A positioning method, device, medium and robot based on laser semantics

Abstract

The application provides a laser semantic-based positioning method. Laser semantic elements including actual electric poles and electric wires are extracted from each laser point cloud frame, and then only the extracted semantic element point cloud is used for registration positioning, so that the positioning speed and positioning accuracy are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robots, in particular to a positioning method and device based on laser semantics, a medium and a robot. BACKGROUND

[0002] Autonomous mobile robots require the robot to have the ability to walk autonomously, and the prerequisite for achieving this ability is that the robot knows its own location. Therefore, the positioning technology of autonomous mobile robots is one of the hot technologies in recent years. The positioning methods widely used in autonomous navigation mobile robots at present include laser positioning method, GPS positioning method, visual positioning method, etc. The patrol robot is a specific application of the automatic mobile robot, mainly used for power station patrol, photographing, video recording and other work. Compared with ordinary autonomous mobile robots, the patrol robot is equipped with an integrated gimbal camera device on the robot body.

[0003] Because the laser radar has high precision and can work all-weather compared with the camera, the patrol robot currently adopts the laser positioning method. The patrol robot is mainly used in the substation scene, and is required to automatically reach the specified patrol point and take pictures and recognize the dial / indicator light, and to alarm the abnormal dial and indicator light. In order to achieve high recognition rate, the positioning accuracy of the patrol robot is required to be relatively high, otherwise the dial may be not fully or not well taken, thereby affecting the recognition accuracy. The registration accuracy of ordinary laser point cloud is not easy to meet the accuracy requirement, and it is also time-consuming.

[0004] At present, the positioning technology of the patrol robot is basically based on laser radar, and the registration between the laser point cloud frames before and after the movement (all points emitted by the multi-line radar once) is directly used to obtain the pose (displacement and angle) of the robot before and after the movement. Directly using point cloud frame registration will increase the calculation, and at the same time, due to the sparseness of the substation scene (usually only power poles, lines, meter boxes, etc.), the registration accuracy of the laser point cloud frame directly cannot meet the requirements of the substation patrol robot.

[0005] The background description provided herein is for the purpose of generally presenting the context of the disclosure. The subject matter of the background description is not prior art to the present application and is not admitted to be prior art by inclusion in this section. SUMMARY

[0006] In view of the above technical problems in the related art, the present application provides a positioning method based on laser semantics, which comprises the following steps:

[0007] S1, acquiring any two frames of laser radar point cloud data in the process of robot walking, denoted as the first frame Scan_i and the second frame Scan_j;

[0008] S2, input the first frame Scan_i and the second frame Scan_j into the pole semantic extraction network respectively, and perform point cloud semantic extraction on Scan_i and Scan_j to obtain point clouds of all poles;

[0009] S3, input the first frame Scan_i and the second frame Scan_j into the wire semantic extraction network respectively, and perform point cloud semantic extraction on Scan_i and Scan_j to obtain point clouds of all wires;

[0010] S4, combine the pole point clouds and the wire point clouds of the first frame as the semantic point clouds of the first frame, combine the pole point clouds and the wire point clouds of the second frame as the semantic point clouds of the second frame, take the semantic point clouds of the first frame as the source point clouds of an iterative closest point (ICP) point cloud registration algorithm, take the semantic point clouds of the second frame as the target point clouds of the ICP point cloud registration algorithm, and use the ICP point cloud registration algorithm to obtain the motion pose of the robot.

[0011] Specifically, step S2 further includes:

[0012] S21, perform up-down clustering on the input first frame Scan_i laser point cloud, randomly select a point P1 within a first preset height, take the point P1 as a center point, construct a first length-width-height region, if the number of points in the first length-width-height region is greater than a first threshold value, retain all points in the first length-width-height region, and mark the first length-width-height region as a first candidate pole A_i;

[0013] S22, perform up-down clustering on the input second frame Scan_j laser point cloud, randomly select a point P2 within a first preset height, take the point P2 as a center point, construct a first length-width-height region, if the number of points in the first length-width-height region is greater than a first threshold value, retain all points in the first length-width-height region, and mark the first length-width-height region as a second candidate pole A_j;

[0014] S23, use the (Z, X) data of all point clouds in the first candidate pole A_i and the second candidate pole A_j to perform two-dimensional straight line fitting through least squares, when the error after two-dimensional straight line fitting is less than a third threshold value, the first candidate pole A_i or the second candidate pole A_j is a pole, otherwise, discard the point cloud in the first candidate pole A_i or the second candidate pole A_j, and obtain the pole point cloud of the first frame or / and the pole point cloud of the second frame.

[0015] Specifically, step S3 includes:

[0016] S31, the first frame Scan_i laser point cloud input is clustered up and down, a point p is randomly selected, a second length-width-height region is constructed with the point p as the center point, if the number of points in the second length-width-height region is greater than a second threshold value, all points in the second length-width-height region are retained and marked as a first candidate wire B_i;

[0017] S32, the second frame Scan_i laser point cloud input is clustered up and down, a point p is randomly selected, a second length-width-height region is constructed with the point p as the center point, if the number of points in the second length-width-height region is greater than a second threshold value, all points in the second length-width-height region are retained and marked as a second candidate wire B_j;

[0018] S33, all point clouds in the first candidate wire B_i and the second candidate wire B_j are respectively fitted by least squares using (Z, X) data of the points, when the error after two-dimensional straight line fitting is less than a fourth threshold value, the first candidate wire B_i or the second candidate wire B_j is a wire, otherwise, the point cloud in the first candidate wire B_i or the second candidate wire B_j is discarded, and the wire point cloud of the first frame and / or the wire point cloud of the second frame is obtained;

[0019] Specifically, the third threshold value is 0.2*n, wherein n is all point clouds in the first candidate wire pole A_i or the second candidate wire pole A_j.

[0020] Specifically, the fourth threshold value is 0.1*n, wherein n is all point clouds in the first candidate wire B_i or the second candidate wire B_j.

[0021] In a second aspect, another embodiment of the present application discloses a positioning device based on laser semantics, which comprises the following units:

[0022] An original laser point cloud frame acquisition unit is configured to acquire any two frames of laser radar point cloud data in the process of robot walking, denoted as a first frame Scan_i and a second frame Scan_j.

[0023] A wire pole semantic extraction unit is configured to input the first frame Scan_i and the second frame Scan_j into a wire pole semantic extraction network respectively, and perform point cloud semantic extraction on Scan_i and Scan_j to obtain point clouds of all wire poles.

[0024] A wire extraction unit is configured to input the first frame Scan_i and the second frame Scan_j into a wire semantic extraction network respectively, and perform point cloud semantic extraction on Scan_i and Scan_j to obtain point clouds of all wires.

[0025] The pose acquisition unit is configured to merge the power pole point cloud and the power line point cloud of the first frame as a first frame semantic point cloud, merge the power pole point cloud and the power line point cloud of the second frame as a second frame semantic point cloud, and use the ICP point cloud registration algorithm to acquire the motion pose of the robot by taking the first frame semantic point cloud as a source point cloud of the ICP point cloud registration algorithm and taking the second frame semantic point cloud as a target point cloud of the ICP point cloud registration algorithm.

[0026] Specifically, the power pole semantic extraction unit further includes:

[0027] The first frame power pole extraction unit is configured to perform up-down clustering on the input first frame Scan_i laser point cloud, randomly select a point P1 within a first preset height, construct a first length-width-height region with the point P1 as a center point, and if the number of points in the first length-width-height region is greater than a first threshold value, retain all points in the first length-width-height region and mark the points as a first candidate power pole A_i.

[0028] The second frame power pole extraction unit is configured to perform up-down clustering on the input second frame Scan_j laser point cloud, randomly select a point P2 within a first preset height, construct a first length-width-height region with the point P2 as a center point, and if the number of points in the first length-width-height region is greater than a first threshold value, retain all points in the first length-width-height region and mark the points as a second candidate power pole A_j.

[0029] The power pole confirmation unit is configured to perform two-dimensional straight line fitting on all point clouds in the first candidate power pole A_i and the second candidate power pole A_j by using (Z, X) data of the points through least squares, and when an error after the two-dimensional straight line fitting is less than a third threshold value, the first candidate power pole A_i or the second candidate power pole A_j is a power pole, otherwise, discard the point cloud in the first candidate power pole A_i or the second candidate power pole A_j, and acquire the power pole point cloud of the first frame and / or the power pole point cloud of the second frame.

[0030] Specifically, the power line extraction unit further includes:

[0031] The first frame power line extraction unit is configured to perform up-down clustering on the input first frame Scan_i laser point cloud, randomly select a point p, construct a second length-width-height region with the point p as a center point, and if the number of points in the second length-width-height region is greater than a second threshold value, retain all points in the second length-width-height region and mark the points as a first candidate power line B_i.

[0032] The second frame wire extraction unit is used for up-down clustering of the input second frame Scan_i laser point cloud, a point p is randomly selected, a second length-width-height region is constructed with the point p as the center point, if the number of points in the second length-width-height region is greater than a second threshold value, all points in the second length-width-height region are retained and marked as a second candidate wire B_j;

[0033] The wire confirmation unit is used for two-dimensional straight line fitting of (Z, X) data of all point clouds in the first candidate wire B_i and the second candidate wire B_j through least squares, when the error after the two-dimensional straight line fitting is less than a fourth threshold value, the first candidate wire B_i or the second candidate wire B_j is a wire, otherwise, the point cloud in the first candidate wire B_i or the second candidate wire B_j is discarded, and the wire point cloud of the first frame and / or the wire point cloud of the second frame is obtained.

[0034] Specifically, the third threshold value is 0.2*n, where n is all point clouds in the first candidate wire pole A_i or the second candidate wire pole A_j.

[0035] Specifically, the fourth threshold value is 0.1*n, where n is all point clouds in the first candidate wire B_i or the second candidate wire B_j.

[0036] In a third aspect, another embodiment of the present application discloses a robot, which comprises a central processor, a memory, and a laser radar, the memory stores instructions, and the processor executes the instructions to implement the above-mentioned laser semantic positioning method.

[0037] In a fourth aspect, another embodiment of the present application discloses a non-volatile memory, which stores instructions, and the processor executes the instructions to implement the above-mentioned laser semantic positioning method.

[0038] The present application extracts laser semantic elements including actual wire poles and wires from each laser point cloud frame, and then only uses the extracted semantic element point cloud for registration and positioning, so that the positioning speed and positioning accuracy are improved. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0040] Figure 1 It is a laser semantic positioning method flowchart provided by an embodiment of the present application.

[0041] Figure 2 is a power pole extraction method flowchart provided by an embodiment of the present application;

[0042] Figure 3 is a power pole extraction method flowchart provided by an embodiment of the present application;

[0043] Figure 4 is a positioning device schematic diagram based on laser semantics provided by an embodiment of the present application;

[0044] Figure 5 is a power pole extraction unit schematic diagram provided by an embodiment of the present application;

[0045] Figure 6 is a power pole extraction unit schematic diagram provided by an embodiment of the present application;

[0046] Figure 7 is a positioning device schematic diagram based on laser semantics provided by an embodiment of the present application. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present application. Embodiment one

[0048] Reference Figure 1 The embodiment discloses a positioning method based on laser semantics, which comprises the following steps:

[0049] S1, acquiring any two frames of laser radar point cloud data in the process of robot walking, denoted as a first frame Scan_i and a second frame Scan_j;

[0050] The robot of the embodiment is configured with a laser radar, and the robot can autonomously navigate according to the laser radar when working. Specifically, the robot of the embodiment can autonomously walk according to a planned route when patrolling, or can manually control the robot to walk, for example, receive control instructions of a control personnel. Preferably, the robot of the embodiment generally autonomously navigates according to a planned route.

[0051] The two frames of laser radar point cloud data of the embodiment are two frames of continuous laser point cloud data. In one embodiment, the time of acquiring the first frame Scan_i is earlier than the time of acquiring the second frame Scan_j. In another embodiment, the time of acquiring the first frame Scan_i is later than the time of acquiring the second frame Scan_j.

[0052] The robot of the embodiment is used for patrol of a power station. The robot for power station patrol usually only has power poles, point lines and electric meter boxes due to the sparseness of the power station scene; and it is difficult to meet the requirements of the power station patrol robot by directly using the laser point cloud frame registration accuracy.

[0053] S2, input the first frame Scan_i and the second frame Scan_j into the power pole semantic extraction network respectively, perform point cloud semantic extraction on Scan_i and Scan_j to obtain point clouds of all power poles;

[0054] The embodiment first uses the power pole semantic extraction network to extract and process all power poles in the laser point cloud.

[0055] Specifically, referring to Figure 2 The step S2 of the embodiment further includes:

[0056] S21, perform up-down clustering on the input first frame Scan_i laser point cloud, randomly select a point P1 within a first preset height, construct a first length-width-height region with the point P1 as the center point, if the number of points in the first length-width-height region is greater than a first threshold value, retain all points in the first length-width-height region and mark them as a first candidate power pole A_i;

[0057] Specifically, the first preset height of the embodiment is 3m;

[0058] For the power pole of the power plant, it is an elongated object, and it is also an elongated object in the laser point cloud data. Therefore, the embodiment sets a first length-width-height of 1m*1m*6m for detecting the candidate power pole. Specifically, the first threshold value is 100.

[0059] S22, perform up-down clustering on the input second frame Scan_j laser point cloud, randomly select a point P2 within the first preset height, construct a first length-width-height region with the point P2 as the center point, if the number of points in the first length-width-height region is greater than the first threshold value, retain all points in the first length-width-height region and mark them as a second candidate power pole A_j;

[0060] S23, use the (Z, X) data of all point clouds in the first candidate power pole A_i and the second candidate power pole A_j to perform two-dimensional straight line fitting by least squares, when the error after two-dimensional straight line fitting is less than a third threshold value, the first candidate power pole A_i or the second candidate power pole A_j is a power pole, otherwise discard the point cloud in the first candidate power pole A_i or the second candidate power pole A_j, and obtain the power pole point cloud of the first frame and / or the power pole point cloud of the second frame;

[0061] A specific third threshold value is 0.2*n, where n is all point clouds in the first candidate power pole A_i or the second candidate power pole A_j.

[0062] For all point clouds in the first candidate power pole A_i and the second candidate power pole A_j, a two-dimensional straight line fitting is performed by least squares using (Z, X) data of points

[0063] z= Kx +B

[0064] Calculate the error after fitting, where n represents all point clouds in the first candidate power pole A_i and the second candidate power pole A_j; when the error is less than 0.2*n, the first candidate power pole A_i or the second candidate power pole A_j is a power pole, otherwise discard the point cloud in the first candidate power pole A_i or the second candidate power pole A_j.

[0065] where Z represents the value of the point in the robot height direction, X represents the value of the point in the robot forward direction, K and B represent the slope and intercept of the fitted straight line respectively, is the measurement value of the i-th point in the Z direction, is the measurement value of the i-th point in the X direction.

[0066] S3, input the first frame Scan_i and the second frame Scan_j into the power line semantic extraction network respectively, and perform point cloud semantic extraction on Scan_i and Scan_j to obtain all power line point clouds;

[0067] Specifically, referring to Figure 3 The step S3 of the embodiment includes:

[0068] S31, perform up-down clustering on the input first frame Scan_i laser point cloud, randomly select a point p, take the point p as the center point, construct a second length-width-height region, if the number of points in the second length-width-height region is greater than a second threshold value, retain all points in the second length-width-height region and mark them as the first candidate power line B_i;

[0069] Specifically, the second length-width-height of the embodiment is a 3m*3m*1m region.

[0070] The second threshold value is 30.

[0071] Specifically, the first preset height of the embodiment is 3m;

[0072] Specifically, because the height of the power line of the power station is generally about 3m, in order to remove the interference of irrelevant objects, before randomly selecting a point p in step S31, points with a height lower than 2.5m are removed.

[0073] S32, the second frame Scan_i laser point cloud input is clustered up and down, a point p is randomly selected, a second length-width-height region is constructed with the point p as the center point, if the number of points in the second length-width-height region is greater than a second threshold value, all points in the second length-width-height region are retained and marked as a second candidate power line B_j;

[0074] S33, all point clouds in the first candidate power line B_i and the second candidate power line B_j are respectively fitted by least squares using (Z, X) data of the points to obtain a two-dimensional straight line, when the error after two-dimensional straight line fitting is less than a fourth threshold value, the first candidate power line B_i or the second candidate power line B_j is a power line, otherwise, the point cloud in the first candidate power line B_i or the second candidate power line B_j is discarded, and the power line point cloud of the first frame and / or the power line point cloud of the second frame is obtained.

[0075] The fourth threshold value is specifically 0.1*n, wherein n is all point clouds in the first candidate power line B_i or the second candidate power line B_j.

[0076] All point clouds in the first candidate power line B_i and the second candidate power line B_j are respectively fitted by least squares using (Z, X) data of the points to obtain a two-dimensional straight line

[0077] z= Kx +B

[0078] The error after fitting is calculated, Wherein n represents all point clouds in the first candidate power line B_i and the second candidate power line B_j; when the error err is less than 0.1*n, the first candidate power line B_i or the second candidate power line B_j is a power line, otherwise, the point cloud in the first candidate power line B_i or the first candidate power line B_j is discarded.

[0079] Wherein Z represents the value of the point in the robot height direction, X represents the value of the point in the robot forward direction, K and B respectively represent the slope and intercept of the fitted straight line, is the measurement value of the i-th point in the Y direction, is the measurement value of the i-th point in the X direction.

[0080] S4, the power line pole point cloud and the power line point cloud of the first frame are merged as the first frame semantic point cloud, the power line pole point cloud and the power line point cloud of the second frame are merged as the second frame semantic point cloud; and the first frame semantic point cloud is used as a source point cloud source of an iterative closest point algorithm ICP point cloud registration algorithm, the second frame semantic point cloud is used as a target point cloud target of the ICP point cloud registration algorithm, and the motion pose of the robot is obtained by using the ICP point cloud registration algorithm.

[0081] The iterative closest point algorithm ICP is the most commonly used point cloud registration. Point cloud registration refers to inputting two point clouds P s(source) and P t (target), output a transform T, such that the degree of coincidence of T(P s ) and T(P t ) is as high as possible, thereby estimating the pose. The general algorithmic flow of ICP is:

[0082] point cloud preprocessing; 2. matching; 3. weighting; 4. eliminating unreasonable corresponding point pairs; 5. calculating loss; 6. minimizing loss to solve the current optimal transform; 7. returning to step 2 for iteration until convergence.

[0083] The ICP registration algorithm of the embodiment is the existing technology, and the embodiment will not be described in detail.

[0084] The embodiment extracts laser semantic elements including actual power poles and power lines for each laser point cloud frame, and then only performs registration positioning through the extracted semantic element point cloud, thereby improving positioning speed and positioning accuracy. Embodiment Two

[0085] Reference Figure 4 , the embodiment discloses a positioning device based on laser semantics, which comprises the following units:

[0086] An original laser point cloud frame acquisition unit is configured to acquire any two frames of laser radar point cloud data in the process of robot walking, denoted as a first frame Scan_i and a second frame Scan_j.

[0087] The robot of the embodiment is provided with a laser radar, and the robot can perform autonomous navigation according to the laser radar when working. Specifically, the robot of the embodiment can autonomously walk according to a planned route when performing patrol, or the robot can be manually controlled to walk, for example, to receive control instructions of a control personnel. Preferably, the robot of the embodiment generally autonomously navigates according to a planned route.

[0088] The two frames of laser radar point cloud data of the embodiment are two frames of continuous laser point cloud data. In one embodiment, the first frame Scan_i is acquired earlier than the second frame Scan_j. In another embodiment, the first frame Scan_i is acquired later than the second frame Scan_j.

[0089] The robot of the embodiment is used for patrol of a power station. The robot for power station patrol is usually provided with only power poles, point lines, and electric meter boxes due to the sparseness of the power station scene, and it is difficult to meet the requirements of the power station inspection robot by directly using the laser point cloud frame registration accuracy.

[0090] The power pole semantic extraction unit is configured to input the first frame Scan_i and the second frame Scan_j into a power pole semantic extraction network respectively, and perform point cloud semantic extraction on Scan_i and Scan_j to obtain point clouds of all power poles.

[0091] The power pole semantic extraction unit is configured to input the first frame Scan_i and the second frame Scan_j into a power pole semantic extraction network respectively, and perform point cloud semantic extraction on Scan_i and Scan_j to obtain point clouds of all power poles.

[0092] Specifically, referring to Figure 5 The power pole semantic extraction unit of the embodiment further includes the following units:

[0093] The first frame power pole extraction unit is configured to perform up-down clustering on the input first frame Scan_i laser point cloud, randomly select a point P1 within a first preset height, construct a first length-width-height region with the point P1 as the center point, and if the number of points in the first length-width-height region is greater than a first threshold, retain all points in the first length-width-height region and mark them as a first candidate power pole A_i.

[0094] Specifically, the first preset height of the embodiment is 3 m.

[0095] For the power pole of a power plant, it is an elongated object, and it is also an elongated object in the laser point cloud data. Therefore, the embodiment sets a first length-width-height of 1 m*1 m*6 m for detecting the candidate power pole. Specifically, the first threshold is 100.

[0096] The second frame power pole extraction unit is configured to perform up-down clustering on the input second frame Scan_j laser point cloud, randomly select a point P2 within the first preset height, construct a first length-width-height region with the point P2 as the center point, and if the number of points in the first length-width-height region is greater than the first threshold, retain all points in the first length-width-height region and mark them as a second candidate power pole A_j.

[0097] The power pole confirmation unit is configured to perform two-dimensional straight line fitting on all point clouds in the first candidate power pole A_i and the second candidate power pole A_j respectively by least squares using (Z, X) data of the points.

[0098] z= Kx +B

[0099] Calculate the error after fitting, where n represents all point clouds in the first candidate power pole A_i and the second candidate power pole A_j; when the error err is less than 0.2*n, the first candidate power pole A_i or the second candidate power pole A_j is a power pole, otherwise discard the point cloud in the first candidate power pole A_i or the second candidate power pole B_i.

[0100] wherein Z represents a value of a point in a robot height direction, X represents a value of a point in a robot advancing direction, K and B respectively represent a slope and an intercept of a fitting straight line, is a measured value of the i-th point in the Z direction, is a measured value of the i-th point in the X direction.

[0101] The wire extraction unit is configured to input the first frame Scan_i and the second frame Scan_j into a wire semantic extraction network respectively, and perform point cloud semantic extraction on Scan_i and Scan_j to obtain point clouds of all wires.

[0102] Specifically, the wire extraction unit of the embodiment further includes the following units:

[0103] The first frame wire extraction unit is configured to perform up-down clustering on the input first frame Scan_i laser point cloud, randomly select a point p, construct a second length-width-height region with the point p as the center point, and if the number of points in the second length-width-height region is greater than a second threshold value, retain all points in the second length-width-height region and mark them as first candidate wires B_i.

[0104] Specifically, the second length-width-height of the embodiment is a 3m*3m*1m region.

[0105] The second threshold value is 30.

[0106] Specifically, the first preset height of the embodiment is 3m.

[0107] Specifically, because the height of the power line of the power station is generally about 3m, in order to remove the interference of irrelevant objects, before the random selection of the point p in step S31, the points with a height lower than 2.5m are removed.

[0108] The second frame wire extraction unit is configured to perform up-down clustering on the input second frame Scan_i laser point cloud, randomly select a point p, construct a second length-width-height region with the point p as the center point, and if the number of points in the second length-width-height region is greater than a second threshold value, retain all points in the second length-width-height region and mark them as second candidate wires B_j.

[0109] The wire confirmation unit is configured to perform two-dimensional straight line fitting on all point clouds in the first candidate wires B_i and the second candidate wires B_j respectively by using (Z, X) data of the points through least squares.

[0110] z= Kx +B

[0111] Calculate the error after fitting, Wherein n represents all point clouds in the first candidate electric wire B_i and the second candidate electric wire B_j; when the error is less than 0.1*n, the first candidate electric wire B_i or the second candidate electric wire B_j is an electric wire, otherwise the point clouds in the first candidate electric wire B_i or the second candidate electric wire B_j are discarded.

[0112] Wherein Z represents the value of the point in the robot height direction, X represents the value of the point in the robot forward direction, K and B represent the slope and intercept of the fitted straight line respectively, is the measured value of the i th point in the Y direction, is the measured value of the i th point in the X direction.

[0113] The pose acquisition unit combines the electric pole point cloud and the electric wire point cloud of the first frame as a first frame semantic point cloud, combines the electric pole point cloud and the electric wire point cloud of the second frame as a second frame semantic point cloud, takes the first frame semantic point cloud as a source point cloud source of an iterative closest point algorithm ICP point cloud registration algorithm, takes the second frame semantic point cloud as a target point cloud target of the ICP point cloud registration algorithm, and acquires the motion pose of the robot by using the ICP point cloud registration algorithm.

[0114] The iterative closest point algorithm ICP is the most commonly used point cloud registration. Point cloud registration refers to inputting two point clouds P s (source) and P t (target), and outputting a transformation T, so that the coincidence degree of T(P s ) and T(P t ) is as high as possible, thereby being used for estimating the pose. The general algorithm flow of ICP is as follows:

[0115] Point cloud preprocessing; 2. Matching; 3. Weighting; 4. Eliminating unreasonable corresponding point pairs; 5. Calculating loss; 6. Minimizing loss to solve the current optimal transformation; 7. Returning to step 2 for iteration until convergence.

[0116] The ICP registration algorithm of the embodiment is the existing technology, and will not be described herein.

[0117] The embodiment extracts the laser semantic elements including the actual electric pole and the electric wire from each laser point cloud frame, and then only performs registration and positioning by using the extracted semantic element point cloud, thereby improving the positioning speed and positioning accuracy. Embodiment three

[0118] Reference Figure 7 , Figure 7Fig. 1 is a structural schematic diagram of a laser semantic based positioning device according to an embodiment. The laser semantic based positioning device 20 according to the embodiment includes a processor 21, a memory 22, and a computer program stored in the memory 22 and executable on the processor 21. The processor 21 implements the steps in the above method embodiments when executing the computer program. Alternatively, the processor 21 implements the functions of the modules / units in the above apparatus embodiments when executing the computer program.

[0119] The computer program can be divided into one or more modules / units, which are stored in the memory 22 and executed by the processor 21 to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the laser semantic based positioning device 20. For example, the computer program can be divided into the modules in Embodiment Two, and the specific functions of the modules are described in the working process of the apparatus in the above embodiments, which will not be repeated here.

[0120] The laser semantic based positioning device 20 can include, but is not limited to, the processor 21 and the memory 22. Those skilled in the art can understand that the schematic diagram is only an example of the laser semantic based positioning device 20, and does not limit the laser semantic based positioning device 20, which can include more or fewer components than the diagram, or combine certain components, or different components, for example, the laser semantic based positioning device 20 can also include an input / output device, a network access device, a bus, etc.

[0121] The processor 21 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic components, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor and the like. The processor 21 is the control center of the laser semantic based positioning device 20, which connects all parts of the laser semantic based positioning device 20 through various interfaces and lines.

[0122] The memory 22 can be used to store the computer programs and / or modules, and the processor 21 realizes various functions of the laser semantic positioning device 20 by running or executing the computer programs and / or modules stored in the memory 22, and calling the data stored in the memory 22. The memory 22 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required for a function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), and the like. In addition, the memory 22 can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0123] When the modules / units of the laser semantic positioning device 20 are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such understanding, the present application realizes all or part of the processes in the above-mentioned embodiment methods, and can also be completed by instructing related hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor 21, the steps of the above-mentioned various method embodiments can be realized. The computer program includes computer program code, which can be in the form of source code, object code, an executable file, or some intermediate form, etc. The computer-readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a U disk, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the computer-readable medium can include or exclude contents according to the requirements of legislation and patent practice in a jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0124] It should be noted that the apparatus embodiments described above are merely illustrative, and the units described as separate units can or can not be physically separate, and the units shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. In addition, the connection relationship between the modules in the apparatus embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0125] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A laser semantic-based positioning method, comprising the following steps: S1, obtaining any two frames of laser radar point cloud data in the process of robot walking, denoted as a first frame Scan_i and a second frame Scan_j; S2, inputting the first frame Scan_i and the second frame Scan_j into a power pole semantic extraction network respectively, performing point cloud semantic extraction on Scan_i and Scan_j to obtain point clouds of all power poles; step S2 further comprises: S21, performing up-down clustering on the input first frame Scan_i laser point cloud, randomly selecting a point P1 within a first preset height, taking the point P1 as a center point, constructing a first length-width-height region, if the number of points in the first length-width-height region is greater than a first threshold, retaining all points in the first length-width-height region and marking them as a first candidate power pole A_i; S22, performing up-down clustering on the input second frame Scan_j laser point cloud, randomly selecting a point P2 within a first preset height, taking the point P2 as a center point, constructing a first length-width-height region, if the number of points in the first length-width-height region is greater than a first threshold, retaining all points in the first length-width-height region and marking them as a second candidate power pole A_j; S23, using (Z, X) data of all points in the first candidate power pole A_i and the second candidate power pole A_j to perform two-dimensional straight line fitting through least squares, when the error after two-dimensional straight line fitting is less than a third threshold, the first candidate power pole A_i or the second candidate power pole A_j is a power pole, otherwise, discarding the point cloud in the first candidate power pole A_i or the second candidate power pole A_j, obtaining power pole point clouds of the first frame and / or power pole point clouds of the second frame; S3, inputting the first frame Scan_i and the second frame Scan_j into a power line semantic extraction network respectively, performing point cloud semantic extraction on Scan_i and Scan_j to obtain point clouds of all power lines; S4, merging the power pole point clouds and the power line point clouds of the first frame as first frame semantic point clouds, merging the power pole point clouds and the power line point clouds of the second frame as second frame semantic point clouds; and taking the first frame semantic point clouds as a source point cloud source of an iterative closest point (ICP) point cloud registration algorithm, taking the second frame semantic point clouds as a target point cloud target of the ICP point cloud registration algorithm, and using the ICP point cloud registration algorithm to obtain the motion pose of the robot.

2. The method of claim 1, wherein step S3 comprises: S31, performing up-down clustering on the input first frame Scan_i laser point cloud, randomly selecting a point p, taking the point p as a center point, constructing a second length-width-height region, if the number of points in the second length-width-height region is greater than a second threshold, retaining all points in the second length-width-height region and marking them as a first candidate power line B_i; S32, the input second frame Scan_i laser point cloud is clustered up and down, a point p is randomly selected, a second length-width-height region is constructed with the point p as the center point, if the number of points in the second length-width-height region is greater than a second threshold value, all points in the second length-width-height region are retained and marked as a second candidate power line B_j; S33, all point clouds in the first candidate power line B_i and the second candidate power line B_j are respectively fitted by least square with the data of the points to obtain a two-dimensional straight line, when the error after the two-dimensional straight line fitting is less than a fourth threshold value, the first candidate power line B_i or the second candidate power line B_j is a power line, otherwise, the point cloud in the first candidate power line B_i or the second candidate power line B_j is discarded, and the power line point cloud of the first frame and / or the power line point cloud of the second frame is obtained.

3. The method of claim 2, wherein the third threshold value is 0.2*n, wherein n is the number of all point clouds in the first candidate power pole A_i or the second candidate power pole A_j.

4. The method of claim 3, wherein the fourth threshold value is 0.1*n, wherein n is the number of all point clouds in the first candidate power line B_i or the second candidate power line B_j.

5. A positioning device based on laser semantics, comprising the following units: an original laser point cloud frame acquisition unit, configured to acquire any two frames of laser radar point cloud data in a robot walking process, denoted as a first frame Scan_i and a second frame Scan_j; The power pole semantic extraction unit is configured to input the first frame Scan_i and the second frame Scan_j into a power pole semantic extraction network respectively, and perform point cloud semantic extraction on Scan_i and Scan_j to obtain point clouds of all power poles. the power pole semantic extraction unit further comprises: a first frame power pole extraction unit, configured to cluster up and down the input first frame Scan_i laser point cloud, randomly select a point P1 within a first preset height, construct a first length-width-height region with the point P1 as the center point, if the number of points in the first length-width-height region is greater than a first threshold value, all points in the first length-width-height region are retained and marked as a first candidate power pole A_i; a second frame power pole extraction unit, configured to cluster up and down the input second frame Scan_j laser point cloud, randomly select a point P2 within a first preset height, construct a first length-width-height region with the point P2 as the center point, if the number of points in the first length-width-height region is greater than a first threshold value, all points in the first length-width-height region are retained and marked as a second candidate power pole A_j; a power pole confirmation unit, configured to respectively fit all point clouds in the first candidate power pole A_i and the second candidate power pole A_j by least square with the (Z, X) data of the points to obtain a two-dimensional straight line, when the error after the two-dimensional straight line fitting is less than a third threshold value, the first candidate power pole A_i or the second candidate power pole A_j is a power pole, otherwise, the point cloud in the first candidate power pole A_i or the second candidate power pole A_j is discarded, and the power pole point cloud of the first frame and / or the power pole point cloud of the second frame is obtained; a power line extraction unit, configured to input the first frame Scan_i and the second frame Scan_j into a power line semantic extraction network respectively, and perform point cloud semantic extraction on Scan_i and Scan_j to obtain all power line point clouds. The pose acquisition unit is configured to combine the power pole point cloud and the power line point cloud of the first frame as a first frame semantic point cloud, combine the power pole point cloud and the power line point cloud of the second frame as a second frame semantic point cloud, and use the first frame semantic point cloud as a source point cloud of an iterative closest point (ICP) point cloud registration algorithm, use the second frame semantic point cloud as a target point cloud of the ICP point cloud registration algorithm, and acquire the motion pose of the robot by using the ICP point cloud registration algorithm.

6. The apparatus of claim 5, wherein the power line extraction unit further comprises: a first frame power line extraction unit configured to perform up-down clustering on the input first frame Scan_i laser point cloud, randomly select a point p, construct a second length-width-height region with the point p as a center point, and retain all points in the second length-width-height region and mark the points as a first candidate power line B_i if a number of points in the second length-width-height region is greater than a second threshold value; a second frame power line extraction unit configured to perform up-down clustering on the input second frame Scan_i laser point cloud, randomly select a point p, construct a second length-width-height region with the point p as a center point, and retain all points in the second length-width-height region and mark the points as a second candidate power line B_j if a number of points in the second length-width-height region is greater than a second threshold value; a power line confirmation unit configured to perform two-dimensional straight line fitting on all point clouds in the first candidate power line B_i and the second candidate power line B_j by using data of the points through least squares, and determine that the first candidate power line B_i or the second candidate power line B_j is a power line when an error after the two-dimensional straight line fitting is less than a fourth threshold value, or discard the point clouds in the first candidate power line B_i or the second candidate power line B_j to obtain the power line point cloud of the first frame and / or the power line point cloud of the second frame.

7. The apparatus of claim 6, wherein the third threshold value is 0.2*n, and n is a number of all point clouds in the first candidate power pole A_i or the second candidate power pole A_j.

8. The apparatus of claim 7, wherein the fourth threshold value is 0.1*n, and n is a number of all point clouds in the first candidate power line B_i or the second candidate power line B_j.

Images