Navigation method of inspection robot applied to natural gas power plant

Abstract

The present invention relates to a navigation method for an inspection robot applied to a natural gas power plant. The present invention solves the problems of the prior art. The key points of its technical solution are as follows: Step 1: a detection point is set at a location where feedback is required in the natural gas power plant, and at least one feedback content exists at the detection point. A host computer establishes an electronic map according to the target natural gas power plant; feasible roads and detection point locations are set in the electronic map of the natural gas power plant; Step 2: the host computer synchronizes the electronic map to the inspection robot; Step 3: the host computer sets the inspection route and target detection points and inspection items as required, and then sends them to the inspection robot; Step 4: the inspection robot performs inspection according to the sent inspection route, and the control unit determines the current position of the inspection robot according to the feedback signal of the feedback device and the feedback content of the detection point, and performs real-time detection according to the sensor group, and performs inspection according to the real-time detection situation.

Description

Technical Field

[0001] The invention belongs to an intelligent inspection device and relates to an inspection robot navigation method applied to a natural gas power plant. Background Art

[0002] Robots have basic features such as perception, decision-making, and execution. They can assist or even replace humans to complete dangerous, heavy, and complex work, improve work efficiency and quality, serve human life, and expand or extend the scope of human activities and capabilities. Therefore, some complicated work, such as intelligent inspection of the main plant of a power plant, can be handed over to robots for completion. Nowadays, robots are used in many inspection scenarios. For example, on June 4, 2021, Shenzhen Zhonghe Puda Measurement Technology Co., Ltd. disclosed a fire monitoring robot for dynamic fire scenes (Chinese patent application number CN202021950562.X). This application discloses a fire monitoring robot for dynamic fire scenes, including a shell, a smoke sensor, a camera, an MCU controller, and a main controller. The smoke sensor, the MCU controller, and the main controller are all installed in the inner cavity of the shell. The MCU controller and the camera are electrically connected to the main controller. The smoke sensor is electrically connected to the MCU controller. A round hole is opened at the top of the front panel of the shell, and the camera is inserted into the round hole. The shell is provided with a power interface and a transmission interface. The bottom panel of the shell is provided with a mounting hole, and the shell is suitable for being screwed to the mounting plate through the mounting hole.

[0003] Another example is a substation robot inspection system and its inspection method disclosed in application number CN202011119444.9. The substation robot inspection system includes a background control server and an inspection robot that establish a network connection based on 5G communication technology. Through the use of the fifth generation of communication technology, the communication rate and stability problems between the robot and the background server are solved, and the performance of the background server can be fully utilized to achieve the effect of real-time image parsing and processing. The inspection method includes task scheduling, path planning, robot autonomous navigation, pan-tilt camera control, real-time image file return and real-time recognition technology, which can realize continuous inspection of the robot during the automatic inspection process, reduce the number of stops, and improve the inspection efficiency and inspection range under the same battery capacity. At the same time, with the support of high-performance industrial control computers, the recognition time is short, the accuracy is high, and the recognition range is wide, which fully utilizes the intelligence of the robot, improves the inspection efficiency, and enhances the user experience. Therefore, it is completely feasible and necessary to develop an intelligent inspection device for the main plant of a natural gas power plant based on robot technology to replace manual inspection. Summary of the invention

[0004] The present invention solves the problems of low efficiency and slow speed of manual inspection and difficulty in inspection in narrow space in the prior art, and provides an inspection robot navigation method applied to natural gas power plants.

[0005] The technical solution adopted by the present invention to solve its technical problem is: a navigation method for an inspection robot applied to a natural gas power plant, which is suitable for a natural gas power plant, comprises a host computer and at least one inspection robot, each of which comprises at least a control unit, a sensor group, a motion controller, a motion device, a feedback device and a communication unit, the control unit is connected to the motion device through the motion controller, the motion device is provided with a feedback device, the feedback signal of the feedback device is input into the motion controller, the control unit is connected to the host computer through the communication unit, the sensor group is electrically connected to the control unit, and the following steps are performed:

[0006] Step 1: The natural gas power plant sets a detection point at a location where feedback is required, and at least one feedback content exists at the detection point. The host computer establishes an electronic map according to the target natural gas power plant; the electronic map of the natural gas power plant is set with feasible routes and detection point locations;

[0007] Step 2: The host computer synchronizes the electronic map to the inspection robot;

[0008] Step 3: The host computer sets the inspection route, target inspection points, and inspection items as needed and sends them to the inspection robot;

[0009] Step 4: The inspection robot conducts inspection according to the inspection route issued. The control unit determines the current position of the inspection robot according to the feedback signal of the feedback device and the feedback content of the detection point, and uses the sensor group to conduct real-time detection, and conducts inspection in combination with the real-time detection situation.

[0010] The technology of the inspection robot inspecting the main plant of the natural gas power plant according to the drawn inspection map can be completed by setting the original digital map to be transmitted to the robot, or by the robot's autonomous drawing. The robot's autonomous drawing method requires the configuration of radar and other devices, and then the overall line is detected by manual control. In the process of self-service drawing, the automatic traversal method can also be used to obtain the status of the overall line, but the traversal method takes a long time and is subject to greater interference. Furthermore, even in the process of manual inspection, there are obstacles. At this time, the user can delete and modify it in the subsequent revision process. The setting of the detection point is both the target point of data collection and the detection point of data feedback and position feedback. Through the real-time feedback of the inspection robot, the user can clearly understand the current status and position of the inspection robot in the upper computer without the need to rely on other positioning system devices. Furthermore, the inspection robot can also make further position verification through the feedback of its own motion device to form a positioning system with redundant position detection. This positioning system can ensure the navigation accuracy of the inspection robot. Therefore, no matter which of the above methods is adopted, it is necessary to ensure that the inspection map is true, and a number of inspection points are set in the inspection map. At the inspection points, the inspection robot will perform corresponding inspection actions. The inspection actions include collecting field data, collecting equipment information and other operations. The specific operations may be wireless reception of data, shooting and displaying data, using sensors to measure data, etc., and there is at least one feedback content at the inspection point. Many of the inspection points do not collect field data or collect equipment information. In this case, the feedback is simply location data feedback. The feedback of the inspection point can be completed by comparing the boundaries of the physical objects, that is, the robot is required to take photos at a fixed angle, and compare the boundaries of the taken photos with the boundaries of the preset photos, so as to further collect field data, collect equipment information or realize simple location data feedback according to the shooting content, so as to ensure the stable and intelligent operation of the robot. In addition to the above content, the feedback of the detection point in the present invention can also be achieved by presetting a QR code at a fixed position at the detection point. This position is very convenient for the inspection robot to scan the QR code, and the inspection robot has a higher priority for scanning the QR code. After scanning the preset QR code, the inspection robot performs the corresponding action at the detection point according to the action instruction of the preset QR code. Therefore, the feedback efficiency can be increased in the above way. The inspection combined with the real-time detection described in the present invention specifically includes, through real-time detection, judging whether there are obstacles on the current line, if there are no obstacles, continuing the inspection, and if there are obstacles, executing the corresponding subsequent content.

[0011] If the preset QR code is not set, the action should be recorded when it is executed for the first time. In this case, corresponding calibration is required when it is executed for the first time, and the control of the inspection robot is also more complicated.

[0012] By setting up the device of the present invention, intelligent inspection can be realized in the main plant of a natural gas power plant. The types of robots that can be selected by the present invention are not very restricted. As long as the above-mentioned equipment exists, the intelligent inspection task should be able to be completed. Through the above-mentioned settings, especially the layout and design of the detection points, it is ensured that the inspection robot can maintain a stable route while obtaining inspection data, and cooperate with a series of sensors including gas sensors and thermal imaging sensors to promote the completion of intelligent inspection of the main plant of a natural gas power plant.

[0013] Preferably, the sensor group includes a camera and a distance detector. When the inspection robot performs real-time detection based on the sensor group and finds that there is an obstacle on the target line, it first stops and makes a preliminary judgment on the obstacle. The judgment process includes object identification. If the object is determined to be a fixed obstacle in the object identification, the detour step is started. If the object is determined to be a mobile obstacle in the object identification, the robot stops and waits or hands over the control to the host computer. The distance detector can be the ranging radar described above, or any other ranging device.

[0014] Preferably, the detour step includes, in the inspection robot, determining the inspection route according to the projection area of ​​the object to be identified in the image taken at a fixed distance and a fixed focal length, and according to the actual proportion of the projection area of ​​the object to be identified in the target route; if the actual proportion of the projection area of ​​the object to be identified in the target route is less than a preset value, the inspection robot reduces its speed and detours on the target route at the set speed; if the actual proportion of the projection area of ​​the object to be identified in the target route is greater than the preset value, the detour is performed by replanning the route.

[0015] Preferably, the detection points include detection points for feedback of device information or detection points only for feedback of location data, and the method of re-planning the route includes automatically selecting all detection points for feedback of device information that have not been inspected in this inspection route, and then selecting an optional route based on the position of the detection points for feedback of device information and the current identified object position using the shortest path algorithm.

[0016] Preferably, if during the object recognition process, a highlighted image with a highlighted area covering more than a set ratio detected by the control unit appears, the control unit transmits the highlighted image to the recognition unit, and the recognition unit recognizes it through the following recognition steps:

[0017] Recognition step 1: Decompose the brightness component of the input image and calculate the mean value of the regional brightness;

[0018] Identification step 2: weighted adjustment is performed on the color data of each region according to the mean value of the regional brightness, and the difference between the weighted adjusted color data and the adjacent color data is threshold judged, and the color data that meets the set threshold is set to 1, otherwise it is set to 0;

[0019] In the third identification step, the color data and brightness data of each area are fused according to a set ratio to form a boundary response, and a boundary image is made according to the boundary response, and the boundary image is used for object identification.

[0020] Preferably, the control unit verifies the current location data according to the acquired location data to infer the shooting position of the current highlight image and the position of the shooting target, and adjusts the inspection route of the inspection robot according to the position of the inspection robot on the inspection map, selects the side of the shooting target as the next target position of the inspection robot, and reshoots the shooting target after the inspection robot is configured at the target position to form a secondary highlight image of the same target, and transmits it to the recognition unit, which recognizes it again through the recognition step. The recognition unit uses the boundary response obtained by the initial recognition of the same target as the front boundary image, and the recognition unit scales the boundary response obtained by the second recognition of the same target into a side boundary image proportional to the front boundary image. If it is difficult to identify what the object is after the boundary of the object is identified in the highlight environment, it can be directly uploaded to the host computer for identification. In order to facilitate manual identification, the shooting angle can be adjusted to form a front and side form to obtain a more recognizable form, or these images can be uploaded to the host computer, and the operator of the host computer makes manual judgments.

[0021] As preferably, the side of the front boundary image is merged with the side of the side boundary image, the front boundary image and the side boundary image are perpendicular to each other, the front boundary of all front boundary images is divided into a number of front boundary segments of the same length, all front boundary segments are extended backward, the side boundary of all side boundary images is divided into a number of side boundary segments of the same length, all side boundary segments are extended to the center side of the front boundary image, the longest length of the front boundary segment and the side boundary segment extension is a set value, and all front boundary segments and side boundary segments are retained. The picture content with any intersection is used as a recognition image. When the host computer displays, the center line of the recognition image is the side of the front boundary image, and it is also the side of the side boundary image. Through the above form, the acquisition of a quasi-stereoscopic recognition image is achieved, which is more convenient and intuitive for judging and analyzing the content of an object. The side in the present invention refers to the side of the photo (canvas) where the rectangular image is located.

[0022] Preferably, in the recognition step 1, the input image I(x,y) is decomposed into a brightness component L(x,y), and the mean value L of the regional brightness is calculated with each pixel of L(x,y) as the center.avg (x,y),

[0023] If the point (x, y) is the location that needs to be enhanced, the brightness adjustment response L is obtained by weighted calculation of the brightness adjustment factor a. rod (x, y), the value range of the brightness adjustment factor a is greater than 0 and less than or equal to 1.

[0024] Preferably, in the second recognition step, for the input image I(x,y), each pixel is decomposed into several color responses, which are encoded to form several color channels, and then a convolution operation is performed on the difference between the opposite responses and the Gaussian derivative function Gau(x,y).

[0025] Preferably, in the identification step three, the convolution operation result and the brightness adjustment response are fused according to a set ratio to obtain a boundary response.

[0026] The substantial effect of the present invention is that the setting of the detection point is not only the target point of data collection but also the detection point of data feedback and position feedback. Through the real-time feedback of the inspection robot, the user can clearly understand the current state and position of the inspection robot in the host computer, which can ensure the navigation accuracy of the inspection robot. At the detection point, the inspection robot will perform the corresponding inspection action, and can further collect field data, collect equipment information or realize simple location data feedback according to the shooting content, so as to ensure the stable and intelligent operation of the robot. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A schematic diagram of a process of the present invention;

[0028] Figure 2 A schematic diagram of a framework structure of the present invention with an identification unit;

[0029] Figure 3 A schematic diagram of the effect of the shortest path algorithm in the present invention;

[0030] Figure 4 A schematic diagram of a process for identifying the boundary of a single image according to the present invention;

[0031] Figure 5 This is a schematic diagram showing a comparison of the effects of two boundary recognition and splicing images according to the present invention;

[0032] Figure 6 A schematic diagram of a robot that can be used in the present invention;

[0033] Figure 7 Schematic diagram of a robot that can be used in the present invention. DETAILED DESCRIPTION

[0034] The technical solution of the present invention is further described in detail below through specific embodiments.

[0035] Embodiment 1:

[0036] The technical solution adopted by the present invention to solve the technical problem is: a navigation method for an inspection robot applied to a natural gas power plant (see attached Figure 1 and attached Figure 2 ), applicable to natural gas power plants, including a host computer and at least one inspection robot (see Appendix Figure 6 and attached Figure 7 ), each inspection robot at least includes a control unit, a sensor group, a motion controller, a motion device, a feedback device and a communication unit, the control unit is connected to the motion device through the motion controller, the motion device is equipped with a feedback device, the feedback signal of the feedback device is input into the motion controller, the control unit is connected to the host computer through the communication unit, the sensor group is electrically connected to the control unit, and the following steps are performed:

[0037] Step 1: The natural gas power plant sets a detection point at a location where feedback is required, and at least one feedback content exists at the detection point. The host computer establishes an electronic map according to the target natural gas power plant; the electronic map of the natural gas power plant is set with feasible routes and detection point locations;

[0038] Step 2: The host computer synchronizes the electronic map to the inspection robot;

[0039] Step 3: The host computer sets the inspection route, target inspection points, and inspection items as needed and sends them to the inspection robot;

[0040] Step 4: The inspection robot conducts inspection according to the inspection route issued. The control unit determines the current position of the inspection robot according to the feedback signal of the feedback device and the feedback content of the detection point, and uses the sensor group to conduct real-time detection, and conducts inspection in combination with the real-time detection situation.

[0041] The technology of the inspection robot inspecting the main plant of the natural gas power plant according to the drawn inspection map can be completed by setting the original digital map to be transmitted to the robot, or by the robot's autonomous drawing. The robot's autonomous drawing method requires the configuration of radar and other devices, and then the overall line is detected by manual control. In the process of self-service drawing, the automatic traversal method can also be used to obtain the status of the overall line, but the traversal method takes a long time and is subject to greater interference. Furthermore, even in the process of manual inspection, there are obstacles. At this time, the user can delete and modify it in the subsequent revision process. The setting of the detection point is both the target point of data collection and the detection point of data feedback and position feedback. Through the real-time feedback of the inspection robot, the user can clearly understand the current status and position of the inspection robot in the upper computer without the need to rely on other positioning system devices. Furthermore, the inspection robot can also make further position verification through the feedback of its own motion device to form a positioning system with redundant position detection. This positioning system can ensure the navigation accuracy of the inspection robot. Therefore, no matter which of the above methods is adopted, it is necessary to ensure that the inspection map is true, and a number of inspection points are set in the inspection map. At the inspection points, the inspection robot will perform corresponding inspection actions. The inspection actions include collecting field data, collecting equipment information and other operations. The specific operations may be wireless reception of data, shooting and displaying data, using sensors to measure data, etc., and there is at least one feedback content at the inspection point. Many of the inspection points do not collect field data or collect equipment information. In this case, the feedback is simply location data feedback. The feedback of the inspection point can be completed by comparing the boundaries of the physical objects, that is, the robot is required to take photos at a fixed angle, and compare the boundaries of the taken photos with the boundaries of the preset photos, so as to further collect field data, collect equipment information or realize simple location data feedback according to the shooting content, so as to ensure the stable and intelligent operation of the robot. In addition to the above content, the feedback of the detection point in the present invention can also be achieved by presetting a QR code at a fixed position at the detection point. This position is very convenient for the inspection robot to scan the QR code, and the inspection robot has a higher priority for scanning the QR code. After scanning the preset QR code, the inspection robot performs the corresponding action at the detection point according to the action instruction of the preset QR code. Therefore, the feedback efficiency can be increased in the above way. The inspection combined with the real-time detection described in the present invention specifically includes, through real-time detection, judging whether there are obstacles on the current line, if there are no obstacles, continuing the inspection, and if there are obstacles, executing the corresponding subsequent content.More specifically, the sensor group includes a camera and a distance detector. When the inspection robot performs real-time detection based on the sensor group and finds that there is an obstacle on the target line, it first stops and makes a preliminary judgment on the obstacle. The judgment process includes object identification. If it is determined to be a fixed obstacle in the object identification, the detour step is started. If it is determined to be a mobile obstacle in the object identification, the robot stops and waits or hands over the control to the host computer. When the robot stops and makes a preliminary judgment on the obstacle first, if it is a mobile obstacle, it may move away by itself. At this time, it can continue to operate as usual. If it is determined to be an immovable obstacle, the pneumatic detour step will be performed.

[0042] The detour step includes: judging the inspection route according to the projection area of ​​the object to be identified in the image taken by the inspection robot at a fixed distance and fixed focal length, and according to the actual proportion of the projection area of ​​the object to be identified in the target route; if the actual proportion of the projection area of ​​the object to be identified in the target route is less than the preset value, the inspection robot reduces the speed to detour on the target route at the set speed; if the actual proportion of the projection area of ​​the object to be identified in the target route is greater than the preset value, the detour is performed by replanning the route. The projection area here only looks at the projection area of ​​the outermost contour. The specific detection method is that the inspection robot horizontally shoots the target at a specified height, and at the set height, the contour of the obstacle contained in any position on the horizontal line is obtained, and its vertical projection on the route is obtained, and the specific degree of occupation of the route is judged according to this projection. Then, the inspection robot judges whether it is necessary to detour according to the width of the route and the specific position and area occupied. The detection points include detection points for feedback of device information or detection points for feedback of location data only. The method of re-planning the route includes automatically selecting all detection points for feedback of device information that have not been inspected in the current inspection route, and then selecting an optional route using the shortest path algorithm according to the location of the detection point for feedback of device information and the current location of the identified object (see Appendix Figure 3 ).

[0043] If during the object recognition process, a highlighted image with a highlighted area coverage exceeding the set ratio detected by the control unit appears, the control unit transmits the highlighted image to the recognition unit, which performs the following recognition steps (see Appendix Figure 4 ) for identification,

[0044] Recognition step 1: Decompose the brightness component of the input image and calculate the mean value of the regional brightness;

[0045] Identification step 2: weighted adjustment is performed on the color data of each region according to the mean value of the regional brightness, and the difference between the weighted adjusted color data and the adjacent color data is threshold judged, and the color data that meets the set threshold is set to 1, otherwise it is set to 0;

[0046] In the third identification step, the color data and brightness data of each area are fused according to a set ratio to form a boundary response, and a boundary image is made according to the boundary response, and the boundary image is used for object identification.

[0047] Preferably, the control unit verifies the current location data according to the acquired location data to infer the shooting position of the current highlight image and the position of the shooting target, and adjusts the inspection route of the inspection robot according to the position of the inspection robot on the inspection map, selects the side of the shooting target as the next target position of the inspection robot, and reshoots the shooting target after the inspection robot is configured at the target position to form a secondary highlight image of the same target, and transmits it to the recognition unit, which recognizes it again through the recognition step. The recognition unit uses the boundary response obtained by the initial recognition of the same target as the front boundary image, and the recognition unit scales the boundary response obtained by the second recognition of the same target into a side boundary image proportional to the front boundary image. If it is difficult to identify what the object is after the boundary of the object is identified in the highlight environment, it can be directly uploaded to the host computer for identification. In order to facilitate manual identification, the shooting angle can be adjusted to form a front and side form to obtain a more recognizable form, or these images can be uploaded to the host computer, and the operator of the host computer makes manual judgments.

[0048] As preferably, the side of the front boundary image is merged with the side of the side boundary image, the front boundary image and the side boundary image are perpendicular to each other, the front boundary of all front boundary images is divided into a number of front boundary segments of the same length, all front boundary segments are extended backward, the side boundary of all side boundary images is divided into a number of side boundary segments of the same length, all side boundary segments are extended to the center side of the front boundary image, the longest length of the front boundary segment and the side boundary segment extension is a set value, and all front boundary segments and side boundary segments are retained. The picture content with any intersection is used as a recognition image. When the host computer displays, the center line of the recognition image is the side of the front boundary image, and it is also the side of the side boundary image. Through the above form, the acquisition of a quasi-stereoscopic recognition image is achieved, which is more convenient and intuitive for judging and analyzing the content of an object. The side in the present invention refers to the side of the photo (canvas) where the rectangular image is located.

[0049] Preferably, in the recognition step 1, the input image I(x,y) is decomposed into a brightness component L(x,y), and the mean value L of the regional brightness is calculated with each pixel of L(x,y) as the center. avg (x,y),

[0050]

[0051] In the above formula, (i, j) represents the coordinates in the area, step represents the moving step length,

[0052] If the point (x, y) is located at a position that needs to be enhanced, the brightness adjustment response L is obtained. rod (x,y),

[0053]

[0054] In the above formula, a is an artificially set brightness adjustment factor, 0<a≤1.

[0055] Preferably, in the recognition step 2, for the input image I(x, y), each pixel is decomposed into several color responses, which are encoded to form several color channels, and then the difference between the opposite responses is convolved with the Gaussian derivative function Gau(x, y), and the result is recorded as

[0056]

[0057] Among them, * represents the convolution operator;

[0058] γ and σ are set manually, and the preset values ​​are γ = 0.5 and σ = 1.5; C(x, y) and S(x, y) represent opposite color responses, and the subscript C + / S - represents the corresponding color channel, k(x,y) represents the opposing color response weight, if the opposing color difference is greater than the threshold, k(x,y)=0 is set; if the opposing color difference is less than the threshold, k(x,y)=1 is set. The specific setting rules of k(x,y) are as follows:

[0059] Using the following formula, Calculate L rod The local brightness mean of (x,y) and will After nonlinear activation processing, it serves as a dynamic adjustment factor of k(x,y).

[0060]

[0061] The image I(x,y) is dynamically encoded using k(x,y).

[0062] Preferably, in the identification step three, the and L rod (x,y), get the boundary response I border (x,y),

[0063]

[0064] P1 and P2 are the color dynamic encoding fusion ratio and brightness fusion ratio respectively.

[0065] In this embodiment, since highlights are inevitable when taking photos, such as highlights caused by the reflective surface of a flash in the dark, highlights caused by flickering lights, etc., these highlights will affect the acquisition of effective information in the picture. For example, when meter reading is performed by taking photos, since highlights will affect the display of meter data, at this time, more accurate meter data can be obtained through the image contour recognition method provided in this embodiment.

[0066] In this embodiment, the control unit verifies the current location data based on the acquired location data to infer the shooting location of the current highlighted image and the location of the shooting target, and adjusts the inspection route of the inspection robot according to the position of the inspection robot on the inspection map, and selects the side of the shooting target as the next target position of the inspection robot. After the inspection robot is configured at the target position, it re-shoots the shooting target to form a secondary highlighted image of the same target, and transmits it to the recognition unit, which recognizes it by going through the recognition steps again. The recognition unit uses the boundary response obtained by the initial recognition of the same target as the front boundary image, and the recognition unit scales the boundary response obtained by the secondary recognition of the same target into a side boundary image that is proportional to the front boundary image. If an image that does not conform to the preset image boundary appears, then the recognition of the captured image is problematic. For example, after the boundary of an object is recognized in a high-light environment, it is still difficult to identify what the object is. Therefore, a more recognizable form can be obtained by adjusting the shooting angle to form the front and side. Furthermore, the side of the front boundary image can be merged with the side of the side boundary image. The front boundary image and the side boundary image are perpendicular to each other. The front boundaries of all front boundary images are divided into several front boundary segments of the same length. All front boundary segments extend backward. The side boundaries of all side boundary images are divided into several side boundary segments of the same length. All side boundary segments extend toward the center of the front boundary image. The longest length of the front boundary segment and the side boundary segment is the set value. All the image contents with any intersection between the front boundary segment and the side boundary segment are retained as the recognition image. When displayed on the host computer, the center line of the recognition image is the side of the front boundary image, which is also the side of the side boundary image. Through the above form, the acquisition of a quasi-stereoscopic recognition image is realized, which is more convenient and intuitive for judging and analyzing the content of the object. The side in this embodiment refers to the side of the photo (canvas) where the rectangular image is located. Figure 5For objects with the same front boundary, when the side boundaries are different, the content displayed in the recognition image is very different, and the display of the recognition image can provide users with better observation effects. This technology is more useful in abnormal conditions such as fire and highlight, because it is difficult to know the specific actual phenomenon from the highlight image simply by looking at the scene picture taken, and the method in this embodiment can obtain better scene images, which provides great help for on-site command and judgment.

[0067] In summary, the detection point setting of this embodiment is both the target point for data collection and the detection point for data feedback and position feedback. Through the real-time feedback of the inspection robot, the user can clearly understand the current state and position of the inspection robot in the host computer, which can ensure the navigation accuracy of the inspection robot. At the detection point, the inspection robot will perform the corresponding inspection action, and can further collect field data, collect equipment information or realize simple location data feedback according to the shooting content, so as to ensure the stable and intelligent operation of the robot.

[0068] The above-described embodiment is only a preferred solution of the present invention and does not limit the present invention in any form. There are other variations and modifications without exceeding the technical solution described in the claims.

Claims

1. A navigation method for an inspection robot applied to a natural gas power plant, characterized in that: Applicable to natural gas power plants, including a host computer and at least one inspection robot, each of the inspection robots at least includes a control unit, a sensor group, a motion controller, a motion device, a feedback device and a communication unit, the control unit is connected to the motion device through the motion controller, the motion device is equipped with a feedback device, the feedback signal of the feedback device is input into the motion controller, the control unit is connected to the host computer through the communication unit, the sensor group is electrically connected to the control unit, and the following steps are performed: Step 1: The natural gas power plant sets a detection point at a location where feedback is required, and at least one feedback content exists at the detection point. The host computer establishes an electronic map according to the target natural gas power plant; the electronic map of the natural gas power plant is set with feasible roads and detection point locations; Step 2: The host computer synchronizes the electronic map to the inspection robot; Step 3: The host computer sets the inspection route, target inspection points, and inspection items as needed and sends them to the inspection robot; Step 4: The inspection robot conducts inspection according to the inspection route issued, and the control unit determines the current position of the inspection robot according to the feedback signal of the feedback device and the feedback content of the detection point, and conducts real-time detection according to the sensor group, and conducts inspection according to the real-time detection situation; The sensor group includes a camera and a distance detector. When the inspection robot performs real-time detection according to the sensor group and finds that there is an obstacle on the target road, it first stops and makes a preliminary judgment on the obstacle. The judgment process includes object recognition. If the object is determined to be a fixed obstacle in the object recognition, the detour step is started. If the object is determined to be a mobile obstacle in the object recognition, the robot stops and waits or hands over the control right to the host computer. The detour step includes determining the inspection route according to the projection area of ​​the object to be identified in the image taken by the inspection robot at a fixed distance and a fixed focal length, and according to the actual proportion of the projection area of ​​the object to be identified in the target route; if the actual proportion of the projection area of ​​the object to be identified in the target route is less than a preset value, the inspection robot reduces the speed to detour on the target route at a set speed; if the actual proportion of the projection area of ​​the object to be identified in the target route is greater than the preset value, the inspection robot detours by replanning the route; The detection points include detection points for feeding back device information or detection points only for feeding back location data. The method of re-planning the route includes automatically selecting all detection points for feeding back device information that have not been inspected in the current inspection route, and then selecting an optional route using the shortest path algorithm based on the location of the detection point for feeding back device information and the location of the currently identified object; If during the object recognition process, the control unit detects that the highlighted area covers a highlighted image exceeding the set ratio, the control unit transmits the highlighted image to the recognition unit, which recognizes it through the following recognition steps: Recognition step 1: Decompose the brightness component of the input image and calculate the mean value of the regional brightness; Identification step 2: weighted adjustment is performed on the color data of each region according to the mean value of the regional brightness, and the difference between the weighted adjusted color data and the adjacent color data is threshold judged, and the color data that meets the set threshold is set to 1, otherwise it is set to 0; Recognition step three: the color data and brightness data of each area are fused according to a set ratio to form a boundary response, and a boundary image is made according to the boundary response, and the boundary image is used to identify the object; The control unit verifies the current location data based on the acquired location data to infer the shooting location of the current highlighted image and the location of the shooting target, and adjusts the inspection route of the inspection robot according to the position of the inspection robot on the inspection map, selects the side of the shooting target as the next target position of the inspection robot, configures the inspection robot at the target position and then shoots the shooting target again to form a secondary highlighted image of the same target, and transmits it to the recognition unit, which recognizes it by going through the recognition steps again. The recognition unit uses the boundary response obtained by the initial recognition of the same target as the front boundary image, and scales the boundary response obtained by the secondary recognition of the same target into a side boundary image that is proportional to the front boundary image.

2. The inspection robot navigation method applied to a natural gas power plant according to claim 1 is characterized in that: The side edges of the front boundary image are merged with the side edges of the side boundary image. The front boundary image and the side boundary image are perpendicular to each other. The front boundaries of all front boundary images are divided into several front boundary segments of the same length. All front boundary segments extend backward. The side boundaries of all side boundary images are divided into several side boundary segments of the same length. All side boundary segments extend toward the center of the front boundary image. The longest length of the front boundary segment and the side boundary segment is the set value. All picture contents with any intersection between the front boundary segment and the side boundary segment are retained as recognition images.

3. The inspection robot navigation method applied to a natural gas power plant according to claim 2 is characterized in that: In the recognition step 1, the input image I(x,y) is decomposed into the brightness component L(x,y), and the mean value Lavg(x,y) of the regional brightness is calculated with each pixel of L(x,y) as the center. If the point (x, y) is a position that needs to be enhanced, the brightness adjustment response Lrod(x, y) is obtained by weighted calculation of the brightness adjustment factor, and the value range of the brightness adjustment factor is greater than 0 and less than or equal to 1.

4. The inspection robot navigation method applied to a natural gas power plant according to claim 3 is characterized in that: In the second recognition step, for each pixel of the input image, several color responses are decomposed, which are encoded to form several color channels, and then a convolution operation is performed on the difference between the opposite responses and the Gaussian derivative function.

5. The inspection robot navigation method applied to a natural gas power plant according to claim 4 is characterized in that: In the identification step three, the convolution operation result and the brightness adjustment response are fused according to a set ratio to obtain a boundary response.

Images