaller tension coefficient makes the curve smoother, which is suitable for describing slowly bending cables. If α is set to 0.7, the curve will be closer to the marker point, which is suitable for representing areas of sharp bending.

[0090] When determining control points, geometric constraints are established for each marker point Pi, including positional constraints and tangent constraints. For each point in the marker point sequence, the positional difference vector between its preceding and following points is calculated as the approximate tangent direction at that point. For example, the tangent direction of marker point P3 can be calculated using (P4-P2) / ||P4-P2||, with the result being (0.53, 0.78, 0.32).

[0091] Based on these constraints and tension coefficients, the positions of the control points are calculated. For each marker point Pi, at least two control points Ci,1 and Ci,2 need to be determined. The method for calculating the control points is as follows: using the marker point as a reference, offset it a certain distance forward or backward along the tangent direction. The offset distance is related to the distance between adjacent marker points and the tension coefficient. For example, C3,1 can be calculated as P3 minus the tangent direction multiplied by d2,3×α / 3, resulting in (20.7, 23.8, 5.7); C3,2 is P3 plus the tangent direction multiplied by d3,4×α / 3, resulting in (23.9, 27.4, 6.3).

[0092] After the control points are initially determined, the weight coefficients of the control points are optimized by the least squares method so that the generated cubic spline curve best fits all the marked points. The error function is set as the sum of squared Euclidean distances between the position of the spline curve at the parameter value of the marked point and the actual position of the marked point. The weights of the control points are adjusted by an iterative optimization algorithm (such as gradient descent) until the error drops below the threshold (such as 0.001) or reaches the maximum number of iterations (such as 100 times).

[0093] After optimization, a continuous and smooth cubic spline curve is obtained, which represents the real-time deformation profile of the fiber optic temperature sensing cable. This curve can be represented by the parametric equation C(t), where t∈[0,1].

[0094] For temperature measurement point location, the initial distance ratio between each temperature measurement point and adjacent marker points is first determined. After fiber deformation, the new position of the temperature measurement point is located on the spline curve using an arc length parameterization method. The local arc length variation caused by fiber deformation is considered to compensate for the temperature measurement point position. The fiber elongates or compresses in certain sections, causing changes in the actual arc length ratio. The compensation coefficient is calculated by comparing the arc length ratio between adjacent marker points before and after deformation. The adjusted ratio coefficient is used to recalculate the parameter values, which are then substituted into the parametric equation to obtain the real-time three-dimensional coordinates of the temperature measurement point. This process is repeated for all temperature measurement points on the cable to obtain complete real-time three-dimensional coordinate information of the temperature measurement points, providing an accurate basis for subsequent spatial location analysis of temperature data.

[0095] In one optional implementation, based on the initial distance ratio between the temperature measuring point and adjacent reflective marker points, arc length parameterized interpolation positioning is performed on a continuous smooth deformation curve. The local arc length change caused by fiber deformation is calculated, and real-time compensation for the temperature measuring point position is performed, including:

[0096] Based on the initial distance ratio between the temperature measuring point and the adjacent reflective marker point, a continuous smooth deformation curve is fitted using the non-uniform rational B-spline method. The deformation curve is then represented by arc length parameterization. The curvature value at the curve node is calculated. Based on the curvature value, the parameter interval is divided using an adaptive step size, and the correspondence between the parameter and the arc length is established.

[0097] Calculate the local deformation characteristics of the optical fiber, establish a mapping relationship between the local deformation characteristics and the position offset based on the correspondence, calculate the compensation amount of the temperature measuring point position using the piecewise cubic Hermit interpolation method, iteratively optimize the temperature measuring point position in the parameter space until the preset accuracy requirements are met, and realize real-time compensation of the temperature measuring point position.

[0098] In a fiber optic distributed temperature measurement system, several reflective markers are set as position references. Assuming N reflective markers are deployed on the fiber, their initial positions are known and fixed. The temperature measuring point is located between adjacent reflective markers. The initial distance ratio between the temperature measuring point and its adjacent reflective markers can be obtained through initial calibration. For example, if temperature measuring point i is located between reflective markers j and j+1, the initial distance ratio can be expressed as the ratio of the distance from temperature measuring point i to reflective marker j to the total distance between reflective marker j and j+1, denoted as ri,j. In practical applications, reflective markers can be placed every 5 meters, with the temperature measuring point located between two adjacent reflective markers. If the temperature measuring point is 2 meters from the previous reflective marker and 3 meters from the next reflective marker, then its initial distance ratio is 0.4.

[0099] When the optical fiber deforms, to ensure the spatial positioning accuracy of the temperature measurement data, compensation is needed for the position of the temperature measurement points. A non-uniform rational B-spline method is used to fit a continuous smooth deformation curve. A sequence of control points is constructed based on the position coordinates of the reflective markers. Appropriate node vectors and weighting coefficients are set. The control points can be points in three-dimensional space, determined by the spatial coordinates of the reflective markers. For example, in a specific application scenario, 10 control points can be used, with open uniform node vectors and weighting coefficients set to 1.0 for each. By adjusting the positions of the control points, the spline curve is made to accurately pass through all the reflective markers.

[0100] The spline curve is represented by arc length parameterization, establishing a correspondence between parameters and arc length. Since the curve parameters and arc length are usually not linearly related, it is necessary to calculate the curvature values ​​at the curve nodes and adaptively divide the parameter intervals based on the curvature values. In regions with large curvature, the parameter intervals should be smaller to ensure the accuracy of arc length calculation; in regions with small curvature, the parameter intervals can be appropriately increased. For example, when the curvature of the curve is greater than 0.5 at a certain point, the parameter interval at that point is subdivided into one-quarter of the original interval; when the curvature is between 0.1 and 0.5, it is subdivided into half; when the curvature is less than 0.1, the original parameter interval remains unchanged.

[0101] Arc length can be calculated using numerical integration methods, such as Gaussian quadrature or Simpson's method. For each divided parameter interval, the corresponding arc length increment is calculated, and the total arc length from the curve's starting point to the current parameter point is accumulated. A lookup table of parameter values ​​and corresponding arc lengths is then established for subsequent interpolation positioning. For example, the parameter interval [0,1] can be divided into 100 sub-intervals, and the arc length corresponding to each parameter value can be calculated, forming a lookup table containing 101 data points.

[0102] Calculating the local deformation characteristics of optical fiber mainly includes changes in local elongation and curvature. Local elongation can be calculated by the ratio of the change in arc length between adjacent reflective markers before and after deformation. For example, if the arc length between reflective markers j and j+1 before deformation is Lj,j+1, and after deformation it becomes L'j,j+1, then the local elongation is L'j,j+1 / Lj,j+1. In practical applications, if the original arc length is 5 meters and it becomes 5.2 meters after deformation, then the local elongation is 1.04, indicating that the optical fiber segment has been stretched by 4%.

[0103] Based on the mapping relationship between local deformation characteristics and positional offset, a piecewise cubic Hermitian interpolation method is used to calculate the positional compensation of the temperature measurement points. Hermitian interpolation considers not only the position of the data points but also derivative information, which can generate a smoother interpolation curve. When constructing the Hermitian interpolation, the positional information comes from the spatial coordinates of the reflective marker points, and the derivative information can be approximated by the difference between adjacent data points. For example, for the deformed curve, the tangent vector can be calculated at each reflective marker point and used as the derivative condition for the Hermitian interpolation.

[0104] The position of the temperature measurement point is iteratively optimized within the parameter space. The iteration process starts from the initial position of the temperature measurement point, calculates the position offset based on local elongation and curvature changes, updates the position of the temperature measurement point in the parameter space, and recalculates the corresponding spatial coordinates and local deformation characteristics. This process is repeated until the position change is less than a preset threshold. The preset accuracy threshold can be set to 0.01 meters, meaning that when the change in position compensation is less than 1 centimeter, the convergence condition is considered met, and real-time compensation of the temperature measurement point position is completed.

[0105] In practical applications, optical fibers are laid out along complex structures, such as the casing of power equipment. Assuming the initial state of the optical fiber is an "S" shape along the surface of a power transformer, its shape changes when the transformer's temperature rises, causing the casing to expand. Using the method described above, the actual position of the temperature measurement point after deformation can be accurately calculated, ensuring that the temperature data correctly corresponds to the actual monitoring location. This improves the spatial positioning accuracy of the temperature measurement system and provides reliable data support for equipment condition monitoring.

[0106] In one optional implementation, a distance-attenuation-based weight initialization method is used to assign initial weights to the temperature measurement points of each temperature field dataset. The weight coefficients are dynamically adjusted according to the magnitude of the local temperature gradient, and the optimal weight combination is determined through iterative optimization to generate three-dimensional temperature field mesh data, including:

[0107] The local feature scale is determined based on the spatial density distribution of temperature measurement points. The local feature scale is then substituted into the attenuation function to calculate the distance attenuation coefficient. The distance attenuation coefficient is corrected by combining the measurement uncertainty information, and initial weights are assigned to the temperature measurement points of each temperature field dataset.

[0108] The weighted average temperature is calculated using the distance matrix between grid points and temperature measurement points to construct the initial value of the temperature field; the initial value of the temperature field is then fitted by least squares to generate a local temperature distribution function.

[0109] The temperature gradient vector is solved based on the local temperature distribution function, and the weighting coefficients are dynamically adjusted for the temperature gradient vector.

[0110] An iterative optimization method is used to alternately update the weight coefficients and the local temperature distribution function until the temperature field changes in adjacent iterations meet the convergence condition. Based on the final determined weight coefficients and the temperature data in the temperature field dataset, a three-dimensional temperature field mesh is generated.

[0111] like Figure 2 As shown, the method includes:

[0112] The temperature field dataset contains multiple temperature measurement points, each with three-dimensional spatial coordinates (x, y, z) and a corresponding temperature value T and measurement uncertainty u. For example, in an industrial furnace, 50 temperature measurement points are arranged, with spatial coordinates ranging from (x1, y1, z1) to (x...). 50 ,y 50 ,z 50 The corresponding temperature readings are T1 to T. 50 The measurement uncertainty is u1 to u 50 .

[0113] The local feature scale is determined based on the spatial density distribution of temperature measurement points. For each temperature measurement point, the average distance between it and its k nearest neighbors (usually k is 5 to 10) is calculated as the local feature scale d of that point. For example, for temperature measurement point (x3, y3, z3), if the distances of its five nearest neighbors are 15mm, 18mm, 22mm, 25mm and 27mm respectively, then the local feature scale d3 of that point is 21.4mm.

[0114] The distance attenuation coefficient is calculated by substituting the local feature scale into the attenuation function. Using a Gaussian attenuation function, the distance attenuation coefficient between any two temperature measuring points i and j can be calculated as the ratio of the Euclidean distance between the two points to the local feature scale. For example, if the distance between temperature measuring points i and j is 30 mm, and the local feature scale of point i is 20 mm, then the distance attenuation coefficient can be expressed as exp(-30...). 2 / 20 2 ), approximately 0.144.

[0115] The distance attenuation coefficient is corrected, and the smaller the measurement uncertainty, the greater the weight of the temperature measurement point should be. For example, for temperature measurement point i, its measurement uncertainty is 0.5℃ and its reference uncertainty is 1.0℃, then the uncertainty correction coefficient is 1.0 / 0.5 = 2.0. Multiplying this coefficient by the distance attenuation coefficient yields the corrected attenuation coefficient.

[0116] Initial weights are assigned to the temperature measurement points in each temperature field dataset. The corrected distance attenuation coefficients are normalized to obtain the initial weights for each temperature measurement point. For example, if the corrected attenuation coefficient for a certain temperature measurement point is 0.288, and the sum of the corrected attenuation coefficients for all temperature measurement points is 5.76, then the initial weight of that temperature measurement point is 0.288 / 5.76 = 0.05.

[0117] Construct a three-dimensional temperature field mesh to be solved, and determine the mesh resolution according to the application requirements. For example, divide a space of 1000mm×800mm×600mm into a mesh of 50×40×30, and calculate the temperature value of each mesh point.

[0118] Calculate the distance matrix between grid points and temperature measurement points. For each grid point, calculate its Euclidean distance to all temperature measurement points to form a distance matrix. For example, for a grid point (100mm, 200mm, 300mm), calculate its distance to all 50 temperature measurement points to obtain a distance vector containing 50 elements.

[0119] Based on the distance matrix and initial weights, the weighted average temperature of each grid point is calculated to construct the initial value of the temperature field. For example, if a grid point is 25mm, 30mm, and 40mm away from the three nearest temperature measurement points, with corresponding temperatures of 85℃, 83℃, and 87℃, and initial weights of 0.06, 0.05, and 0.03, then the weighted average temperature of this grid point is (85×0.06+83×0.05+87×0.03) / (0.06+0.05+0.03)=84.86℃.

[0120] The initial temperature field is fitted using least-squares to generate a local temperature distribution function. A certain number of temperature measurement points are selected near each grid point, and a polynomial function is used to fit the local temperature distribution. For example, 10 temperature measurement points are selected around a certain grid point, and a second-order polynomial is used for fitting to obtain the temperature distribution function near that point.

[0121] The temperature gradient vector is obtained by solving the local temperature distribution function. This is done by calculating the partial derivatives of the temperature distribution function in the x, y, and z directions. For example, a temperature gradient vector at a certain grid point might be (-0.15℃ / mm, 0.25℃ / mm, 0.10℃ / mm), indicating that at that point, the temperature decreases by 0.15℃ per millimeter in the x-direction, increases by 0.25℃ per millimeter in the y-direction, and increases by 0.10℃ per millimeter in the z-direction.

[0122] The weighting coefficients of the temperature gradient vector are dynamically adjusted. In regions with large temperature gradients, the weight of temperature measurement points in that direction is increased. Specifically, the cosine of the angle between the line connecting a temperature measurement point and a grid point and the direction of the temperature gradient is calculated. The smaller the angle, the greater the weight gain. For example, if the angle between the line connecting a temperature measurement point and a grid point and the direction of the temperature gradient is 30°, then the weight gain coefficient is cos(30°) = 0.866. The adjusted weights are obtained by multiplying the original weights by (1 + 0.866).

[0123] By repeatedly performing the above steps, the temperature of the grid point is recalculated using the updated weights in each iteration, and then the temperature distribution function is refitted and the temperature gradient is calculated, and the weights are adjusted again. For example, after 5 iterations, the temperature value of a certain grid point gradually converges from the initial 84.86℃ to 85.23℃.

[0124] The maximum change in temperature between two consecutive iterations is compared with a preset threshold (e.g., 0.01℃). If the change is less than the threshold, the iteration is considered to have converged. For example, the maximum change in temperature between the 6th and 5th iterations is 0.008℃, which is less than the threshold of 0.01℃, so the iteration terminates.

[0125] Based on the final determined weighting coefficients and the temperature data in the temperature field dataset, a three-dimensional temperature field mesh is generated. The final weights are then applied to each temperature measurement point to calculate the temperature values ​​for all mesh points, forming a complete three-dimensional temperature field dataset. This data can be further used for applications such as temperature field visualization, thermal stress analysis, or process control.

[0126] This invention relates to a cable-type heat-sensing fire detection system with intelligent positioning function, the system comprising:

[0127] The first unit is used to lay multiple fiber optic temperature sensing cables along multiple preset longitudinal paths in the space to be measured. The surface of the fiber optic temperature sensing cables is set with reflective marking points at preset intervals, and multiple temperature measuring points are set at preset intervals on each fiber optic temperature sensing cable.

[0128] The second unit is used to acquire images of reflective markers, extract the two-dimensional coordinate information of reflective markers in the images, calculate the real-time three-dimensional spatial position of reflective markers using a multi-view three-dimensional reconstruction algorithm, obtain the real-time deformation profile of the fiber optic temperature sensing cable using a cubic spline curve fitting method, and calculate the real-time three-dimensional coordinate information of the temperature measuring point based on the relative distance between the real-time three-dimensional spatial position of the reflective markers and the temperature measuring point.

[0129] The third unit is used to associate temperature data with the real-time three-dimensional coordinate information of the corresponding temperature measurement point to obtain a temperature field dataset.

[0130] The fourth unit is used to assign initial weights to the temperature measurement points of each temperature field dataset using a distance decay-based weight initialization method. The weight coefficients are dynamically adjusted according to the magnitude of the local temperature gradient, and the optimal weight combination is determined through iterative optimization to generate three-dimensional temperature field mesh data.

[0131] The fifth unit is used to provide temperature warnings and abnormal temperature location information based on the three-dimensional temperature field grid data and a preset warning temperature threshold.

[0132] A third aspect of the present invention,

[0133] An electronic device is provided, comprising:

[0134] processor;

[0135] Memory used to store processor-executable instructions;

[0136] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0137] Fourth aspect of the present invention,

[0138] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0139] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cable type temperature sensing fire detection method with intelligent positioning function, characterized in that, The method comprises the following steps: a plurality of optical fiber temperature sensing cables are arranged along a plurality of longitudinal paths preset in a measured space, and reflective marker points are arranged on the surfaces of the optical fiber temperature sensing cables at a preset interval, and a plurality of temperature measurement points are arranged on each optical fiber temperature sensing cable at a preset distance; two-dimensional coordinate information of the reflective marker points in the image is extracted, real-time three-dimensional spatial positions of the reflective marker points are calculated by using a multi-view three-dimensional reconstruction algorithm, a real-time deformation profile of the optical fiber temperature sensing cable is obtained by using a cubic spline curve fitting method, and real-time three-dimensional coordinate information of the temperature measurement points is calculated according to a relative distance relationship between the real-time three-dimensional spatial positions of the reflective marker points and the temperature measurement points; temperature data is associated with the real-time three-dimensional coordinate information of the corresponding temperature measurement points to obtain a temperature field data set; an initial weight is assigned to each temperature measurement point of each temperature field data set by using a weight initialization method based on distance attenuation, a weight coefficient is dynamically adjusted according to a local temperature gradient, an optimal weight combination is determined by iterative optimization, and a three-dimensional temperature field grid data is generated; temperature warning and abnormal temperature positioning information are given according to the three-dimensional temperature field grid data and in combination with a preset warning temperature threshold.

2. The method of claim 1, wherein, The two-dimensional coordinate information of the reflective marker points in the image is extracted, and the real-time three-dimensional spatial positions of the reflective marker points are calculated by using a multi-view three-dimensional reconstruction algorithm, which comprises the following steps: a region growing algorithm is used to extract a complete candidate reflective region with the reflective marker point region as a seed point, the centroid coordinates of the candidate reflective region are calculated, and two-dimensional coordinates and confidence scores of the reflective marker points are obtained; an intrinsic matrix and distortion coefficients of a camera are obtained, a relative pose between cameras is calculated based on epipolar geometry constraints, and a unified spatial coordinate system is established; the two-dimensional coordinates of the reflective marker points are projected into the unified spatial coordinate system, a corresponding relationship of the reflective marker points under different perspectives is established, a real-time three-dimensional spatial position of the reflective marker points is output by combining a triangulation method and an observation iterative optimization method.

3. The method of claim 2, wherein, The two-dimensional coordinates of the reflective marker points are projected into the unified spatial coordinate system, the corresponding relationship of the reflective marker points under different perspectives is established, and the real-time three-dimensional spatial position of the reflective marker points is output by combining a triangulation method and an observation iterative optimization method, which comprises the following steps: the two-dimensional coordinates of the reflective marker points are projected into the unified spatial coordinate system to establish a projection relationship equation of two-dimensional image coordinates and three-dimensional spatial coordinates, an initial solution is obtained by using a linear least squares method, a spatial position continuity constraint term is introduced, and three-dimensional coordinates are iteratively optimized by combining observation weights; a covariance matrix of the three-dimensional coordinates is calculated based on a Jacobian matrix of an observation equation, and a coordinate component standard deviation is obtained; a re-projection error of the reflective marker points under each perspective is calculated according to the covariance matrix and the coordinate component standard deviation; observation data with a re-projection error greater than a preset error threshold is removed, and the observation data after screening is used to perform optimization solving again until a preset convergence condition is met, and a final three-dimensional spatial position of the reflective marker points is output.

4. The method of claim 1, wherein, The real-time deformation profile of the optical fiber temperature sensing cable is obtained by using a cubic spline curve fitting method, and the real-time three-dimensional coordinate information of the temperature measurement points is calculated according to the relative distance relationship between the real-time three-dimensional spatial positions of the reflective marker points and the temperature measurement points, which comprises the following steps: Sort the three-dimensional spatial positions of the retro-reflective marker points according to the laying path of the fiber-optic temperature cable to construct a marker point sequence; Calculate the node vector distribution according to the distance between adjacent marker points in the marker point sequence and the curvature variation, control the smoothness of the curve shape by introducing an adjustable tension coefficient, determine the control point spatial distribution in combination with the geometric constraints at the marker points, optimize the control point weight coefficient by using the least square method, and generate a continuous smooth deformation curve of the fiber-optic temperature cable; According to the initial distance proportion relationship between the temperature measuring points and the adjacent retro-reflective marker points, perform arc length parameterized interpolation positioning on the continuous smooth deformation curve, calculate the local arc length variation caused by the fiber deformation, and perform real-time compensation on the temperature measuring point positions to output real-time three-dimensional coordinate information of the temperature measuring points.

5. The method of claim 4, wherein, According to the initial distance proportion relationship between the temperature measuring points and the adjacent retro-reflective marker points, perform arc length parameterized interpolation positioning on the continuous smooth deformation curve, calculate the local arc length variation caused by the fiber deformation, and perform real-time compensation on the temperature measuring point positions to output real-time three-dimensional coordinate information of the temperature measuring points. According to the initial distance proportion relationship between the temperature measuring points and the adjacent retro-reflective marker points, perform arc length parameterized interpolation positioning on the continuous smooth deformation curve, calculate the local arc length variation caused by the fiber deformation, and perform real-time compensation on the temperature measuring point positions to output real-time three-dimensional coordinate information of the temperature measuring points. Calculate the local deformation characteristics of the fiber, establish a mapping relationship between the local deformation characteristics and the position offset according to the corresponding relationship, calculate the temperature measuring point position compensation amount by using the piecewise cubic Hermite interpolation method, and iteratively optimize the temperature measuring point positions in the parameter space until the preset accuracy requirement is met, to realize real-time compensation of the temperature measuring point positions.

6. The method of claim 1, wherein, The initial weight of each temperature field data set is assigned by using a weight initialization method based on distance attenuation, the weight coefficient is dynamically adjusted according to the local temperature gradient, and the optimal weight combination is determined by iterative optimization to generate three-dimensional temperature field grid data, including: The local feature scale is determined based on the spatial density distribution of the temperature measuring points, the distance attenuation coefficient is calculated by substituting the local feature scale into the attenuation function, and the initial weight of each temperature field data set is assigned by modifying the distance attenuation coefficient based on the measurement uncertainty information; The weighted average temperature is calculated by using the distance matrix between the grid points and the temperature measuring points to construct the temperature field initial value; the temperature field initial value is least square fitted to generate a local temperature distribution function; The temperature gradient vector is solved based on the local temperature distribution function, and the weight coefficient is dynamically adjusted for the temperature gradient vector; The weight coefficient and the local temperature distribution function are alternately updated by using an iterative optimization method until the temperature field changes of adjacent iterations meet the convergence condition, and the three-dimensional temperature field grid data is generated according to the finally determined weight coefficient and the temperature data in the temperature field data set.

7. A cable-type temperature-sensing fire detection system with intelligent positioning functionality for implementing the method according to any one of claims 1 to 6, characterized in that including: The first unit is configured to arrange a plurality of fiber-optic temperature cables along a plurality of longitudinal paths in a measured space, and set retro-reflective marker points on the surfaces of the fiber-optic temperature cables at a preset interval, and set a plurality of temperature measuring points on each fiber-optic temperature cable at a preset distance; The second unit is configured to collect a retro-reflective marker point image, extract two-dimensional coordinate information of the retro-reflective marker point in the image, calculate real-time three-dimensional spatial positions of the retro-reflective marker point by using a multi-view three-dimensional reconstruction algorithm, obtain a real-time deformation profile of the fiber-optic temperature cable by using a cubic spline curve fitting method, and calculate real-time three-dimensional coordinate information of the temperature measurement point according to a relative distance relationship between the real-time three-dimensional spatial positions of the retro-reflective marker point and the temperature measurement point. The third unit is configured to associate temperature data with real-time three-dimensional coordinate information of the corresponding temperature measurement point to obtain a temperature field data set. The fourth unit is configured to assign an initial weight to each temperature measurement point of the temperature field data set by using a weight initialization method based on distance attenuation, dynamically adjust a weight coefficient according to a local temperature gradient size, determine an optimal weight combination by iterative optimization, and generate three-dimensional temperature field grid data. The fifth unit is configured to give temperature warning and abnormal temperature positioning information according to the three-dimensional temperature field grid data in combination with a preset warning temperature threshold.

8. An electronic device, comprising: The computer program instructions are executed by the processor to implement the method in any one of claims 1 to 6. The computer program instructions are executed by the processor to implement the method in any one of claims 1 to 6. ​ ​ 9. A computer-readable storage medium having stored thereon computer program instructions, wherein, ​

Images