A distance parameter identification method for infrared temperature measurement equipment of power equipment

Abstract

The present invention seeks to protect a method for identifying distance parameters of infrared temperature measurement equipment for electric equipment, which can accurately identify the shooting distance between a handheld infrared thermal imager and the electric equipment during infrared temperature measurement, including preprocessing the infrared image, marking the electric equipment area in the infrared image, automatically identifying the equipment type based on the SSD algorithm, obtaining the minimum adjacent rectangular frame of the equipment based on image processing, identifying the pixel width of the target equipment, and realizing automatic identification of the equipment distance in the image based on an improved monocular distance measurement algorithm. The technical method of the present invention can automatically identify the equipment distance using the pixel width of the electric equipment in the infrared image, solving the problem of difficulty in distance identification caused by changes in shooting angles and incomplete equipment shooting in infrared imaging of electric equipment.

Description

Technical Field

[0001] The invention relates to the technical field of infrared distance measurement, and in particular to a distance parameter identification method for infrared temperature measurement equipment of electric power equipment. Background Art

[0002] Substation power equipment will inevitably fail during long-term operation. When a failure occurs, the most obvious feature is abnormal temperature. Infrared thermal imagers can detect the operating temperature of equipment without power outage or contact with the equipment, and monitor the operating status of equipment. They are suitable for real-time monitoring of the temperature of high-voltage live parts. Therefore, in recent years, infrared thermal imagers have been widely used in power equipment status monitoring.

[0003] However, despite the increasing application of thermal imaging in power equipment monitoring, there are relatively few studies on the potential defects and limitations of thermal imaging. There are many factors that affect infrared imaging, among which the shooting distance is one of the main factors affecting the effect of infrared imaging. Accurate measurement of the shooting distance is one of the main methods to improve the accuracy of infrared imaging detection equipment faults.

[0004] In view of the problems existing in the current infrared live detection, the present invention mainly relates to a method for identifying distance parameters of infrared temperature measurement equipment for electric equipment. In view of the problem that the shooting distance setting of the current thermal imager products suitable for electrical detection is inaccurate during use, resulting in abnormal infrared image temperature measurement data, an improved algorithm for monocular ranging of infrared imaging of electric equipment based on target pixel width recognition is proposed, which realizes automatic identification of equipment distance by using the pixel width of the electric equipment in the infrared image, and solves the problem of difficulty in distance recognition caused by shooting angle changes and incomplete equipment shooting in infrared imaging of electric equipment. The results show that the improved algorithm can meet the monocular ranging requirements of infrared images of electric equipment.

[0005] The present invention mainly utilizes the fact that the pixel width of the target device is not affected by the shooting angle when the infrared image is taken, and improves the monocular distance measurement algorithm accordingly. Since there is an angle between the camera and the horizontal and vertical directions during shooting, the imaging coordinate system and the pixel coordinate system are not parallel. Therefore, the target plane is approximately perpendicular to the optical axis, and the pixel width is used to solve the distance of the device in combination with the relationship between the imaging coordinate system and the world coordinate system. Summary of the invention

[0006] The present invention aims to solve the above problems of the prior art. A distance parameter identification method for infrared temperature measurement equipment of power equipment is proposed. The technical solution of the present invention is as follows:

[0007] A method for identifying distance parameters of infrared temperature measuring equipment for electric power equipment comprises the following steps:

[0008] Input the infrared image of the power equipment, and perform preprocessing on the infrared image of the power equipment to remove redundant information and enhance the data set;

[0009] Labelme software was used to mark the power equipment area in the infrared image;

[0010] Target detection based on the SSD (Single Shot Multi-Frame Detector) algorithm to obtain automatic device type identification;

[0011] The minimum adjacent rectangular frame of the power equipment is obtained through the image processing algorithm in OpenCV, the vertex coordinates of the minimum adjacent rectangular frame are output, and the pixel width of the target device for identification is calculated;

[0012] Based on the improved monocular ranging algorithm, the automatic recognition of the device distance in the image is realized. It is proposed to approximately make the target plane perpendicular to the optical axis, and there is only a relative position change on the Z axis between the world coordinate system and the camera coordinate system. The distance of the device can be solved by using the relationship between the imaging coordinate system and the world coordinate system.

[0013] Furthermore, the preprocessing of the infrared image of the power equipment including removing redundant information and enhancing the data set specifically includes:

[0014] The OS module in Python is used to write a batch processing program to remove redundant information including text and symbols from the original thermal image, and the data set is enhanced by re-adjusting the temperature range setting of the imaging to complete the preprocessing of the sample image.

[0015] Furthermore, the labelme software is used to mark the power equipment area in the infrared image, specifically including:

[0016] The Labelme software annotates the device type corresponding to the image, which is used to analyze the image features of different devices, evaluate the designed deep learning model through device classification problems, and train and test the target recognition model by annotating the device location. The deep learning model is a convolutional neural network for image recognition, which consists of several convolutional layers, activation layers, pooling layers and fully connected layers. The convolutional layer is used for dimensionality reduction and feature extraction, the activation layer is used to simulate arbitrary functions and enhance the representation ability of the network, the pooling layer is used to reduce the amount of calculation and improve the generalization ability, and the fully connected layer is used for feature classification and improve the output quality.

[0017] Furthermore, the use of the SSD algorithm for target detection to obtain automatic device type identification specifically includes:

[0018] The SSD target test model structure is divided into two parts: the first part is the target basic feature extraction network; the second part is the multi-scale feature detection network, which is used to implement multi-scale feature extraction on the feature layer extracted in the first part; the SSD target detection algorithm process consists of two steps: one is target positioning, that is, giving the position of the target object in the picture; the other is classification detection, which gives the probability that each candidate belongs to a specific category.

[0019] Furthermore, the target test model specifically includes:

[0020] After the image is input, it passes through the forward network propagation, and all the candidate boxes in the region will generate category probability prediction values ​​and position offset prediction values. According to the set threshold, the boxes with probability prediction values ​​lower than the threshold are deleted, that is, it is considered that there are no objects in these boxes. Then, redundant boxes are removed by non-maximum suppression in units of categories to obtain the detection box that matches the target to be detected.

[0021] Furthermore, the method of obtaining the minimum adjacent rectangular frame of the power equipment through the image processing algorithm in OpenCV, outputting the vertex coordinates of the minimum adjacent rectangular frame, and calculating the pixel width of the target device for identification specifically includes:

[0022] First, the SSD target detection frame is obtained, the coordinates of the device box vertices are read and the image outside the box is removed. Then, the device area in the image is extracted based on the mask technology. The noise interference including the wired power around the device is removed by performing an erosion and dilation operation on the image. Finally, the image is contour fitted to identify the minimum adjacent rectangle of the device and the coordinates of the four vertices of the rectangle are output. The pixel width is equal to the short side length of the minimum adjacent rectangle, so that the distance of the device can be calculated.

[0023] Furthermore, the mask technology-based extraction of the device area portion in the image specifically includes: first, binarizing the image so that the image contains only black and white colors, then filtering and denoising the image to obtain a clearer device outline, and finally, using an image segmentation method to segment the image to obtain the image mask area and extract the device area portion in the image.

[0024] The pixel width is equal to the short side length of the smallest adjacent rectangular frame, so the distance of the device can be calculated. The target pixel width calculation formula is:

[0025]

[0026] Among them, dx and dy are the parameters obtained by camera calibration, and u, u0, v, and v0 are the parameters identified in the infrared image.

[0027] Furthermore, the automatic recognition of the device distance in the image based on the improved monocular ranging algorithm specifically includes: based on the similar triangle principle in the geometric ranging algorithm, the final distance between the camera and the target is calculated as follows:

[0028]

[0029] Where D is the actual width of the device, w is the target pixel width, f, dx, dy are the parameters obtained by camera calibration, and u, u0, v, v0 are the parameters identified in the infrared image.

[0030] The advantages and beneficial effects of the present invention are as follows:

[0031] The innovation of the present invention is mainly achieved by steps 5 to 7 of the claims, which adopt an improved monocular ranging algorithm and utilize the relationship between the world coordinate system, the imaging coordinate system and the pixel coordinate system to realize the automatic recognition of the device distance in the image. Steps 1 to 4 pave the way for subsequent steps, mainly to output the vertex coordinates of the minimum adjacent rectangular frame and provide data support for the subsequent implementation of the improved monocular ranging algorithm. Since the thermal imager products currently suitable for electrical inspection need to input the shooting distance when in use, the inspectors can usually only set the average distance by visual inspection, resulting in inaccurate infrared image data. Based on this, an improved monocular ranging algorithm is proposed to realize automatic recognition of distance. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic diagram of the infrared image preprocessing process provided by the preferred embodiment of the present invention;

[0033] Figure 2 It is the structure diagram of the SSD model;

[0034] Figure 3 It is the pixel width process of identifying the device;

[0035] Figure 4 It is the correspondence between actual width and pixel width;

[0036] Figure 5 It is a flow chart of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.

[0038] The technical solution of the present invention to solve the above technical problems is:

[0039] like Figure 5As shown, a method for identifying distance parameters of infrared temperature measurement equipment for power equipment can accurately identify the shooting distance between a handheld infrared thermal imager and the power equipment during infrared temperature measurement, including preprocessing the infrared image, marking the power equipment area in the infrared image, automatically identifying the equipment type based on the SSD algorithm, obtaining the minimum adjacent rectangular frame of the equipment based on image processing, identifying the pixel width of the target equipment, and realizing automatic identification of the equipment distance in the image based on an improved monocular ranging algorithm.

[0040] 1. Power equipment target detection and type recognition based on SSD algorithm

[0041] The present invention preprocesses the infrared image dataset based on the construction of the infrared image dataset of power equipment. The redundant information such as text and symbols of the original thermal image is removed by writing a batch processing program, and the dataset is enhanced by re-adjusting the temperature range setting of the imaging to complete the preprocessing of the sample image. The processing process is as follows: Figure 1 As shown in the figure. After completing the preprocessing of the infrared image, the Labelme software is used to mark the power equipment area in the infrared image. The Labelme software marks the device type corresponding to the image, which is used to analyze the image features of different devices, and evaluates the designed deep learning model through the device classification problem, and uses the marked device location to train and test the target recognition model.

[0042] The single-stage target detection model represented by the SSD (Single Shot Multi-Frame Detector) algorithm achieves a good compromise between computational accuracy, computational speed, and computational complexity. The SSD target test model structure is divided into two parts: the first part is the target basic feature extraction network; the second part is the multi-scale feature detection network, which can implement multi-scale feature extraction on the feature layer extracted in the previous part. The SSD target detection algorithm process consists of two steps: one is target positioning, which gives the position of the target object in the image; the other is classification detection, which gives the probability that each candidate belongs to a specific category. Its architecture diagram is shown below. Figure 2 shown.

[0043] 2. Target device pixel width identification based on image processing algorithm

[0044] For infrared images, the present invention uses the image processing algorithm in OpenCV to identify the pixel width of the target device. First, the SSD target detection frame is obtained, the device box coordinates are read and the image outside the box is removed. Then, the device area in the image is extracted based on the mask technology. The image is opened by corrosion and expansion operations to remove noise interference such as wired electricity around the device. Finally, the image is contour fitted to identify the minimum adjacent rectangle of the device, and the coordinates of the four vertices of the rectangle are output. It can be found that the pixel width is equal to the short side length of the minimum adjacent rectangle, so the distance of the device can be calculated. The specific implementation steps are as follows: Figure 3 shown.

[0045] 3. Automatic identification of device distance based on improved monocular ranging algorithm

[0046] In view of the problem that the optical axis is not parallel to the ground due to the angle between the lens and the horizontal plane when shooting existing infrared images, the present invention proposes that the target plane can be approximately perpendicular to the optical axis, and there is only a relative position change on the Z axis between the world coordinate system and the camera coordinate system. In view of the problem that the pixel width of the target detection frame is not the pixel width corresponding to the actual width of the device due to the tilted shooting angle of the device, it is proposed to solve the transformation between the imaging coordinate system and the pixel coordinate system by the target pixel width. It can be found that even if the angle of the device is tilted, since the overall appearance of the power equipment is cylindrical, no matter from which angle it is photographed, the maximum width of the device will not change; even if the device is not fully photographed, its width can still be reflected. The corresponding relationship between the actual width of the device and the pixel width in the present invention is as follows: Figure 4 shown.

[0047] Based on the similar triangle principle in the geometric ranging algorithm, the final distance between the camera and the target is calculated as follows:

[0048]

[0049] Where D is the actual width of the device, w is the target pixel width, f, dx, dy are the parameters obtained by camera calibration, and u, u0, v, v0 are the parameters identified in the infrared image.

[0050] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0051] The above embodiments should be understood to be only used to illustrate the present invention and not to limit the protection scope of the present invention. After reading the contents of the present invention, technicians can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A method for identifying distance parameters of infrared temperature measuring equipment for electric power equipment, characterized in that: The following steps are involved: Inputting the infrared image of the power equipment, and performing preprocessing on the infrared image of the power equipment by removing redundant information and enhancing the data set; Labelme software was used to mark the power equipment area in the infrared image; Target detection based on the SSD single-shot multi-frame detector algorithm obtains automatic device type identification; The minimum adjacent rectangular frame of the power equipment is obtained through the image processing algorithm in OpenCV, the vertex coordinates of the minimum adjacent rectangular frame are output, and the pixel width of the target device for identification is calculated; Automatic recognition of the device distance in the image is achieved based on the improved monocular ranging algorithm. The target plane is set perpendicular to the optical axis. There is only a relative position change on the Z axis between the world coordinate system and the camera coordinate system. The distance of the device is solved by using the relationship between the imaging coordinate system and the world coordinate system. The method of obtaining the minimum adjacent rectangular frame of the power equipment by the image processing algorithm in OpenCV, outputting the vertex coordinates of the minimum adjacent rectangular frame, and calculating the pixel width of the target device for identification specifically includes: First, get the SSD target detection box, read the coordinates of the device box vertices and remove the image outside the box. Then, extract the device area in the image based on the mask technology. Remove the noise interference including the wired power around the device by performing the erosion and expansion operation on the image. Finally, perform contour fitting on the image to identify the minimum adjacent rectangle of the device and output the coordinates of the four vertices of the rectangle. The pixel width is equal to the short side length of the minimum adjacent rectangle, so that the distance of the device can be calculated. The method of extracting the device area portion in the image based on the mask technology specifically includes: First, the image is binarized so that it contains only black and white colors. Then, the image is filtered and denoised to obtain a clearer device outline. Finally, the image is segmented using an image segmentation method to obtain the image mask area and extract the device area in the image. The pixel width is equal to the short side length of the smallest adjacent rectangular frame, so the distance of the device can be calculated. The target pixel width calculation formula is: Among them, dx and dy are the parameters obtained by camera calibration, and u, u0, v, and v0 are the parameters identified in the infrared image; The improved monocular ranging algorithm is used to realize automatic recognition of the device distance in the image, specifically including: based on the similar triangle principle in the geometric ranging algorithm, the final distance between the camera and the target is calculated as follows: Where D is the actual width of the device, w is the target pixel width, f, dx, dy are the parameters obtained by camera calibration, and u, u0, v, v0 are the parameters identified in the infrared image.

2. A distance parameter identification method for infrared temperature measuring equipment of electric power equipment according to claim 1, characterized in that: The preprocessing of the infrared image of the power equipment, including removing redundant information and enhancing the data set, specifically includes: A batch processing program was written using the OS module in Python to remove redundant information including text and symbols from the original thermal image. The data set was enhanced by readjusting the temperature range settings for imaging to complete the preprocessing of the sample image.

3. A distance parameter identification method for infrared temperature measuring equipment of electric power equipment according to claim 1, characterized in that: The labelme software is used to mark the power equipment area in the infrared image, specifically including: The Labelme software annotates the device type corresponding to the image, which is used to analyze the image features of different devices, and evaluates the designed deep learning model through device classification problems. The device location is annotated to train and test the target recognition model. The deep learning model is a convolutional neural network for image recognition, which consists of several convolutional layers, activation layers, pooling layers and fully connected layers. The convolutional layer is used for dimensionality reduction and feature extraction, the activation layer is used to simulate arbitrary functions and enhance the representation ability of the network, the pooling layer is used to reduce the amount of calculation and improve the generalization ability, and the fully connected layer is used for feature classification and improve the output quality.

4. A distance parameter identification method for infrared temperature measuring equipment of electric power equipment according to claim 1, characterized in that: The use of the SSD algorithm for target detection to obtain automatic device type identification specifically includes: The SSD target test model structure is divided into two parts: the first part is the target basic feature extraction network; the second part is the multi-scale feature detection network, which is used to implement multi-scale feature extraction on the feature layer extracted in the first part; the SSD target detection algorithm process consists of two steps: one is target positioning, that is, giving the position of the target object in the picture; the other is classification detection, which gives the probability that each candidate belongs to a specific category.

5. A method for identifying distance parameters of infrared temperature measuring equipment for electric power equipment according to claim 4, characterized in that: The target test model specifically includes: After the picture is input, it passes through the forward network propagation, and all region candidate boxes will generate category probability prediction values ​​and position offset prediction values. According to the set threshold, the boxes with probability prediction values ​​lower than the threshold are deleted, that is, it is considered that there are no targets in these boxes. Then, redundant boxes are removed by non-maximum suppression in units of categories to obtain the detection box that matches the target to be detected.

Images