CN109994036B - Sand table manufacturing method based on machine learning - Google Patents
Sand table manufacturing method based on machine learning Download PDFInfo
- Publication number
- CN109994036B CN109994036B CN201910239668.4A CN201910239668A CN109994036B CN 109994036 B CN109994036 B CN 109994036B CN 201910239668 A CN201910239668 A CN 201910239668A CN 109994036 B CN109994036 B CN 109994036B
- Authority
- CN
- China
- Prior art keywords
- landform
- sand table
- projection
- information
- laser holographic
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B25/00—Models for purposes not provided for in G09B23/00, e.g. full-sized devices for demonstration purposes
- G09B25/06—Models for purposes not provided for in G09B23/00, e.g. full-sized devices for demonstration purposes for surveying; for geography, e.g. relief models
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Educational Administration (AREA)
- Educational Technology (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a sand table manufacturing method based on machine learning, belongs to the technical field of sand table manufacturing, and solves the problems that in the prior art, the sand table manufacturing period is long, the display is lack of authenticity, and real-time adjustment cannot be performed. The method comprises the following steps: constructing a laser holographic projection prediction model and a landform information prediction model, and training; adjusting projection parameters and landform information of the laser holographic projector in real time by using the trained model; the laser holographic projector performs sand table projection imaging according to the projection parameter value; and the mechanical arm carries out sand table landform manufacturing according to the landform information. The sand table manufacturing process is quick and efficient, the laser holographic projection can truly display the state of the simulated landform, and the mechanical arm quickly piles up and adjusts the landform; the model trained by machine learning can be used for solving the optimal projection parameters and the landform information in real time, the projection state and the landform are corrected and adjusted automatically in real time, and the manufactured sand table achieves the optimal projection size and effect.
Description
Technical Field
The invention relates to the technical field of sand table manufacturing, in particular to a sand table manufacturing method based on machine learning.
Background
Military sand tables typically contain two main parts: one is a bottom tray and the other is a simulation of a mountain river. At present, the manufacture of mountains and rivers mainly adopts two materials, one is a plastic simulation product, and the other is sand.
The sand table that uses plastic simulation article as the owner needs to take the form of special customization, at first carries out the drawing of 3D drawing to true physiognomy, carries out the shape that 3D printed or moulded plastics to the drawing after that, makes out simulation article preparation, puts these simulation articles on the tray again for show the physiognomy to other people. On one hand, the sand table is long in preparation period, and the preparation of landforms needs to be carried out in a 3D printing or injection molding mode and needs to prepare scenes; in a sudden state, the landform is not prepared in time, cannot be adjusted in time, and cannot meet the condition of tracking the change of a real scene. On the other hand, the exhibition area is limited, as the landform is manufactured by 3D printing or injection molding glue, the size of the plastic part is required to a certain extent, the cost is correspondingly increased for the size exceeding the conventional size, the reusability is lacked after the plastic part is used, each landform is unique, and the existing landform cannot be applied to other areas for exhibition, so that the waste is caused.
The sand table mainly using sand is characterized in that required sand is covered on a tray firstly, then each characteristic is measured and marked according to an existing drawing (such as a mountain and a river), finally, the sand is piled or dug according to a marking point, the piled part represents a mountain, and the dug part represents a river. On one hand, the sand table is long in preparation and manufacturing period, real landforms need to be scaled in equal proportion, and marking is carried out in the sand table; meanwhile, the whole process of the production needs manual work for producing landform, such as sand piling to represent mountains and sand digging to represent lakes. On the other hand, in the use stage of the scene, firstly, the accuracy is lacked, different landforms can be represented in the same display mode, and people are prone to causing mishaps, for example, a lake and a mountain depression terrain can be simultaneously represented in a sand pit form, and the marking is needed for distinguishing; secondly, the reality is lacked, the color of the sand is single, and the real scene cannot be displayed; finally, the adjustment is inconvenient, the sand needs to be adjusted manually when the landform is changed in use, the difficulty is increased and the efficiency is low according to the complexity of the change and the increase of the size.
Disclosure of Invention
In view of the above analysis, the present invention aims to provide a sand table making method based on machine learning, so as to solve the problems of long making period, lack of authenticity in display and incapability of real-time adjustment of the existing sand table.
The purpose of the invention is mainly realized by the following technical scheme:
the sand table manufacturing method based on machine learning is characterized by comprising the following steps of:
constructing a laser holographic projection prediction model and a landform information prediction model, and training;
adjusting projection parameters and landform information of the laser holographic projector in real time by using the trained model;
the laser holographic projector performs sand table projection imaging according to the projection parameter value;
and the mechanical arm carries out sand table landform manufacturing according to the landform information.
The invention has the following beneficial effects:
the sand table manufacturing process is quick and efficient, and the instantaneity problem and the authenticity problem of the traditional sand table can be effectively solved through machine learning, laser holographic projection and mechanical arm manufacturing; laser holographic projection can be more real show the state of simulation landform (for example, mountain river etc.), uses the arm can be quick pile up the topography and adjust. Simultaneously through constantly monitoring external environment, when sand table size and outside show environment change, utilize the model that machine learning trained to seek optimum projection parameter and landform information in real time, carry out automatic correction and adjustment to the state of projection and landform, the sand table of making reaches best projection size and effect.
On the basis of the scheme, the invention is further improved as follows:
further, the method also comprises the step of collecting laser projection data and landform manufacturing data, wherein the step of collecting the laser projection data comprises the following steps:
dividing the internal area of the sand table, and taking the intersection point of each divided sub-area as a reference point;
installing a binocular camera and a laser holographic projector, locking the whole area of the sand table, and positioning all the datum points;
operating the laser holographic projector, and matching with a calibration tool, and sequentially performing data acquisition on the reference points to obtain a data set required by training a laser holographic projection prediction model;
the data set includes a planar position coordinate, a vertical position coordinate, a focus range, and a sharpness of the fiducial.
Further, the collecting of the geomorphologic data includes:
scanning the contour of the landform, and collecting the corners;
capturing length, width and height information of the landform key points according to the positions of the corners;
and decomposing the parts between the adjacent key points, and acquiring length, width and height data of each decomposed part to obtain complete landform data for training a landform information prediction model.
Further, the data acquisition of the reference points includes:
collecting the length and the width: operating a laser holographic projector to project a cross-shaped cursor, sequentially projecting the cross-shaped cursor at the position of each datum point, adjusting the position of a calibration tool until the cross-shaped cursor in the calibration tool is overlapped with the cross-shaped cursor position, and recording the plane position coordinates of the datum points;
and (3) carrying out depth and depth size collection: and operating the laser holographic projector to project a cross cursor, sequentially projecting the cross cursor at the position of each datum point, keeping the state that the cross cursor in the calibration tool is overlapped with the cross cursor position, moving the calibration tool upwards/downwards until the cross cursor reaches the optimal display effect in a binocular camera display interface, and obtaining the vertical height, the focusing range and the definition information of the position as the vertical position coordinate, the focusing range and the definition of the datum point.
Further, when the positions of the cross in the calibration tool and the cross of the cursor overlap each other in the long and wide dimension collection, the position coordinates of the reference point in the plane are recorded by further calibration using the laser range sensor.
Further, training the constructed laser holographic projection prediction model and the landform information prediction model comprises the following steps:
initializing the network weight and the bias, and randomly selecting a first input sample in the data set;
activating forward propagation by a neuron, performing weight and bias processing on the input sample through a hidden layer, and solving a result and an error of an output layer;
performing back propagation according to the error, and adjusting the network weight and the bias;
and judging whether the training is finished or not according to a preset finishing condition.
Further, the real-time adjustment of the projection parameters of the laser holographic projector by using the trained model comprises: and acquiring the length, width, height and depth information of the sand table, marking a high-complexity area, determining the weight of the high-complexity area, and calculating the focusing range and definition by using the trained model.
Further, after the length, width, height and depth information of the sand table is obtained, searching and matching are carried out in the established existing database, and when the matching is consistent, the corresponding focusing range and the corresponding definition information are directly selected.
Further, adjusting the landform information in real time by using the trained model, comprising: and determining key points of each corner of the contour, solving the distance between adjacent corners, screening and combining the key points according to the distance, and performing detail supplement on parts among the screened key points by using the trained model.
Further, still include: real landform information is monitored in real time, when the landform changes, the interval time of the landform change is compared with a preset reference, and when the interval time of the landform change is smaller than the reference, the actual landform information is ignored; when the interval time of the landform change is greater than the reference, on one hand, key points are collected in a landform change area, landform data are supplemented, and landform building is performed again; and on the other hand, parameter prediction is carried out on the high-complexity region of the landform again to obtain a new focusing range and definition parameters, and projection display is carried out.
In the invention, the technical schemes can be combined with each other to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.
FIG. 1 is a flow chart of a sand table making method based on machine learning according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a sand table area partitioning and calibrating tool according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating the gathering of geomorphic data according to an embodiment of the present invention;
FIG. 4 is a diagram illustrating a neural network structure according to an embodiment of the present invention;
FIG. 5 is a flowchart illustrating training a constructed prediction model according to an embodiment of the present invention;
FIG. 6 is a flowchart illustrating a method for determining projection parameters according to an embodiment of the present invention;
FIG. 7 is a diagram illustrating a method for obtaining topographic data in accordance with an embodiment of the present invention;
FIG. 8 is a diagram illustrating a key point merge result according to an embodiment of the present invention;
FIG. 9 is a schematic diagram of a laser projector according to an embodiment of the present invention;
FIG. 10 is a schematic view of a robotic arm according to an embodiment of the present invention.
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.
The invention discloses a sand table manufacturing method based on machine learning. As shown in fig. 1, the method comprises the following steps:
s1, constructing a laser holographic projection prediction model and a landform information prediction model, and training;
step S2, adjusting the projection parameters and the landform information of the laser holographic projector in real time by using the trained model;
step S3, the laser holographic projector carries out sand table projection imaging according to the projection parameter value;
and step S4, the mechanical arm carries out sand table landform manufacturing according to the landform information.
Compared with the prior art, the sand table manufacturing method based on machine learning provided by the embodiment has the advantages that the sand table manufacturing process is quick and efficient, and the instantaneity problem and the authenticity problem of the traditional sand table can be effectively solved through machine learning, laser holographic projection and mechanical arm manufacturing; laser holographic projection can be more real show the state of simulation landform (for example, mountain river etc.), uses the arm can be quick pile up the topography and adjust. Simultaneously through constantly monitoring external environment, when sand table size and outside show environment change, utilize the model that machine learning trained to seek optimum projection parameter and landform information in real time, carry out automatic correction and adjustment to the state of projection and landform, the sand table of making reaches best projection size and effect.
In particular, the present invention relates to a method for producing,
in step S1, a laser holographic projection prediction model and a landform information prediction model are constructed and trained to obtain a trained laser holographic projection prediction model and a trained landform information prediction model;
in the sand table manufacturing process, the advantages of artificial intelligence, a laser holographic projection technology and mechanical arm manufacturing are fully exerted, the artificial intelligence automatically and immediately obtains an optimal parameter solution through machine learning, the laser holographic projection simulates and shows an actual scene more truly, and the mechanical arm replaces manual work to automatically perform terrain manufacturing and adjustment. In this embodiment, machine learning uses a supervised learning algorithm, a data set containing a large amount of data (the data set has a target value) is first obtained, and the data set is processed by selecting a suitable algorithm, and in order to make the value returned by the prediction model as close to the target value as possible, the weights of the features are continuously adjusted in the training process until the value returned by the prediction model meets the accuracy requirement; specifically, the method can comprise the following steps:
s101, collecting data, constructing a database required by a training model, and collecting the data in a network sharing database or a manual collection mode, preferably, in the manual collection mode, wherein the collection mode comprises laser projection data collection and geomorphologic making data collection;
step S10101, collecting laser projection data. Firstly, dividing the internal area of the sand table, and taking the intersection point of each divided sub-area as a reference point; then, installing a binocular camera and a laser holographic projector, locking the whole area of the sand table, and positioning all the datum points; and finally, operating the laser holographic projector, and matching with a calibration tool to sequentially acquire data of the reference points to obtain a data set required by the training model.
In this embodiment, a rectangular sand table is taken as an example, and as shown in fig. 2, a sand table frame prepared in advance is laid (when the sand table frame is not prepared, bricks can be used for piling or digging 15 to 25 CM down on the flat ground) and is matched with the current place as much as possible, sand soil with appropriate humidity of 3 to 5CM is laid in the sand table frame (the sand soil is picked up by hand to be easy to grasp into a cluster and to be loose when being thrown down), and the sand table frame is flattened and compacted by a wood plate to be used as a reference surface of the sand table. The sand table area is divided (automatically divided by a computer) firstly, the dividing mode can be selected according to the shape of the sand table and the projection display precision requirement, illustratively, a rectangular sand table is divided into 20 sub-rectangles according to a 4-by-5 matrix, and 12 intersection points of the sub-rectangles in the sand table are used as reference points. After the area division is completed, a binocular camera and a laser holographic projector are installed at the position capable of covering the sand table, the binocular camera is used for locking the whole area of the sand table (generally installed right above the central point of the rectangular sand table), the 12 datum points are located, the laser holographic projector is operated (generally installed at the position forming an angle of 45 degrees with the plane of the sand table), so that the projection definition on the datum plane of the sand table is optimal, a calibration tool is matched for carrying out sand table size data acquisition (including length, width and depth dimensions), and a data set of sand table size-12 point positions-focusing power/definition is obtained in a gathering mode.
Collecting the length and the width: and operating the laser holographic projector to project a cross cursor, projecting the cross cursor at the position of the 1 point, and enabling the color of the cursor to be calibrated to be red. A calibrator holds the calibration tool by hand, the cross position in the calibration tool and the cross position of a cursor are overlapped, the focal length value of a binocular camera is kept fixed, calibration is carried out by using a laser ranging sensor (installed right above a sand table), and the position coordinate of a point 1 is recorded (the surface of the sand table is used as a reference surface). It should be noted that, the laser ranging sensor is used for further calibration, so that the size data on the local details can be corrected, and the acquired position information of the reference is more accurate; if the position of the "1" point is calibrated, the color of the cursor changes to green, indicating that the position of the point has been collected. And when the position of the point 1 is calibrated, the cross cursor of the laser holographic projector is moved to the position of the point 2, the calibration mode is the same as the above, and the rest is repeated until all the data of the 12 reference points are collected, and the length and the width of the sand table are collected.
Collecting depth and depth dimensions; and on the basis of completing the collection of the length and the width, the height and the depth of the sand table are collected. And operating the laser holographic projector to project a cross cursor, sequentially projecting the cross cursor at the position of each datum point, keeping the state that the cross cursor in the calibration tool is overlapped with the cross position of the cursor, moving the calibration tool upwards/downwards until the cross cursor reaches the optimal display effect in a binocular camera display interface, and obtaining the vertical height (relative to a datum plane), the focusing range and definition information of the position as the vertical position coordinate, the focusing range and the definition of the datum point.
Specifically, height gathering: and operating the laser projector to project a cross cursor, projecting the cross cursor at the position of a 1 point, and enabling the color of the cursor to be calibrated to be red. The calibration staff holds the calibration tool by hand, overlaps the position of the cross in the calibration tool and the position of the cross of the cursor and moves upwards at the same time, and notes that the cross and the cursor are required to keep an overlapped state in the process of moving upwards. With the 'ten' in the calibration tool being continuously lifted up, the position of the '1' point is calibrated and positioned through the binocular camera, the best display effect (the clearest display) is found on the height, and the obtained height value of the point is the height coordinate of the '1' point (the binocular camera is automatically positioned); the cursor color changes from red to green at this time, indicating that the data collection at the "1" dot height is complete. When the height of the position of the point 1 is collected, the cursor moves to the position of the point 2, the collection mode is the same, and the like is repeated until the data of the point 12 are collected completely, and the height of the sand table is collected completely. Depth collection: and operating the laser projector to project a cross cursor, projecting the cross cursor at the position of a 1 point, and enabling the color of the cursor to be calibrated to be red. The calibration personnel digs a pit in the area of the point 1, then holds the calibration tool to overlap the cross in the calibration tool and the cross of the cursor, and moves downwards towards the bottom of the pit, and the cross and the cursor are required to keep an overlapped state in the process of moving downwards. As the "ten" in the calibration tool is moved down continuously, the best display effect is found in depth by the position of the binocular camera at the "1" point. The cursor color changes from red to green at this time, indicating that data collection for the "1" point depth is complete. When the position depth of the point 1 is collected, the cursor moves to the position of the point 2, the collection mode is the same as the above, and so on, and the depth collection of the sand table is finished until the data of the point 12 are completely collected; and after the height or depth signal is acquired, combining the focusing range and definition of the laser projector on the reference surface, and obtaining the focusing range and definition information of the optimal display position as the focusing range and definition of the corresponding reference point. In addition, when the laser projector performs real environment display, red light, green light and blue light information at the above reference points can be recorded, and data collection statistics are shown in table 1.
Table 1 data collection statistics table:
step S10102, landform making data collection. In this embodiment, a model is built according to the existing landform or the existing sand table landform model, such as data of mountains and rivers, is directly acquired. Specific data are shown in table 2 and table 3; in the scanned contour, collecting the corners, and capturing the length, width and height information of the landform key points (the vertexes of the corners); as shown in fig. 3, according to the real landform state, the length, width and height of the landform are decomposed into small parts (which can be decomposed according to the complexity of the actual landform and the requirement of the display precision), and further, the length, width and height data of the small parts between the corners are collected, i.e. more detailed landform data are obtained to be used as a landform sample of the mountain, so that the prediction model can be learned.
TABLE 2 mountain data
TABLE 3 river data
Step S102, constructing a laser imaging prediction model and a landform information sand table prediction model; through the constructed model, when the size of the sand table and the external environment are changed, the optimal laser holographic projection parameter data and the complete landform information are obtained in real time, and the optimal display effect is achieved. The prediction model can be constructed by various machine learning algorithms, and in the embodiment, the neural networks of the two prediction models are as shown in fig. 4, and numerical information of the focusing power/definition can be automatically obtained by inputting the size of the sand table.
Step S103, training the prediction model constructed in step S102 by using the database obtained in step S101, as shown in fig. 5, specifically:
step S10301, initializing a network weight and a bias; in the initialization phase, a random number is set for the network weights and offsets, and the first input sample is randomly selected from the data set. During the machine learning process, this portion of the network weights and biases are continually adjusted as the actual values differ from the target values.
Step S10302, activating forward propagation by the neuron, performing weight and bias processing on the input sample through a hidden layer, and solving a result and an expected error of an output layer; in the neural network of figure 3 of the drawings,
represents the weight between the ith node of the L-th layer and the jth node of the L + 1-th layer, and the weight between the L1 and the L2 layer is expressed as Ll
The weight between the L2 and L3 layers is
A bias term representing the ith node of the l +1 th layer; by using
Represents the input value of the jth node of the l +1 th layer. When l is equal to 1, the ratio of the total of the two,
and represents the output value of the jth node of the l +1 th layer after the jth node is subjected to the activation function theta (x). The formula is as follows:
thus, one training is completed, and an output result h is obtainedW,b(x)。
And step S10303, performing back propagation according to the error, and adjusting the network weight and the bias. Wherein, for the output layer, Errj＝Oj(1-Oj)(Tj-Oj) For the hidden layer, Errj＝Oj(1-Oj)∑Errkwjk(ii) a Weight update, Δ Wij＝(l)ErrjOj，Wij’＝Wij+ΔWij(ii) a Offset update, Δ θj＝(l)Errj，θj＝j+Δθj。
Step S10304, determining whether training is finished, and determining whether training is finished according to a preset training finishing condition, for example, when the update of the weight is lower than a certain threshold, the predicted error rate is lower than a certain threshold, and a preset certain number of cycles is reached, finishing training, otherwise, continuing to select a new sample for training.
Step S2, adjusting the projection parameters and the topographic information of the laser projector by using the trained model, as shown in fig. 6;
step S201, adjusting projection parameters of the laser projector. When the length, width, height and depth of the sand table are changed by using the trained laser projection prediction model, only size information of the sand table or the length, width, height and depth data of 12 datum points are required to be recorded, and parameters for automatically calculating the focusing range and definition, such as the parameters of x2 and y2 in table 1, are recorded. And the parameters (red, green, blue, focus range, sharpness) are packaged and transmitted to the laser projector through a signal line. The laser projector projects on the sand table according to the projected parameter data. It should be noted that, in order to improve the real-time performance, when the length, width, height, and depth of the sand table change, the existing database constructed may be searched and matched first, and when the matching is consistent, the corresponding focusing range and definition information are directly selected.
It should be emphasized that, in order to improve the reality of the display of the laser projection sand table, when acquiring projection parameters such as the focusing power and the definition, a region with high complexity may be selected so as to improve the prediction effect of the prediction model, which is specifically as follows:
and S20101, marking a high-complexity area, comparing the high-complexity area of the real landform to be simulated with an area of the current sand table, and marking a point location area which accords with the high complexity in the sand table. For example, the high complexity region is 1 point or 4, 5 points … (note that if a specific region is required, the region is marked as a high complexity region by human before the processing terminal performs automatic alignment).
And S20102, determining the weight of the high-complexity area, and weighting the key factors such as the position, the number and the requirement degree with high complexity. Considering that the complexity of each region in the simulated real environment is different, corresponding key factors need to be synthesized so as to confirm a certain point or certain points with the highest requirements on definition and focusing range, and achieve the best projection effect.
And S20103, calculating projection parameters by using the trained model, and processing the focus range and the definition value contained in the covered area by using the trained prediction model according to the length, width, height and depth of the sand table to find the most suitable focus range and definition value. For example, if the weighted selected regions are the locations of point 1 and point 2, as shown in Table 4, then the X1, X2, and Y1, Y2 values are processed to obtain the corresponding X and Y values.
TABLE 4 Focus Range and sharpness solution
Step S202, adjusting the landform information. The trained landform information prediction model is utilized, when the landform changes, the contour of the landform is analyzed, data of a plurality of key points are selected, landform data to be presented are captured in the established landform information database, or correction is carried out according to the result after machine learning and an expected value, finally, the purpose that the details in the landform data can be automatically supplemented according to the contour of the mountain range is achieved, and the effect of being the same as that of a real landform environment is achieved. In particular, the amount of the solvent to be used,
step S20201, determining contour key points, collecting corners in the scanned contour, and taking the vertexes of all corners as key points; as shown in fig. 7, coordinates of point 1 to point N are collected;
step S20202, starting from the first corner, calculating the distances between all adjacent corners, such as distance 1-2, distance 2-3 …;
step S20203, selecting the key points according to the distances, setting a distance value x as a reference, comparing all the distances with the reference, and if the distances are smaller than the reference, discarding the distances, and if the distances are larger than the reference, retaining the distances; for example, if the distance 4-5 is less than the reference value x, point 5 is discarded. Then taking the distance 4-6, comparing with the reference value x, if the distance 4-6 is smaller than the reference value x, abandoning … and so on.
Step S20204, after the screening, if the obtained key points are discontinuous, merging the key points, and performing corresponding processing on discontinuous portions for merging. For example, in fig. 7, point 5 is discarded, and points 4 and 6 are processed on the x and y axes to yield a new point 6 position. After merging, the positions of all points are reordered in order, as shown in fig. 8, for example, the position of a new point 6 is changed to a point 5, and the sorted continuous key points are ready for the next adjustment.
In step S20205, the missing details in the contour between the key points after the filtering and merging are supplemented, and as shown in fig. 9, the missing data of x2, x4, x6, x8, and h2 are supplemented by using the prediction model.
And step S3, the laser holographic projector carries out sand table projection imaging in real time according to the obtained projection parameter values. As shown in fig. 9, in the present embodiment, a three-primary-color laser technology is adopted, and a laser holographic projector uses a laser beam to transmit a picture, wherein optical components of the laser projector mainly include a red, green, blue three-color light valve, a beam combining X-prism, a projection lens, and a driving light valve. There are red, green and blue lasers in the laser projector. The laser is expanded by the corresponding optical element and the processing chip in the machine and then transmitted to the X prism to integrate the three beams of laser, and then the integrated laser is transmitted to the projection curtain by the projection objective, so that the display process of the whole laser projector is completed.
Step S4, the mechanical arm carries out sand table landform manufacturing according to the landform information;
utilize machine learning to carry out automatic regulation control to preparation sand table in-process arm, efficiency is higher, can snatch the data of some key points to the profile of geomorphic according to the geomorphic shape, supplements the detail of other deletions automatically, provides the arm with the complete geomorphic data after supplementing. The mechanical arm quickly piles up and adjusts the landform according to complete landform data, so that the time for manually establishing a model for the landform is saved. Wherein, arm, as shown in fig. 10, includes: moving arms, a chassis, a laser distance sensor and a dexterous hand; wherein, one end of the movable arm is provided with a dexterous hand, the other end is arranged on the chassis, and the movable arm carries out omnibearing rotation by taking a connecting point with the chassis as an axis; the chassis is provided with a moving component (such as a wheel); the dexterous hand can realize the all-round rotation; the number of the laser distance sensors is at least two, and the laser distance sensors are respectively arranged on the dexterous hand and the chassis and used for monitoring obstacles in the moving process. First, the base point of the mechanical arm is determined: before the mechanical arm works, the mechanical arm is placed at a certain position on the periphery of the sand table, and the coordinate of the position is recorded into the control terminal to be used as a position base point of the mechanical arm. Then, the moving position of the mechanical arm is determined, the control terminal calculates the complexity of the whole sand table landform, arranges the sand table landform according to the complexity, and sends the positions of the complexity area to the mechanical arm in sequence (the complexity is high- > low). Finally, the mechanical arm moves through the chassis moving part according to the sequence of the base point position, the high-complexity region position and the high-low complexity region position; at the in-process that removes, measure the barrier through the range finding inductor, prevent the collision, when every removes a assigned position, carry out the piling task through dexterous hand, finish until all position landforms pile up, then the arm playback, remove original reference position, to this, the piling up work of landform is accomplished to the arm.
It should be noted that, in order to solve the problem that the traditional sand table has poor instantaneity, and when the landform changes, real-time adjustment cannot be performed, so as to reflect the real landform. The method in this embodiment further comprises the steps of: real landform information is monitored in real time, latest landform information is obtained, and when the landform changes, whether the constructed landform information and projection are updated or not is judged according to preset conditions;
specifically, a time interval is preset as a reference (the time interval can be set according to actual precision requirements, hardware conditions, sand table scale and the like), when the change of the landform is monitored, the interval time of the landform change is compared with the reference, and if the interval time of the landform change is smaller than the reference, the interval time of the landform change is ignored; if the interval time of the landform change is greater than the reference, on one hand, collecting key points in a new landform change area, and supplementing landform data so as to facilitate the stacking of the mechanical arm; on the other hand, the method can perform parameter prediction again on the complexity area of the landform to obtain a new focusing range and definition parameters for projection display.
Considering the time cost required by the mechanical arm to re-stack the terrain, wherein the interval time of change of the terrain is greater than the reference, after the new terrain is collected, in the embodiment, different stacking modes are further adopted according to the degree of change of the terrain, and specifically, the collected new critical point and the recently collected critical point parameter are compared in a preset error range (specific numerical values can be set according to conditions such as actual display effect, hardware equipment and real-time requirement); when the updated key point parameter is larger than the error range, the mode of 'reforming' is adopted, and when the updated key point parameter is smaller than the error range, the mode of 'small-range correction' is adopted.
In the remodeling mode, the landform is smoothed and remodeled, and the mechanical arm firstly smoothes the landform of the area into an initial state according to initial parameters (a height Z coordinate in the initial state and an X, Y axis coordinate); then, the obtained latest geomorphic information is utilized to be re-tiled (step S202 is executed). In the short-range correction mode, the region with changed features is further used as a reference, the key points of the contour in a certain range around the region are collected, the latest feature information is obtained, and the contour is re-tiled (step S202).
Artificial intelligence, a laser holographic imaging technology and mechanical arm landform manufacturing are added in a military sand table, and the problems brought by the existing sand table can be solved. The laser holographic imaging technique can solve the problems of accuracy and authenticity. On the basis, interaction functions such as distance measurement and position marking can be added, and more information on the battlefield can be acquired in real time.
Those skilled in the art will appreciate that all or part of the flow of the method implementing the above embodiments may be implemented by hardware associated with computer program instructions, and the program may be stored in a computer readable storage medium. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
Claims (8)
1. A sand table manufacturing method based on machine learning is characterized by comprising the following steps:
constructing a laser holographic projection prediction model and a landform information prediction model, and training;
adjusting projection parameters and landform information of the laser holographic projector in real time by using the trained model;
carrying out sand table projection imaging by the laser holographic projector according to the numerical value of the projection parameter;
the mechanical arm carries out sand table landform manufacturing according to the landform information;
further comprising performing laser projection data gathering, the performing laser projection data gathering comprising:
dividing the internal area of the sand table, and taking the intersection point of each divided sub-area as a reference point;
installing a binocular camera and a laser holographic projector, locking the whole area of the sand table, and positioning all the datum points;
operating the laser holographic projector, and matching with a calibration tool, and sequentially performing data acquisition on the reference points to obtain a data set required by training a laser holographic projection prediction model; wherein, the data acquisition of each reference point comprises:
collecting the length and the width: operating a laser holographic projector to project a cross-shaped cursor, sequentially projecting the cross-shaped cursor at the position of each datum point, adjusting the position of a calibration tool until the cross-shaped cursor in the calibration tool is overlapped with the cross-shaped cursor position, and recording the plane position coordinates of the datum points;
and (3) carrying out depth and depth size collection: operating a laser holographic projector to project a cross cursor, sequentially projecting the cross cursor at the position of each datum point, keeping the overlapping state of the cross and the cross of the cursor in a calibration tool, moving the calibration tool upwards/downwards until the cross cursor reaches the optimal display effect in a binocular camera display interface, and obtaining the vertical height, the focusing range and the definition information of the position as the vertical position coordinate, the focusing range and the definition of the datum point;
the data set includes a planar position coordinate, a vertical position coordinate, a focus range, and a sharpness of the fiducial.
2. The method of claim 1, further comprising performing a geomorphologic data collection, the performing the geomorphologic data collection comprising:
scanning the contour of the landform, and collecting the corners;
capturing length, width and height information of the landform key points according to the positions of the corners;
and decomposing the parts between the adjacent key points, and acquiring length, width and height data of each decomposed part to obtain complete landform data for training a landform information prediction model.
3. The method of claim 2, wherein during the length and width dimension gathering, when the cross in the calibration tool and the cross in the cursor overlap, further calibration is performed by a laser range sensor to record the planar position coordinates of the reference point.
4. The method of claim 3, wherein the training of the constructed laser holographic projection prediction model and the landform information prediction model comprises:
initializing the network weight and the bias, and randomly selecting a first input sample in the data set;
activating forward propagation by a neuron, performing weight and bias processing on the input sample through a hidden layer, and solving a result and an error of an output layer;
performing back propagation according to the error, and adjusting the network weight and the bias;
and judging whether the training is finished or not according to a preset finishing condition.
5. The method of claim 4, wherein the real-time adjustment of the projection parameters of the laser holographic projector by using the trained model comprises: and acquiring the length, width, height and depth information of the sand table, marking a high-complexity area, determining the weight of the high-complexity area, and calculating the focusing range and definition by using the trained model.
6. The method according to claim 4 or 5, characterized in that after the information of the length, width, height and depth of the sand table is obtained, the existing database is searched for matching, and when the matching is consistent, the corresponding focusing range and definition information are directly selected.
7. The method of claim 6, wherein adjusting the geomorphic information in real time using the trained models comprises: and determining key points of each corner of the contour, solving the distance between adjacent corners, screening and combining the key points according to the distance, and performing detail supplement on parts among the screened key points by using the trained model.
8. The method of claim 7, further comprising: real landform information is monitored in real time, when the landform changes, the interval time of the landform change is compared with a preset reference, and when the interval time of the landform change is smaller than the reference, the actual landform information is ignored; when the interval time of the landform change is greater than the reference, on one hand, key points are collected in a landform change area, landform data are supplemented, and landform building is performed again; and on the other hand, parameter prediction is carried out on the high-complexity region of the landform again to obtain a new focusing range and definition parameters, and projection display is carried out.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910239668.4A CN109994036B (en) | 2019-03-27 | 2019-03-27 | Sand table manufacturing method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910239668.4A CN109994036B (en) | 2019-03-27 | 2019-03-27 | Sand table manufacturing method based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109994036A CN109994036A (en) | 2019-07-09 |
| CN109994036B true CN109994036B (en) | 2020-12-29 |
Family
ID=67131782
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910239668.4A Active CN109994036B (en) | 2019-03-27 | 2019-03-27 | Sand table manufacturing method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109994036B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110428757B (en) * | 2019-07-25 | 2021-07-13 | 湖北安心智能科技有限公司 | Projection sand table system control method and device and readable storage medium |
| CN110444100B (en) * | 2019-07-25 | 2021-10-15 | 湖北安心智能科技有限公司 | Projection sand table capable of automatically constructing terrain and projection system |
| CN115131350B (en) * | 2022-08-30 | 2022-12-16 | 南京木木西里科技有限公司 | Large-depth-of-field observation and surface topography analysis system |
| CN115512083B (en) * | 2022-09-20 | 2023-04-11 | 广西壮族自治区地图院 | Multi-inclination-angle numerical control sand table self-adaptive projection method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016078029A1 (en) * | 2014-11-19 | 2016-05-26 | 刘海燕 | Urban planning model display system |
| CN205487097U (en) * | 2016-01-18 | 2016-08-17 | 江苏达科智能科技有限公司 | Electron sand table operation panel |
| CN206115829U (en) * | 2016-05-18 | 2017-04-19 | 杭州贝赞成教育科技有限公司 | Mutual equipment of sand table |
Family Cites Families (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104764415A (en) * | 2014-12-31 | 2015-07-08 | 中铁宝桥集团有限公司 | Steel bridge structure three-dimensional measurement and detection system and measurement method |
| CN105096730A (en) * | 2015-06-23 | 2015-11-25 | 南京光锥信息科技有限公司 | Mechanism and method for configuring 2.5D (2.5-dimensional) array terrains |
| CN105045389B (en) * | 2015-07-07 | 2018-09-04 | 深圳水晶石数字科技有限公司 | A kind of demenstration method of interactive sand table system |
| CN206040084U (en) * | 2016-05-11 | 2017-03-22 | 深圳市盈天下广告有限公司 | Interactive sand table of holographic vision developments |
| CN106094721B (en) * | 2016-07-01 | 2018-06-01 | 西安工业大学 | The processing method of three-dimensional live topographic map sand table |
| CN107705241B (en) * | 2016-08-08 | 2021-08-03 | 国网新疆电力公司 | Sand table construction method based on tile terrain modeling and projection correction |
| CN106340064B (en) * | 2016-08-25 | 2019-02-01 | 北京大视景科技有限公司 | A kind of mixed reality sand table device and method |
| CN106353957A (en) * | 2016-09-29 | 2017-01-25 | 南京仁光电子科技有限公司 | Laser-projector-based interaction device and method |
| CN106448430A (en) * | 2016-11-10 | 2017-02-22 | 杭州绿建建筑模型有限公司 | A technique for manufacturing a three-dimensional sand table model |
| WO2018140348A1 (en) * | 2017-01-26 | 2018-08-02 | The Charles Stark Draper Laboratory, Inc. | Method and apparatus for light field generation |
| CN106980366A (en) * | 2017-02-27 | 2017-07-25 | 合肥安达创展科技股份有限公司 | Landform precisely catches system and fine high definition projection imaging system |
| CN106782026B (en) * | 2017-03-30 | 2023-07-18 | 南京工业职业技术学院 | Array stacking rod type three-dimensional sand table forming machine |
| CN208422342U (en) * | 2018-08-08 | 2019-01-22 | 深圳市大峡谷科技发展有限公司 | The projection of dual-projection three-dimensional imaging digital sand table controls equipment |
-
2019
- 2019-03-27 CN CN201910239668.4A patent/CN109994036B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016078029A1 (en) * | 2014-11-19 | 2016-05-26 | 刘海燕 | Urban planning model display system |
| CN205487097U (en) * | 2016-01-18 | 2016-08-17 | 江苏达科智能科技有限公司 | Electron sand table operation panel |
| CN206115829U (en) * | 2016-05-18 | 2017-04-19 | 杭州贝赞成教育科技有限公司 | Mutual equipment of sand table |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109994036A (en) | 2019-07-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109994036B (en) | Sand table manufacturing method based on machine learning | |
| US11915501B2 (en) | Object detection method and apparatus, electronic device, and storage medium | |
| US11875537B2 (en) | Multi view camera registration | |
| CN110009985B (en) | Sand table making devices based on machine learning | |
| US7196730B2 (en) | Method and system for complete 3D object and area digitizing | |
| CN110146099B (en) | Synchronous positioning and map construction method based on deep learning | |
| KR101347260B1 (en) | Drawing image system be able to input the numeric data to the aviation image | |
| CN112132972A (en) | Three-dimensional reconstruction method and system for fusing laser and image data | |
| CN107967463B (en) | Virtual face recognition method based on synthetic image and deep learning | |
| CN110334701B (en) | Data acquisition method based on deep learning and multi-vision in digital twin environment | |
| CN106097443B (en) | City indoor and outdoor integrated three-dimensional scenario building and spatially adaptive air navigation aid | |
| CN109255302A (en) | Object recognition methods and terminal, mobile device control method and terminal | |
| CN114998536A (en) | Model generation method and device based on novel basic mapping and storage medium | |
| CN106705849A (en) | Calibration method of linear-structure optical sensor | |
| CN107230233A (en) | The scaling method and device of telecentric lens 3-D imaging system based on bundle adjustment | |
| CN113963044A (en) | RGBD camera-based intelligent loading method and system for cargo box | |
| CN109856946B (en) | Method for manufacturing laser holographic projection sand table | |
| CN108375985A (en) | A kind of soil three-dimensional planning and designing platform and its design method | |
| CN116758234A (en) | Mountain terrain modeling method based on multipoint cloud data fusion | |
| CN102693543A (en) | Method for automatically calibrating Pan-Tilt-Zoom in outdoor environments | |
| CN109856945B (en) | Laser holographic projection sand table manufacturing device | |
| CN106384331A (en) | High-definition image synthetic method based on precision displacement | |
| CN116704112A (en) | 3D scanning system for object reconstruction | |
| CN104200469A (en) | Data fusion method for vision intelligent numerical-control system | |
| CN112489190B (en) | Full-automatic indoor scanning method, system and computer readable storage medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |