An inspection robot assisted positioning method

Abstract

The application provides an inspection robot auxiliary positioning method, which is based on a system comprising a robot, a collection module installed on the robot and a jetson module. In the method of the application, a positioning position identification pattern is arranged on the moving track of the inspection robot, then the position identification pattern is identified based on a lightweight model of deep learning, and finally the position coordinate information is returned to the robot system. When the existing positioning system loses positioning, it can be adjusted. Moreover, the position identification pattern is composed of circular reflective stickers, and the circular reflective stickers are composed of solid and hollow circular stickers. Compared with the feature description point method, the position error caused by similar feature description points in a similar scene can be avoided, and the processor's algorithm burden can be reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to a kind of inspection robot auxiliary positioning method. BACKGROUND

[0002] When existing inspection robot executes task, system usually needs to be set as inspection path according to map in advance to go to task point, and the system obtains the data of odometer installed in the tire of the inspection robot as positioning information when the inspection robot moves, if obstacle is encountered when going to task point, the obstacle is detected by laser detector arranged on the vehicle body and the pose is adjusted to avoid the obstacle, in this case, the data recorded by odometer is different in the case of no obstacle and the case of having obstacle in a section, so it is inaccurate to still use the odometer data without obstacle to determine the positioning information in the case of having obstacle, at this time, in the prior art, when designing a route to a target task point, the system usually collects obstacles in the route first, and then adjusts the data collected by the odometer when the robot avoids the obstacle, and the data is used as positioning information, however, in the actual process, the obstacles in the route may change (for example, the position and size may change), if the adjustment standard of the pre-collected case is still used to adjust the positioning information at this time, the result is often inaccurate, in addition, if wheel slip occurs in the journey, it will also cause the odometer to be inaccurate, and thus the positioning information is inaccurate.

[0003] The present application uses auxiliary positioning technology to correct the positioning of the robot based on the existing positioning method, specifically, when the auxiliary positioning technology obtains the positioning information, the robot system modifies the existing positioning information to the positioning information obtained by the auxiliary positioning technology, and the present application provides an auxiliary positioning method for inspection robot. SUMMARY

[0004] To achieve the above purpose, the present application provides an auxiliary positioning method for inspection robot, and the specific technical solutions are as follows:

[0005] A positioning method for inspection robot, which is based on a system including a robot, a collection module installed on the robot, and a jetson module, and includes the following steps:

[0006] S1: the collection module collects position identification pattern information arranged on the inspection path and transmits the information to the jetson module; wherein the position identification pattern information is multiple, distributed along the inspection path, and the position identification pattern information records the position information thereof;

[0007] S2: the jetson module processes the position identification pattern information based on a lightweight model of convolutional neural network to obtain the category serial number of the position identification pattern.

[0008] S3: The jetson module obtains the robot coordinate position (x, y) from the pre-stored category file based on the category serial number.

[0009] S4: The jetson module obtains the contour points of the position identification pattern, sorts the contour points of the whole image by area, obtains the contour with the largest area, obtains the pixel width P of the contour, and combines the focal length F of the collection module and the actual width W of the position identification pattern to calculate the distance D between the robot and the position identification pattern.

[0010] S5: The preliminary angle is obtained according to the function of the target contour, and the final angle angle is obtained by using the length-width WH determination principle, if W is greater than H, angle is 180°-preliminary angle relative to the horizontal x-axis direction, otherwise, angle is 90°-preliminary angle relative to the horizontal x-axis direction; finally, the position coordinates (x, y, z, angle) are obtained, x is the x-axis coordinate, y is the y-axis coordinate, z is the distance d coordinate, and angel is the rotation angle of the position identification pattern relative to the positive direction of the robot.

[0011] Further, the position identification pattern information is composed of circular reflective stickers, the circular reflective stickers are composed of solid and hollow circular stickers, and different position identification patterns have different positions of the circular reflective stickers.

[0012] Further, the lightweight model of the convolutional neural network is a MobileNetV3 model, the parameter amount of the model is 5.14MB, the whole network is composed of one convolutional layer, 13 Bneck_conf blocks and two fully connected layers; the position identification pattern enters from the first layer of the network, then the convolution operation of each layer is performed to obtain the final feature map, then the final feature map is classified and scored through the fully connected layer, the class with the highest score is obtained and the category serial number is output.

[0013] Further, the contour points are obtained by sequentially performing image color space conversion, Gaussian filter noise removal and edge detection processing using a canny operator on the position identification pattern.

[0014] Further, the function of the target contour is an imutils.grab_contours function.

[0015] The beneficial effects of the present application are that the positioning position identification pattern is arranged on the moving track of the inspection robot, then the position identification pattern is recognized based on the lightweight model of deep learning, and finally the position coordinate information is returned to the robot system, and when the existing positioning system loses positioning, it can be adjusted. DETAILED DESCRIPTION​

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application.

[0017] In the present application, the terms "upper", "lower", "left", "right", "front", "back", "top", "bottom", "inner", "outer", "middle", "vertical", "horizontal", "lateral", "longitudinal", and the like indicate the orientation or positional relationship. These terms are mainly used for better description of the present application and its embodiments, and are not used to limit the indicated devices, elements or components to have a specific orientation, or to be constructed and operated in a specific orientation.

[0018] In addition, the above-mentioned partial terms may be used to represent other meanings in addition to the orientation or positional relationship, for example, the term "upper" may also be used to represent a certain dependent relationship or connection relationship in some cases. Those skilled in the art can understand the specific meaning of these terms in the present application according to the specific situation.

[0019] In addition, the terms "mount", "set", "provided with", "connect", "connected" should be understood broadly. For example, it can be fixedly connected, detachably connected, or integrally constructed; it can be mechanically connected, or electrically connected; it can be directly connected, or indirectly connected through an intermediate medium, or internal communication between two devices, elements or components. Those skilled in the art can understand the specific meaning of the above terms in the present application according to the specific situation.

[0020] In addition, the terms "first", "second", and the like are mainly used to distinguish different devices, elements or components (the specific types and structures may be the same or different), and are not used to indicate or imply the relative importance and quantity of the indicated devices, elements or components. Unless otherwise specified, the meaning of "multiple" is two or more.

[0021] It should also be understood that the terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification and the appended claims of the present application, unless otherwise clear from the context, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0022] It should be further understood that the term "and / or" used in the description and claims of the application means one or more of the associated listed items as well as all possible combinations of the items and includes these combinations.

[0023] The present application provides a positioning method for a patrol robot, which is based on a system comprising a robot, a collection module installed on the robot, and a jetson module, wherein the method comprises the following steps:

[0024] S1: The collection module collects position identification pattern information arranged on a patrol path and transmits the information to the jetson module; wherein there are multiple position identification pattern information, which are distributed along the patrol path, and the position identification pattern information corresponds to record the position information where it is located;

[0025] Specifically, the collection module is an infrared camera collection module, and the position identification pattern is composed of circular reflective stickers, which are composed of solid and hollow circular stickers. Different position identification patterns have different positions of the circular reflective stickers. The infrared camera collection module emits infrared light to the position identification pattern, and the position identification pattern reflects the light to the infrared camera collection module for collection after receiving the light.

[0026] S2: The jetson module processes the position identification pattern information based on a lightweight model of a convolutional neural network to obtain a category serial number of the position identification pattern;

[0027] Specifically, the lightweight model of the convolutional neural network is a MobileNetV3 model, the parameter amount of the model is 5.14MB, and the entire network is composed of one convolutional layer, 13 Bneck_conf blocks (depthwise separable convolution), and two fully connected layers. The position identification pattern enters from the first convolutional layer, passes through the 13 Bneck_conf blocks in turn, obtains the final feature map, and then classifies and scores the final feature map through the fully connected layer to obtain the highest score of one category and output the category serial number.

[0028] S3: The jetson module obtains the robot coordinate position (x, y) from a pre-stored category file based on the category serial number;

[0029] S4: The jetson module obtains the contour points of the position identification pattern based on the position identification pattern, sorts the contour points of the entire image by area, obtains the contour with the largest area, obtains the pixel width P of the contour, and combines the focal length F of the collection module and the actual width W of the position identification pattern to calculate the distance D between the robot and the position identification pattern: ;

[0030] Specifically, the acquisition of the contour point is to sequentially perform image color space conversion, Gaussian filtering to remove noise, and edge detection processing using a Canny operator on the position identification pattern.

[0031] S5: The initial angle is obtained by solving the target contour, inputting a binary image of a label pattern, then finding each contour in the label image through the imutils.grab_contours function, sorting the contour areas, taking the maximum contour as the target contour, and then taking the angle value of the target contour to obtain the initial angle. The final angle angle is obtained by using the length-width WH determination principle, wherein H is the height of the target contour, if W is greater than H, angle is 180°-initial angle relative to the horizontal x-axis direction, otherwise, angle is 90°-initial angle relative to the horizontal x-axis direction; is the final obtained position coordinate (x, y, z, angle), x is the x-axis coordinate, y is the y-axis coordinate, z is the z-axis in the coordinate system, the value taken is the distance D value, and angel is the rotation angle of the position identification pattern relative to the positive direction of the robot.

[0032] Preferably, the function of the target contour is the imutils.grab_contours function, specifically, a binary image of a position identification pattern, then finding each contour in the position identification pattern through the imutils.grab_contours function, sorting the contour areas, taking the maximum contour as the target contour, and then taking the angle value of the target contour to obtain the initial angle.

[0033] In the present application, the following advantages exist:

[0034] The position identification pattern is arranged on the moving track of the inspection robot, then the position identification pattern is identified based on the lightweight model of deep learning, and finally the position coordinate information is returned to the robot system. When the existing positioning system loses positioning, it can be adjusted. Moreover, the position identification pattern is composed of circular reflective stickers, and the circular reflective stickers are composed of solid and hollow circular stickers. Compared with the feature description point method, the position error caused by similar feature description points in a similar scene can be avoided, and the processor's algorithm burden can be reduced.

[0035] Compared with the traditional template matching method, the label category recognition method based on deep learning has higher robustness. The traditional matching method cannot solve the matching problem of label images obtained at different angles and distances of the camera, and needs to spend a lot of preparation work to establish a template library, and after changing different indoor scenes, the position identification patterns in the template library need to be changed or increased, which cannot adapt to different indoor scenes, and the practicability is very low. The label category recognition based on deep learning can adapt to different indoor scenes with a trained model, and does not need to change the model. Moreover, the model based on deep learning can well learn the position identification patterns with differences obtained at different angles and distances, has higher generalization, and has higher recognition accuracy.

[0036] The mobileNetv3 lightweight model is adopted, the model parameter quantity is reduced by 15% compared with the original model, the running speed is increased by 9%, and the model is more suitable for edge deployment, and the resource occupation of hardware is reduced.

[0037] The jetson edge chip is used as the main core processor, has high flexibility, and has certain computing power support, and can meet the real-time detection demand of the inspection robot.

[0038] It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other to obtain new embodiments without conflict.

[0039] The above is only a preferred embodiment of the present application, and does not limit the present application in any form, and the protection scope of the present application should be based on the protection scope of the claims. Although the present application has been disclosed as above, it is not intended to limit the present application, and any person skilled in the art can make some changes or modifications to the equivalent embodiments with the disclosed technical content without departing from the scope of the technical scheme of the present application. Any simple modification, equivalent change and modification made on the basis of the technical essence of the present application to the above embodiments are still within the scope of the technical scheme of the present application.

Claims

1. A method for positioning a patrol robot, the method being based on a system comprising a robot, a collecting module mounted on the robot, a jetson module, characterized in that, The method comprises the following steps: S1: The collection module collects position identification pattern information arranged on the inspection path and transmits the information to the jetson module; wherein the position identification pattern information is multiple, is distributed along the inspection path, and the position identification pattern information corresponds to record the position information where it is located; S2: The jetson module processes the position identification pattern information based on a light weight model of convolutional neural network to obtain a category serial number of the position identification pattern; the light weight model of convolutional neural network is a MobileNetV3 model, the parameter quantity of the model is 5.14MB, the whole network is composed of one convolutional layer, 13 Bneck_conf blocks and two fully connected layers; the position identification pattern enters from the first layer of the network, then is processed through convolution operation of each layer to obtain a final feature map, then the final feature map is classified and scored through the fully connected layer, the highest score of one category is obtained and the category serial number is outputted; S3: The jetson module obtains the robot coordinate position (x, y) from the pre-stored category file based on the category serial number; S4: the jetson module acquires the contour points of the position identification pattern, sorts the contour points of the whole image by area, acquires the contour with the largest area, acquires the pixel width P of the contour, combines the focal length F of the acquisition module, the actual width W of the position identification pattern, and calculates the distance D between the robot and the position identification pattern: ; S5: The preliminary angle is obtained according to the function of the target contour, the final angle angle is obtained by using the WH determination principle, H is the height of the target contour, if W is greater than H, angle is 180°-preliminary angle relative to the horizontal x axis direction, otherwise, angle is 90°-preliminary angle relative to the horizontal x axis direction; finally, the position coordinate (x, y, z, angle) is obtained, x is the x axis coordinate, y is the y axis coordinate, z is the z axis in the coordinate system, the value is the value of the distance D, and angel is the rotation angle of the position identification pattern relative to the positive direction of the robot.

2. The method of claim 1, wherein: The position identification pattern information is composed of circular reflective stickers, the circular reflective stickers are composed of solid and hollow circular stickers, and different position identification patterns have different positions of the circular reflective stickers.

3. The method of claim 1, wherein: The contour point is obtained by sequentially performing image color space conversion, Gaussian filter noise removal and edge detection processing using a canny operator on the position identification pattern.

4. The method of claim 1, wherein: The function of the target contour is an imutils.grab_contours function.