A machine vision-based power inspection robot detection method and system

Abstract

The application relates to the technical field of power detection, in particular to a power inspection robot detection method and system based on machine vision. The application constructs a target region three-dimensional model through pre-acquisition of region basic data, and obtains historical defect data to mark a detection key region, then uses a genetic algorithm to optimize an initial path and key detection points, and removes redundant acquisition points, so that the accuracy of a visual detection path is improved; meanwhile, a path acquisition collaborative analysis module is used to analyze visual data in real time, generate an optimization instruction to dynamically adjust the path, and avoid the low efficiency and missing detection problems caused by independent operation of the visual and navigation; the application collects dynamic data of detection such as definition, vibration acceleration, environmental wind speed and environmental light through a visual perception module, analyzes and outputs a comprehensive fluctuation value through an environmental adaptability module, obtains a corresponding fluctuation value interval, generates corresponding environmental adaptability regulation instructions, and avoids the influence caused by motion blur and environmental mutation.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of power detection, in particular to a power inspection robot detection method and system based on machine vision. BACKGROUND

[0002] The scale of the power system continues to expand, the number of transmission lines and substations increases rapidly, the traditional manual inspection is low in efficiency and high in labor intensity, the safety risk is high in high-altitude, high-cold and other high-risk environments, and defects are easy to be missed due to human factors, meanwhile, the intelligent power grid construction policy promotes and the equipment aging problem is highlighted, higher requirements are put forward for the accuracy and real-time performance of the inspection, the inspection technology develops from the early simple equipment relying on mechanical arms and remote control to the model with autonomous inspection ability, and now the AI, multi-sensor and big data technologies are combined to realize intelligent defect identification and fault diagnosis, and the application scenarios are expanded from transmission lines to substations and new energy stations, forming multiple types of equipment such as wheeled and rail-mounted.

[0003] At present, the visual acquisition module and the navigation module of the existing power inspection robot are often independently operated when detecting, the planning of the path will affect the detection angle, repeated scanning or missing of the key area, the overall efficiency is low, and the visual detection is easy to be affected by the movement of the robot, resulting in motion blur, environmental mutation and the like, and the detection accuracy is reduced. SUMMARY

[0004] The application provides a power inspection robot detection method and system based on machine vision, which is used to solve the above technical problems.

[0005] The first aspect of the application provides a power inspection robot detection method based on machine vision, which comprises the following steps:

[0006] Step 1: model construction based on pre-acquired regional basic data to obtain a target region three-dimensional model, and marking a detection key region in the target region three-dimensional model;

[0007] As a further improvement of the application, step 1.1, establishing a target region three-dimensional model: identifying the regional basic data of the target region in the database to obtain the device parameters and geographic information of the inspection target of the target detection region; and obtaining the pre-scanning data of the laser radar, obtaining the structure model of each device in the inspection target region based on the device structure data of the pre-scanning data, and collecting the structure model of each device, the corresponding device parameters and the geographic coordinates to obtain the target region three-dimensional model of the target region.

[0008] Further, step 1.2, key region detection, a preset key model comparison library is obtained, each model corresponding to the target region three-dimensional model is matched with the key model comparison library, and the model matched with the key model comparison library is marked as a detection key region; the key model comparison library is a high-incidence position of equipment defects obtained through historical defect occurrence data of the equipment.

[0009] Step 2: The path planning analysis mechanism is used to plan the path of the target region three-dimensional model obtained, and the visual detection path is output;

[0010] As a further improvement of the application, the path planning analysis mechanism has the following specific planning steps:

[0011] Step 2.1: The initial state path is output by initial path analysis;

[0012] Step 2.2: The detection key region and the key point to be optimized corresponding to each detection region are obtained, and the corresponding preset key region detection standard is obtained, and the key detection point is simulated and output by the population evolution of the genetic algorithm, and the detection key region, the key point to be optimized, and the key region detection standard are simulated and output by the population evolution of the genetic algorithm.

[0013] Step 2.3: The key detection point is matched with the initial state path, the collection point matched with the key detection point is marked as a redundant detection point, and the initial state path corresponding to the redundant detection point is removed to obtain the visual detection path.

[0014] Further, the initial state path is output by initial path analysis, which is specifically: based on a preset time interval division, a path optimization period is obtained, a target region three-dimensional model of the current path optimization period is obtained, and each device model is obtained. The detection region and the corresponding visual angle shielding region and the optimization target are obtained, the initial path is output by analyzing each detection region, visual angle shielding region and optimization target by the population evolution of the genetic algorithm, the collection ability parameters of each power inspection robot are obtained, that is, the visual collection accuracy and the field of view angle of visual collection, the visual collection accuracy and the field of view angle are matched with the initial path, and the collection points corresponding to each detection region are obtained based on the coverage range of the field of view angle; the initial path and the collection points are matched to obtain the initial state path.

[0015] Step 3: Based on the visual detection path, the visual collection is performed to obtain the corresponding visual collection data in real time, and the visual collection data is input into the path collection collaborative analysis module, and then the collaborative optimization instruction is output; the collaborative optimization instruction is executed in step 4;

[0016] As a further improvement of the application, the path collection collaborative analysis module has the following specific execution steps:

[0017] The visual image of each collection point of the current corresponding to-be-detected region is obtained by identifying the visual collection data, each visual image is matched with the pre-stored device standard visual image corresponding to the to-be-detected region to obtain the visual angle occlusion region of each visual image; meanwhile, the visual images corresponding to adjacent collection points are matched to obtain the picture repetition region by picture superposition matching;

[0018] The visual angle occlusion ratio value is obtained by calculating the proportion of the visual angle occlusion region and the device standard visual image, and the visual angle occlusion ratio value is compared with the pre-set occlusion ratio threshold value, when the occlusion ratio threshold value is exceeded, the visual angle occlusion ratio value is output as an occlusion defect value; the detection key region corresponding to the visual angle occlusion region and the device standard visual image is matched, when the visual angle occlusion region covers the detection key region, the visual angle occlusion ratio value is directly output as an occlusion defect value, and the pre-set key missing multiple is obtained, then the occlusion defect value and the key missing multiple are multiplied to obtain a key region defect value;

[0019] The repetition region ratio value is obtained based on the picture repetition region, the repetition region ratio value is divided into a plurality of continuous repetition region ratio intervals according to the pre-set ratio separation value, the interval minimum value of each repetition region ratio interval is obtained, the interval minimum values of each interval are arranged in ascending order to obtain an interval ascending sequence number; each repetition region ratio interval is assigned a repetition defect coefficient, the value of which is greater than zero, and the value of each repetition defect coefficient increases with the increase of the interval ascending sequence number; the repetition region ratio value corresponding to the current adjacent collection point is matched with each repetition region ratio interval, the corresponding repetition region ratio interval is output and the corresponding repetition defect coefficient is obtained, and the repetition defect coefficient and the repetition region ratio value are multiplied to obtain a repetition defect value.

[0020] The occlusion defect value, the key region defect value and the repetition defect value are normalized and the values are taken, a comprehensive defect value is calculated and output by using a pre-set comprehensive calculation formula; the comprehensive defect value is executed to generate an analysis of optimization instructions, and a cooperative optimization instruction is output.

[0021] Further, the comprehensive defect value is executed to generate an analysis of optimization instructions, which specifically analyzes as follows: when the comprehensive defect value exceeds the pre-set comprehensive defect threshold value, the comprehensive defect value is identified, when the occlusion comprehensive defect value and the repetition defect value are both not zero, the corresponding repetition region is marked as a to-be-optimized region signal based on the repetition defect value, the occlusion comprehensive defect value is identified, if the key region defect value is not zero, a key detection point optimization instruction of the corresponding key region is generated, otherwise the corresponding occlusion defect key point is marked as a signal;

[0022] When the occlusion comprehensive defect value is zero, the corresponding repeated area is marked as a to-be-optimized area signal based on the repeated defect value; the cooperative optimization instruction includes the to-be-optimized area signal, the key detection point optimization instruction and the occlusion defect key marking signal.

[0023] Step 4: generating a detection optimization strategy based on the cooperative optimization instruction analysis and executing.

[0024] As a further improvement of the application, the detection optimization strategy is generated based on the cooperative optimization instruction analysis and executed, which specifically comprises: obtaining the to-be-optimized area signal, the key detection point optimization instruction and the occlusion defect key marking signal by identifying the cooperative optimization instruction; obtaining the repeated area based on the to-be-optimized area signal, marking the two adjacent detection points corresponding to the repeated area as to-be-optimized targets, and executing step 2.1; obtaining the perspective occlusion area based on the occlusion defect key marking signal, and executing step 2.1 on the perspective occlusion area; obtaining the key detection point of the corresponding key area based on the key detection point optimization instruction, marking the key detection point as a to-be-optimized key point, and executing step 2.2.

[0025] The second aspect of the application provides a machine vision-based power inspection robot detection system, which comprises a visual perception module, an environmental adaptability module and a detection execution module.

[0026] The visual perception module is used for collecting data through the data perception unit of the power inspection robot to obtain detection dynamic data; the detection dynamic data includes a definition parameter, motion perception data and environmental perception data.

[0027] The environmental adaptability module is used for performing environmental adaptability analysis on the detection dynamic data to obtain environmental adaptability regulation instructions.

[0028] As a further improvement of the application, the detection dynamic data is analyzed to obtain environmental adaptability regulation instructions, and the specific analysis steps are as follows:

[0029] The detection dynamic data is identified based on data recognition technology to obtain the definition parameter, the motion perception data and the environmental perception data.

[0030] The definition corresponding to each preset frame number is obtained based on the definition parameter, a preset definition reference value is obtained and is compared with the definition of each frame to identify the definition drop, when the definition is less than the definition reference value, the definition drop value is output, when the definition drop value exceeds a preset drop ratio, the corresponding frame number is marked as a mutation frame, and the total number of mutation frames is counted; the definition value and the total number of mutation frames are input into a preset visual fluctuation calculation formula to obtain a visual fluctuation value.

[0031] The preset sample number of vibration acceleration and robot posture is obtained based on motion sensing data, a preset acceleration safety threshold is acquired, the vibration acceleration of each sample is compared with the acceleration safety threshold, the sample exceeding the acceleration safety threshold is marked as an abnormal sample number, and the acceleration abnormality ratio is obtained by ratio calculation of the abnormal sample number and the preset sample number;The robot posture corresponding to each sample is compared with the preset standard posture to obtain the posture error angle;The acceleration abnormality ratio and the posture error angle are input into the preset motion fluctuation quantization formula for calculation to output the motion fluctuation value;

[0032] The environmental wind speed and environmental light in the preset sampling period are obtained based on environmental sensing data, the preset controllable wind speed threshold, the limit wind speed threshold and the light mutation threshold are acquired;The maximum environmental wind speed in the sampling period is acquired and marked as the instantaneous maximum wind speed;The light mutation intensity corresponding to each sample in the sampling period is acquired, the sample exceeding the light mutation threshold is recorded as a light mutation sample, and the light mutation ratio is obtained by ratio calculation of the number of light mutation samples and the total number of samples in the sampling period;The values of the instantaneous maximum wind speed and the light mutation ratio are input into the preset environmental fluctuation quantization formula for calculation to output the environmental fluctuation value;

[0033] The visual fluctuation value, the motion fluctuation value and the environmental fluctuation value are normalized and the values are taken, and the comprehensive fluctuation value is calculated;The comprehensive fluctuation value is divided into three fluctuation intervals, i.e. low fluctuation interval, medium fluctuation interval and high fluctuation interval according to the preset fluctuation value interval;The current corresponding comprehensive fluctuation value is matched with each fluctuation interval, when the comprehensive fluctuation value is in the low fluctuation interval, the environmental adaptability control instruction is output as maintaining the acquisition strategy;When in the medium fluctuation interval, the instruction is to increase the shutter speed;When in the high fluctuation interval, the corresponding maximum fluctuation value is acquired, i.e. the instruction is to increase the shutter speed when the visual fluctuation value is maximum, the instruction is to start the wind-avoiding strategy when the motion fluctuation value is maximum, and the instruction is to reduce the moving speed and reset the robot posture when the environmental fluctuation value is maximum.

[0034] The detection execution module is used for inputting the environmental adaptability control instruction into the robot control unit and controlling execution.

[0035] The technical scheme provided by the application has the beneficial effects compared with the prior art:

[0036] 1、The application constructs a target area three-dimensional model by pre-acquiring area basic data, and acquires historical defect data to mark detection key areas, and then optimizes the initial path and key detection points and removes redundant acquisition points by using a genetic algorithm, thereby improving the accuracy of the visual detection path;At the same time, the path acquisition collaborative analysis module analyzes the visual data in real time to generate an optimization instruction to dynamically adjust the path, thereby avoiding the low efficiency and missed detection caused by independent running of vision and navigation.

[0037] 2、The application collects dynamic data of definition, vibration acceleration, environmental wind speed and environmental light through a visual perception module, an environmental adaptability module analyzes and outputs a comprehensive fluctuation value, and obtains a corresponding fluctuation value interval, generates a corresponding environmental adaptability regulation instruction, and avoids the influence of motion blur and environmental mutation. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. The following drawings are not drawn in scale, and the focus is on showing the main idea of the present application.

[0039] Figure 1 A method step diagram of a power inspection robot detection method based on machine vision of the present application;

[0040] Figure 2 An execution block diagram of the present application for generating a detection optimization strategy based on collaborative optimization instruction analysis and execution;

[0041] Figure 3 A principle block diagram of a power inspection robot detection system based on machine vision of the present application. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0043] For the convenience of understanding, the specific process of the embodiments of the present application will be described below. Please refer to Figures 1-2 In the embodiments of the present application, one embodiment of a power inspection robot detection method based on machine vision includes:

[0044] Step 1: inspection data preparation, model construction based on pre-acquired regional basic data to obtain a target region three-dimensional model, mark the detection key region on the target region three-dimensional model, which is specifically:

[0045] Step 1.1, establish a target region three-dimensional model: identify the regional basic data of the target region collected in the database to obtain the device parameters, geographic information of the inspection target of the target detection region; and obtain the pre-scanning data of the laser radar, obtain the structure model of each device in the inspection target region based on the device structure data of the pre-scanning data, and collect the structure model of each device and the corresponding device parameters and geographic coordinates to obtain the target region three-dimensional model of the target monitoring region.

[0046] It should be noted that the inspection target is a power transmission line, a substation device, etc.; the device parameters include wire type, insulator size, tower height, and substation device size, etc.; the geographic information includes latitude, longitude, and altitude, etc.

[0047] Step 1.2, mark the key area for detection: obtain the preset key model comparison library, match the models corresponding to the target area three-dimensional model with the key model comparison library, and mark the models matched with the key model comparison library as the key area for detection; the key model comparison library is a high-risk part of equipment defects obtained through historical defect occurrence data of equipment, and the high-risk part of equipment defects includes but is not limited to wire joints, insulator strings, and lightning arresters.

[0048] Step 2: path planning analysis, the obtained target area three-dimensional model is analyzed by a preset path planning analysis mechanism, and a visual detection path is output;

[0049] The path planning analysis mechanism has the following specific planning steps:

[0050] Step 2.1: initial path output, based on the preset time interval division to obtain the path optimization period, obtain the target area three-dimensional model of the current path optimization period, and obtain the detection area of each device model and the corresponding visual angle shielding area and optimization target, analyze the initial path of each detection area, visual angle shielding area and optimization target by population evolution simulation of genetic algorithm, obtain the collection capability parameters of each power inspection robot, i.e. visual collection accuracy and field of view angle, and superimpose the visual collection accuracy and field of view angle on the initial path, obtain the collection points corresponding to each detection area based on the coverage range of the field of view angle; superimpose the initial path and the collection points to obtain the initial state path.

[0051] The power inspection robot includes but is not limited to a drone, a ground wheeled / track type robot, and a track type robot; the analysis process of the genetic algorithm is to encode each detection area and historical path node as a chromosome, generate a new path through crossover and mutation operations, and output the initial path in combination with the reference coverage rate of the detection area, the visual angle deviation threshold, and the path smoothness.

[0052] Step 2.2: key area path output, obtain the detection key area corresponding to each detection area and the optimization key point, and obtain the corresponding preset key area detection standard (i.e. detection standard accuracy, visual sample collection amount), and similarly, simulate the key detection point by population evolution simulation of genetic algorithm to simulate natural selection on the detection key area, optimization key point, and key area detection standard.

[0053] Step 2.3: Path redundancy optimization, coinciding the key detection points with the initial state path, marking the acquisition points coinciding with the key detection points as redundant detection points, removing the redundant detection points corresponding to the initial state path to obtain the visual detection path.

[0054] Step 3: Cooperative control analysis, based on the initial detection path, performing real-time acquisition of corresponding visual acquisition data, and inputting the visual acquisition data into the path acquisition cooperative analysis module, then outputting the cooperative optimization instruction; executing step 4 on the cooperative optimization instruction.

[0055] The specific execution steps of the path acquisition cooperative analysis module are as follows:

[0056] Identifying the visual images of each acquisition point of the current corresponding detection area from the visual acquisition data, and matching each visual image with the pre-stored device standard visual image corresponding to the detection area to obtain the visual angle occlusion area of each visual image (including but not limited to the visual angle offset occlusion area caused by too high acquisition visual angle offset, the visual angle occlusion area caused by having an occlusion in the visual angle, and the blur area caused by dynamic blur); at the same time, matching the visual images corresponding to adjacent acquisition points to obtain the picture repeat area;

[0057] Calculating the visual angle occlusion ratio value by comparing the visual angle occlusion area with the device standard visual image, and comparing the visual angle occlusion ratio value with the pre-set occlusion ratio threshold value, when the occlusion ratio threshold value is exceeded, the visual angle occlusion ratio value is output as the occlusion defect value; matching the visual angle occlusion area with the detection key area corresponding to the device standard visual image, when the visual angle occlusion area covers the detection key area, the visual angle occlusion ratio value is directly output as the occlusion defect value, and the pre-set key missing multiple value (its value is greater than 1) is obtained, then the occlusion defect value and the key missing multiple value are multiplied to obtain the key area defect value;

[0058] Based on the repeat area, the repeat area ratio value is obtained, the repeat area ratio value is divided into multiple continuous repeat area ratio intervals according to the pre-set ratio separation value, the interval minimum value of each repeat area ratio interval is obtained, and the interval minimum values of each interval are arranged in ascending order to obtain the interval ascending order number; each repeat area ratio interval is assigned a repeat defect coefficient, whose value is greater than zero, and the value of each repeat defect coefficient increases with the increase of the interval ascending order number; matching the repeat area ratio value corresponding to the current adjacent acquisition point with each repeat area ratio interval, outputting the corresponding repeat area ratio interval and obtaining the corresponding repeat defect coefficient, and multiplying the repeat defect coefficient and the repeat area ratio value to obtain the repeat defect value.

[0059] The three values of the shielding defect value, the key area defect value and the repeated defect value are normalized and the values are taken, and a preset comprehensive calculation formula is used The comprehensive defect value Qzh is calculated and outputted; wherein, respectively represented as the shielding defect value, the key area defect value and the repeated defect value; is the shielding comprehensive defect value; are preset weight factors, and the values are greater than zero; the optimization instruction generation analysis is performed on the comprehensive defect value, and the cooperative optimization instruction is outputted;

[0060] The optimization instruction generation analysis is performed on the comprehensive defect value, and the specific analysis is that when the comprehensive defect value exceeds a preset comprehensive defect threshold value, the comprehensive defect value is identified, when the shielding comprehensive defect value and the repeated defect value are both not zero, the corresponding repeated area is marked as a to-be-optimized area signal based on the repeated defect value, the shielding comprehensive defect value is identified, if the key area defect value is not zero, the key detection point optimization instruction of the corresponding key area is generated, otherwise the corresponding shielding defect key point marking signal is marked;

[0061] When only the shielding comprehensive defect value is zero, the corresponding repeated area is marked as a to-be-optimized area signal based on the repeated defect value; the cooperative optimization instruction includes the to-be-optimized area signal, the key detection point optimization instruction and the shielding defect key point marking signal.

[0062] Step 4: detection execution, the detection optimization strategy is analyzed and generated based on the cooperative optimization instruction and executed, and the specific process is that the cooperative optimization instruction is identified to obtain the to-be-optimized area signal, the key detection point optimization instruction and the shielding defect key point marking signal; the repeated area is obtained based on the to-be-optimized area signal, the two adjacent detection points corresponding to the repeated area are marked as to-be-optimized targets, and step 2.1 is executed; the perspective shielding area is obtained based on the shielding defect key point marking signal, and step 2.1 is executed on the perspective shielding area; the key detection point of the corresponding key area is obtained based on the key detection point optimization instruction, the key detection point of the key area is marked as a to-be-optimized key point, and step 2.2 is executed.

[0063] Please refer to Figure 3 The application also provides a power inspection robot detection system based on machine vision, which comprises:

[0064] A visual perception module is configured to acquire detection dynamic data through a data perception unit (i.e., a visual acquisition unit, a motion perception unit and an environment perception unit) of the power inspection robot; the detection dynamic data comprises a definition parameter, motion perception data and environment perception data.

[0065] The visual acquisition unit collects to obtain the definition parameter corresponding to the preset frame number; the motion perception unit obtains the motion perception data of the power inspection robot, i.e. vibration acceleration and robot posture; and the environment perception unit collects to obtain the environment perception data (environmental wind speed and environmental light).

[0066] The environment adaptability module is configured to perform environment adaptability analysis on the detection dynamic data to obtain an environment adaptability control instruction.

[0067] The environment adaptability analysis on the detection dynamic data includes the following specific analysis steps:

[0068] The detection dynamic data is identified based on a data recognition technology to obtain the definition parameter, the motion perception data and the environment perception data.

[0069] The definition corresponding to each preset frame number is obtained based on the definition parameter, a preset definition reference value is obtained and compared with the definition corresponding to each frame to identify the decrease amplitude, when the definition is less than the definition reference value, the definition decrease amplitude is output, when the definition exceeds a preset decrease amplitude ratio, the corresponding frame number is marked as a mutation frame, the total number of mutation frames is counted, and the definition value and the total number of mutation frames are input into a preset visual fluctuation calculation formula to calculate a visual fluctuation value S; wherein, represents the definition corresponding to the i-th frame, and nt represents the total number of mutation frames. represents the preset frame number, . is the maximum definition within the preset frame number, is the maximum decrease amplitude of single-frame definition.

[0070] The vibration acceleration and the robot posture of a preset sample number are obtained based on the motion perception data, a preset acceleration safety threshold is obtained, the vibration acceleration of each sample is compared with the acceleration safety threshold, the sample exceeding the acceleration safety threshold is marked as an abnormal sample number, the abnormal sample number is compared with the preset sample number to calculate an acceleration abnormality ratio, the robot posture corresponding to each sample is compared with a preset standard posture to obtain a posture error angle, and the acceleration abnormality ratio and the posture error angle are input into a preset motion fluctuation quantification formula to calculate and output a motion fluctuation value Y; wherein, is the acceleration abnormality ratio, is the posture error sum of the preset sample number M, is the posture error angle of the m-th sample, are both preset weight factors, and the values are both in [0, 1].

[0071] The environment wind speed and the ambient light in a preset sampling period are obtained based on the environment perception data, a preset controllable wind speed threshold, a limit wind speed threshold and a light mutation threshold are obtained, a maximum environment wind speed in the sampling period is obtained and is marked as an instantaneous maximum wind speed, light mutation intensities corresponding to samples in the sampling period are obtained, samples exceeding the light mutation threshold are marked as light mutation samples, and a light mutation proportion is obtained by ratio calculation of the number of the light mutation samples and the total number of the samples in the sampling period; values of the instantaneous maximum wind speed and the light mutation proportion are input into a preset environment fluctuation quantification formula to calculate and output an environment fluctuation value H; respectively, the instantaneous maximum wind speed and the light mutation proportion; respectively, the limit wind speed threshold and the controllable wind speed threshold.

[0072] The visual fluctuation value, the motion fluctuation value and the environment fluctuation value are normalized and values thereof are taken, and are calculated to obtain a comprehensive fluctuation value Bo; wherein, is a harmonic mean of the three fluctuation values, used to emphasize the influence of a smaller value and avoid being pulled down by a single item, is a minimum value, used to avoid a zero denominator; is a weight of the superimposed maximum fluctuation value, used to highlight the dominant fluctuation factor, is a preset highlight weight factor, taking a value in [0, 1]; the comprehensive fluctuation value is divided into three fluctuation intervals, i.e., a low fluctuation interval, a medium fluctuation interval and a high fluctuation interval, according to a preset fluctuation value interval; when the comprehensive fluctuation value is in the low fluctuation interval, an environment adaptability control instruction output is to maintain a collection strategy; when in the medium fluctuation interval, the output instruction is to increase a shutter speed; when in the high fluctuation interval, a corresponding maximum fluctuation value is obtained, i.e., when the visual fluctuation value is maximum, the instruction is to increase the shutter speed, when the motion fluctuation value is maximum, the instruction is to start a wind-avoiding strategy (for example, pause action), and when the environment fluctuation value is maximum, the instruction is to reduce a moving speed and reset a robot posture.

[0073] A detection execution module is configured to input the environment adaptability control instruction into a robot control unit and perform control execution.

[0074] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limiting them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features thereof; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting a robot for power inspection based on machine vision, characterized in that, The method comprises the following steps: Step 1: model construction based on pre-acquired regional basic data to obtain a three-dimensional model of a target region, and marking a detection key region on the three-dimensional model of the target region; Step 2: path planning of the acquired three-dimensional model of the target region through a preset path planning analysis mechanism, and output of a visual detection path; Step 3: real-time acquisition of corresponding visual acquisition data based on the visual detection path for visual acquisition, input of the visual acquisition data into a path acquisition collaborative analysis module, and then output of a collaborative optimization instruction obtained; The specific execution steps of the path acquisition collaborative analysis module are as follows: The visual acquisition data is identified to obtain visual images of each acquisition point of the current corresponding detection region, the visual images are matched with the pre-stored equipment standard visual images corresponding to the detection region to obtain visual angle occlusion regions of the visual images, and meanwhile, the visual images corresponding to adjacent acquisition points are matched to obtain picture repetition regions; The visual angle occlusion regions are calculated with the equipment standard visual images to obtain visual angle occlusion proportion values, and the visual angle occlusion proportion values are compared with a preset occlusion proportion threshold value, and when the visual angle occlusion proportion values exceed the occlusion proportion threshold value, the visual angle occlusion proportion values are output as occlusion defect values; The visual angle occlusion regions are matched with the detection key regions corresponding to the equipment standard visual images, when the visual angle occlusion regions cover the detection key regions, the visual angle occlusion proportion values are directly output as occlusion defect values, a preset key missing multiple is acquired, and then the occlusion defect values and the key missing multiple are multiplied to obtain key region defect values; Based on the picture repetition regions, repetition region proportion values are obtained, the repetition region proportion values are divided into multiple continuous repetition region proportion intervals according to a preset proportion separation value, interval minimum values of each repetition region proportion interval are acquired, the interval minimum values of each interval are arranged in ascending order to obtain interval ascending serial numbers, each repetition region proportion interval is assigned a repetition defect coefficient, the repetition defect coefficient has a value greater than zero, and the value of each repetition defect coefficient increases with the increase of the interval ascending serial number; the repetition region proportion value corresponding to the current adjacent acquisition point is matched with each repetition region proportion interval, the corresponding repetition region proportion interval is output and the corresponding repetition defect coefficient is obtained, and the repetition defect coefficient and the repetition region proportion value are multiplied to obtain a repetition defect value; The occlusion defect values, the key region defect values and the repetition defect values are normalized and the values are taken, a preset comprehensive calculation formula is used for calculation and a comprehensive defect value is output; the comprehensive defect value is analyzed according to the optimization instruction to generate an analysis, and a collaborative optimization instruction is output; The comprehensive defect value is analyzed according to the optimization instruction to generate an analysis, and a collaborative optimization instruction is output; When the only occlusion comprehensive defect value is zero, the corresponding repeat area is marked as a to-be-optimized area signal based on the repeat defect value; the cooperative optimization instruction includes the to-be-optimized area signal, the key detection point optimization instruction and the occlusion defect key marking signal Step 4: generating a detection optimization strategy based on the cooperative optimization instruction analysis and executing The cooperative optimization instruction analysis generates a detection optimization strategy and executes, which is specifically: identifying the cooperative optimization instruction to obtain the to-be-optimized area signal, the key detection point optimization instruction and the occlusion defect key marking signal; obtaining a repeat area based on the to-be-optimized area signal, and recording two adjacent detection points corresponding to the repeat area as to-be-optimized targets, and executing step 2.1; Obtaining a perspective occlusion area based on the occlusion defect key marking signal, executing step 2.1 on the perspective occlusion area, obtaining a key detection point of a corresponding key area based on the key detection point optimization instruction, marking the key detection point as a to-be-optimized key point, and executing step 2.2; The path planning analysis mechanism specifically plans the following steps: Step 2.1: outputting an initial state path through initial path analysis; Step 2.2: obtaining a detection key area corresponding to each to-be-detected area and a to-be-optimized key point, and obtaining a corresponding preset key area detection standard, simulating natural selection on the detection key area, the to-be-optimized key point and the key area detection standard through population evolution of a genetic algorithm to output a key detection point; Step 2.3: superimposing the key detection point and the initial state path, marking a collection point that coincides with the key detection point as a redundant detection point, and removing the redundant detection point corresponding to the initial state path to obtain a visual detection path. 2.The machine vision-based power inspection robot detection method of claim 1, wherein, Based on the pre-collected area basic data, a target area three-dimensional model is constructed to mark a detection key area, which is specifically: Step 1.1, establishing a target area three-dimensional model: identifying the area basic data of the collection target area in the database to obtain the device parameters and geographic information of the inspection target of the target detection area; And obtaining the pre-scanning data of the laser radar, obtaining the structure model of each device in the inspection target area based on the device structure data of the pre-scanning data, and collecting the structure model of each device and the corresponding device parameters and geographic coordinates to obtain the target area three-dimensional model of the target area; Step 1.2, marking a detection key area: obtaining a preset key model comparison library, superimposing and matching each model corresponding to the target area three-dimensional model and the key model comparison library, and marking the model matched with the key model comparison library as a detection key area; the key model comparison library is a device defect high-risk position obtained through device historical defect occurrence data. 3.The method of claim 2, wherein, The initial state path is output by initial path analysis, specifically: a path optimization period is obtained based on a preset time interval division, a target area three-dimensional model of the current path optimization period is acquired, and each device model is obtained. The detection area, the corresponding visual angle blocking area and the optimization target are analyzed and output by the initial path of the population evolution simulation of the genetic algorithm to natural selection, the collection capability parameters of each power inspection robot are obtained, that is, the visual collection accuracy and the field of view angle of visual collection, the visual collection accuracy and the field of view angle are overlapped with the initial path, and the collection points corresponding to each detection area are obtained based on the coverage range of the field of view angle; the initial path and the collection points are overlapped to obtain the initial state path.

4. A machine vision-based power inspection robot detection system, characterized by, The visual perception module, the environmental adaptability module and the detection execution module are included, so that the power inspection robot detection system based on machine vision executes the power inspection robot detection method based on machine vision as claimed in any one of claims 1-3.

5. The machine vision-based power inspection robot detection system of claim 4, wherein, The visual perception module is used for collecting data by the data perception unit of the power inspection robot to obtain detection dynamic data; the detection dynamic data includes definition parameters, motion perception data and environment perception data; The environmental adaptability module is used for performing environmental adaptability analysis on the detection dynamic data to obtain environmental adaptability regulation instructions; The detection execution module is used for inputting the environmental adaptability regulation instructions into the robot control unit and performing control execution.

6. The machine vision-based power inspection robot detection system of claim 5, wherein, The environmental adaptability analysis on the detection dynamic data obtains environmental adaptability regulation instructions, and the specific analysis steps are as follows: The detection dynamic data is identified based on data recognition technology to obtain definition parameters, motion perception data and environment perception data; Based on the definition parameters, the definition corresponding to each preset frame number is obtained, a preset definition reference value is acquired and is compared with the definition of each frame to identify the definition drop, when the definition is less than the definition reference value, the definition drop value is output, when the definition drop value exceeds the preset drop ratio, the corresponding frame number is marked as a mutation frame, and the total number of mutation frames is counted; the definition value and the total number of mutation frames are input into a preset visual fluctuation calculation formula to obtain a visual fluctuation value; Based on the motion perception data, the vibration acceleration and the robot posture of a preset sample number are obtained, a preset acceleration safety threshold is acquired, the vibration acceleration of each sample is compared with the acceleration safety threshold, the samples exceeding the acceleration safety threshold are marked as abnormal sample numbers, and the abnormal sample numbers are compared with the preset sample number to obtain an acceleration abnormality ratio; The robot posture corresponding to each sample is compared with a preset standard posture to obtain an attitude error angle; The acceleration abnormality ratio and the attitude error angle are input into a preset motion fluctuation quantification formula to calculate and output a motion fluctuation value; Based on the environment perception data, the environmental wind speed and the environmental light in a preset sampling period are obtained, and a preset controllable wind speed threshold, a limit wind speed threshold and a light mutation threshold are obtained. The maximum environmental wind speed in the sampling period is obtained and marked as the instantaneous maximum wind speed. The light mutation intensity corresponding to each sample in the sampling period is obtained, and the samples exceeding the light mutation threshold are recorded as light mutation samples. The number of light mutation samples is calculated by ratio with the total number of samples in the sampling period to obtain the light mutation proportion. The values of the instantaneous maximum wind speed and the light mutation proportion are input into a preset environment fluctuation quantization formula for calculation to output an environment fluctuation value. The visual fluctuation value, the motion fluctuation value and the environment fluctuation value are normalized and the values are taken, and the comprehensive fluctuation value is calculated by using. The comprehensive fluctuation value is divided into three fluctuation intervals, i.e. low fluctuation interval, medium fluctuation interval and high fluctuation interval, according to a preset fluctuation value interval. The current corresponding comprehensive fluctuation value is matched with each fluctuation interval. When the comprehensive fluctuation value is in the low fluctuation interval, the output environmental adaptability control instruction is to keep the acquisition strategy. When in the medium fluctuation interval, the output instruction is to increase the shutter speed. When in the high fluctuation interval, the corresponding maximum fluctuation value is obtained, i.e. the instruction is to increase the shutter speed when the visual fluctuation value is maximum, the instruction is to start the wind avoidance strategy when the motion fluctuation value is maximum, and the instruction is to reduce the moving speed and reset the machine posture when the environment fluctuation value is maximum.

Images