Industrial gas leakage detection method based on infrared thermal imaging

Abstract

The invention relates to the field of image processing, in particular to an industrial gas leakage detection method based on infrared thermal imaging, which comprises the following steps of: acquiring video streams by using a plurality of infrared thermal imagers, and constructing a gas basic geometric voxel grid through a background difference and view cone intersection method; thirdly, the sight line penetration depth and a statistical variable coefficient are calculated, and adaptive normalized radiation feature mapping based on statistical moments is constructed so as to solve errors caused by traditional homogeneous hypothesis; then, measuring and calculating a three-dimensional form compactness factor, generating a heterogeneous space density field weight in combination with the relative depth, and fitting a real diffusion gradient; and finally, calibrating a physical mapping coefficient by using single-point sensor data, and performing integral calculation on the real-time mass flow in combination with a three-dimensional optical flow method. According to the invention, the complex density gradient in the gas can be effectively reduced, and high-precision three-dimensional visual display and quantitative flow monitoring of gas leakage are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image processing, and particularly relates to an industrial gas leakage detection method based on infrared thermal imaging. BACKGROUND

[0002] Industrial gas leakage is a major safety hazard in energy production, and real-time monitoring and quantitative evaluation of it is crucial. Detection technology based on infrared thermal imaging is widely used in the discovery and positioning of industrial gas leakage because of its non-contact and visual characteristics, which can capture invisible gas clouds to the human eye. In order to further evaluate the severity of the leakage, it is usually necessary to reconstruct the three-dimensional spatial distribution of the gas according to the two-dimensional infrared image, and then calculate its mass flow.

[0003] In existing three-dimensional reconstruction techniques, using multiple cameras for synchronous shooting and adopting the view frustum intersection method is the main means to construct a three-dimensional model. This algorithm extracts the gas binary mask under each viewing angle, calculates the common intersection of all viewing frustums in three-dimensional space, and generates a three-dimensional basic geometric voxel grid of the gas. However, the model generated based on the view frustum intersection method is usually considered as a solid block with homogeneous interior, that is, it is assumed that the density distribution inside the three-dimensional grid is uniform and consistent, thereby ignoring the physical properties of the gas as a fluid in a free diffusion state. In fact, the gas cloud usually presents a complex density gradient inside. Simply considering the gas as a homogeneous body cannot reflect its true internal density decay law, resulting in significant errors in subsequent mass integration and flow estimation, and it is difficult to meet the needs of high-precision quantitative monitoring. SUMMARY

[0004] In view of the problem that the frustum intersection method will have significant errors in quality integration and flow estimation, the application provides an industrial gas leakage detection method based on infrared thermal imaging, comprising: acquiring infrared video streams synchronously collected by a plurality of calibrated infrared thermal imagers in a monitoring area, and performing background difference on each video stream to extract a binary mask of the gas; using the internal and external parameter matrices of each camera, constructing a basic geometric voxel grid in three-dimensional space through the frustum intersection method; projecting the normalized radiation feature back to the basic geometric voxel grid, and multiplying it by the corresponding adaptive spatial weight to obtain a relative density field representing gas leakage; the normalized radiation feature is positively correlated with the radiation intensity of the corresponding pixel of the binary mask, and is negatively correlated with the length of the line-of-sight path passing through the basic geometric voxel grid; the normalized radiation feature also includes a correction term, which is positively correlated with the coefficient of variation of the radiation intensity of all gas pixels at the corresponding viewing angle and the difference between the pixel radiation intensity and the mean value; the adaptive spatial weight is positively correlated with the relative depth of the voxel to the nearest surface of the grid, and is negatively correlated with the compactness degree factor; the compactness degree factor is used to control the decay rate of the adaptive spatial weight along the surface of the voxel grid to the inside.

[0005] The application overcomes the defects of lack of depth information in traditional two-dimensional infrared detection and difficulty in processing semi-transparent medium in traditional three-dimensional reconstruction. The basic grid is constructed by using the frustum intersection method, and the back projection of the normalized radiation feature and the adaptive spatial weight are combined, so that the three-dimensional relative density field of the gas can be recovered from the two-dimensional images of multiple angles. Especially, the radiation feature correction which is negatively correlated with the length of the line-of-sight path and positively correlated with the coefficient of variation, and the weight distribution based on depth and compactness degree, effectively solve the ambiguity problem of internal density distribution reconstruction of the gas, and improve the accuracy and robustness of the three-dimensional spatial distribution detection of the leaked gas.

[0006] Further, the calculation method of the adaptive spatial weight is specifically: ; Among them represents the voxel corresponding adaptive spatial weight; represents the relative depth of the nearest surface of the voxel ; represents the exponential function; represents the compactness degree factor.

[0007] The application provides a specific adaptive spatial weight calculation model, which enables the reconstruction algorithm to dynamically adjust the density contribution value according to the distance of the gas cloud to the surface, retains the edge contour information, and to a certain extent, simulates the density decay or accumulation law of the gas from the outside to the inside, so as to obtain a more physical fact-based internal structure estimation of the gas.

[0008] Further, the calculation method of the shape compactness factor is specifically: ; Wherein represents the shape compactness factor; represents the total volume of the basic geometric voxel grid; represents the outer surface area of the basic geometric voxel grid.

[0009] Further, the calculation method of the normalized radiation feature is specifically: ; Wherein represents the normalized radiation feature; represents the radiation intensity of the corresponding pixel; represents the line-of-sight path length; represents the hyperbolic tangent function; represents the mean value of the radiation intensity of all gas pixels; represents the standard deviation of the radiation intensity of all gas pixels; represents the coefficient of variation, which is the ratio of the standard deviation to the mean value, used to characterize the dispersion degree of the gas concentration.

[0010] By introducing the hyperbolic tangent function and the radiation intensity coefficient of variation as correction terms, this method not only eliminates the interference of the line-of-sight path length on the integrated radiation intensity, but also uses the intensity fluctuation caused by gas turbulence to enhance the feature saliency of the dynamic gas region, effectively suppresses the background thermal noise, and makes the extracted features more truly reflect the relative density of the gas itself.

[0011] Further, the background difference includes modeling the background using a mixture Gaussian model; when the difference between the current frame pixel intensity of the infrared thermal imager and the background model exceeds a set threshold, the corresponding pixel is marked as foreground and included in the binary mask.

[0012] The present application uses a mixture Gaussian model for background modeling, which can adapt to the complex and variable thermal environment in industrial sites compared to the simple inter-frame difference method. It can more accurately separate the dynamic gas leakage region from the complex background and generate a high-quality binary mask, providing accurate input data for subsequent three-dimensional geometric reconstruction and reducing the false detection rate.

[0013] Further, the measured concentration value of a single-point gas concentration sensor arranged in the monitoring space is collected; the relative density value corresponding to the position of the single-point gas concentration sensor in the relative density field is extracted; and a physical calibration coefficient is calculated based on the ratio of the measured concentration value to the relative density value; the single-point gas concentration sensor is a laser backscattering point detector.

[0014] Further, the three-dimensional velocity field of the basic geometric voxel mesh is calculated; a virtual cross section is defined downstream of the leakage source; the relative density, velocity component, cross section micro-element area, and physical calibration coefficient of each voxel passing through the virtual cross section are integrated to obtain the real-time mass flow rate of the gas leakage.

[0015] This invention utilizes a three-dimensional optical flow method to directly estimate the velocity field of a gas in voxel space. Compared to the two-dimensional optical flow method, it can capture the diffusion, ascent, and turbulent motion of the gas in three-dimensional space. This not only reconstructs the true trajectory of the gas but also provides an accurate velocity vector input for calculating mass flow rate, thus improving the accuracy of flow rate estimation.

[0016] Furthermore, the method for calculating the three-dimensional velocity field is the three-dimensional optical flow method.

[0017] Furthermore, the infrared thermal imager is a long-wave infrared thermal imager.

[0018] Furthermore, it also includes using a ray projection volume rendering algorithm to visualize the relative density field, and mapping different relative density values ​​to preset colors and transparency.

[0019] The technical effects of this invention are as follows: This invention constructs a basic geometric voxel mesh and utilizes a combination of normalized radiation feature backprojection and adaptive spatial weighting to accurately reconstruct the relative density field of a gas. Combined with single-point sensor calibration, it achieves the conversion from qualitative imaging to quantitative concentration. Furthermore, it calculates the leakage mass flow rate using a three-dimensional optical flow method, solving the problem of traditional infrared detection's difficulty in obtaining three-dimensional spatial distribution and accurate quantification. Attached Figure Description

[0020] Figure 1 This is a schematic flowchart illustrating an embodiment of the industrial gas leak detection method based on infrared thermal imaging in this invention. Figure 2 This is a schematic illustration of an infrared thermal image of a gas in an industrial environment; Figure 3 This is a schematic diagram illustrating a three-dimensional scatter plot of the basic geometric voxel mesh constructed based on the view frustum intersection method in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the geometric penetration depth distribution in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the normalized radiation characteristic distribution in an embodiment of the present invention; Figure 6 This is a schematic illustration of a three-dimensional relative density field scatter plot showing regions with a density greater than 0.3 in an embodiment of the present invention. Figure 7 This is a schematic diagram illustrating the density distribution at a virtual cross-section in an embodiment of the present invention; Figure 8 This is a schematic diagram illustrating the velocity field distribution at a virtual cross-section in an embodiment of the present invention; Figure 9 This is a schematic diagram illustrating the visualization effect of a three-dimensional density field rendering of gas leakage based on a ray casting algorithm in an embodiment of the present invention. Detailed Implementation

[0021] 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, not all, of the embodiments of the present invention. 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.

[0022] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0023] Example of an industrial gas leak detection method based on infrared thermal imaging: like Figure 1 As shown, the industrial gas leak detection method based on infrared thermal imaging of the present invention includes: S1. Acquire synchronous video streams using a calibrated infrared thermal imager array and perform background subtraction to generate a binarized gas mask.

[0024] In this embodiment, the hardware acquisition terminal includes components configured at the perimeter of the monitoring area. The platform has been calibrated with a long-wave infrared thermal imager (preferably: FLIR A700, resolution...). Frame rate The system includes a single-point gas concentration sensor (preferably a laser backscattering point detector). To obtain the initial geometric contour of the gas cloud, the four infrared thermal imagers are first controlled to simultaneously acquire video streams of gas leaks in the monitoring area. Figure 2 As shown, it is a raw infrared image of a gas cloud captured by an infrared thermal imager.

[0025] After receiving the raw infrared image, the data processing workstation performs non-uniformity correction (NUC) on it to eliminate fixed-pattern noise and uses a Gaussian smoothing filter to remove random thermal noise.

[0026] Subsequently, a Gaussian Mixture Model (GMM) is used for background modeling, and the current frame is compared with the background model using a difference operation. The brightness difference threshold is set to... (Grayscale value) When the grayscale change of a pixel exceeds this threshold, it is marked as a foreground gas region, thereby extracting the gas binarization mask from each viewpoint.

[0027] Based on the intrinsic and extrinsic parameter matrices of each camera, the system employs the view frustum intersection method to calculate the common intersection of all view masks in 3D space. This embodiment discretizes the 3D space into a matrix with side lengths of... The voxels are used to obtain a basic geometric voxel mesh containing only geometric occupancy information. .at this time It is a solid block with a homogeneous interior, and does not contain density distribution information.

[0028] like Figure 3 As shown, the basic geometric voxel mesh is generated using the view frustum intersection method. The scattered points in the figure represent voxel points in three-dimensional space that are identified as gas. At this point, the mesh only reflects the geometric shape of the gas and does not yet contain density gradient information, thus exhibiting a relatively uniform distribution.

[0029] S2. Calculate the line-of-sight penetration depth and coefficient of variation to obtain normalized radiation characteristics based on statistical moments.

[0030] To address the errors caused by treating gas as a homogeneous body in traditional methods, this embodiment adaptively modulates the nonlinear mapping relationship between two-dimensional brightness and three-dimensional density using statistical features. Specifically, for the... The image from each perspective is first processed using a ray casting algorithm, where rays emitted from the camera's optical center pass through a basic geometric voxel grid. Calculate the physical length of the line segment where the ray cuts off inside the grid to obtain the geometric penetration depth, which is also the line-of-sight path length.

[0031] Then, the infrared radiation intensity of all pixels marked as gas is calculated from the current viewpoint. Calculate its grayscale mean. and standard deviation The dispersion of gas concentration is obtained, specifically: ; here This is a dimensionless statistic. A large value indicates significant concentration differences within the gas cloud, such as during the initial stage of high-pressure injection; a small value indicates a more uniform gas distribution, such as during the final stage of diffusion. The path-dependent normalized radiation characteristics are then calculated using the following method. : ; The above formula is used to address the problem that the linear density assumption cannot reflect the complex gradients inside the gas, where Indicates the radiation intensity of a pixel. Indicates geometric penetration depth. This represents the hyperbolic tangent function, used to limit the magnitude of nonlinear corrections. Note that when... In this case, the gas concentration distribution can be considered uniform, and the nonlinear correction term does not take effect. .

[0032] Part Two above For nonlinear correction terms, using S-shaped property and coefficient of variation of a function As a gain, when the image contrast is high and the pixel brightness is much higher than the mean, the correction term increases significantly, thus giving higher density weights to the bright areas; conversely, when... When the value approaches 0, the formula automatically degenerates into a linear model, thus achieving parameter adaptation.

[0033] like Figure 4 As shown, the calculated geometric penetration depth is illustrated. Its value reflects the length of the line of sight within the gas envelope; Figure 5 This demonstrates the normalized radiation characteristics after statistical moment correction. As can be seen, compared with simple depth information, normalized features more clearly enhance the regions where the radiation intensity inside the gas changes significantly, providing richer texture information for subsequent density field reconstruction.

[0034] S3. Calculate the morphological compactness factor and combine it with the relative depth parameter to calculate the adaptive spatial weight, thereby obtaining the relative density field.

[0035] Furthermore, the macroscopic three-dimensional shape of the gas is used to constrain the diffusion model of the microscopic density, thereby reconstructing the true density gradient. First, a statistical analysis of the basic geometric voxel mesh is performed. The total number of voxels within is used to calculate the total volume. And its outer surface area is calculated using an isosurface extraction algorithm. Then, the morphological compactness factor is calculated, specifically: ; In this embodiment, if the morphological compactness factor A value close to 1 indicates that the gas exhibits a high-pressure jet-like shape resembling a sphere; if the morphology compactness factor is close to 1... A value significantly less than 1, such as close to 0.3, indicates that the gas exhibits a loose, irregular plume-like appearance.

[0036] For each voxel within the grid Calculate its normalized Euclidean distance to the nearest surface. The adaptive spatial weights are calculated as follows: ; The above formula is used to simulate the density decay law under different fluid states; where Indicates the relative depth of a voxel from the surface (0 for the surface, 1 for the center); This represents an exponential function.

[0037] In the index term The rate of decay was controlled. When the gas shape is compacted, i.e. At that time, the exponential term approaches 0, and the weight increases with depth. The change is approximately linear, consistent with the characteristics of a high-pressure gas mass; when the gas shape is loose, that is... When significantly less than 1, As the weight of the edge region increases, it decays exponentially, resulting in a thin edge that reflects a free-diffusion state.

[0038] Finally, the system will use the features from step S2 Back-projection into three-dimensional space, and utilizing The relative density value of each voxel is obtained by weighting. Specifically, these include: ; in Represents the basic geometric voxel mesh The first in Individual factors; This represents the calculated adaptive spatial weights; This indicates the total number of infrared thermal imagers; An index representing the viewpoint; This represents a projection mapping function used to project voxels in three-dimensional space. Projected to the The coordinate mapping process on the two-dimensional imaging plane of a camera; Then it means in the first In the image from each viewpoint, the normalized radiative eigenvalues ​​at the corresponding projection coordinates are calculated based on the method in step S2.

[0039] like Figure 6 As shown, a scatter plot of the relative density field in three-dimensional space at a specified density is displayed, which intuitively constructs the three-dimensional morphology of the gas cloud.

[0040] S4. Collect data from a single physical sensor, calibrate the field mapping coefficients, and calculate the real-time mass flow rate by combining the optical flow velocity field integral.

[0041] First, read the spatial location. Measured density value of a single-point gas concentration sensor at a given location Simultaneously, the numerical values ​​at the corresponding locations are extracted from the reconstructed relative density field. Next, the physical mapping coefficients are calculated. This converts the relative density field into an absolute physical density field, and further utilizes the three-dimensional optical flow method to calculate the voxel velocity field. Finally, a virtual cross-section was selected downstream of the leak source. Calculate the real-time mass flow rate through this cross section. : ; in The area of ​​the cross-sectional element is represented as follows: , This represents the cross-sectional normal vector. Through the above integral calculation based on the voxel-level density field, the difference in contribution to the flow rate between the high-concentration core region and the low-concentration edge region can be effectively distinguished, avoiding the errors of traditional volume estimation methods.

[0042] like Figure 7 and Figure 8 As shown, to facilitate flow integration, the system extracted data from a virtual cross-section (Y=50mm). Figure 7 The density distribution on this cross section is shown. Figure 8 The velocity field vector on this cross section is displayed. By performing a dot product and integration of these two fields on the cross section, high-precision mass flow rate calculation can be achieved.

[0043] S5. Render three-dimensional voxel data to visualize the internal concentration gradient distribution of the gas and output quantitative flow monitoring results.

[0044] Based on the final calculated physical density field data, a ray-projection rendering technique is used for visualization. The system uses a preset color transfer function to map low-density areas to highly transparent blue and high-density areas to opaque red. On the user interface, operators can not only observe the three-dimensional cloud of gas leakage and its high-concentration core from any angle, but also read the mass flow rate calculated in step S4 in real time. (unit: This provides intuitive and quantitative data support for industrial safety handling.

[0045] The final output effect is as follows Figure 9 As shown, the calculated three-dimensional density field is transformed into a semi-transparent color cloud map through ray projection volume rendering technology.

Claims

1. An industrial gas leak detection method based on infrared thermal imaging, characterized in that, The method includes: acquiring infrared video streams synchronously collected by multiple calibrated infrared thermal imagers within the monitoring area, and performing background subtraction on each video stream to extract a binary mask of the gas; using the intrinsic and extrinsic parameter matrices of each camera, constructing a basic geometric voxel mesh in three-dimensional space through the frustum intersection method; The normalized radiation features are back-projected onto the basic geometric voxel grid and multiplied with the corresponding adaptive spatial weights to obtain the relative density field characterizing the gas leakage. The normalized radiation feature is positively correlated with the radiation intensity of the corresponding pixel in the binarized mask and negatively correlated with the length of the line-of-sight path through the basic geometric voxel grid. The normalized radiation feature also includes a correction term, which is positively correlated with the coefficient of variation of the radiation intensity of all gas pixels under the corresponding viewpoint and the difference between the pixel radiation intensity and the mean. The adaptive spatial weights are positively correlated with the relative depth of the voxel to the nearest surface of the mesh and negatively correlated with the morphological compactness factor; the morphological compactness factor is used to control the decay rate of the adaptive spatial weights along the surface of the voxel mesh inward.

2. The industrial gas leak detection method based on infrared thermal imaging according to claim 1, characterized in that, The specific method for calculating the adaptive spatial weights is as follows: ； in Voxel representation Corresponding adaptive spatial weights; Voxel representation The relative depth of the nearest surface; Represents an exponential function; This represents the morphological compactness factor.

3. The industrial gas leak detection method based on infrared thermal imaging according to claim 2, characterized in that, The specific method for calculating the morphological compactness factor is as follows: ； in This represents the morphological compactness factor; This represents the total volume of the basic geometric voxel mesh; This represents the outer surface area of ​​the basic geometric element mesh.

4. The industrial gas leak detection method based on infrared thermal imaging according to claim 1, characterized in that, The specific method for calculating the normalized radiation characteristics is as follows: ； in This represents the normalized radiation characteristic; This indicates the radiation intensity of the corresponding pixel; Indicates the length of the line-of-sight path; Represents the hyperbolic tangent function; This represents the average radiant intensity of all gas pixels. This represents the standard deviation of the radiative intensity of all gas pixels; The coefficient of variation is the ratio of the standard deviation to the mean, used to characterize the dispersion of gas concentration.

5. The industrial gas leak detection method based on infrared thermal imaging according to claim 4, characterized in that, The background subtraction includes modeling the background using a Gaussian mixture model; When the difference between the current frame pixel intensity of the infrared thermal imager and the background model exceeds a set threshold, the corresponding pixel is marked as the foreground and included in the binarization mask.

6. The industrial gas leak detection method based on infrared thermal imaging according to claim 1, characterized in that, Collect measured concentration values ​​from single-point gas concentration sensors deployed within the monitoring space; Extract the relative density value in the relative density field corresponding to the location of the single-point gas concentration sensor; The physical calibration coefficient is calculated based on the ratio of the measured concentration value to the relative density value. The single-point gas concentration sensor is a laser backscattering point detector.

7. The industrial gas leak detection method based on infrared thermal imaging according to claim 6, characterized in that, Calculate the three-dimensional velocity field of the basic geometric voxel mesh; Define a virtual cross section downstream of the leakage source; The real-time mass flow rate of the gas leak is obtained by integrating the relative density, velocity component, cross-sectional micro-element area, and physical calibration coefficient of each voxel passing through the virtual cross-section.

8. The industrial gas leak detection method based on infrared thermal imaging according to claim 7, characterized in that, The method for calculating the three-dimensional velocity field is the three-dimensional optical flow method.

9. The industrial gas leak detection method based on infrared thermal imaging according to claim 1, characterized in that, The infrared thermal imager is a long-wave infrared thermal imager.

10. The industrial gas leak detection method based on infrared thermal imaging according to claim 1, characterized in that, It also includes using a ray projection volume rendering algorithm to visualize the relative density field, and mapping different relative density values ​​to preset colors and transparency.

Images