CN-122023413-A - Equipment thermal defect three-dimensional quantitative detection method, system and equipment based on visible light and infrared fusion

CN122023413ACN 122023413 ACN122023413 ACN 122023413ACN-122023413-A

Abstract

The application provides a three-dimensional quantitative detection method, a system and equipment for equipment thermal defects based on visible light and infrared fusion, which are used for responding to trigger pulses to acquire original visible light images, infrared radiation images and sparse laser ranging point data at the same physical time; the method comprises the steps of inputting a preprocessed visible light image into a monocular depth estimation network to generate a dense relative depth map representing a relative depth relation of a scene, extracting relative depth values corresponding to pixel coordinates of laser ranging points in the dense relative depth map, solving global scale factors and system offset, transforming space point coordinates in an absolute scale three-dimensional point cloud into an infrared sensor coordinate system, constructing a three-dimensional fusion point cloud model integrating temperature and geometric attributes according to an infrared sensor internal parameter, screening the three-dimensional fusion point cloud model according to a preset physical temperature threshold to obtain a overheat point set, and dividing the overheat point set into at least one defect point cloud cluster by applying a three-dimensional Euclidean space clustering algorithm.

Inventors

YUAN MINGXU
Zhou Binsen
YUAN HONG
LI DONGMING
ZHAO ZHIMIN
MAO KEXIANG
DING ZHENG
WANG XIN
ZHOU XIAOBO
Gan Zaixu

Assignees

国网四川雅安电力（集团）股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. The three-dimensional quantitative detection method for the thermal defects of the equipment based on the fusion of visible light and infrared light is applied to detection of the power equipment to be detected, and is characterized by comprising the following steps: responding to the trigger pulse, and driving a visible light sensor, an infrared sensor and a laser sensor to acquire original visible light images, infrared radiation images and sparse laser ranging point data at the same physical time; extracting pixel coordinates of laser spots in the visible light image and the infrared image, taking the pixel coordinates as a space association control point, and optimizing a joint calibration external parameter matrix among the visible light sensor, the infrared sensor and the laser sensor in real time through a beam method adjustment algorithm; converting the dense relative depth map into an absolute depth map with a real physical size, and back-projecting to generate an absolute-scale three-dimensional point cloud; transforming the space point coordinates in the absolute scale three-dimensional point cloud to an infrared sensor coordinate system by applying the optimized joint calibration external parameter matrix; Projecting the transformed points to an infrared image plane according to the internal parameters of the infrared sensor, obtaining absolute temperature values of corresponding pixel positions through bilinear interpolation, binding the absolute temperature values as attributes to corresponding space points, and constructing a three-dimensional fusion point cloud model integrating temperature and geometric attributes; Screening the three-dimensional fusion point cloud model according to a preset physical temperature threshold to obtain an overheat point set; And dividing the overheat point set into at least one defect point cloud cluster by using a three-dimensional Euclidean space clustering algorithm.
2. The method for three-dimensional quantitative detection of thermal defects of equipment based on fusion of visible light and infrared light according to claim 1, further comprising, before optimizing the joint calibration extrinsic matrix among the visible light sensor, the infrared sensor and the laser sensor in real time by a beam method adjustment algorithm: respectively carrying out radial distortion correction and tangential distortion correction on the original visible light image and the infrared radiation image; And improving the extraction precision of the centroid of the laser spot in the image by applying a characteristic enhancement algorithm so as to improve the geometric consistency of different mode data in the three-dimensional space mapping.
3. The method for three-dimensional quantitative detection of thermal defects of equipment based on fusion of visible light and infrared light according to claim 2, wherein the method for optimizing the joint calibration extrinsic matrix among the visible light sensor, the infrared sensor and the laser sensor in real time through a beam method adjustment algorithm comprises the following steps: In the detection process, monitoring the re-projection error of the laser light spot in the visible light and infrared image in real time; when the re-projection error exceeds a preset deviation threshold due to environmental vibration or temperature change, an external parameter correction flow is automatically triggered, and rotation and translation parameters between the sensors are dynamically updated by minimizing the spatial distance error of the spatial correlation control points.
4. The method for three-dimensional quantitative detection of thermal defects of a device based on fusion of visible light and infrared light according to claim 3, further comprising: inputting the preprocessed visible light image into a monocular depth estimation network to generate a dense relative depth map representing the relative depth relation of a scene; Extracting a relative depth value corresponding to the pixel coordinate of the laser ranging point in the dense relative depth map, constructing a linear scale recovery model by combining an absolute distance true value provided by the laser ranging point, and solving a global scale factor and a system offset; extracting a depth median value in a neighborhood window taking pixel coordinates of the laser spots as a center from the dense relative depth map as a sampling observation value; Performing linear regression analysis on a plurality of sampling observation values and corresponding laser ranging absolute distance true values, and calculating slope and intercept of mapping the relative depth value into absolute depth; the slope and intercept are the global scale factor and system offset, respectively.
5. The method for three-dimensional quantitative detection of thermal defects of a device based on fusion of visible light and infrared light according to claim 4, further comprising: For each defective point cloud cluster, determining a defective three-dimensional centroid position by calculating an arithmetic average value of all point cloud coordinates of the defective point cloud cluster, and fitting a defective surface approximate plane by principal component analysis to calculate an actual projection area of the defective point cloud cluster; calculating the average value of the coordinates of each space point in the single defect point cloud cluster to obtain a three-dimensional space coordinate representing the physical center of the thermal defect; extracting the main plane direction of the defect point cloud cluster by a principal component analysis method, projecting the point cloud onto the main plane to restore the unfolded form of the defect surface, and calculating the area of a closed area of the projected polygon as the actual physical area of the closed area; and comparing a preset equipment operation risk rule with the actual physical area, the highest temperature and the relative temperature rise value in the point cloud cluster, and judging the emergency level of the thermal defect.
6. The method for three-dimensional quantitative detection of thermal defects of equipment based on fusion of visible light and infrared light according to claim 1, further comprising: Non-uniformity correction is carried out on the original response data of the infrared sensor so as to eliminate response difference among detector pixels; combining a preset blackbody calibration mapping table to convert the corrected gray scale value into an initial temperature value; And acquiring the material emissivity of the surface of the equipment to be detected, and compensating and correcting the initial temperature value by combining the environmental background temperature parameter acquired in real time to acquire the real absolute temperature distribution of the surface of the equipment.
7. The method for three-dimensional quantitative detection of thermal defects of equipment based on fusion of visible light and infrared light according to claim 1, further comprising: the operation load current data of the power equipment to be tested is called; carrying out load level correction on the absolute temperature value according to the running load current data so as to obtain normalized temperature rise data under standard load; Similar defect parameters of the same equipment in the historical inspection record are searched, and the degradation rate of the current thermal defect is calculated; and carrying out decision analysis on the thermal defects according to the current environmental meteorological parameters and preset equipment importance degree weights, so as to distribute corresponding overhaul work orders.
8. The method for three-dimensional quantitative detection of thermal defects of equipment based on fusion of visible light and infrared light according to claim 1, further comprising: extracting a three-dimensional model slice, three-dimensional space position coordinates, an actual physical area and a risk assessment conclusion of the thermal defect, and generating a structured diagnosis report with a space-time index label; and projecting the three-dimensional fusion point cloud model to an augmented reality interface of the mobile terminal to realize virtual-real combination plotting of the thermal defect on the real equipment space position.
9. A three-dimensional quantitative detection system for thermal defects of equipment, which is applied to the three-dimensional quantitative detection method for thermal defects of equipment based on fusion of visible light and infrared rays according to any one of claims 1 to 8, and is characterized in that the system comprises: The signal trigger module is used for generating trigger pulses; The information extraction module is used for responding to the trigger pulse, driving a visible light sensor, an infrared sensor and a laser sensor to acquire original visible light images, infrared radiation images and sparse laser ranging point data at the same physical time, extracting pixel coordinates of laser spots in the visible light images and the infrared images, taking the pixel coordinates as space association control points, and optimizing a joint calibration external reference matrix among the visible light sensor, the infrared sensor and the laser sensor in real time through a beam method adjustment algorithm; The coordinate transformation module is used for inputting the preprocessed visible light image into a monocular depth estimation network to generate a dense relative depth map representing the relative depth relation of a scene, extracting relative depth values corresponding to pixel coordinates of laser ranging points in the dense relative depth map, constructing a linear scale recovery model by combining absolute distance true values provided by the laser ranging points, solving a global scale factor and a system offset, converting the dense relative depth map into an absolute depth map with a real physical size according to the global scale factor and the system offset, and back projecting to generate an absolute scale three-dimensional point cloud; The defect diagnosis module is used for screening the three-dimensional fusion point cloud model according to a preset physical temperature threshold value to obtain an overheat point set, dividing the overheat point set into at least one defect point cloud cluster by applying a three-dimensional Euclidean space clustering algorithm, determining the three-dimensional mass center position of the defect by calculating the arithmetic average value of all point cloud coordinates of the defect point cloud cluster aiming at each defect point cloud cluster, and fitting a defect surface approximate plane by principal component analysis to calculate the actual projection area of the defect point cloud cluster.
10. An electronic device, comprising: Processor, and A memory having stored thereon computer readable instructions for controlling the processor to perform the method for three-dimensional quantitative detection of thermal defects of a device based on fusion of visible light and infrared according to any one of claims 1 to 8.

Description

Equipment thermal defect three-dimensional quantitative detection method, system and equipment based on visible light and infrared fusion Technical Field The application relates to the technical field of power inspection, in particular to a method, a system and equipment for three-dimensional quantitative detection of thermal defects of equipment based on fusion of visible light and infrared. Background With the rapid increase of the operation scale and the automatic inspection requirements of power system equipment, higher requirements are put on online, non-contact and quantifiable defect identification. The operation and maintenance frequency is improved, the unmanned aerial vehicle and the robot patrol are popularized, and the dependence on digital twin and predictive maintenance is achieved, so that the requirements of positioning accuracy, traceability and automatic decision can not be met by a pure two-dimensional thermal image 'hot spot' and meanwhile, engineering constraint requirements of cost, power consumption and portability require that patrol equipment is low in cost and easy to deploy, and the equipment has absolute measurement capability available for engineering. In the related technology, the main flow method generally follows a paradigm of reconstructing three-dimensional geometry first and then registering and superposing two-dimensional infrared thermal images, but the method is easy to generate space mapping misalignment under the conditions of sparse texture, shielding or field disturbance, and often stays only on a visual superposition layer, and stable one-to-one mapping from temperature pixels to three-dimensional surface points cannot be constructed, so that the real three-dimensional position and physical size of the thermal defects cannot be reliably and automatically derived. Disclosure of Invention In view of the foregoing, it is desirable to provide a method, a system and a device for three-dimensional quantitative detection of thermal defects of a device based on fusion of visible light and infrared light, which can overcome at least one of the above defects. In a first aspect, an embodiment of the present application provides a method for three-dimensionally quantifying and detecting a thermal defect of a device based on fusion of visible light and infrared light, which is applied to detecting a power device to be detected, and the method includes: responding to the trigger pulse, and driving a visible light sensor, an infrared sensor and a laser sensor to acquire original visible light images, infrared radiation images and sparse laser ranging point data at the same physical time; Extracting pixel coordinates of laser spots in the visible light image and the infrared image, taking the pixel coordinates as space association control points, and optimizing a joint calibration external parameter matrix among the visible light sensor, the infrared sensor and the laser sensor in real time through a beam method adjustment algorithm; Converting the dense relative depth map into an absolute depth map with a real physical size, and back-projecting to generate an absolute-scale three-dimensional point cloud; transforming the space point coordinates in the absolute scale three-dimensional point cloud to an infrared sensor coordinate system by applying the optimized joint calibration external parameter matrix; Projecting the transformed points to an infrared image plane according to the internal parameters of the infrared sensor, obtaining absolute temperature values of corresponding pixel positions through bilinear interpolation, binding the absolute temperature values as attributes to corresponding space points, and constructing a three-dimensional fusion point cloud model integrating temperature and geometric attributes; Screening the three-dimensional fusion point cloud model according to a preset physical temperature threshold to obtain an overheat point set; And dividing the overheat point set into at least one defect point cloud cluster by using a three-dimensional Euclidean space clustering algorithm. In an embodiment, before optimizing the joint calibration extrinsic matrix among the visible light sensor, the infrared sensor and the laser sensor in real time by using a beam method adjustment algorithm, the method further comprises: respectively carrying out radial distortion correction and tangential distortion correction on the original visible light image and the infrared radiation image; And improving the extraction precision of the centroid of the laser spot in the image by applying a characteristic enhancement algorithm so as to improve the geometric consistency of different mode data in the three-dimensional space mapping. In an embodiment, the optimizing the joint calibration extrinsic matrix among the visible light sensor, the infrared sensor and the laser sensor in real time through a beam method adjustment algorithm includes: In the detection process, monitoring the