CN111578946A - Laser navigation AGV repositioning method and device based on deep learning - Google Patents
Laser navigation AGV repositioning method and device based on deep learning Download PDFInfo
- Publication number
- CN111578946A CN111578946A CN202010461902.0A CN202010461902A CN111578946A CN 111578946 A CN111578946 A CN 111578946A CN 202010461902 A CN202010461902 A CN 202010461902A CN 111578946 A CN111578946 A CN 111578946A
- Authority
- CN
- China
- Prior art keywords
- agv
- repositioning
- map
- laser
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/20—Instruments for performing navigational calculations
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
- G01S17/02—Systems using the reflection of electromagnetic waves other than radio waves
- G01S17/06—Systems determining position data of a target
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
- G01S17/88—Lidar systems specially adapted for specific applications
- G01S17/93—Lidar systems specially adapted for specific applications for anti-collision purposes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Theoretical Computer Science (AREA)
- Electromagnetism (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Automation & Control Theory (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a laser navigation AGV repositioning method and device based on deep learning, which comprises the following steps: constructing a global contour map of an AGV operation scene; collecting and using peripheral obstacle information of the AGV at each position in a running scene to form a data set, and training to obtain a repositioning convolutional neural network; collecting distance information of objects around the AGV needing to be repositioned to form a local map, inputting the local map into the repositioning convolutional neural network, and calculating to obtain the maximum possible position of the AGV in the global contour map; feature matching is performed in a local range around the maximum possible position to produce a pinpoint location. The method has the advantages that the time consumed by one-time forward calculation of the repositioning convolutional neural network is short, the precise positioning is only used for local searching feature matching, and the calculation complexity is low, so the calculation speed is extremely high, the problem of the calculation complexity in the conventional repositioning method for global feature matching can be solved, and the time consumed in the whole repositioning process can be greatly reduced.
Description
Technical Field
The invention relates to the technical field of laser positioning, in particular to a laser navigation AGV repositioning method and device based on deep learning.
Background
The automatic navigation and positioning technology of the current industrially mature mobile robot (AGV) is a laser positioning technology, and by installing a laser radar on the AGV, the accurate distance between objects (such as a wall surface, equipment and the like) in the AGV running environment and the AGV can be obtained through scanning, the objects are mapped to a map according to the position relation while the moving path of the AGV is accurately calculated, a two-dimensional contour map of the robot running environment can be established, and the contour characteristics of the peripheral environment are measured to be matched with the map in the automatic navigation and walking process of the robot so as to calculate the position of the robot.
The first step in laser positioning of the mobile robot is repositioning, which results in an initial position of the mobile robot that must accurately coincide with the actual position of the robot, which would otherwise result in subsequent map matching being unable to be performed correctly, or in unpredictable erroneous position values.
The repositioning problem is one of the technical difficulties of positioning the mobile robot by adopting laser, and the reason is that the moving scene of the mobile robot is often very large and may reach 10000 square meters (100 meters by 100 meters), the feature matching of a whole map is required during repositioning, namely, all possible positions of the whole map are searched in a traversing way and are sequentially matched with the map features which can be seen nearby the positions, the calculated amount is very large, tens of seconds may be consumed, and the system practicability is very poor.
Based on the reasons, in actual engineering at present, an operator is often required to judge an approximate position of a robot in a map in advance, the robot is input as an initial value of relocation, then the mobile robot starts a markerless positioning algorithm to be matched with scene features near the approximate position, and then autonomous navigation walking is started after an accurate initial positioning value is obtained. However, this method has a high precision requirement for the approximate position given by the operator, and if the operator inputs the position incorrectly due to inexperience or carelessness, the initial positioning value calculated by the robot is often completely wrong, and an autonomous navigation walking fault is caused.
In summary, in the field of laser positioning of mobile robots, a fast and accurate automatic repositioning method is lacking.
Disclosure of Invention
The invention aims to provide a laser navigation AGV repositioning method and device based on deep learning, so as to solve the problem of computational complexity in the global feature matching of the traditional initial positioning method.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
in a first aspect, an embodiment of the present invention provides a laser navigation AGV repositioning method based on deep learning, including:
constructing a global contour map of an AGV operation scene;
collecting peripheral obstacle information of each position of an AGV in a running scene to form a training data set, and obtaining a repositioning convolutional neural network by using the data set;
collecting distance information of objects around the AGV to form a local map, inputting the local map into the repositioning convolutional neural network, and calculating to obtain the maximum possible position of the AGV in the global contour map;
feature matching is performed in a local range around the maximum possible position to produce a pinpoint location.
Further, the global contour map records the position information of the object which can be detected by the laser radar in the AGV running scene.
Further, a method for constructing the global contour map of the AGV running scene adopts a laser SLAM method.
Further, the laser SLAM method includes:
controlling the AGV to walk around in a running scene, acquiring obstacle information around the AGV in real time by a laser radar installed on the AGV in the process, defining the acquired data information each time as a laser frame, calculating the relative position movement of the AGV at each frame acquisition time according to the laser frames, synthesizing the obstacle information in the environment recorded in the frames into a map, and completing the construction of the global contour map.
Further, the relocation convolutional neural network mainly comprises a plurality of alternating convolutional layers and pooling layers, and then a full connection layer is added, and finally a SOFTMAX layer is added.
Further, the training process of relocating the convolutional neural network includes:
rasterizing the two-dimensional outline map to obtain an M multiplied by N two-dimensional grid map;
collecting each collected laser frame f1,f2,…,fnAnd recording the positions of the map corresponding to the AGV when the laser frame is acquired, judging that the positions are positioned at the specific grid positions in the two-dimensional grid map, and recording the positions as p1,p2,…,pnWhere any one of p1(i-0, 1, …, n) stores a data pair<j,k>(j is less than or equal to M, k is less than or equal to N) and represents that the grid is positioned on the jth row and the kth column in the two-dimensional grid map;
will f is1,f2,…,fnAs a training input sample set, p1,p2,…,pnAnd generating n training samples as an output label set, training the repositioning convolutional neural network to obtain a group of optimized network weights, and finishing the training of the repositioning convolutional neural network.
Further, a gradient back propagation method is adopted to train the repositioning convolution neural network.
Further, performing feature matching in a local range around the maximum possible position to produce a pinpoint location, comprising:
taking the coordinate of the maximum possible position on the global contour map as an initial position p of fine positioning calculation;
collecting a laser frame by using a laser radar, and recording all the sensed local contour feature sets as Ec;
based on the initial position p, setting a certain range around p as a search area p _ search, and determining a set phi (p) of all possible positions in the search range by a preset search step1,p2,p3,…pn}；
For any possible position p 'epsilon phi, projecting Ec to a coordinate system of a two-dimensional contour map according to a coordinate conversion relation to obtain a contour feature set after coordinate transformation as Ec, comparing the matching degree of the Ec and all contour feature sets Emap in the two-dimensional contour map, and calculating the matching score of the position p';
and taking the position with the highest matching score in the phi set including the initial position p as the accurate positioning position of the current AGV. In a second aspect, an embodiment of the present invention provides a laser navigation AGV repositioning device based on deep learning, including:
the map construction module is used for constructing a global contour map of the AGV operation scene;
the neural network construction module is used for acquiring peripheral obstacle information of each position of the AGV in a running scene to form a training data set, and obtaining a repositioning convolutional neural network by using the data set;
the rough positioning module is used for acquiring distance information of objects around the AGV to form a local map, inputting the local map into the repositioning convolutional neural network, and calculating to obtain the maximum possible position of the AGV in the global contour map;
a fine positioning module for performing feature matching in a local range around the maximum possible position to generate a fine positioning position.
Further, the fine positioning module comprises:
the initial setting submodule is used for taking the coordinate of the maximum possible position on the two-dimensional contour map as an initial position p of fine positioning calculation;
the acquisition submodule is used for acquiring a laser frame by using a laser radar, and recording all the perceived local contour feature sets as Ec;
a search determination module for setting a certain range around p as a search based on the initial position pThe region p _ search determines the set phi of all possible positions in the search range as { p } in a preset search step1,p2,p3,…pn}；
The matching calculation submodule is used for projecting Ec to a coordinate system of a global contour map according to a coordinate conversion relation for any possible position p 'epsilon phi to obtain a contour feature set after coordinate transformation as Ec, comparing the matching degree of the Ec and all contour feature sets Emap in the two-dimensional contour map, and calculating the matching score of the position p';
and the accurate positioning position determining module is used for acquiring all matching scores, and taking the position with the highest matching score in the phi set including the initial position p as the accurate positioning position of the current AGV.
According to the embodiment of the invention, the method of the invention does not need an operator to manually give an approximate initial position, so that the positioning navigation fault caused by the error of the operator can be effectively avoided. The repositioning process of the invention is fast, because the time consumed by one-time forward calculation of the repositioning convolutional neural network is very short, the precise positioning step only carries out local search feature matching, and the calculation complexity is not large, therefore, the method has extremely fast calculation speed, can avoid the problem of the calculation complexity when the traditional initial positioning method carries out global feature matching, and can greatly reduce the time consumed by the whole initial positioning process.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:
FIG. 1 is a flow chart of a deep learning based laser navigation AGV repositioning method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a deep learning based laser navigation AGV relocation apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "comprises" and "comprising," and any variations thereof, in the description and claims of the present invention and the above-described drawings, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
In accordance with an embodiment of the present invention, there is provided a deep learning based laser navigation AGV repositioning method, it being noted that the steps illustrated in the flowchart of the drawings may be performed in a computer system, such as a set of computer executable instructions, and that while a logical order is illustrated in the flowchart, in some cases the steps illustrated or described may be performed in an order different than that described herein.
FIG. 1 is a deep learning based laser navigation AGV repositioning method according to an embodiment of the present invention, as shown in FIG. 1, including the following steps:
step S101, constructing a global contour map of an AGV running scene;
step S102, collecting peripheral obstacle information of the AGV at each position in a running scene to form a training data set, and obtaining a repositioning convolutional neural network by using the data set;
step S103, collecting distance information of objects around the AGV to form a local map, inputting the local map into the repositioning convolutional neural network, and calculating to obtain the maximum possible position of the AGV in the global contour map;
and step S104, performing feature matching in a local range near the maximum possible position to generate a precise positioning position.
Through the above embodiment of the present invention, the method of the present invention can be divided into two phases, i.e., a map building phase in step S101 and step S102, and a relocation phase in step S103 and step S104. The method of the invention does not need the operator to manually give the approximate initial position, thus effectively avoiding the positioning navigation fault caused by the error of the operator. The repositioning process of the invention is fast, because the time consumed by one-time forward calculation of the repositioning convolutional neural network is very short, the precise positioning step only carries out local search feature matching, and the calculation complexity is not large, therefore, the method has extremely fast calculation speed, can avoid the problem of the calculation complexity when the traditional initial positioning method carries out global feature matching, and can greatly reduce the time consumed by the whole initial positioning process.
According to the embodiment of the invention, the global contour map records the position information of the object which can be detected by the laser radar in the AGV running scene. Without loss of generality, the outline characteristics of common objects such as walls, equipment, cabinets, desks and the like can be recorded in a two-dimensional outline map as long as the objects can be detected by a laser radar.
According to the above embodiment of the present invention, a common laser SLAM (simultaneous localization and mapping) method may be adopted in the method for constructing the global contour map of the AGV operation scene, and specifically, the laser SLAM method includes: controlling the AGV to walk around in a running scene, acquiring obstacle information around the AGV in real time by a laser radar installed on the AGV in the process, defining the acquired data information each time as a laser frame, calculating the relative position movement of the AGV at each frame acquisition time according to the laser frames, synthesizing the obstacle information in the environment recorded in the frames into a map, and completing the construction of the global contour map.
Common laser SLAM processes here include GMAP PING, HECTOR, Cartographer, and the like.
According to the embodiment of the invention, the repositioning convolutional neural network belongs to a deep learning network, the network inputs local contour information which can be detected by laser of the AGV at any position, and outputs the possibility that the AGV is located at each position in a global map when the local contour information is collected. The invention adopts the repositioning CNN to simulate a human recognition map and determine the behavior characteristics of the position of the positioning, namely, a possible area which is most consistent with the local contour characteristics of the AGV periphery is quickly found in a global map by observing the local contour characteristics. That is, the relocation CNN records a mapping relationship of "local contour features" → "seeing the positions of these features", through training, and the AGV realizes autonomous location through the recognition and understanding of the environment information.
The network structure of the relocation CNN adopts a CNN (deep convolutional neural network) common architecture, namely the relocation CNN mainly comprises a plurality of alternating convolutional layers and pooling layers, wherein the network structures such as the specific layer number of the convolutional layers and the pooling layers, the number of convolutional cores in each convolutional layer and the like need to be specifically designed and determined according to scene characteristics, then a full-connection layer is added to nonlinearly superpose all extracted local characteristics, finally a SOFTMAX (soft maximum possible value) layer normalization result is added, and the final output is the probability value of each position.
Further, the relocation CNN needs to complete training in the mapping stage without loss of generality, and the training process of the relocation convolutional neural network includes the following steps:
step S201, performing rasterization processing on the two-dimensional contour map to obtain a two-dimensional grid map, and assuming that the whole map is divided into M multiplied by N two-dimensional grid maps without loss of generality;
step S202, collecting each laser frame f collected in the process of drawing by the laser SLAM method1,f2,…,fnAnd recording the positions of the map corresponding to the AGV when the laser frame is acquired, judging that the positions are positioned at the specific grid positions in the two-dimensional grid map, and recording the positions as p1,p2,…,pnWhere any one of p1(i-0, 1, …, n) stores a data pair<j,k>(j. ltoreq. M, k. ltoreq. N), indicating that it is located at the second place in the two-dimensional grid mapj rows and k columns of grids;
step S203, converting f1,f2,…,fnAs a training input sample set, p1,p2,…,pnAnd generating n training samples as an output label set, training the repositioning convolutional neural network to obtain a group of optimized network weights, and finishing the training of the repositioning convolutional neural network. In this embodiment, a gradient back propagation method is used to train the relocation convolutional neural network. The number n of training samples and the training ending condition can be adjusted and optimized according to the actual application effect.
According to the above embodiment of the present invention, the relocation stage consists of two steps, i.e. coarse positioning (step S103) and fine positioning (step S104).
Specifically, the rough positioning step is to calculate the possibility of the AGV being located at each position in the global map by using the relocation CNN. Specifically, a laser radar is used for collecting a group of detected local contour information frames at the moment needing repositioning, a repositioning deep learning network is input, and a forward calculation is carried out on repositioning CNN once to obtain the probability value of the AGV in each grid position in a two-dimensional grid map at the moment.
According to the above embodiment of the present invention, the fine positioning step, i.e. performing feature matching in a local range around the maximum possible position to generate a fine positioning position, includes:
step S301, selecting the coordinate of the grid center point with the highest possibility in the global contour map as the initial position p of the fine positioning calculation according to the probability value of the AGV in each grid obtained by the calculation of the repositioning CNN;
step S302, collecting a laser frame by using a laser radar, and recording all the sensed local contour feature sets as Ec;
step S303, based on the initial position p, setting a certain range around p as a search area p _ search, and determining all possible position sets phi ═ { p } in the search range by a preset search step1,p2,p3,…pn}；
Step S304, for any possible position p 'epsilon phi, projecting Ec to a coordinate system of a two-dimensional contour map according to a coordinate conversion relation to obtain a contour feature set after coordinate transformation as Ect, comparing the matching degree of the Ect and all contour feature sets Emap in the two-dimensional contour map, and calculating the matching score of the position p';
step S305, the position with the highest matching score in the phi set including the initial position p is taken as the accurate positioning position of the current AGV.
Without loss of generality, in step S301, the probability values of the grids may be sorted first, and several grid center points with the highest probability are selected instead of only one grid with the highest probability, so that there are multiple initial positions p, the calculation process may be performed once for each initial position p, the obtained best position result is compared again, and the position with the highest matching score is selected as the accurate positioning position of the current mobile robot.
According to another aspect of the embodiment of the present invention, there is also provided a deep learning based laser navigation AGV repositioning apparatus, as shown in fig. 2, the apparatus including:
the map building module 91 is used for building a global contour map of an AGV running scene;
the neural network construction module 92 is used for training and obtaining a repositioning convolutional neural network by using obstacle information around the AGV;
the rough positioning module 93 is used for acquiring distance information of objects around the AGV to form a local map, inputting the local map into the repositioning convolutional neural network, and calculating to obtain the maximum possible position of the AGV in the global contour map;
a fine positioning module 94 for performing feature matching in a local range around the maximum possible position to produce a fine positioning position.
According to the above embodiment of the present invention, the fine positioning module includes:
the initial setting submodule is used for taking the coordinate of the maximum possible position on the two-dimensional contour map as an initial position p of fine positioning calculation;
the acquisition submodule is used for acquiring a laser frame by using a laser radar, and recording all the perceived local contour feature sets as Ec;
a searching determining module, configured to set a certain range around p as a search region p _ search based on the initial position p, and determine, by a preset search step, all possible position sets Φ ═ p in the search range1,p2,p3,…pn}；
The matching calculation submodule is used for projecting Ec to a coordinate system of a global contour map according to a coordinate conversion relation for any possible position p 'epsilon phi to obtain a contour feature set after coordinate transformation as Ec, comparing the matching degree of the Ec and all contour feature sets Emap in the two-dimensional contour map, and calculating the matching score of the position p';
and the accurate positioning position determining module is used for acquiring all matching scores, and taking the position with the highest matching score in the phi set including the initial position p as the accurate positioning position of the current AGV.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
In the above embodiments of the present invention, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.
In the embodiments provided in the present application, it should be understood that the disclosed technology can be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units may be a logical division, and in actual implementation, there may be another division, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, units or modules, and may be in an electrical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic or optical disk, and other various media capable of storing program codes.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.
Claims (10)
1. A laser navigation AGV repositioning method based on deep learning is characterized by comprising the following steps:
constructing a global contour map of an AGV operation scene;
collecting peripheral obstacle information of each position of an AGV in a running scene to form a training data set, and obtaining a repositioning convolutional neural network by using the data set;
collecting distance information of objects around the AGV to form a local map, inputting the local map into the repositioning convolutional neural network, and calculating to obtain the maximum possible position of the AGV in the global contour map;
feature matching is performed in a local range around the maximum possible position to produce a pinpoint location.
2. The laser navigation AGV repositioning method based on deep learning of claim 1, wherein the global contour map records the position information of the object which can be detected by the laser radar in the AGV running scene.
3. The laser navigation AGV repositioning method based on deep learning of claim 1, wherein the method for constructing the global contour map of the AGV running scene is a laser SLAM method.
4. The deep learning-based laser navigation AGV repositioning method according to claim 1, wherein the laser SLAM method comprises:
controlling the AGV to walk around in a running scene, acquiring obstacle information around the AGV in real time by a laser radar installed on the AGV in the process, defining the acquired data information each time as a laser frame, calculating the relative position movement of the AGV at each frame acquisition time according to the laser frames, synthesizing the obstacle information in the environment recorded in the frames into a map, and completing the construction of the global contour map.
5. The deep learning-based laser navigation AGV repositioning method according to claim 1, wherein the repositioning convolutional neural network mainly comprises a plurality of alternating convolutional layers and pooling layers, and then a full connection layer is added, and finally a SOFTMAX layer is added.
6. The deep learning-based laser navigation AGV repositioning method according to claim 1, wherein the training process of repositioning the convolutional neural network comprises:
rasterizing the two-dimensional outline map to obtain an M multiplied by N two-dimensional grid map;
collecting each collected laser frame f1,f2,…,fnAnd recording the positions of the map corresponding to the AGV when the laser frame is acquired, judging that the positions are positioned at the specific grid positions in the two-dimensional grid map, and recording the positions as p1,p2,…,pnWhere any one of p1(i-0, 1, …, n) stores a data pair<j,k>(j is less than or equal to M, k is less than or equal to N) and represents that the grid is positioned on the jth row and the kth column in the two-dimensional grid map;
will f is1,f2,…,fnAs a training input sample set, p1,p2,…,pnAnd generating n training samples as an output label set, training the repositioning convolutional neural network to obtain a group of optimized network weights, and finishing the training of the repositioning convolutional neural network.
7. The deep learning-based laser navigation AGV repositioning method according to claim 1, wherein a gradient back propagation method is adopted to train the repositioning convolutional neural network.
8. The deep learning based laser navigation AGV repositioning method of claim 1 wherein feature matching is performed within a local range around the maximum possible position to produce a fine positioning position comprising:
taking the coordinate of the maximum possible position on the global contour map as an initial position p of fine positioning calculation;
collecting a laser frame by using a laser radar, and recording all the sensed local contour feature sets as Ec;
setting a certain range around p based on the initial position pFor searching the region p _ search, determining the set of all possible positions within the search range with a preset search step ═ p1,p2,p3,…pn}；
For any possible position p 'epsilon phi, projecting Ec to a coordinate system of a two-dimensional contour map according to a coordinate conversion relation to obtain a contour feature set after coordinate transformation as Ec, comparing the matching degree of the Ec and all contour feature sets Emap in the two-dimensional contour map, and calculating the matching score of the position p';
and taking the position with the highest matching score in the phi set including the initial position p as the accurate positioning position of the current AGV.
9. A laser navigation AGV relocation device based on deep learning, characterized in that includes:
the map construction module is used for constructing a global contour map of the AGV operation scene;
the neural network construction module is used for acquiring peripheral obstacle information of each position of the AGV in a running scene to form a training data set, and obtaining a repositioning convolutional neural network by using the data set;
the rough positioning module is used for acquiring distance information of objects around the AGV to form a local map, inputting the local map into the repositioning convolutional neural network, and calculating to obtain the maximum possible position of the AGV in the global contour map;
a fine positioning module for performing feature matching in a local range around the maximum possible position to generate a fine positioning position.
10. The deep learning based laser navigation AGV repositioning apparatus of claim 9, wherein said fine positioning module comprises:
the initial setting submodule is used for taking the coordinate of the maximum possible position on the two-dimensional contour map as an initial position p of fine positioning calculation;
the acquisition submodule is used for acquiring a laser frame by using a laser radar, and recording all the perceived local contour feature sets as Ec;
a searching determining module, configured to set a certain range around p as a search region p _ search based on the initial position p, and determine, by a preset search step, all possible position sets Φ ═ p in the search range1,p2,p3,…pn}；
The matching calculation submodule is used for projecting Ec to a coordinate system of a global contour map according to a coordinate conversion relation for any possible position p 'epsilon phi to obtain a contour feature set after coordinate transformation as Ec, comparing the matching degree of the Ec and all contour feature sets Emap in the two-dimensional contour map, and calculating the matching score of the position p';
and the accurate positioning position determining module is used for acquiring all matching scores, and taking the position with the highest matching score in the phi set including the initial position p as the accurate positioning position of the current AGV.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010461902.0A CN111578946A (en) | 2020-05-27 | 2020-05-27 | Laser navigation AGV repositioning method and device based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010461902.0A CN111578946A (en) | 2020-05-27 | 2020-05-27 | Laser navigation AGV repositioning method and device based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111578946A true CN111578946A (en) | 2020-08-25 |
Family
ID=72112345
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010461902.0A Pending CN111578946A (en) | 2020-05-27 | 2020-05-27 | Laser navigation AGV repositioning method and device based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111578946A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112711012A (en) * | 2020-12-18 | 2021-04-27 | 上海蔚建科技有限公司 | Global position initialization method and system of laser radar positioning system |
| CN113095227A (en) * | 2021-04-13 | 2021-07-09 | 京东数科海益信息科技有限公司 | Robot positioning method and device, electronic equipment and storage medium |
| CN114236553A (en) * | 2022-02-23 | 2022-03-25 | 杭州蓝芯科技有限公司 | Autonomous mobile robot positioning method based on deep learning |
| CN114510044A (en) * | 2022-01-25 | 2022-05-17 | 北京圣威特科技有限公司 | AGV navigation ship navigation method and device, electronic equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9129190B1 (en) * | 2013-12-04 | 2015-09-08 | Google Inc. | Identifying objects in images |
| CN106323273A (en) * | 2016-08-26 | 2017-01-11 | 深圳微服机器人科技有限公司 | Robot relocation method and device |
| CN109556607A (en) * | 2018-10-24 | 2019-04-02 | 上海大学 | A method of quickly processing localization for Mobile Robot " kidnapping " problem |
| CN110189366A (en) * | 2019-04-17 | 2019-08-30 | 北京迈格威科技有限公司 | A kind of laser rough registration method, apparatus, mobile terminal and storage medium |
| CN110456797A (en) * | 2019-08-19 | 2019-11-15 | 杭州电子科技大学 | A kind of AGV relocation system and method based on 2D laser sensor |
| CN110988795A (en) * | 2020-03-03 | 2020-04-10 | 杭州蓝芯科技有限公司 | Mark-free navigation AGV global initial positioning method integrating WIFI positioning |
-
2020
- 2020-05-27 CN CN202010461902.0A patent/CN111578946A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9129190B1 (en) * | 2013-12-04 | 2015-09-08 | Google Inc. | Identifying objects in images |
| CN106323273A (en) * | 2016-08-26 | 2017-01-11 | 深圳微服机器人科技有限公司 | Robot relocation method and device |
| CN109556607A (en) * | 2018-10-24 | 2019-04-02 | 上海大学 | A method of quickly processing localization for Mobile Robot " kidnapping " problem |
| CN110189366A (en) * | 2019-04-17 | 2019-08-30 | 北京迈格威科技有限公司 | A kind of laser rough registration method, apparatus, mobile terminal and storage medium |
| CN110456797A (en) * | 2019-08-19 | 2019-11-15 | 杭州电子科技大学 | A kind of AGV relocation system and method based on 2D laser sensor |
| CN110988795A (en) * | 2020-03-03 | 2020-04-10 | 杭州蓝芯科技有限公司 | Mark-free navigation AGV global initial positioning method integrating WIFI positioning |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112711012A (en) * | 2020-12-18 | 2021-04-27 | 上海蔚建科技有限公司 | Global position initialization method and system of laser radar positioning system |
| CN113095227A (en) * | 2021-04-13 | 2021-07-09 | 京东数科海益信息科技有限公司 | Robot positioning method and device, electronic equipment and storage medium |
| CN113095227B (en) * | 2021-04-13 | 2023-11-07 | 京东科技信息技术有限公司 | Robot positioning method and device, electronic equipment and storage medium |
| CN114510044A (en) * | 2022-01-25 | 2022-05-17 | 北京圣威特科技有限公司 | AGV navigation ship navigation method and device, electronic equipment and storage medium |
| CN114236553A (en) * | 2022-02-23 | 2022-03-25 | 杭州蓝芯科技有限公司 | Autonomous mobile robot positioning method based on deep learning |
| CN114236553B (en) * | 2022-02-23 | 2022-06-10 | 杭州蓝芯科技有限公司 | Autonomous mobile robot positioning method based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111578946A (en) | Laser navigation AGV repositioning method and device based on deep learning | |
| CN109682382B (en) | Global fusion positioning method based on self-adaptive Monte Carlo and feature matching | |
| JP6921341B2 (en) | Navigation methods, devices, devices, and storage media based on ground texture images | |
| CN105792353B (en) | Crowd sensing type WiFi signal fingerprint assisted image matching indoor positioning method | |
| CN106796434A (en) | Ground drawing generating method, self-position presumption method, robot system and robot | |
| CN101467887B (en) | X ray perspective view calibration method in operation navigation system | |
| CN110907947B (en) | Real-time loop detection method in mobile robot SLAM problem | |
| US20120089295A1 (en) | Moving robot and method to build map for the same | |
| CN114549738A (en) | Unmanned vehicle indoor real-time dense point cloud reconstruction method, system, equipment and medium | |
| CN114088081A (en) | Map construction method for accurate positioning based on multi-segment joint optimization | |
| CN112733971B (en) | Pose determination method, device and equipment of scanning equipment and storage medium | |
| CN110751722A (en) | Method and device for simultaneously positioning and establishing image | |
| CN114266821A (en) | Online positioning method and device, terminal equipment and storage medium | |
| CN112150549B (en) | Visual positioning method based on ground texture, chip and mobile robot | |
| EP4076693A1 (en) | Location determination and mapping with 3d line junctions | |
| Al Khawli et al. | Integrating laser profile sensor to an industrial robotic arm for improving quality inspection in manufacturing processes | |
| CN116630662A (en) | Feature point mismatching eliminating method applied to visual SLAM | |
| Landi et al. | Spot the difference: A novel task for embodied agents in changing environments | |
| CN116127405A (en) | Position identification method integrating point cloud map, motion model and local features | |
| Nunez et al. | An algorithm for fitting 2-D data on the circle: applications to mobile robotics | |
| CN114998539A (en) | Smart city sensor terrain positioning and mapping method | |
| Clipp et al. | Adaptive, real-time visual simultaneous localization and mapping | |
| CN115638795B (en) | Indoor multi-source ubiquitous positioning fingerprint database generation and positioning method | |
| CN117058358B (en) | Scene boundary detection method and mobile platform | |
| Chida et al. | Enhanced Encoding with Improved Fuzzy Decision Tree Testing Using CASP Templates |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200825 |
|
| RJ01 | Rejection of invention patent application after publication |