CN110348368A - A kind of artificial intelligence analytic method, computer-readable medium and the system of floor plan - Google Patents
A kind of artificial intelligence analytic method, computer-readable medium and the system of floor plan Download PDFInfo
- Publication number
- CN110348368A CN110348368A CN201910611813.7A CN201910611813A CN110348368A CN 110348368 A CN110348368 A CN 110348368A CN 201910611813 A CN201910611813 A CN 201910611813A CN 110348368 A CN110348368 A CN 110348368A
- Authority
- CN
- China
- Prior art keywords
- feature
- door
- window
- floor plan
- sample graph
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/40—Document-oriented image-based pattern recognition
- G06V30/42—Document-oriented image-based pattern recognition based on the type of document
- G06V30/422—Technical drawings; Geographical maps
Abstract
The present invention relates to a kind of artificial intelligence analytic methods of floor plan, method includes the following steps: step S1: obtaining house type sample graph;Step S2: the building part in house type sample graph is obtained by shape feature, and isolates wall feature and door and window feature from building part;Step S3: based on deep learning models coupling door and window feature one door and window detection model of training, and online difficult sample method for digging is combined to optimize door and window detection model, to detect the feature of door and the feature of window;And step S4: 3D floor plan is precipitated according to the feature of door, the feature of window and wall characteristic solution in house type sample graph.To realize the two-dimentional floor plan of unmanned parsing, artificial parsing cost is reduced, analyzing efficiency is increased, optimize door and window detection model in combination with online difficult sample method for digging, to promote the detection precision and detection speed of door and window detection model.The present invention also provides a kind of computer-readable mediums.The present invention also provides a kind of artificial intelligence resolution systems of floor plan.
Description
[technical field]
The present invention relates to artificial intelligence analytic method, the computers of field of house buildings more particularly to a kind of floor plan can
Read medium and system.
[background technique]
It according to the preliminary stage of floor plan building, is typically necessary and floor plan is parsed, primarily to obtaining
The important elements such as the house type frame of floor plan, the wall in floor plan, door, window, then according to the element combination market of acquisition
In upper existing 3D modeling software, 3D floor plan can be automatically generated.So that house purchaser purchases according to visual 3D floor plan
Room, while promoting house-purchase experience, moreover it is possible to by the house type geological information of extraction, construction personnel be assisted to complete the works such as construction budget
Make.
Traditional floor plan parsing is generally dependent upon manually based on Cad (Computer Aided Design, computer
Computer Aided Design) software manual parsing, high labor cost, analyzing efficiency are low.
[summary of the invention]
Of the existing technology to overcome the problems, such as, the present invention provides a kind of artificial intelligence analytic method of floor plan, calculates
Machine readable medium and system.
The scheme that the present invention solves technical problem is to provide a kind of artificial intelligence analytic method of floor plan, and this method includes
Following steps: step S1: house type sample graph is obtained;Step S2: obtaining the building part in house type sample graph by shape feature,
And wall feature and door and window feature are isolated from building part;Step S3: it is instructed based on deep learning models coupling door and window feature
Practice a door and window detection model, and combines online difficult sample method for digging (OHEM, Online Hard Example Mining)
Optimize door and window detection model, to detect the feature of door and the feature of window;And step S4: according to the feature of door in house type sample graph,
3D floor plan is precipitated in the feature and wall characteristic solution of window.
Preferably, step S3 is based on deep learning models coupling door and window feature one door and window detection model of training, to detect door
Feature and window feature, further include steps of step S31: extract in sample graph area-of-interest (ROI,
Region Of Interest);Step S32: dimension-reduction treatment is carried out to area-of-interest by pond layer;Step S33: by complete
Articulamentum calculates classification of the area-of-interest in sample graph;Step S34: according to frame homing method (Bounding Box
Regression the offset predicted value for) obtaining area-of-interest, is optimized with the position to area-of-interest, and calculates the
One loss function;And step S35: optimize first-loss function in conjunction with online difficult sample method for digging.
Preferably, step S31 extracts the area-of-interest in sample graph, further includes steps of step before
S301: use largely include door and window floor plan as training set, and select using shape circle the spy of door in door and window feature
It seeks peace the feature of window;Step S302: corresponding label is made according to the feature of shape frame feature on the door and window;And step S303:
The feature of door and the feature of window are set as area-of-interest.
Preferably, step S1 obtains house type sample graph, further includes steps of step S101 later: adjustment sample
The pixel of figure;And step S102: sample graph adjusted is cut within setting specification.
Preferably, in above-mentioned steps S2, the closing surrounded in sample graph by line segment is extracted by depth-first search traversal method
Region, to obtain the building part in sample graph.
Preferably, step S2 obtains the building part in house type sample graph by depth-first search traversal method, and from building
Obtaining wall feature and door and window feature in part, it further includes steps of step S21: extracting every line segment in sample graph
The coordinate of middle two-end-point;Step S22: using any line segment coordinate wherein as starting point, and select the coordinate of adjacent segments into
Row connection forms path and is traversed;Step S23: being back to starting point, forms a closed area according to the path of traversal;And step
Rapid S24: the coordinate for choosing a point again is starting point, recycles above-mentioned steps S22- step S23, until owning in traversal sample graph
Line segment, and obtain the building part of sample graph.
Preferably, step is further included steps of after the building part that step S24 is obtained in sample graph
S241: the noise in sample graph is eliminated by connected domain;Step S242: the edge line segment of building part is extracted, and combines and eliminates
Building part after noise reinforces the texture of building part edge line segment;Step S243: it is restored using preset structural element
The thickness of building part;Step S244: according to pixel grey scale, and defining straight features is segmentation threshold, identifies wall feature;
And step S245: the open region in retrieval wall feature, using the basis detected as door and window feature.
The present invention also provides a kind of computer-readable medium, it is stored with computer program in the computer-readable medium,
Wherein, the computer program is arranged to execute the artificial intelligence analytic method of above-mentioned floor plan when operation.
The present invention also provides a kind of artificial intelligence resolution system of floor plan, the artificial intelligence resolution systems of the floor plan
Include: read module, is configured as obtaining house type sample graph;Identification module is configured as obtaining house type sample by shape feature
Building part in this figure, and wall feature and door and window feature are isolated from building part;Door and window detection module, is configured as
Based on deep learning models coupling door and window feature one door and window detection model of training, and combine online difficult sample method for digging optimization
Door and window detection model, to detect the feature of door and the feature of window;And parsing module, it is configured as according to door in house type sample graph
3D floor plan is precipitated in feature, the feature of window and wall characteristic solution.
Preferably, door and window detection module further comprises: extraction module, is configured as extracting interested in sample graph
Region;Pond module is configured as carrying out dimension-reduction treatment to area-of-interest by pond layer;Full link block, is configured as
Classification of the area-of-interest in sample graph is calculated by full articulamentum;Frame regression block is configured as being returned according to frame
Method obtains the offset predicted value of area-of-interest, is optimized with the position to area-of-interest, and calculate first-loss
Function;And optimization module, it is configured as combining formerly difficult sample method for digging optimization first-loss function.
Compared with prior art, artificial intelligence analytic method, computer-readable medium and the house type of floor plan of the invention
The artificial intelligence resolution system of figure has the advantage that
1. obtaining the building part in house type sample graph by shape feature, and wall feature and door and window feature are isolated,
And then pass through the door and window detection model of deep learning model training, and combine online difficult sample method for digging optimization door and window detection
Model parses house type sample finally by the feature combination 3D software detected to detect the feature of the feature and window gone out
Scheme corresponding 3D floor plan, to realize the two-dimentional floor plan of unmanned parsing, reduces artificial parsing cost, increase analyzing efficiency, together
When in conjunction with the first-loss function in online difficult sample method for digging optimization door and window detection model, to promote door and window detection model
Detection precision.
2. by depth-first search traversal method combined shape feature, by the size marking and text note in house type sample graph
It releases and filters out, and then obtain the building part in house type sample graph, interference when door and window is detected to reduce.
3. the artificial intelligence analytic method of floor plan is stored in computer program, in order to computer-readable medium fortune
The row computer program reduces artificial parsing cost, increases analyzing efficiency to realize the two-dimentional floor plan of unmanned parsing.
4. read module, identification module, door and window detection module and solution are arranged in the artificial intelligence resolution system of floor plan
Module is analysed, to realize the two-dimentional floor plan of unmanned parsing, artificial parsing cost is reduced, increases analyzing efficiency.
[Detailed description of the invention]
Fig. 1 is the flow diagram of the artificial intelligence analytic method of first embodiment of the invention floor plan.
Fig. 2 is the process signal in the artificial intelligence analytic method of first embodiment of the invention floor plan before step S1
Figure.
Fig. 3 is the flow diagram of step S2 in the artificial intelligence analytic method of first embodiment of the invention floor plan.
Fig. 4 is that depth-first search traversal method detects wall in the artificial intelligence analytic method of first embodiment of the invention floor plan
The schematic diagram of body characteristics.
Fig. 5 is the flow diagram of step S24 in the artificial intelligence analytic method of first embodiment of the invention floor plan.
Fig. 6 is the flow diagram of step S3 in the artificial intelligence analytic method of first embodiment of the invention floor plan.
Fig. 7 is the feature of door in the artificial intelligence analytic method separation house type sample graph of first embodiment of the invention floor plan
With the schematic diagram before the feature of window.
Fig. 8 is the signal in the artificial intelligence analytic method detection wall characterized openings area of first embodiment of the invention floor plan
Figure.
Fig. 9 is the process signal in the artificial intelligence analytic method of first embodiment of the invention floor plan before step S31
Figure.
Figure 10 is the module diagram of the artificial intelligence resolution system of third embodiment of the invention floor plan.
Figure 11 is that the module of door and window detection module in the artificial intelligence resolution system of third embodiment of the invention floor plan is shown
It is intended to.
Description of symbols: 1, the artificial intelligence resolution system of floor plan;11, read module;12, identification module;13, door
Window detection module;14, parsing module;131, extraction unit;132, pond unit;133, full connection unit;134, frame returns
Unit;135, optimize unit.
[specific embodiment]
In order to make the purpose of the present invention, technical solution and advantage are more clearly understood, below in conjunction with attached drawing and embodiment,
The present invention will be described in further detail.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention,
It is not intended to limit the present invention.
Referring to Fig. 1, first embodiment of the invention provides a kind of artificial intelligence analytic method of floor plan, this method includes
Following steps:
Step S1: house type sample graph is obtained;
Step S2: the building part in house type sample graph is obtained by shape feature, and isolates wall from building part
Body characteristics and door and window feature;
Step S3: it based on deep learning models coupling door and window feature one door and window detection model of training, and combines online difficult
Sample method for digging optimizes door and window detection model, to detect the feature of door and the feature of window;And
Step S4: 3D floor plan is precipitated according to the feature of door, the feature of window and wall characteristic solution in house type sample graph.
Firstly, obtaining house type sample graph based on architectural drawing;Then, pass through depth-first search traversal method combined shape feature
The closed area surrounded in sample graph by line segment is extracted, to obtain the building part in sample graph, and is separated from building part
Wall feature and door and window feature out;In turn, it based on deep learning models coupling door and window feature one door and window detection model of training, and ties
It closes online difficult sample method for digging and optimizes door and window detection model, to detect the feature of door and the feature of window;Finally, according to house type
3D floor plan, i.e., the door that will be detected in house type sample graph is precipitated in the feature of door, the feature of window and wall characteristic solution in sample graph
Feature, the feature of window and wall feature be input in 3D modeling software, to automatically generate the family 3D corresponding with house type sample graph
Type figure.
It is appreciated that architectural drawing can be to be obtained, by architectural drawing by the two dimensional draftings software design such as Cad software
Format is converted to the general format of dxf format or other two dimensional drafting softwares, can be obtained house type sample graph;
The non-building part in house type sample graph can be filtered out by depth-first search traversal method, as size marking,
Textual annotation etc.;Optimize door and window detection model by online difficult sample method for digging, to promote the convergence of door and window detection model
Speed and detection accuracy;
Deep learning model includes but is not limited to convolutional neural networks RCNN (RegionConvolutional Neural
) and Recognition with Recurrent Neural Network RNN (Recurrent Neural Network) etc. Network;
Online difficulty sample method for digging is difficult sample to be filtered out from floor plan sample, and then screening is obtained
These samples are applied to be trained in stochastic gradient descent, excellent to carry out to the first-loss function in door and window detection model
Change, difficult sample can be the higher feature of two or more similarities.
Referring to Fig. 2, step S1 obtains house type sample graph, further include steps of later
Step S101: the pixel of sample graph is adjusted;And
Step S102: sample graph adjusted is cut within setting specification.
First the current size of sample graph and setting specification are compared, to adjust the pixel of sample graph, then will
Sample graph adjusted is cut to setting specification, with the specification of unified samples figure, keeps content information all in sample graph more straight
It sees, to increase the processing speed of the feature of subsequent identification wall feature, the feature of door and window.
Referring to Fig. 3, step S2 obtains building in house type sample graph by depth-first search traversal method combined shape feature
Part is built, and obtains from building part wall feature and door and window feature it is further included steps of
Step S21: the coordinate of two-end-point in every line segment in sample graph is extracted;
Step S22: using any line segment coordinate wherein as starting point, and selecting the coordinate of adjacent segments to be attached,
Path is formed to be traversed;
Step S23: being back to starting point, forms a closed area according to the path of traversal;And
Step S24: the coordinate for choosing a point again is starting point, recycles above-mentioned steps S22- step S23, until traversal sample
All line segments in this figure, and obtain the building part of sample graph.
Firstly, corresponding two-end-point coordinate is extracted in the position based on every line segment in sample graph, depth is then used
First traversal method selects the coordinate of adjacent segments to be attached using any line segment coordinate wherein as starting point, with shape
It is traversed at path, then, is back to starting point, a closed area is formed according to the path of traversal;Finally, choosing one again
The coordinate of point is starting point, recycles above-mentioned steps, until all line segments in traversal sample graph, and obtain the building part of sample graph.
Referring to Fig. 4, there is the letter such as size marking, textual annotation, building part in application scenes, in sample graph
Breath, the two-end-point coordinate of every line segment first in acquisition sample graph, then, using first traversal method with one end in line segment A
Point sets out to other end point, and the coordinate of adjacent segments B is selected to be attached, and is traversed with forming path, is then successively passed through
The coordinate for crossing line segment C and line segment D is finally returned to starting point, completes primary traversal, and according to the path of traversal obtain by line segment A,
B, C, D-shaped at enclosure space P;
And when being traversed to line segment E, F, G, due to can not according to the path of traversal formed enclosure space, line segment E,
F, G is not belonging to building part.
Fig. 5-Fig. 7 is please referred to, is further included steps of after the building part that step S24 is obtained in sample graph
Step S241: the noise in sample graph is eliminated by connected domain;
Step S242: the edge line segment of building part is extracted, and combines the building part eliminated after noise, reinforces building
The texture of part edge line segment;
Step S243: the thickness of building part is restored using preset structural element;
Step S244: according to pixel grey scale, and defining straight features is segmentation threshold, identifies wall feature;And
Step S245: the open region in retrieval wall feature, using the basis detected as door and window feature.
Firstly, passing through the noise in connected domain elimination sample graph in conjunction with the pixel in sample graph;Then, building portion is extracted
The edge line segment divided, and the building part eliminated after noise is combined, reinforce the texture of building part edge line segment;In turn, it adopts
The thickness that building part is restored with preset structural element, can compare with the thickness in former floor plan, extraction and original image
The highest figure of thickness similitude;In turn, according to pixel grey scale, and defining straight features is segmentation threshold, identifies wall spy
Sign, that is, defining solid section and hollow part in building part is segmentation threshold, is split to building part, to identify
Wall feature, as shown in fig. 6, wherein solid section is expressed as concrete walls P1, hollow parts are expressed as brick wall P2, Jin Ergen
Wall feature is identified according to solid section and hollow parts, and having hollow wire body and solid lines in hollow parts is sliding door
Q1, hollow parts have the straight line of connection outside and camber line is vertical hinged door Q2, and having two lines section in hollow parts, then correspondence is expressed as
Windowpane R1, hollow parts have a plurality of line segment of enclosing is then corresponding to be expressed as burglary-resisting window R2 outside;Finally, in retrieval wall feature
Open region that is, after wall feature is divided and comes out, can filter out door and window part using the basis detected as door and window feature
Feature forms open region, as shown in Figure 7 in wall feature.
It is appreciated that the number to pixel in connected domain sets a threshold value, when pixel is less than in a connected domain
The half of threshold value then determines region corresponding to the connected domain for noise, conversely, then determining area corresponding to the connected domain
Domain is building part;To prevent operation during eliminating noise leads to the edge blurry of building part and causes the mistake identified
Difference, it is therefore desirable to which texture reinforcement processing is carried out to the line segment at building part edge;Preset structural element can be the structure of 3*3
Element, the structural element of 4*4, structural element of 5*5 etc..
Referring to Fig. 8, step S3 is based on deep learning models coupling door and window feature one door and window detection model of training, with detection
The feature of door and the feature of window, further include steps of
Step S31: the area-of-interest in sample graph is extracted;
Step S32: dimension-reduction treatment is carried out to area-of-interest by pond layer;
Step S33: classification of the area-of-interest in sample graph is calculated by full articulamentum;
Step S34: the offset predicted value of area-of-interest is obtained according to frame homing method, to area-of-interest
Position optimizes, and calculates first-loss function;And
Step S35: optimize first-loss function in conjunction with online difficult sample method for digging.
Firstly, the area-of-interest in sample graph is extracted by the convolutional layer in deep learning model, so that deep learning
The part for area-of-interest of model is handled, to reduce the processing time;Then, by pond layer to region of interest
Feature in domain carries out dimension-reduction treatment, in order to avoid the feature in area-of-interest is too small, can adjust maximum pond layer (Max
Pool core size);In turn, classification of the feature in sample graph in area-of-interest is calculated by full articulamentum, i.e., it will be complete
Each characteristic point in articulamentum is connect entirely with all characteristic points of preceding layer, to calculate each spy in area-of-interest
The corresponding classification of sign point, such as any classification in windowpane, burglary-resisting window, vertical hinged door or sliding door;In turn, it is returned according to frame
Method obtains the offset predicted value of characteristic point in area-of-interest, is carried out with the position to characteristic point in area-of-interest excellent
Change, for example, a frame frame can be used to select characteristic point, and is indicated with a four-tuple (x, y, w, h), wherein (x, y) table
Show that the centre coordinate of frame, (w, h) indicate the width and height of frame, when frame returns, by carrying out to four-tuple (x, y, w, h)
The transformation such as translation, scaling, the position of the characteristic point after being optimized then calculate the first-loss letter of deep learning model at this time
Number;Finally, online difficult sample method for digging is combined to optimize first-loss function, the training of door and window detection model is completed, simultaneously
The feature and window for the door for detecting door and window detection model more accurately.
It is appreciated that the characteristic point in area-of-interest is the characteristic point extracted from the feature, the feature of window of door,
Feature is by multiple feature point groups at the feature of feature and window including door.
Referring to Fig. 9, step S31 extracts the area-of-interest in sample graph, further include steps of before
Step S301: using largely includes that the floor plan of door and window selects door and window as training set, and using shape circle
The feature of the feature of door and window in feature;
Step S302: corresponding label is made according to the feature of shape frame feature on the door and window;And
Step S303: the feature of door and the feature of window are set as area-of-interest.
Firstly, when training deep learning model, use largely include door and window floor plan as training set, and using shape
Shape circle selects the feature of the feature of door and window in door and window feature;Then, made according to shape frame feature on the door and the feature of window
Corresponding label;Finally, setting the feature of door and the feature of window as area-of-interest according to label.
Second embodiment of the invention provides a kind of computer-readable medium, and computer journey is stored in computer-readable medium
Sequence, wherein computer program is arranged to execute the artificial intelligence analytic method of above-mentioned floor plan when operation.
In accordance with an embodiment of the present disclosure, it may be implemented as computer software journey above with reference to the process of flow chart description
Sequence.For example, embodiment of the disclosure includes a kind of computer program product comprising carry meter on a computer-readable medium
Calculation machine program, the computer program include the program code for method shown in execution flow chart.In such embodiments,
The computer program can be downloaded and installed from network by communications portion, and/or be mounted from detachable media.At this
When computer program is executed by central processing unit (CPU), the above-mentioned function of limiting in the present processes is executed.It needs to illustrate
, computer-readable medium described herein can be computer-readable signal media or computer readable storage medium
Either the two any combination.Computer readable storage medium for example including but be not limited to electricity, magnetic, optical, electromagnetic, infrared
The system of line or semiconductor, device or device, or any above combination.Computer readable storage medium it is more specific
Example can include but is not limited to: have electrical connection, portable computer diskette, hard disk, the random visit of one or more conducting wires
Ask memory (RAM), read-only memory (ROM), erasable programmable read only memory (EPROM or flash memory), optical fiber, portable
Compact disc read-only memory (CD-ROM), light storage device, magnetic memory device or above-mentioned any appropriate combination.?
In the application, computer readable storage medium can be any tangible medium for including or store program, which can be referred to
Enable execution system, device or device use or in connection.And in this application, computer-readable signal media
It may include in a base band or as the data-signal that carrier wave a part is propagated, wherein carrying computer-readable program generation
Code.The data-signal of this propagation can take various forms, including but not limited to electromagnetic signal, optical signal or above-mentioned any
Suitable combination.Computer-readable signal media can also be any computer-readable other than computer readable storage medium
Medium, the computer-readable medium can be sent, propagated or transmitted for being used by instruction execution system, device or device
Or program in connection.The program code for including on computer-readable medium can pass with any suitable medium
It is defeated, including but not limited to: wireless, electric wire, optical cable, RF etc. or above-mentioned any appropriate combination.
The calculating of the operation for executing the application can be write with one or more programming languages or combinations thereof
Machine program code, described program design language include object oriented program language such as Java, Smalltalk, C++,
It further include conventional procedural programming language such as " C " language or similar programming language.Program code can be complete
It executes, partly executed on the user computer on the user computer entirely, being executed as an independent software package, part
Part executes on the remote computer or executes on remote computer or server-side completely on the user computer.It is relating to
And in the situation of remote computer, remote computer can include local area network (LAN) or wide area network by the network of any kind
(WAN) it is connected to subscriber computer, or, it may be connected to outer computer (such as led to using ISP
Cross internet connection).
Referring to Fig. 10, third embodiment of the invention provides a kind of artificial intelligence resolution system 1 of floor plan, including read
Module 11, identification module 12, door and window detection module 13 and parsing module 14;
Read module 11 is configured as obtaining house type sample graph;
Identification module 12 is configured as obtaining the building part in house type sample graph by shape feature, and from building part
In isolate wall feature and door and window feature;
Door and window detection module 13 is configured as based on deep learning models coupling door and window feature one door and window detection model of training,
And online difficult sample method for digging is combined to optimize door and window detection model, to detect the feature of door and the feature of window;And
Parsing module 14 is configured as that 3D is precipitated according to the feature of door, the feature of window and wall characteristic solution in house type sample graph
Floor plan.
Please refer to Figure 11, door and window detection module further comprises: extraction unit 131, pond unit 132, full connection are single
Member 133, frame return unit 134 and optimization unit 135;
Extraction unit 131 is configured as extracting the area-of-interest in sample graph;
Pond unit 132 is configured as carrying out dimension-reduction treatment to area-of-interest by pond layer;
Full connection unit 133 is configured as calculating classification of the area-of-interest in sample graph by full articulamentum;
Frame returns unit 134 and is configured as obtaining the offset predicted value of area-of-interest according to frame homing method,
It is optimized with the position to area-of-interest, and calculates first-loss function;And
Optimization unit 135 is configured as combining formerly difficult sample method for digging optimization first-loss function.
Definitions relevant content in first embodiment is equally applicable to the present embodiment.
Compared with prior art, artificial intelligence analytic method, computer-readable medium and the house type of floor plan of the invention
The artificial intelligence resolution system of figure has the advantage that
1. obtaining the building part in house type sample graph by shape feature, and wall feature and door and window feature are isolated,
And then pass through the door and window detection model of deep learning model training, and combine online difficult sample method for digging optimization door and window detection
Model parses house type sample finally by the feature combination 3D software detected to detect the feature of the feature and window gone out
Scheme corresponding 3D floor plan, to realize the two-dimentional floor plan of unmanned parsing, reduces artificial parsing cost, increase analyzing efficiency, together
When in conjunction with the first-loss function in online difficult sample method for digging optimization door and window detection model, to promote door and window detection model
Detection precision.
2. by depth-first search traversal method combined shape feature, by the size marking and text note in house type sample graph
It releases and filters out, and then obtain the building part in house type sample graph, interference when door and window is detected to reduce.
3. the artificial intelligence analytic method of floor plan is stored in computer program, in order to computer-readable medium fortune
The row computer program reduces artificial parsing cost, increases analyzing efficiency to realize the two-dimentional floor plan of unmanned parsing.
4. read module, identification module, door and window detection module and solution are arranged in the artificial intelligence resolution system of floor plan
Module is analysed, to realize the two-dimentional floor plan of unmanned parsing, artificial parsing cost is reduced, increases analyzing efficiency.
The foregoing is merely present pre-ferred embodiments, are not intended to limit the invention, it is all principle of the present invention it
Any modification made by interior, equivalent replacement and improvement etc. should all be comprising within protection scope of the present invention.
Claims (10)
1. a kind of artificial intelligence analytic method of floor plan, it is characterised in that: method includes the following steps:
Step S1: house type sample graph is obtained;
Step S2: obtaining the building part in house type sample graph by shape feature, and wall spy is isolated from building part
It seeks peace door and window feature;
Step S3: based on deep learning models coupling door and window feature one door and window detection model of training, and online difficult sample is combined
Method for digging optimizes door and window detection model, to detect the feature of door and the feature of window;And
Step S4: 3D floor plan is precipitated according to the feature of door, the feature of window and wall characteristic solution in house type sample graph.
2. the artificial intelligence analytic method of floor plan as described in claim 1, it is characterised in that: step S3 is based on deep learning
Models coupling door and window feature one door and window detection model of training further comprises following step to detect the feature of door and the feature of window
It is rapid:
Step S31: the area-of-interest in sample graph is extracted;
Step S32: dimension-reduction treatment is carried out to area-of-interest by pond layer;
Step S33: classification of the area-of-interest in sample graph is calculated by full articulamentum;
Step S34: the offset predicted value of area-of-interest is obtained, according to frame homing method with the position to area-of-interest
It optimizes, and calculates first-loss function;And
Step S35: optimize first-loss function in conjunction with online difficult sample method for digging.
3. the artificial intelligence analytic method of floor plan as claimed in claim 2, it is characterised in that: step S31 extracts sample graph
In area-of-interest, further include steps of before
Step S301: using largely includes that the floor plan of door and window selects door and window feature as training set, and using shape circle
The feature of middle door and the feature of window;
Step S302: corresponding label is made according to the feature of shape frame feature on the door and window;And
Step S303: the feature of door and the feature of window are set as area-of-interest.
4. the artificial intelligence analytic method of floor plan as described in claim 1, it is characterised in that: step S1 obtains house type sample
Figure, further includes steps of later
Step S101: the pixel of sample graph is adjusted;And
Step S102: sample graph adjusted is cut within setting specification.
5. the artificial intelligence analytic method of floor plan as described in claim 1, it is characterised in that: in above-mentioned steps S2, pass through
Depth-first search traversal method extracts the closed area surrounded in sample graph by line segment, to obtain the building part in sample graph.
6. the artificial intelligence analytic method of floor plan as claimed in claim 5, it is characterised in that: step S2 passes through depth-first
Traversal method combined shape feature obtains the building part in house type sample graph, and wall feature and door are obtained from building part
Window feature its further include steps of
Step S21: the coordinate of two-end-point in every line segment in sample graph is extracted;
Step S22: using any line segment coordinate wherein as starting point, and selecting the coordinate of adjacent segments to be attached, and is formed
Path is traversed;
Step S23: being back to starting point, forms a closed area according to the path of traversal;And
Step S24: the coordinate for choosing a point again is starting point, recycles above-mentioned steps S22- step S23, until traversal sample graph
In all line segments, and obtain the building part of sample graph.
7. the artificial intelligence analytic method of floor plan as claimed in claim 6, it is characterised in that: obtain sample in step S24
It is further included steps of after building part in figure
Step S241: the noise in sample graph is eliminated by connected domain;
Step S242: the edge line segment of building part is extracted, and combines the building part eliminated after noise, reinforces building part
The texture of edge line segment;
Step S243: the thickness of building part is restored using preset structural element;
Step S244: according to pixel grey scale, and defining straight features is segmentation threshold, identifies wall feature;And
Step S245: the open region in retrieval wall feature, using the basis detected as door and window feature.
8. a kind of computer-readable medium, it is characterised in that: it is stored with computer program in the computer-readable medium,
In, the computer program is arranged to the artificial intelligence that perform claim when operation requires floor plan described in any one of 1-7
It can analytic method.
9. a kind of artificial intelligence resolution system of floor plan, it is characterised in that: the artificial intelligence resolution system packet of the floor plan
It includes:
Read module is configured as obtaining house type sample graph;
Identification module is configured as obtaining the building part in house type sample graph by shape feature, and divides from building part
Separate out wall feature and door and window feature;
Door and window detection module is configured as based on deep learning models coupling door and window feature one door and window detection model of training, and ties
It closes online difficult sample method for digging and optimizes door and window detection model, to detect the feature of door and the feature of window;And
Parsing module is configured as that 3D house type is precipitated according to the feature of door, the feature of window and wall characteristic solution in house type sample graph
Figure.
10. the artificial intelligence resolution system of floor plan as claimed in claim 9, it is characterised in that: in door and window detection module into
One step includes:
Extraction module is configured as extracting the area-of-interest in sample graph;
Pond module is configured as carrying out dimension-reduction treatment to area-of-interest by pond layer;
Full link block is configured as calculating classification of the area-of-interest in sample graph by full articulamentum;
Frame regression block is configured as obtaining the offset predicted value of area-of-interest according to frame homing method, to sense
The position in interest region optimizes, and calculates first-loss function;And
Optimization module is configured as combining formerly difficult sample method for digging optimization first-loss function.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910611813.7A CN110348368B (en) | 2019-07-08 | 2019-07-08 | Method, computer readable medium and system for artificial intelligence analysis of house type graph |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910611813.7A CN110348368B (en) | 2019-07-08 | 2019-07-08 | Method, computer readable medium and system for artificial intelligence analysis of house type graph |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110348368A true CN110348368A (en) | 2019-10-18 |
CN110348368B CN110348368B (en) | 2021-06-01 |
Family
ID=68178428
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910611813.7A Active CN110348368B (en) | 2019-07-08 | 2019-07-08 | Method, computer readable medium and system for artificial intelligence analysis of house type graph |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110348368B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111091618A (en) * | 2019-12-17 | 2020-05-01 | 吉林吉动盘古网络科技股份有限公司 | Combined indoor layout method and system based on house type data driving |
CN111339336A (en) * | 2020-03-11 | 2020-06-26 | 贝壳技术有限公司 | Analysis method and system for house type data |
CN111985291A (en) * | 2020-02-19 | 2020-11-24 | 广东三维家信息科技有限公司 | Method and device for identifying door in house type graph and electronic equipment |
CN111985518A (en) * | 2020-02-18 | 2020-11-24 | 广东三维家信息科技有限公司 | Door and window detection method and model training method and device thereof |
CN112541214A (en) * | 2020-12-03 | 2021-03-23 | 深圳市矩阵室内装饰设计有限公司 | Method and device for processing house type design drawing and house type design drawing processing equipment |
CN112750113A (en) * | 2021-01-14 | 2021-05-04 | 深圳信息职业技术学院 | Glass bottle defect detection method and device based on deep learning and linear detection |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105279787A (en) * | 2015-04-03 | 2016-01-27 | 北京明兰网络科技有限公司 | Method for generating three-dimensional (3D) building model based on photographed house type image identification |
US20180300877A1 (en) * | 2015-12-10 | 2018-10-18 | 3M Innovative Properties Company | Method for automatic tooth type recognition from 3d scans |
CN109740506A (en) * | 2018-12-29 | 2019-05-10 | 广东三维家信息科技有限公司 | A kind of house type image-recognizing method and device |
-
2019
- 2019-07-08 CN CN201910611813.7A patent/CN110348368B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105279787A (en) * | 2015-04-03 | 2016-01-27 | 北京明兰网络科技有限公司 | Method for generating three-dimensional (3D) building model based on photographed house type image identification |
US20180300877A1 (en) * | 2015-12-10 | 2018-10-18 | 3M Innovative Properties Company | Method for automatic tooth type recognition from 3d scans |
CN109740506A (en) * | 2018-12-29 | 2019-05-10 | 广东三维家信息科技有限公司 | A kind of house type image-recognizing method and device |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111091618A (en) * | 2019-12-17 | 2020-05-01 | 吉林吉动盘古网络科技股份有限公司 | Combined indoor layout method and system based on house type data driving |
CN111985518A (en) * | 2020-02-18 | 2020-11-24 | 广东三维家信息科技有限公司 | Door and window detection method and model training method and device thereof |
CN111985291A (en) * | 2020-02-19 | 2020-11-24 | 广东三维家信息科技有限公司 | Method and device for identifying door in house type graph and electronic equipment |
CN111339336A (en) * | 2020-03-11 | 2020-06-26 | 贝壳技术有限公司 | Analysis method and system for house type data |
CN111339336B (en) * | 2020-03-11 | 2021-04-30 | 贝壳找房(北京)科技有限公司 | Analysis method and system for house type data |
CN112541214A (en) * | 2020-12-03 | 2021-03-23 | 深圳市矩阵室内装饰设计有限公司 | Method and device for processing house type design drawing and house type design drawing processing equipment |
CN112750113A (en) * | 2021-01-14 | 2021-05-04 | 深圳信息职业技术学院 | Glass bottle defect detection method and device based on deep learning and linear detection |
Also Published As
Publication number | Publication date |
---|---|
CN110348368B (en) | 2021-06-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110348368A (en) | A kind of artificial intelligence analytic method, computer-readable medium and the system of floor plan | |
CN113254648B (en) | Text emotion analysis method based on multilevel graph pooling | |
CN105550699B (en) | A kind of video identification classification method based on CNN fusion space-time remarkable information | |
CN109858406B (en) | Key frame extraction method based on joint point information | |
Martinovic et al. | Bayesian grammar learning for inverse procedural modeling | |
CN108229504B (en) | Image analysis method and device | |
CN103076892B (en) | A kind of method and apparatus of the input candidate item for providing corresponding to input character string | |
CN109816772B (en) | Processing method for rapidly generating virtual reality scene from CAD house type file | |
CN103544483B (en) | A kind of joint objective method for tracing based on local rarefaction representation and system thereof | |
CN107291684B (en) | Word segmentation method and system for language text | |
CN107424161B (en) | Coarse-to-fine indoor scene image layout estimation method | |
CN103034726B (en) | Text filtering system and method | |
CN110879881B (en) | Mouse track recognition method based on feature component hierarchy and semi-supervised random forest | |
CN104077447A (en) | Urban three-dimensional space vector modeling method based on paper plane data | |
CN110465089B (en) | Map exploration method, map exploration device, map exploration medium and electronic equipment based on image recognition | |
CN109472021A (en) | Critical sentence screening technique and device in medical literature based on deep learning | |
CN112417539A (en) | Method, device and system for designing house type based on language description | |
CN105138983A (en) | Pedestrian detection method based on weighted part model and selective search segmentation | |
CN112507861A (en) | Pedestrian detection method based on multilayer convolution feature fusion | |
CN110717154A (en) | Method and device for processing characteristics of motion trail and computer storage medium | |
CN110991278A (en) | Human body action recognition method and device in video of computer vision system | |
CN104036550A (en) | Laser radar point-cloud interpretation and reconstruction method for building elevations on basis of shape semantics | |
CN109460434A (en) | Data extract method for establishing model and device | |
TWI533145B (en) | Control method, control program and control device | |
CN114802303A (en) | Obstacle trajectory prediction method, obstacle trajectory prediction device, electronic device, and 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 |