CN113375664A

CN113375664A - Autonomous mobile device positioning method based on dynamic point cloud map loading

Info

Publication number: CN113375664A
Application number: CN202110643205.1A
Authority: CN
Inventors: 蒋涛; 蔡明希; 许林; 邹艳玲; 张林帅; 李晨
Original assignee: Chengdu University of Information Technology
Current assignee: Chengdu University of Information Technology
Priority date: 2021-06-09
Filing date: 2021-06-09
Publication date: 2021-09-10
Anticipated expiration: 2041-06-09
Also published as: WO2022257196A1; CN113375664B; LU502363B1

Abstract

The invention discloses an autonomous mobile device positioning method based on dynamic point cloud map loading, which comprises the following steps: firstly, calibrating, namely, performing online calibration and alignment on an Inertial Measurement Unit (IMU) and a laser radar Lidar by constructing an error model of the IMU; secondly, a map is built, a point cloud loading rule is set, the mobile platform carries the calibrated laser radar and an IMU (inertial measurement Unit) to record a bag, and the bag is mapped based on a down-sampling leam algorithm; performing modular partition processing on the constructed map to obtain a plurality of block maps which can be loaded by the robot, and reserving a loading area between adjacent block maps; and step four, mapping the map. The invention provides an autonomous mobile device positioning method based on point cloud map dynamic loading, which eliminates positioning deviation caused by human error when a carrier is moved and matched, greatly improves initial loading speed, ensures that the moving speed and the matching speed of an inspection robot are kept stable in the moving process, and realizes the reliability of full-time positioning.

Description

Autonomous mobile device positioning method based on dynamic point cloud map loading

Technical Field

The invention relates to the field of complex environment robot inspection. More particularly, the present invention relates to a method for locating an autonomous mobile device based on dynamic loading of a point cloud map for use in industrial plants, agricultural parks, and structured environments with significant features.

Background

In the comparatively obvious structured environment of industrial factory district, agricultural garden and characteristic, because the GPS signal is sheltered from, lead to single sensor's positioning effect relatively poor, the limitation is big, can't satisfy the accurate positioning requirement of sub centimeter level. To solve such problems, a positioning method of multi-sensor fusion is often adopted. However, this approach also causes new problems, such as IMU signal offset, which is not corrected by GPS and results in larger and larger accumulated errors; the laser radar has many useless repeated points in the environment with obvious characteristics, the traditional laser point cloud map is too large, the point cloud is repeated, the loading speed is slow, and the matching speed cannot meet the inspection requirement, so that the positioning distortion is caused, and the accident is caused.

Disclosure of Invention

An object of the present invention is to solve at least the above problems and/or disadvantages and to provide at least the advantages described hereinafter.

To achieve these objects and other advantages in accordance with the purpose of the invention, there is provided an autonomous mobile device positioning method based on dynamic loading of a point cloud map, comprising:

firstly, calibrating, namely, performing online calibration and alignment on an Inertial Measurement Unit (IMU) and a laser radar Lidar by constructing an error model of the IMU;

secondly, a map is built, a point cloud loading rule is set, the mobile platform carries the calibrated laser radar and an IMU (inertial measurement Unit) to record a bag, and the bag is mapped based on a down-sampling leam algorithm;

performing modular partition processing on the constructed map to obtain a plurality of block maps which can be loaded by the robot, and reserving a loading area between adjacent block maps;

and step four, mapping the map, namely mapping the modularized map and the downloaded GPS map to obtain a rasterized map which can be used for the robot to walk.

Preferably, in step one, the online calibration and alignment includes:

s10, establishing an IMU calibration error model;

s11, exciting an error term of the error model;

s12, carrying out observability analysis on the error term;

s13, estimating the component parameters of the error items by adopting an extended Kalman filtering algorithm to obtain a corresponding calibration matrix;

and S14, jointly calibrating the Lidar and the IMU to obtain a compensation matrix which can be added to the map construction.

Preferably, in S10, the establishing of the calibration error model is configured to include:

the gyro error model is established as follows:

in the formula, epsilonⁿ＝[ε_E ε_N ε_U]^TRepresenting gyro error eⁿThe components under the navigation coordinate system respectively refer to errors in east, north and sky directions;

a transformation matrix from a carrier coordinate system to a navigation coordinate system;

the output of the gyroscope is represented by components of the gyroscope x-axis, y-axis and z-axis; error matrix delta K of gyro scale factor_G＝diag[δK_Gx δK_Gyδ_KGz]The components of the scale factor error are expressed in the x axis, the y axis and the z axis of the gyroscope; epsilon^b＝[ε_bx ε_by ε_bz]^TThe method is characterized in that the method is a gyro random constant error, and components of the gyro random constant error represent constant errors of an x axis, a y axis and a z axis of a gyro;

the gyro installation error matrix is established as follows:

wherein, the component of the matrix represents the installation error of the gyroscope x axis, y axis and z axis;

scale factor error delta K based on accelerometer_AMounting error matrix delta A, random constant error

The measurement error model of the accelerometer is constructed as follows:

and establishing a corresponding calibration error model by measuring component reference parameters in the error model and the gyro error model.

Preferably, in S11, the error term excitation is performed on the calibration error model by selecting the motion trajectory setting and/or the attitude change of the inertial sensor;

in S12, observability analysis is realized by determining the observability degree based on a selected excitation mode, obtaining an energy density spectrum of a state variable by an optimal interpolation method, and constructing a Kalman filtering equation according to the described error;

in S13, an extended Kalman filtering algorithm is used for analyzing the model characteristics of a system equation and a measurement equation, observed quantities are constructed and output through state vectors, and estimation of unknown states or internal parameters is completed through time updating and measurement updating;

in S14, obtaining external parameters of the Lidar through the combined calibration of the Lidar and the IMU, calculating the transformation R-T parameters between the Lidar and the IMU through the internal parameters of the IMU obtained in S13, and obtaining a compensation matrix which can be added into the map construction through the inversion operation of the parameter matrix through the error calibration matrix of the IMU and the parameter matrix obtained through the combined calibration.

Preferably, in step two, the point cloud loading rule is configured to include:

based on the mapping principle of Loam, connecting all cubes in a grid into a large point cloud in a circulating mode when a window is slid, and then releasing a complete map through topics;

in the circulation, a constant with an initial value of 0 is added to judge whether the number of the current point clouds is smaller than that of the current point clouds, if not, the circulation is started, the laser cloud encirclements are accumulated, and the initial constant is increased by one bit after each circulation; when the constant value is not less than the current point cloud number, jumping out of the cycle;

the down-sampling is configured to include:

in the Loam algorithm, the volume of a point cloud file is reduced by adopting a size filtering mode, the feature dimension is reduced, and effective information is reserved;

the size filtering mode is that after the point cloud scanned by the first frame is stored in the map, after each frame is added into the map, judgment is carried out, whether the overlapping rate of the point of the previous frame and the point of the current frame reaches 50% or more is carried out, if yes, deletion is carried out, and if not, retention is carried out.

Preferably, in step three, the modular partition processing of the map is configured to include:

s30, determining the number of modules for roughly dividing the map;

s31, determining the size of the buffer area between the modules;

and S32, modularly dividing the map according to a preset dividing rule so that the difference between the point cloud number in each module and a preset value is not more than 10%, and the reserved area between the modules meets the size of a buffer area.

Preferably, in S30, the area number division of the map is based on the number of all point clouds contained in the map, and the relation between the original loading time and the number of division modules is calculated to obtain:

in S31, based on the driving speed of the inspection robot and the time required for loading each point cloud module on the industrial personal computer, setting the length of the area loaded by the next modular map between each two adjacent modules as a buffer area;

in S32, the dividing rule includes:

s320, performing area division on the map based on the number of the modules obtained in the S30, and meanwhile setting each corner in the map as an independent module without additionally performing division;

s321, selecting a plurality of modules as starting points to construct a buffer area between adjacent modules in a diffusion mode.

Preferably, the buffer is constructed in a manner including:

s3210, setting the complete loading time to S by taking each corner as a starting point, and selecting a plurality of corners with smaller relative positions as the starting points from a map with a large number of corners S + 1; for the map with the number of corners smaller than S +1, the number judgment is carried out after all the corners are selected to determine a starting point;

s3211, diffusing the selected starting points outwards at the same time, and setting the number of the point clouds in the map as N, so that the number of the point clouds in the diffusion frame reaches the number of the point clouds in the diffusion frame

Triggering a diffusion limit and stopping frame selection;

s3212, comparing the number of the diffused modules with the target number, if the difference is 1, selecting a new starting point between the two modules which are farthest away, performing secondary diffusion, and performing secondary diffusionThe number of point clouds in the second diffusion frame selection is up to

Triggering and stopping when the number of the point clouds is not reached but the reserved distance between the diffusion radius and any diffusion radius of two adjacent modules is equal to 3 meters;

wherein the new starting point is at a position

And l is the distance between two farthest module starting points, r₁And r₂The diffusion radii of the two modules are respectively;

s3213, integrating all circles into a rectangle according to the radius, and retaining the quantitative point cloud in the module by adopting a K-Means clustering algorithm and a probabilistic greedy method;

s3214, calculating the distance between the boundaries of the adjacent modules, expanding if the result is greater than 3m, and quantitatively shifting if the expansion reaches 20% and the requirement cannot be met, so that the spacing distance meets the requirement;

if the result is less than 3 meters, the two modules are forced to average and delete the point clouds directly corresponding to the two modules so as to reserve a sufficient buffer area.

Preferably, in step four, the construction method of the rasterized map is configured to include:

s41, downloading a 22-level high-resolution map, extracting a value of a part, overlapped with a local map, of the downloaded high-resolution map to serve as a grid map with high resolution, calculating respective offset through position transformation of the map resolution and the map, and storing the offset into a yaml file;

and S42, truncating a part corresponding to the point cloud map from the grid map, and mapping the part into the point cloud map according to the conversion relation in the yaml file.

Preferably, the method further comprises the following steps:

step five, in the inspection operation of the robot, when the robot runs on any module, the industrial personal computer loads the current module and two buffer areas connected with the current module;

when the robot runs to any buffer area, judging whether the two module maps connected with the robot complete loading, if not, loading the map, if so, stopping judging, and when the buffer area runs to the next module area, deleting the previous module area and the unilateral buffer area thereof, loading the next buffer area, keeping the matching speed of the system, and repeating the steps until all the maps are traversed.

The invention at least comprises the following beneficial effects: firstly, before a high-precision point cloud map is established, a laser radar and an IMU (inertial measurement Unit) need to be calibrated, a coordinate system is unified, and the accuracy of later positioning is guaranteed;

secondly, when a high-precision algorithm is used for building the map, repeated points are removed through down-sampling, the volume of a map file is reduced, and the loading speed is guaranteed;

thirdly, the point cloud map is divided into areas according to constraint conditions such as the driving speed of the inspection robot and the density degree of the point cloud map, reasonable loading rules are set, and the point cloud matching speed is improved.

Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.

Drawings

FIG. 1 illustrates error types for an INS;

FIG. 2 is a basic procedure for on-line calibration according to the present invention;

FIG. 3 is a flow chart of a process for excitation using an attitude change setting in accordance with the present invention;

FIG. 4 is a map structure constructed by the original loam algorithm;

FIG. 5 is a map structure constructed after processing by the downsampled loam algorithm of the present invention;

FIG. 6 is a map after the present invention employs modular processing;

FIG. 7 is a corresponding grid map obtained after mapping according to the present invention;

in which fig. 4-7 all perform inverse processing on the actual effect graphs.

Detailed Description

The present invention is further described in detail below with reference to the attached drawings so that those skilled in the art can implement the invention by referring to the description text.

The invention provides a positioning strategy based on point cloud map modularization loading and quick matching. Before a high-precision point cloud map is established, a laser radar and an IMU (inertial measurement Unit) need to be calibrated, and a coordinate system is unified; secondly, when a high-precision algorithm is used for building a map, repeated points are removed through down-sampling, and the volume of a map file is reduced; then, mapping the standard GPS point coordinates into a point cloud map; and finally, carrying out regional division on the point cloud map according to the driving speed of the inspection robot, the density degree of the point cloud map and other constraint conditions, and setting a reasonable loading rule.

The invention provides a positioning strategy based on point cloud map modularization loading and quick matching, which comprises the following steps:

1. calibration

The inertial navigation system INSNS senses the motion of a carrier by using an inertial measurement device (such as an accelerometer, a gyroscope, and the like) mounted on the carrier, and outputs attitude and position information of the carrier. The INS system is completely autonomous, high in confidentiality, flexible and capable of outputting multifunctional parameters, but has the problem that errors are rapidly accumulated along with time, navigation accuracy diverges along with time, the INS system cannot work independently for a long time, and the INS system must be continuously calibrated (namely a corresponding error model for calibration is established).

Analyzing and researching errors of the inertial sensor, wherein the errors comprise systematic errors and random errors, establishing a mathematical model of the sensor errors, correcting output raw data by using the model, and improving the accuracy of the raw data, wherein the types of the errors output by the INS are shown in figure 1, the systematic errors are errors carried by the sensor, parameters of the systematic errors can be obtained from a poor file and can be directly used when the model is established, and the random errors are errors caused by installation or other external factors, and the parameters of the random errors need to be estimated, so that calibration or registration is completed.

As shown in fig. 2, the online calibration and alignment of the INS includes the following specific steps:

the gyro error model is established as follows:

the output of the gyroscope is represented by components of the gyroscope x-axis, y-axis and z-axis; error matrix delta K of gyro scale factor_G＝diag[δK_Gx δK_Gy δK_Gz]The components of the scale factor error are expressed in the x axis, the y axis and the z axis of the gyroscope; epsilon^b＝[ε_bx ε_by ε_bz]^TThe method is characterized in that the method is a gyro random constant error, and components of the gyro random constant error represent constant errors of an x axis, a y axis and a z axis of a gyro;

the gyro installation error matrix is:

the components represent the mounting errors of the gyroscope x-axis, y-axis and z-axis. According to the above definition, a complete gyro error model can be constructed.

Scale factor error δ K in combination with accelerometer_AMounting error matrix delta A, random constant error

The measurement error model of the accelerometer is constructed as follows:

in the formula, the component reference parameters are similar to the gyro error model, and the calibration of the error term can be completed by selecting a proper excitation and estimation method through the established calibration error model.

The excitation of the error term, namely, the system error caused by a certain error is reasonably amplified through a specific mode, thereby facilitating the subsequent operation process. The appropriate excitation mode can improve the observability of certain error parameters and the excellent degree of the excitation effect, and greatly influences the calibration effect. The excitation mode is selected from motion track setting, posture change setting and the like. The method of preliminarily formulating attitude change to perform error term excitation, and the steps of adopting attitude change setting as an excitation method are shown in fig. 3.

Observability is closely related to error term excitation, and whether an error term of a strapdown inertial navigation system is excited in a motor mode is generally defined by the observability degree. The observability degree is determined by setting a reasonable motion track and using observability analysis, the energy density spectrum of the state variable is obtained by an optimal interpolation method, and a Kalman filtering equation is constructed according to the described error, so that the observability analysis is realized.

And (3) parameter estimation, namely, generally using a Kalman filtering method to construct reasonable observed quantity by analyzing model characteristics of a system equation and a measurement equation, taking the state vector as output, and finishing estimation of unknown states or parameters by time updating and measurement updating. Although Kalman filtering has wide application, it mainly solves the problem of linear system, and needs to perform deep improvement and study on the problem of nonlinear filtering, and the subject is to adopt Extended Kalman Filtering (EKF) to simplify the tedious calculation in consideration of the balance error term. And after the completion, fixedly installing, carrying out combined calibration by using a laser radar and an IMU (inertial measurement Unit), obtaining external parameters of the Lidar, and converting R-T parameters between the Lidar and the IMU by the internal parameters of the IMU obtained in the previous step.

And carrying out inversion operation on the error calibration matrix of the IMU and the parameter matrix obtained by combined calibration to obtain a compensation matrix, and adding the compensation matrix into the map construction, so that the preparation work of the map construction is finished.

2. Map construction

The traditional loam algorithm can establish a relatively perfect point cloud map, but the problems of too small carrier loading area, too many repeated points and the like exist. The algorithm is first improved.

Firstly, setting a reasonable point cloud loading rule. The mapping principle of leam is to save a point cloud map accumulated around the robot by a sliding window method. At the beginning, the initial position of the robot initializes a 3D mesh, each cube in the mesh being represented by a point cloud. For each incoming scan, the mapping operation will only update a small cube region around the current location, the updated region will be connected to a large point cloud, and finally the down-sampling process is performed and published through topics. When the robot is free to move within the grid, once an arbitrary boundary of the grid is approached, the grid moves in a corresponding opposite direction, thereby losing the map cube accumulated on the other side of the grid. The size of the grid is specified by the constructor LaserMapping in the Loam algorithm. The original was specified as a 50 x 50 meter area. If the area is expanded blindly, the whole algorithm consumes a large amount of memory, and even breaks down directly when the algorithm is serious. A loop can be set to connect all cubes into a large point cloud, instead of the current update area, to publish a complete map by topic. And (3) judging whether the number of the current point clouds is less than the number of the current point clouds by adding a constant with an initial value of 0, if not, entering circulation, accumulating the laser cloud encirclements, and increasing the initial constant by one bit after one circulation. And circulating, and if the constant is not less than the current point cloud number, jumping out of the circulation. By doing so, the grid area is increased, and memory dissipation is avoided.

And secondly, performing down-sampling. Although the above operation avoids memory dissipation to some extent, the point cloud still occupies a large memory because of the disorder and the geometric multiple increase of the number of the point clouds. And in order to relieve loading pressure, performing down-sampling operation on the point cloud. Through adopting suitable size of considering in order to reduce the point cloud file volume, reduce the dimension of characteristic and keep effective information, avoid overfitting to a certain extent, be convenient for transmit and load. The filter size is designed as the following rule: and (3) the map stores each subsequent frame of the point cloud scanned by the first frame, and judges whether the overlapping rate of the point scanned by the first frame and the point scanned by the previous frame reaches 50% or more after the point scanned by the first frame is added into the map, if so, the point scanned by the first frame is deleted, and if not, the point scanned by the first frame is retained. Typically the down-sampled volume can be reduced to less than half of the original volume without a reduction in features. Meanwhile, due to the characteristic of the matching algorithm NDT adopted in the later period, the final matching positioning effect cannot be influenced after the down-sampling operation is carried out. Continuously reading in the point cloud file in the process of drawing, then performing down-sampling operation, and finally storing the point cloud file in a new file.

And thirdly, carrying the calibrated laser radar and the calibrated IMU by using a mobile platform, moving at a constant speed in the structured road, waking up the radar and the IMU, recording a corresponding bag, and finishing recording after a loop is formed in a path.

And fourthly, using a leam algorithm added with down sampling to map the bag of the previous step.

As shown in fig. 4, in the case of not performing the down-sampling operation, the bag packet data is directly used for mapping, and it can be seen that the map only retains about one fourth of the original map, and the point clouds are disordered and the critical areas are seriously overlapped.

As shown in fig. 5, the map is built by the modified algorithm, the map is smoothly and completely loaded, the display of the loop road is very obvious, the point cloud scales of all areas are uniform, the sparsity degree is uniform, the occupied memory is only one third of the memory of the map shown in fig. 4, and the loading speed is obviously improved.

3. Map modular processing

And carrying out partition processing on the established map. In a first step, the number of roughly divided regions is determined. And counting the number of all point clouds contained in the map, and solving the relation between the original loading time and the number of divided areas. For example, the number of point clouds of a map is N, the complete loading time on the industrial personal computer is about S seconds on average, and in order to ensure that the ideal loading speed of each module on the industrial personal computer does not exceed 1 second, the map needs to be divided into S +1 modules on average for dynamic loading.

In the second step, the size of the buffer area is determined. And a loading area must be reserved between every two adjacent modules as a buffer area. Each module is connected with two buffer areas, when the robot runs to one of the buffer areas, whether the map of the two modules connected with the robot finishes loading is immediately judged, if not, the map is loaded, and if yes, the judgment is stopped. In summary, the buffer area is used to ensure smooth and complete loading of the module map. In order to ensure that the inspection robot can finish map matching without distortion in speed-limiting areas such as a campus, a factory and the like under the condition of ensuring safety, the maximum limit driving speed is adopted to be 5km/h, and the maximum limit driving speed is about 1.5m/s after conversion. The ideal loading time consumption of each module is 1 second, and in order to ensure that the robot does not run out of the loading area, the next module finishes loading, so the reserved time is twice of the loading time, namely 2 seconds, and the span of the cache area is 1.5 x 2-3 meters.

And thirdly, dividing the map. And selecting each starting point. In order to ensure the safety of the inspection robot in the driving process, each corner is an independent module and is not divided. So that the start point can be selected at a more specific corner. And selecting a plurality of corners with smaller relative positions for the map with the number of corners larger than S +1, and performing quantity judgment after selecting all the corners for the map with the number of corners smaller than S + 1. Taking fig. 5 as an example, the specific operations are as follows: respectively selecting 4 corner points as starting points, simultaneously starting outward diffusion, and when the number of the framed point clouds reaches

And triggering a diffusion limit and stopping frame selection. At this time, the difference between the obtained module and the target number is 1, a new starting point is selected between the two modules which are farthest away, and the position of the starting point is selected as

Where l is the distance between the starting points of the two farthest modules, and r1 and r2 are the diffusion radii of the two modules, respectively. Of the current moduleThe diffusion stopping conditions are two, one is that when the amount of the framed point clouds reaches to

And stopping triggering, namely stopping triggering when the number of the point clouds is not reached but the reserved distance between the diffusion radius and any diffusion radius of two adjacent modules is equal to 3 meters. Either of the two conditions is satisfied. Subsequently, all circles are integrated into a rectangle by radius. At this point, it is optimized using a modified K-Means clustering algorithm. K-Means requires that after the central point is quantitatively selected, the distance of each data point to the central point is calculated, and the class to which the data point is closest to which central point. The center point in each class is calculated as the new center point. The above steps are repeated until the center of each class does not change much after each iteration. Therefore, under the condition that the central point is selected and the preliminary clustering is finished, iteration is carried out all the time, and the retained point clouds are selected in a probability greedy manner. Points closer to the center point have a greater probability of being retained, while points further from the center point have a lesser probability of being retained. The high-quality points are reserved and trapping in local optimization is avoided through probability greedy operation. The specific operation of probabilistic greedy is: center point of exhaustion C₁The distances from other points and the sum thereof are found, and the probability of each point being retained is:

(j ═ 1,2, … …, n). This operation retains a quantitative point cloud in the module. And finally, calculating the distance between every two module boundaries, if the distance is greater than 3 meters, expanding, and if the expansion reaches 20%, quantitatively shifting to enable the spacing distance to meet the requirement. If the distance is less than 3 meters, the point clouds directly corresponding to the two modules are forcibly and averagely deleted, and a sufficient buffer area is reserved. Finally, the difference between the point cloud number of each module and a preset value is not more than 10%, and each reserved area meets the requirement of 3-meter interval, so that the map construction can be completed.

In practical operation, the construction of the map may further include: the first step divides the approximate area. And dividing the built map by taking the point cloud sparsity degree as the maximum basis. In order to ensure the safety of the inspection robot in the driving process, each corner is an independent module and is not divided. The point cloud quantity difference in each module is not more than 10% so as to ensure that all time is uniform when modular loading is pursued. For example, the local map is uniformly divided into five areas according to the sparsity degree of the point cloud, and the areas are distinguished by different colors. And secondly, calculating the running distance of the inspection robot. A loading area must be reserved between every two adjacent modules, namely a black area in fig. 6. In a restricted speed area such as a campus and a factory, the running speed of the robot is usually 5km/h and is equal to 1.5m/s after being converted. The time required by loading each point cloud module on the industrial personal computer is about 1s, so that 2s of loading time is reserved for ensuring safe driving, and the next module is loaded when the robot does not run out of a loading area. Therefore, the length of the loading area is set to be 3m in the map, when the robot runs to the loading area, the system starts to judge whether two areas adjacent to the current loading area are loaded, if not, the next area module is started to be loaded, and if so, the judgment is stopped and the operation is continued. And after the loading area finishes driving, deleting the previous module, keeping the matching speed of the system, and repeating the steps in a circulating way.

4. Map mapping

And downloading a 22-level high-resolution map, and extracting a part of the downloaded high-resolution map, which is overlapped with the local map, to obtain a value as a grid map with high resolution. A special point is chosen as the origin, which is typically chosen as the lower left corner. And calculating each offset through position transformation of the map resolution and the map, storing the offset into a yaml file, and finally obtaining the grid map as shown in fig. 7.

And intercepting a part corresponding to the point cloud map from the manufactured grid map, and mapping the part into the point cloud map according to the conversion relation in the yaml file.

To this end, a complete cut modular laser point cloud map containing the gps signal is created.

5. Robot patrol operation based on map

The constructed map is shown in fig. 6. When the robot runs on any module, the industrial personal computer only loads the current module and two buffer areas connected with the current module, when the robot runs to any buffer area, whether the two module maps connected with the industrial personal computer finish loading or not is immediately judged, if not, the map is loaded, and if yes, the judgment is stopped. And when the robot crosses into the next area, deleting the previous area and the unilateral buffer area thereof, loading the next buffer area, and repeating the steps in a circulating way until all the modules are traversed. The positioning strategy or method based on point cloud map modularization loading quick matching has the following advantages that:

1. the laser radar and the IMU are calibrated, so that human errors are eliminated, map deviation is reduced, the wall body is more completely constructed, and positioning deviation caused by the human errors is eliminated when the carrier is matched in a moving mode.

2. The map is subjected to down-sampling and modular division processing, so that the map is completely constructed, the features are increased, the interference is reduced, the memory is reduced, and the initial loading speed is greatly improved. Originally, a map which needs 5 seconds of loading time can be loaded in only 1 second, the precision is not reduced, and the grade is improved. The moving speed and the matching speed of the inspection robot in the moving process are guaranteed to be stable, and the reliability of full-time positioning is realized.

The above scheme is merely illustrative of a preferred example, and is not limiting. When the invention is implemented, appropriate replacement and/or modification can be carried out according to the requirements of users.

The number of apparatuses and the scale of the process described herein are intended to simplify the description of the present invention. Applications, modifications and variations of the present invention will be apparent to those skilled in the art.

While embodiments of the invention have been disclosed above, it is not intended to be limited to the uses set forth in the specification and examples. It can be applied to all kinds of fields suitable for the present invention. Additional modifications will readily occur to those skilled in the art. It is therefore intended that the invention not be limited to the exact details and illustrations described and illustrated herein, but fall within the scope of the appended claims and equivalents thereof.

Claims

1. An autonomous mobile device positioning method based on point cloud map dynamic loading is characterized by comprising the following steps:

2. The method for autonomous mobile device positioning based on point cloud map dynamic loading of claim 1 wherein in step one, the online calibration and alignment comprises:

s10, establishing an IMU calibration error model;

s11, exciting an error term of the error model;

s12, carrying out observability analysis on the error term;

3. The method for autonomous mobile device positioning based on dynamic loading of point cloud map of claim 1 wherein in S10, the establishing of the calibration error model is configured to include:

the gyro error model is established as follows:

the gyro installation error matrix is established as follows:

The measurement error model of the accelerometer is constructed as follows:

4. The method for positioning an autonomous mobile device based on dynamic point cloud map loading of claim 2 wherein in S11, the error term excitation is performed on the calibration error model by selecting the motion trajectory setting and/or attitude change of the inertial sensor;

5. The method of claim 1, wherein in step two, the point cloud loading rules are configured to include:

the down-sampling is configured to include:

6. The method of claim 1, wherein in step three, modular partitioning of the map is configured to comprise:

s30, determining the number of modules for roughly dividing the map;

s31, determining the size of the buffer area between the modules;

7. The method of claim 6, wherein in S30, the partition of the area number of the map is based on the number of all point clouds contained in the map, and the relation between the original loading time and the number of the partition modules is calculated to obtain:

in S32, the dividing rule includes:

8. The method of claim 7, wherein the buffer is constructed in a manner comprising:

Triggering a diffusion limit and stopping frame selection;

s3212, comparing the number of the diffused modules with the target number, if the difference is 1, selecting a new starting point between two modules which are farthest away, performing secondary diffusion, and selecting the point cloud number in the secondary diffusion frame to reach the point cloud number

wherein the new starting point is at a position

9. The method of claim 1, wherein in step four, the construction of the rasterized map is configured to include:

10. The method of claim 1, further comprising: