CN108537272A - Method and apparatus for detection and analysis position in storehouse - Google Patents
Method and apparatus for detection and analysis position in storehouse Download PDFInfo
- Publication number
- CN108537272A CN108537272A CN201810306459.2A CN201810306459A CN108537272A CN 108537272 A CN108537272 A CN 108537272A CN 201810306459 A CN201810306459 A CN 201810306459A CN 108537272 A CN108537272 A CN 108537272A
- Authority
- CN
- China
- Prior art keywords
- mark
- storehouse
- warehouse
- network
- video image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
- G06Q10/087—Inventory or stock management, e.g. order filling, procurement or balancing against orders
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Physics & Mathematics (AREA)
- Tourism & Hospitality (AREA)
- Finance (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Entrepreneurship & Innovation (AREA)
- General Business, Economics & Management (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Accounting & Taxation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of method and apparatus for detection and analysis position in storehouse, including：Obtain the first video image of the predeterminable area in each warehouse at least one warehouse, wherein predeterminable area is provided at least one mark of the position in storehouse for distinguishing each warehouse；It is identified using the first video image of first network identification model pair, to obtain recognition result, wherein, recognition result includes the position in storehouse information in each warehouse, first network identification model is trained to multiple first sample images of predeterminable area using artificial intelligence approach, and the extraction feature of first network identification model includes the feature of at least one mark.
Description
Technical field
The present invention relates to machine learning field more particularly to a kind of method and apparatus for detection and analysis position in storehouse.
Background technology
With the development of modern logistics industry, storage sector has also had been extended to each community, especially a two wires city
City.Since the space size in warehouse in storage sector and singular bearing capacity are not quite similar, when being in red-letter day or order peak period,
" wharf explosion " seems to have become the inevitable problem that each warehouse encounters.Moreover, " wharf explosion " can cause a large amount of express mails to be detained
In the starting station or terminal, cause consumer that can not receive express mail on time, no matter so all to logistics company or consumer
It can cause certain loss.
The problem of for " wharf explosion ", existing technology mostly use manual method and find urgent " wharf explosion " state.
But manual method is relied on to find urgent " wharf explosion " state poor in timeliness, position in storehouse can not be monitored in real time,
It results even in processing lag, is inefficient, the measure intervened in advance and dispatched let alone is taken before also not having wharf explosion.
Invention content
The present invention found for above-mentioned dependence manual method urgent " wharf explosion " state there are the problem of, provide a kind of use
In the method and apparatus of detection and analysis position in storehouse.
In a first aspect, a kind of method for detecting position in storehouse is provided, including：Obtain each storehouse at least one warehouse
First video image of the predeterminable area in library, wherein predeterminable area is provided at least one of the position in storehouse for distinguishing each warehouse
A mark；It is identified using the first video image of first network identification model pair, to obtain recognition result, wherein identification knot
Fruit includes the position in storehouse information in each warehouse, and first network identification model is multiple the to predeterminable area using artificial intelligence approach
What one sample image was trained, the extraction feature of first network identification model includes the feature of at least one mark.
In the first mode in the cards, artificial intelligence approach includes machine learning method, and the above method further includes：
Obtain predeterminable area multiple first sample images, wherein by manually according to it is at least one mark blocked by cargo the case where be
The corresponding position in storehouse information of multiple first sample image labelings；Using machine learning method to multiple first sample figures by mark
As being trained to obtain first network identification model.
With reference to first aspect the first mode in the cards is utilizing machine in second of mode in the cards
Before device learning method is trained to obtain first network identification model to multiple first sample images by mark, by artificial
It is multiple first sample image labeling pedestrian informations the case where human interference according at least one mark, the above method also wraps
It includes：Multiple second sample images are trained using machine learning method to obtain the second Network Recognition model, wherein by artificial
It is identified whether by human interference for multiple second sample images mark is at least one；Utilize the second Network Recognition Model Identification first
At least one of video image identifies whether that by human interference, the second network model includes shot and long term memory network model；Such as
At least one mark of fruit is by human interference, it is determined that recognition result is invalid；Or, if at least one mark is not artificially done
It disturbs, it is determined that recognition result is effective.
Second of mode in the cards with reference to first aspect, it is above-mentioned also to wrap in the third mode in the cards
It includes：Identify that the preset range of at least one mark whether there is pedestrian using first network identification model；Wherein, the second net is utilized
Network identification model identifies that at least one of first video image is identified whether by human interference, including：If there is pedestrian, profit
It is identified whether by human interference at least one of second first video image of Network Recognition Model Identification.
The third mode in the cards with reference to first aspect, in the 4th kind of mode in the cards, it is above-mentioned at least
One mark the case where human interference is included artificially blocking mark, artificial scribble mark and artificial at least one torn in identifying
Kind.
Any of the above described a kind of possible realization method with reference to first aspect, in the 5th kind of mode in the cards, on
It states and multiple first sample images by mark is being trained to obtain first network identification model using machine learning method
Later, further include：Obtain the second video image of the predeterminable area in each warehouse at least one warehouse, wherein preset areas
Domain is provided at least one mark of the position in storehouse for distinguishing each warehouse；Based on the second video image, first network is identified
Model is tested, to obtain the test result of the second video image；If test result includes the position in storehouse information of mistake, for
Second video image marks correct position in storehouse information, and using the second video image after mark as sample image to first network
Identification model is trained again, to obtain updated first network identification model.
With reference to first aspect, second of possible realization of the first of first aspect mode in the cards, first aspect
Mode, the third mode in the cards, the 4th kind of mode in the cards of first aspect of first aspect, at the 6th kind
In mode in the cards, after video image is identified using Network Recognition model, further include：By recognition result reality
When be uploaded to dispatch server, so as to position in storehouse information carry out position in storehouse analysis.
With reference to first aspect, second of possible realization of the first of first aspect mode in the cards, first aspect
Mode, the third mode in the cards, the 4th kind of mode in the cards of first aspect of first aspect, at the 7th kind
It is at least one to be identified as number, Quick Response Code, letter, color or user-defined identification in mode in the cards.
With reference to first aspect, second of possible realization of the first of first aspect mode in the cards, first aspect
Mode, the third mode in the cards, the 4th kind of mode in the cards of first aspect of first aspect, at the 8th kind
In mode in the cards, at least one mark is distributed on the wall of first area or on holder.
Second aspect provides a kind of method of analysis position in storehouse, including：Recognition result is received from server, wherein identification
The result is that the video image of the predeterminable area in each warehouse at least one warehouse is identified using Network Recognition model
It obtains, recognition result includes the position in storehouse information in each warehouse, and Network Recognition model is using artificial intelligence approach to preset areas
What multiple sample images in domain were trained, the extraction feature of Network Recognition model includes the feature of at least one mark；
Early warning is carried out based on recognition result.
In the first mode in the cards, the above method further includes：To the position in storehouse information in each warehouse of storage
Historical data is analyzed, and position in storehouse analysis result is obtained；The logistics in each warehouse is scheduled according to position in storehouse analysis result.
It is above-mentioned to include based on recognition result progress early warning in second of mode in the cards：To each storehouse of storage
The historical data of the position in storehouse information in library is analyzed, and position in storehouse analysis result is obtained；Early warning is carried out according to position in storehouse analysis result.
It is above-mentioned to include based on recognition result progress early warning in the third mode in the cards：Reach in position in storehouse information
Early warning is carried out when predetermined threshold value.The third aspect provides a kind of device for detecting position in storehouse, which is characterized in that including：It obtains
Module, the first video image of the predeterminable area for obtaining each warehouse at least one warehouse, wherein predeterminable area is set
It is equipped at least one mark of the position in storehouse for distinguishing each warehouse；Identification module, for utilizing first network identification model pair
First video image is identified, to obtain recognition result, wherein and recognition result includes the position in storehouse information in each warehouse, and first
Network Recognition model is trained to multiple first sample images of predeterminable area using artificial intelligence approach, first
The extraction feature of Network Recognition model includes the feature of at least one mark.
In the first mode in the cards, artificial intelligence approach includes machine learning method, and acquisition module is additionally operable to
Obtain predeterminable area multiple first sample images, wherein by manually according to it is at least one mark blocked by cargo the case where be
The corresponding position in storehouse information of multiple first sample image labelings；Training module, for utilizing machine learning method to by mark
Multiple first sample images are trained to obtain first network identification model.
In conjunction with the first mode in the cards of the third aspect machine is being utilized in second of mode in the cards
Before device learning method is trained to obtain first network identification model to multiple first sample images by mark, by artificial
It is multiple first sample image labeling pedestrian informations the case where human interference according at least one mark, training module is additionally operable to
Multiple second sample images are trained using machine learning method to obtain the second Network Recognition model, wherein by being manually more
A second sample image mark is at least one to be identified whether by human interference；Above-mentioned apparatus further includes determination module, at least one
When a mark is by human interference, for determining that recognition result is invalid；Or, when at least one mark is not by human interference, use
In determining that recognition result is effective；Identification module is additionally operable to utilize in second the first video image of Network Recognition Model Identification at least
One identifies whether by human interference, and the second network model includes shot and long term memory network model.
In conjunction with second of mode in the cards of the third aspect, in the third mode in the cards, identification module
It is additionally operable to identify that the preset range of at least one mark whether there is pedestrian using first network identification model；There are pedestrians
When, identification module be specifically used for using at least one of second first video image of Network Recognition Model Identification identify whether by
Human interference.
It is at least one in the 4th kind of mode in the cards in conjunction with the third mode in the cards of the third aspect
It identifies and is blocked mark including artificial the case where human interference, artificial scribble identifies and artificially tear at least one of mark.
It is obtained in the 5th kind of mode in the cards in conjunction with a kind of any of the above described possible realization method of the third aspect
Modulus block is additionally operable to obtain the second video image of the predeterminable area in each warehouse at least one warehouse, wherein preset areas
Domain is provided at least one mark of the position in storehouse for distinguishing each warehouse；Identification module, be based on the second video image, for pair
First network identification model is tested, to obtain the test result of the second video image；Training module, if test result packet
The position in storehouse information of mistake is included, then marks correct position in storehouse information for the second video image, and for the second video after marking
Image is trained first network identification model as sample image again, and mould is identified to obtain updated first network
Type.
In conjunction with second of possible realization of the first mode, the third aspect in the cards of the third aspect, the third aspect
Mode, the third mode in the cards, the 4th kind of mode in the cards of the third aspect of the third aspect, at the 6th kind
In mode in the cards, uploading module, for recognition result to be uploaded to dispatch server in real time, so as to position in storehouse information into
It analyzes firm position position.
In conjunction with second of possible realization of the first mode, the third aspect in the cards of the third aspect, the third aspect
Mode, the third mode in the cards, the 4th kind of mode in the cards of the third aspect of the third aspect, at the 7th kind
It is at least one to be identified as number, Quick Response Code, letter, color or user-defined identification in mode in the cards.
In conjunction with second of possible realization of the first mode, the third aspect in the cards of the third aspect, the third aspect
Mode, the third mode in the cards, the 4th kind of mode in the cards of the third aspect of the third aspect, at the 8th kind
In mode in the cards, at least one mark is distributed on the wall of first area or on holder.
Fourth aspect provides a kind of device of analysis position in storehouse, which is characterized in that including：Receiving module is used for from service
Device receives recognition result, wherein recognition result is using Network Recognition model to the pre- of each warehouse at least one warehouse
If what the video image in region was identified, recognition result includes the position in storehouse information in each warehouse, and Network Recognition model is
Multiple sample images of predeterminable area are trained using artificial intelligence approach, the extraction feature of Network Recognition model
Include the feature of at least one mark；Alarm module, for carrying out early warning when position in storehouse information reaches predetermined threshold value.
In the first mode in the cards, artificial intelligence approach includes machine learning method, and above-mentioned apparatus further includes：
The historical data of analysis module, the position in storehouse information for each warehouse to storage is analyzed, and position in storehouse analysis result is obtained；It adjusts
Module is spent, for being scheduled to the logistics in each warehouse according to position in storehouse analysis result.
In second of mode in the cards, above-mentioned apparatus further includes analysis module, for each warehouse to storage
The historical data of position in storehouse information analyzed, obtain position in storehouse analysis result；Alarm module, which is specifically used for being analyzed according to position in storehouse, to be tied
Fruit carries out early warning.
In the third mode in the cards, alarm module is specifically used for carrying out when position in storehouse information reaches predetermined threshold value
Early warning.
The technical solution provided according to an embodiment of the invention, by using the trained first network of artificial intelligence approach
The first video image of predeterminable area is identified in identification model, the position in storehouse information of Current warehouse is obtained, so as to basis
Whether the position in storehouse information obtains Current warehouse " wharf explosion "." wharf explosion " state for relying on manual report warehouse can be thus broken away from,
And the specific location and location status in warehouse can be known in time, it might even be possible to the position of accurate precognition " wharf explosion ".
Description of the drawings
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for
For those of ordinary skill in the art, without creative efforts, it can also be obtained according to these attached drawings other
Attached drawing, wherein：
Fig. 1 is the schematic diagram of the system according to an embodiment of the invention for position in storehouse detection and analysis.
Fig. 2 is the schematic flow chart according to an embodiment of the invention for detecting the method for position in storehouse.
Fig. 3 is the Digital ID pasted in warehouse according to an embodiment of the invention.
Fig. 4 is the schematic flow chart according to an embodiment of the invention for analyzing the method for position in storehouse.
Fig. 5 is the flow diagram according to an embodiment of the invention for detecting the method for position in storehouse.
Fig. 6 is trained flow diagram according to an embodiment of the invention.
Fig. 7 is the schematic diagram according to an embodiment of the invention for detecting the device of position in storehouse.
Fig. 8 is the schematic diagram according to an embodiment of the invention for analyzing the device of position in storehouse.
Fig. 9 is the block diagram according to an embodiment of the invention for detecting the computer equipment of position in storehouse.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation describes, it is clear that the described embodiments are merely a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art obtained without making creative work it is all its
Its embodiment, shall fall within the protection scope of the present invention.
Currently, artificial intelligence has been widely used in multiple fields, but the application in storage sector is also rarely found.
The technical solution of the embodiment of the present invention is to be based on machine Learning Theory, to realize the detection and analysis to position in storehouse.Below with reference to
Specific embodiment is described.
Fig. 1 is the schematic diagram of the system 100 according to an embodiment of the invention for position in storehouse detection and analysis.
System 100 may include being located at least one of warehouse 110 camera, back-end server 120 and dispatch service
Device 130.
Specifically, warehouse 110 can be positioned at the independent warehouse of one or more of different places.It can be in each independence
At least one camera is installed, to shoot at least one mark institute being provided in this warehouse for distinguishing position in storehouse in warehouse
Video image in region, wherein whether the number of camera can be obtained completely according to the video image of above-mentioned zone in warehouse
Determination is fetched, can be 1, can also be multiple.
Back-end server 120 obtains the video image that camera takes by way of wired or wireless network, so
The position in storehouse in warehouse is digitized, and digitized position in storehouse information is passed through by the video image inputted afterwards by artificial intelligence analysis
Wired or wireless network sends dispatch server 130 to, or can also digitized position in storehouse information be uploaded to high in the clouds.Its
In, the mode of neural network (Neural Networks, NNs), certainly, this hair may be used in above-mentioned artificial intelligence program's analysis
The mode of bright analysis image is not limited only to this, can also use other modes, such as other image recognition modes.
Dispatch server 130 can according to digitized position in storehouse information to will " wharf explosion " warehouse carry out early warning, even
Storage analysis can be carried out to historical data, sum up the geographical location with wharf explosion risk and corresponding time, both subsidiary streams are public
Department arranges scheduling.Wherein, dispatch server 130 may include the ends PC, mobile terminal.
Fig. 2 is the schematic flow chart according to an embodiment of the invention for detecting the method for position in storehouse.This method can be with
It is executed by the back-end server 120 of Fig. 1.
210, obtain the first video image of the predeterminable area in each warehouse at least one warehouse, wherein preset areas
Domain is provided at least one mark of the position in storehouse for distinguishing each warehouse.
220, it is identified using the first video image of first network identification model pair, to obtain recognition result, wherein know
Other result includes the position in storehouse information in each warehouse, and first network identification model is using artificial intelligence approach to the more of predeterminable area
What a first sample image was trained, the extraction feature of Network Recognition model includes at least one mark.
Wherein, artificial intelligence approach may include the methods of machine learning and image recognition.
In an embodiment of the present invention, by using the trained first network identification model of artificial intelligence approach to default
First video image in region is identified, and obtains the position in storehouse information of Current warehouse, to obtain current storehouse according to position in storehouse information
Library whether " wharf explosion "." wharf explosion " state for relying on manual report warehouse can be thus broken away from, and can know warehouse in time
Specific location and position in storehouse situation, it might even be possible to the position of accurate precognition " wharf explosion ".
Specifically, including that can identify the position in storehouse in warehouse 110 at least in step 210, in the first video image
One mark, the first video image can in real time be acquired by the camera in warehouse 110.Wherein, at least one mark
Can be the good number of pre-production, Quick Response Code, letter, color or the marks such as self-defined, its position and number can bases
The size in warehouse and the display case of cargo determine.By taking Digital ID as an example, Fig. 3 is warehouse according to an embodiment of the invention
The Digital ID of middle stickup.Different numbers represents different position in storehouse situations in Fig. 3, digital " 8 " represent cargo have reached it is " quick-fried
The state in storehouse ".The good Digital ID as shown in Figure 3 of pre-production can be attached to the predeterminable area in the warehouse for needing to monitor
On wall or on holder, what back-end server 120 can obtain the predeterminable area of current monitor in real time by camera 110 includes
The video image of numbers above mark.
In a step 220, first network identification model can identify that at least one of first video image identifies, and
Export its corresponding recognition result.Wherein, recognition result may include the first video image and each warehouse 110 for outlining mark
Position in storehouse information, and position in storehouse information can be the information such as position and the position in storehouse situation of above at least one mark, at least one mark
The classification of knowledge is also that can represent warehouse position in storehouse situation.Different marks for indicating different positions in storehouse, for example, in warehouse
Predeterminable area is provided with 8 marks such as Fig. 3.If the cargo stacked in warehouse has sheltered from digital " 1 ", and digital " 2 " do not have
Have and blocked by cargo, then show warehouse occupied 20%, at this moment, position in storehouse information indicates (or the remaining storehouse of warehouse occupied 20%
Library 80%).
It is understood that if the cargo stored in warehouse shelters from some mark, the first identification network mould
Type can only just recognize the mark not being blocked, and the mark being exposed at this time has also just reacted warehouse residue position in storehouse
Situation or the position in storehouse situation occupied, to monitor the current position in storehouse in each warehouse in real time.Specifically, blocking can be divided into completely
It blocks and not exclusively blocks.When some mark is blocked completely, this mark of the basic None- identified of first network identification model；Certain
A mark is when not exclusively being blocked, the case where according to blocking, what some identified categories still can be identified, this when
Still can be regarded as this mark not to be blocked.Certainly, the order of accuarcy of recognition result can be according in above-mentioned at least one mark
The distance between each mark determines.
For example, the different numbers in Fig. 3 can represent different positions in storehouse.If the cargo stored in warehouse shelters from
Digital " 1 ", then remaining number is all exposed, the first identification network model can identify these digital positions and its
The position in storehouse of representative.And can export at this time outline it is above be exposed number the first video image, these number positions with
And at this time and minimum in corresponding position in storehouse situation, or these numbers that only selection output is exposed number " 2 " it is right
The position in storehouse situation answered.
According to an embodiment of the invention, artificial intelligence approach includes machine learning method, and the above method further includes：It obtains pre-
If multiple first sample images in region, wherein by manually being blocked by cargo according at least one mark the case where is multiple the
The corresponding position in storehouse information of one sample image labeling；Using machine learning method to being carried out by multiple first sample images of mark
Training obtains first network identification model.
Specifically, first network identification model can be trained using any one of artificial intelligence approach method
It arrives, it can be trained model, can also select to acquire multiple first sample images and be trained to obtain.
It needs to acquire multiple first sample images first during training first network identification model, and to multiple the
At least one of one sample image mark is labeled, and the content of mark includes position and the position in storehouse feelings of at least one mark
Condition, wherein multiple first sample images can acquire in advance, for example, can be taken the photograph by one or more of warehouse 110
As head acquires the multiframe picture in its entirety for the predeterminable area for containing at least one mark as multiple first sample images in advance.
It in one embodiment, can be by multiple first sample figures before being trained multiple first sample images
A large amount of similar training datas are generated as being imitated by image algorithm, so that the first network identification model trained
Recognition effect is more accurate.For example, it is uniformly zoomed in and out, and expand 1 times by flip horizontal.Certainly, scaling
It can be adjusted according to actual demand.Multiple first sample images can also be subjected to the rotation of different angle to expand trained number
According to.Wherein, the position of at least one mark for the mark for including for multiple first sample images can also zoom in and out, turn over
Turn and rotation is expanded.
Depth is utilized for example, the training data after being expanded can be added in the neural network of responsible target detection
The method of study realizes the learning process of end-to-end (end-to-end), to obtain first network identification model.Neural network
Convolutional neural networks (Convolutional Neural Network, CNN) can be selected, for example, can be first with convolution
Neural network CNN automatically extracts the feature of at least one mark, then, passes through gradient descent method combination loss function (Loss
Function mode) trains CNN to obtain first network identification model, wherein the feature packet of at least one mark of CNN extractions
Include color, profile, the texture etc. of mark.The above entire training process can be full automatic, not need manual intervention.
Based on the embodiment of the present invention, the first network identification model obtained using machine learning method can be used for fine
It identifies the position in storehouse information in warehouse 110, and is entire roboticized work, can be broken away from storage field rely on manual report in this way
The state of warehouse position in storehouse.
According to an embodiment of the invention, multiple first sample images by marking are being carried out using machine learning method
It, can also be above-mentioned by being manually multiple first sample image labeling pedestrian informations before training obtains first network identification model
Method further includes：Multiple second sample images are trained using machine learning method to obtain the second Network Recognition model,
In by being manually that multiple second sample images mark at least one is identified whether by human interference；Utilize the second Network Recognition model
Identify that at least one of first video image is identified whether by human interference, the second network model includes shot and long term memory network
Model；If at least one mark is by human interference, it is determined that recognition result is invalid；Or, if it is at least one mark not by
Human interference, it is determined that recognition result is effective.
Specifically, inevitably having the presence of pedestrian in the warehouse of stack goods, the various actions of pedestrian also can be to extremely
A few mark interferes.For example, the case where human interference may include artificially blocking mark, artificial scribble mark and artificial
Tear at least one of mark.
So the can be trained to obtain to multiple first sample images by mark using machine learning method
It is multiple first sample image labeling pedestrians the case where human interference according at least one mark before one Network Recognition model
Information, wherein pedestrian information for example may include position and the height of pedestrian.In this way, first network identification model utilizes machine
What the method for study learnt is not only just the position in storehouse information of at least one mark, further includes pedestrian information.
First video image is input in above-mentioned first network identification model, obtained recognition result includes not only just extremely
The recognition result of a few mark, further includes the recognition result of pedestrian, for example, the recognition result of pedestrian include pedestrian position and
Highly.The process of pedestrian information training is similar at least one process of mark training, and details are not described herein.
In another embodiment, it can utilize machine learning method re -training one that can identify pedestrian in the above warehouse
The Network Recognition model of information will outline mark after being identified using the first video image of first network identification model pair
The first video image known inputs the recognition result that this model obtains including position in storehouse information and pedestrian information.Wherein, this mould
The training process of type is similar to the training process of first network identification model, and only marked content includes the above pedestrian information, is carried
The feature for being characterized in pedestrian is taken, details are not described herein for detailed content.
Due to there is the interference of pedestrian, it will can simultaneously include the first video figure of at least one mark and pedestrian's recognition result
It is again identified that as being transmitted to the second Network Recognition model, judges that at least one of first video image is identified whether by people
For interference.
Specifically, the second Network Recognition model can utilize trained model, can also select to acquire multiple the
Two sample images are trained to obtain.
It needs to acquire multiple second sample images first during the second Network Recognition model of training, and to multiple the
At least one of two sample images identify whether to be labeled by human interference, wherein multiple second sample images can be
It acquires in advance, for example, can in advance be acquired containing at least one mark by one or more of warehouse 110 camera
The multiframe picture in its entirety of predeterminable area is as multiple second sample images.Second network model can be shot and long term memory network
(Long Short Term Memory, LSTM) model.
Based on the embodiment of the present invention, the presence based on pedestrian, training to have obtained using machine learning method can identify
The first network identification model of position in storehouse information and pedestrian information, meanwhile, the addition of the second Network Recognition model eliminates artificial do
The case where disturbing so that the result of position in storehouse information is more accurate.
According to an embodiment of the invention, the above method further includes：At least one mark is identified using first network identification model
The preset range of knowledge whether there is pedestrian；Wherein, at least one in second the first video image of Network Recognition Model Identification is utilized
It is a to identify whether by human interference, including：If there is pedestrian, using in second the first video image of Network Recognition Model Identification
At least one identify whether by human interference.
Specifically, the recognition result being identified using the first video image of first network identification model pair is not only
It may include the position in storehouse information in each warehouse, the pedestrian information in each warehouse can also be included.If the pedestrian in recognition result
In the preset range of at least one mark, then it may determine where there is pedestrians.It specifically, can be at least one mark
Centered on any one mark in knowledge, deviates and belong to preset range within the pre-determined distance of center.If within a preset range
There are pedestrians, so that it may be seen with the presence or absence of artificial dry so that the first transmission of video images to the second Network Recognition model to be identified
It disturbs.
It is of course also possible to judge whether the pedestrian in recognition result has stopped admittedly in the preset range of at least one mark
The fixed time if having stopped regular time, then the first transmission of video images to the second Network Recognition model is identified.
Specifically, regular time can be 2 seconds, 3 seconds or other preset fixed time periods.
In another embodiment, if first network identification model identifies that the preset range of at least one mark has row
People, then being compared it may determine that extremely with the position of at least one mark and position in storehouse situation according to the height of pedestrian and position
Few one identifies whether to be blocked.If at least one mark is blocked by pedestrian, it is determined that recognition result is invalid；Or, if extremely
A few mark is not blocked by pedestrian, it is determined that recognition result is effective.
Based on the embodiment of the present invention, the judgement with the presence or absence of pedestrian has been done first, and only this condition meets ability profit
It is identified whether by human interference at least one of second first video image of Network Recognition Model Identification.Reduce mould in this way
Type handles the quantity of picture, improves processing speed.
According to an embodiment of the invention, multiple first sample images by marking are being carried out using machine learning method
After training obtains first network identification model, further include：Obtain the predeterminable area in each warehouse at least one warehouse
Second video image, wherein predeterminable area is provided at least one mark of the position in storehouse for distinguishing each warehouse；Based on second
Video image tests first network identification model, to obtain the test result of the second video image；If test result
Include the position in storehouse information of mistake, be then that the second video image marks correct position in storehouse information, and by the second video figure after mark
As being trained again to first network identification model as sample image, to obtain updated first network identification model.
Specifically, the second video image is entirely used for test first network identification model, it can be acquired in advance,
Can also be to obtain in real time.
The content of test result can be consistent with the content of above-mentioned recognition result, and details are not described herein.The standard of test result
Exactness can be gone to execute by special supervision mechanism, can also be by manually being checked.If test result includes correct storehouse
Position information illustrates that the first network identification model precision trained is high；If test result includes the position in storehouse information of mistake,
Can be that the second video image marks correct position in storehouse information, content may include at least one mark in the second video image
Then position and position in storehouse situation are trained first network identification model as sample image, again to be updated
First network identification model afterwards.
For example, there may come a time when that light is bad in warehouse, this this may result in phenomena such as the second video image obscures.Such as
Fruit carries out test to first network identification model using the second fuzzy video and is possible to obtain the position in storehouse information of mistake.Such as figure
Shown in 3, if the position in storehouse of cargo has had reached at digital " 6 ", mark " 7 " and " 8 " is exposed, then position in storehouse at this time is answered
This is " 7 ".But if the second video image is fuzzy, first network identification model None- identified " 7 " and " 8 ", then exporting
Position in storehouse information just will include " wharf explosion ".At this point, have special supervision mechanism remind this test result be it is wrong, then
Can be that the second video image marks correct position in storehouse information, content may include position and position in storehouse situation, then as
Sample image is trained first network identification model again, to obtain updated first network identification model.
For example, there are relative positions at least one mark, if the test that first network identification model is tested
As a result relative position is wrong, then can also be trained again to first network identification model with the aforedescribed process.Wherein,
The wrong mark that should occur in position relatively above in the second video image that can refer to of relative position appears in opposite lower section
Position or the mark that should occur in opposite lower position appear in position relatively above.
Based on the embodiment of the present invention, by increasing test phase to first network identification model, and to include mistake
Second video image of position in storehouse information marks correct position in storehouse information, is trained again, obtains updated first network and knows
Other model.So that updated first network identification model rate of false alarm reduces, precision higher.
According to an embodiment of the invention, after being identified using the first video image of first network identification model pair,
Further include：Recognition result is uploaded to dispatch server in real time, to carry out position in storehouse analysis to position in storehouse information.
Specifically, warehouse can be obtained after being identified using the first video image of first network identification model pair
Then position in storehouse information is uploaded to dispatch server 130 by position in storehouse information in real time, can ensure that dispatch server 130 is real-time in this way
Know the position in storehouse in each warehouse 110, and auxiliary dispatching.
Fig. 4 is the schematic flow chart according to an embodiment of the invention for analyzing the method for position in storehouse.
This method is executed in system 100 in Fig. 1 by dispatch server 130.
410, receive recognition result from server.
Recognition result is that back-end server utilizes Network Recognition model to the default of each warehouse at least one warehouse
What the video image in region was identified, recognition result includes the position in storehouse information in each warehouse, and Network Recognition model is profit
Manually intelligent method is trained multiple sample images of predeterminable area, the extraction feature packet of Network Recognition model
Include at least one mark.
420, early warning is carried out based on recognition result.
Specifically, may include based on recognition result progress early warning：It is carried out when position in storehouse information reaches predetermined threshold value pre-
It is alert；Can also include：The historical data of the position in storehouse information in each warehouse of storage is analyzed, position in storehouse analysis result is obtained；
Early warning is carried out according to position in storehouse analysis result.
Dispatch server 130 can receive the position in storehouse information that the above back-end server 120 transmits, wherein position in storehouse is believed
Breath can be identified to obtain using Network Recognition model, and Network Recognition model can be CNN models.When position in storehouse information reaches
When to predetermined threshold value, dispatch server 130 can carry out real-time early warning, such as can alarm in advance, can especially not have also
Have and people is reminded to take intervention in advance and dispatch before " wharf explosion ", improves the efficiency of scheduling.As shown in figure 3, predetermined threshold value can
Can also be artificially to set only the case where remaining " 8 " to be that the cargo in warehouse all shelters from digital " 7 " number below
Other fixed situations, the present invention are not limited herein.Certainly, dispatch server 130 can also be periodically to the history of position in storehouse information
Data carry out storage analysis and summary and provide the geographical location of wharf explosion risk and corresponding time, realize early warning.For example, dispatch service
Device 130 can carry out storage analysis every 2h to the position in storehouse information received, and generate history report, may include in report
The average value of the position of at least one mark and position in storehouse information sometime.Can thus sum up which area which
What time period a warehouse can easy to produce the risk of wharf explosion, and then remind logistics company in advance etc..
According to an embodiment of the invention, the above-mentioned method for analyzing position in storehouse further includes：To the storehouse in each warehouse of storage
The historical data of position information is analyzed, and position in storehouse analysis result is obtained；According to position in storehouse analysis result to the logistics in each warehouse into
Row scheduling.
Specifically, historical data can be before specific location warehouse 110 in accurate position in storehouse letter of determining time
Breath.Position in storehouse analysis result can be that some regional or some position warehouse is easy in sometime point or certain time period
" wharf explosion ".Dispatch server 130 is analyzed and can be obtained by the historical data of the position in storehouse information in each warehouse to storage
To the specific when and where of " wharf explosion ", so that the analysis and judgement of dispatch server 130 are finer, so that adjusting
Efficiency is spent to improve.
Fig. 5 is the flow diagram according to an embodiment of the invention for detecting the method for position in storehouse.
This method is executed in system 100 in Fig. 1 by back-end server 120.
505, multiple first sample images by mark are trained using machine learning method, obtain first network
Identification model.
For example, the content of multiple first sample image labelings includes position in storehouse information and pedestrian information, first network identifies mould
The extraction feature of type includes the feature of at least one mark and the feature of pedestrian.
510, using machine learning method to being labeled at least one identify whether by multiple second samples of human interference
Image is trained, and obtains the second Network Recognition model.
515, obtain the first video image.
For example, the first video image can scale according to a certain percentage or flip horizontal processing, to expand identification number
According to raising accuracy of identification.
520, it is identified using the first video image of first network identification model pair, to obtain recognition result.
For example, recognition result may include position in storehouse information and pedestrian information.
525, using the second Network Recognition model to include the first video image of position in storehouse information and pedestrian information again
It is identified, to determine at least one in the first video image identify whether by human interference.
530, if at least one mark is by human interference in the first video image, then it is determined that recognition result is invalid.
For example, at least one mark is included artificial blocking mark, artificial scribble mark and artificially the case where human interference
Tear at least one of mark.
535, if at least one mark not by human interference, recognition result is effective, defeated in the first video image
Go out position in storehouse information.
540, position in storehouse information is uploaded to dispatch server 130 in real time, to carry out position in storehouse analysis to position in storehouse information.
In order to make it easy to understand, introducing the detailed process of the step 505 in Fig. 5 below in conjunction with Fig. 6.
Fig. 6 is trained flow diagram according to an embodiment of the invention.
602, obtain multiple first sample images.
For example, multiple first sample images can in advance be acquired using camera.
605, multiple first sample images by mark are trained using machine learning method, to obtain the first net
Network identification model.
For example, the content of multiple first sample image labelings includes position in storehouse information and pedestrian information, first network identifies mould
The extraction feature of type includes the feature of at least one mark and the feature of pedestrian.
610, obtain the second video image.
612, first network identification model is tested, to obtain the test result of the second video image.
For example, test result may include position in storehouse information.
615, judge whether test result includes the position in storehouse information of mistake.
620, if test result does not include the position in storehouse information of mistake, first network identification model has passed through test
Stage obtains the higher first network identification model for being directly used in identification of precision.
625, if test result includes the position in storehouse information of mistake, using the second video image after mark as sample image
First network identification model is trained again, to obtain updated first network identification model.
For example, the second video image after mark includes the correct position in storehouse information for being the second video image mark again.
Described above is the methods according to the ... of the embodiment of the present invention for detection and analysis position in storehouse, with reference to Fig. 7 to Fig. 9
Device according to the ... of the embodiment of the present invention for detection and analysis position in storehouse is described.
Fig. 7 is the schematic diagram according to an embodiment of the invention for detecting the device 700 of position in storehouse.
Device 700 includes：Acquisition module 710 and identification module 720.
Acquisition module 710, the first video image of the predeterminable area for obtaining each warehouse at least one warehouse,
Wherein, predeterminable area is provided at least one mark of the position in storehouse for distinguishing each warehouse；Identification module 720, for utilizing
The first video image of first network identification model pair is identified, to obtain recognition result, wherein recognition result includes each storehouse
The position in storehouse information in library, first network identification model be using artificial intelligence approach to multiple first sample images of predeterminable area into
Row training obtains, and the extraction feature of first network identification model includes the feature of at least one mark.
Based on the embodiment of the present invention, this is used to detect the device of position in storehouse by using artificial intelligence approach trained
The first video image of predeterminable area is identified in one Network Recognition model, can obtain the position in storehouse information of Current warehouse, together
When can also obtain the state of " wharf explosion " urgent in Current warehouse.Dependence can be broken away from by being used to detect the device of position in storehouse using this
" wharf explosion " state in manual report warehouse, it might even be possible to the position of accurate precognition " wharf explosion ".
Optionally, as another embodiment, device 700 further includes training module 730.
Acquisition module 710 is additionally operable to obtain multiple first sample images of predeterminable area, wherein by artificial according at least
The case where one mark is blocked by cargo is the corresponding position in storehouse information of multiple first sample image labelings；Training module 730, is used for
Multiple first sample images by mark are trained to obtain first network identification model using machine learning method.
Optionally, as another embodiment, device 700 further includes determination module 740.
Training module 730 is additionally operable to be trained to obtain the second net to multiple second sample images using machine learning method
Network identification model, wherein being identified whether by human interference by being manually that multiple second sample images mark is at least one.Judge mould
Block 740, when at least one mark is by human interference, for determining that recognition result is invalid；Or, it is at least one mark not by
When human interference, for determining that recognition result is effective.；Identification module 720 is additionally operable to utilize the second Network Recognition Model Identification the
At least one of one video image identifies whether that by human interference, the second network model includes shot and long term memory network model.
Optionally, as another embodiment, identification module 720 is additionally operable to identify at least one using first network identification model
The preset range of a mark whether there is pedestrian；When there are pedestrian, identification module 720 is specifically used for utilizing the second Network Recognition
At least one of first video image of Model Identification is identified whether by human interference.
According to an embodiment of the invention, at least one mark is included the artificial mark, artificially of blocking the case where human interference
Scribble mark tears at least one of mark with artificial.
Optionally, as another embodiment, acquisition module 710 is additionally operable to obtain each warehouse at least one warehouse
Second video image of predeterminable area, wherein predeterminable area is provided at least one mark of the position in storehouse for distinguishing each warehouse
Know；Identification module 720 is based on the second video image, for testing first network identification model, to obtain the second video
The test result of image；Training module 730, if test result includes the position in storehouse information of mistake, for the second video image mark
Note correct position in storehouse information, and for the second video image after marking as sample image to first network identification model again
It is secondary to be trained, to obtain updated first network identification model.
Optionally, as another embodiment, device 700 further includes uploading module 750.
Uploading module 750, for recognition result to be uploaded to dispatch server in real time, to carry out position in storehouse to position in storehouse information
Analysis.
According to an embodiment of the invention, at least one to be identified as number, Quick Response Code, letter, color or user-defined identification.
According to an embodiment of the invention, at least one mark is distributed on the wall of first area or on holder.
Be used to detect above the above-mentioned modules in the device of position in storehouse operation and function can referring to figs. 1 to 6 side
The specific descriptions of method embodiment part, in order to avoid repeating, this will not be repeated here.
Fig. 8 is the schematic diagram according to an embodiment of the invention for analyzing the device 800 of position in storehouse.
Device 800 includes：Receiving module 810 and alarm module 820.
Receiving module 810, for receiving recognition result from server, wherein recognition result is to utilize Network Recognition model
The video image of the predeterminable area in each warehouse at least one warehouse is identified, recognition result includes each
The position in storehouse information in warehouse, Network Recognition model are trained to multiple sample images of predeterminable area using artificial intelligence approach
It obtains, the extraction feature of Network Recognition model includes the feature of at least one mark；Alarm module 820, for based on identification
As a result early warning is carried out.
Optionally, as another embodiment, device 800 further includes analysis module 830 and scheduler module 840.
The historical data of analysis module 830, the position in storehouse information for each warehouse to storage is analyzed, and position in storehouse is obtained
Analysis result；Scheduler module 840, for being scheduled to the logistics in each warehouse according to position in storehouse analysis result.
Optionally, as another embodiment, device 800 further includes analysis module 830.
The historical data of analysis module 830, the position in storehouse information for each warehouse to storage is analyzed, and position in storehouse is obtained
Analysis result；Alarm module 820 is specifically used for carrying out early warning according to position in storehouse analysis result.
According to an embodiment of the invention, alarm module 820 is specifically used for carrying out when position in storehouse information reaches predetermined threshold value pre-
It is alert.
The operation and function for being used to detect the above-mentioned modules in the device of position in storehouse above can be with reference chart 1 and Fig. 4
The specific descriptions of embodiment of the method part, in order to avoid repeating, this will not be repeated here.
Fig. 9 is the block diagram according to an embodiment of the invention for detecting the computer equipment 900 of position in storehouse.
With reference to Fig. 9, device 900 is including processing component 910 and by the memory resource representated by memory 920.Processing group
Part 910 further comprises one or more processors, memory 920 for store can by the instruction of the execution of processing component 910,
Such as application program.The application program stored in memory 920 may include one or more modules, each of which corresponds to one
Group instruction.In addition, processing component 910 is configured as executing instruction, to execute the above-mentioned method for detecting position in storehouse.
Device 900 can also include power supply module, be configured as the power management of executive device 900.Device 900 may be used also
To include wired or radio network interface, it is configured as device 900 being connected to network.Device 900 can also include input
Export (I/O) interface.Device 900 can be operated based on the operating system for being stored in memory 920, such as Windows
ServerTM、Mac OS XTM、UnixTM、LinuxTM、FreeBSDTMDeng.
The embodiment of the present invention additionally provides a kind of non-transitorycomputer readable storage medium, when the instruction in storage medium
When being executed by the processor of above-mentioned apparatus 900 so that above-mentioned apparatus 900 is able to carry out a kind of method for detecting position in storehouse, packet
It includes：Obtain the first video image of the predeterminable area in each warehouse at least one warehouse, wherein predeterminable area setting is useful
In at least one mark for the position in storehouse for distinguishing each warehouse；Known using the first video image of first network identification model pair
Not, to obtain recognition result, wherein recognition result includes the position in storehouse information in each warehouse, and first network identification model is to utilize
Artificial intelligence approach is trained multiple first sample images of predeterminable area, the extraction of first network identification model
Feature includes the feature of at least one mark.
Those of ordinary skill in the art may realize that use described in conjunction with the examples disclosed in the embodiments of the present disclosure
In the step of detecting position in storehouse, can be realized with the combination of electronic hardware or computer software and electronic hardware.These functions
It is implemented in hardware or software actually, depends on the specific application and design constraint of technical solution.Professional technique
Personnel specifically can realize described function to each using distinct methods, but this realization is it is not considered that super
Go out the scope of the present invention.
It is apparent to those skilled in the art that for convenience and simplicity of description, the method for foregoing description
It with the specific work process of device, can refer to corresponding processes in the foregoing method embodiment, details are not described herein.
In several embodiments provided herein, it should be understood that disclosed device and method can pass through it
Its mode is realized.For example, the embodiment of the device for detecting position in storehouse described above is only schematical, for example,
The division of the unit, only a kind of division of logic function, formula that in actual implementation, there may be another division manner, for example (,) it is multiple
Unit or component can be combined or can be integrated into another system, or some features can be ignored or not executed.It is another
Point, shown or discussed mutual coupling, direct-coupling or communication connection can be by some interfaces, device or
The INDIRECT COUPLING of unit or communication connection can be electrical, machinery or other forms.
The unit illustrated as separating component may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, you can be located at a place, or may be distributed over multiple
In network element.Some or all of unit therein can be selected according to the actual needs to realize the mesh of this embodiment scheme
's.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, it can also
It is that each unit physically exists alone, it can also be during two or more units be integrated in one unit.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in a computer read/write memory medium.Based on this understanding, technical scheme of the present invention is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be expressed in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be
People's computer, server or network equipment etc.) it performs all or part of the steps of the method described in the various embodiments of the present invention.
And storage medium above-mentioned includes：USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), arbitrary access are deposited
The various media that can store program ver-ify code such as reservoir (RAM, Random Access Memory), magnetic disc or CD.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (10)
1. a kind of method for detecting position in storehouse, which is characterized in that including：
Obtain the first video image of the predeterminable area in each warehouse at least one warehouse, wherein the predeterminable area is set
It is equipped at least one mark of the position in storehouse for distinguishing each warehouse；
First video image is identified using first network identification model, to obtain recognition result, wherein the knowledge
Other result includes the position in storehouse information in each warehouse, and the first network identification model is using artificial intelligence approach to described
What multiple first sample images of predeterminable area were trained, the extraction feature of the first network identification model includes institute
State the feature of at least one mark.
2. according to the method described in claim 1, it is characterized in that, the artificial intelligence approach includes machine learning method, institute
The method of stating further includes：
Obtain multiple first sample images of the predeterminable area, wherein by manually according at least one mark by cargo
The case where blocking is the corresponding position in storehouse information of the multiple first sample image labeling；
The multiple first sample image by mark is trained to obtain described first using the machine learning method
Network Recognition model.
3. according to the method described in claim 2, it is characterized in that, in the utilization machine learning method to the institute by mark
It states before multiple first sample images are trained to obtain the first network identification model, by being manually the multiple first sample
This image labeling pedestrian information,
The method further includes：
Multiple second sample images are trained using the machine learning method to obtain the second Network Recognition model,
In at least one identified whether by human interference by being manually that the multiple second sample image mark is described；
It at least one identifies whether artificially to be done using described in the first video image described in the second Network Recognition Model Identification
It disturbs, second network model includes shot and long term memory network model；
If at least one mark is by human interference, it is determined that the recognition result is invalid；Or,
If at least one mark is not by human interference, it is determined that the recognition result is effective.
4. according to the method described in claim 3, it is characterized in that, further including：
Identify that the preset range of at least one mark whether there is pedestrian using the first network identification model；
Wherein, described at least one to be identified whether using described in the first video image described in the second Network Recognition Model Identification
By human interference, including：
If there is pedestrian, at least one mark in the first video image described in the second Network Recognition Model Identification is utilized
Whether by human interference,
Wherein described at least one mark is included artificially blocking mark, artificial scribble mark and artificially tearing the case where human interference
Pull at least one of mark.
5. method according to any one of claim 2 to 4, which is characterized in that utilize the machine learning side described
Method is trained after obtaining the first network identification model the multiple first sample image by mark, also wraps
It includes：
Obtain the second video image of the predeterminable area in each warehouse at least one warehouse, wherein the predeterminable area is set
It is equipped at least one mark of the position in storehouse for distinguishing each warehouse；
Based on second video image, the first network identification model is tested, to obtain the second video figure
The test result of picture；
If the test result includes the position in storehouse information of mistake, correct position in storehouse letter is marked for second video image
Breath, and the first network identification model is trained again using the second video image after mark as sample image, with
Obtain updated first network identification model.
6. method according to claim 1 to 4, which is characterized in that in utilization Network Recognition model to described
After video image is identified, further include：
The recognition result is uploaded to dispatch server in real time, to carry out position in storehouse analysis to the position in storehouse information.
7. method according to any one of claims 1 to 4, which is characterized in that it is described it is at least one be identified as number,
Quick Response Code, letter, color or user-defined identification；
At least one mark is distributed on the wall of the first area or on holder.
8. a kind of method of analysis position in storehouse, which is characterized in that including：
Recognition result is received from server, wherein the recognition result is using Network Recognition model at least one warehouse
The video image of predeterminable area in each warehouse be identified, the recognition result includes the storehouse in each warehouse
Position information, the Network Recognition model is trained to multiple sample images of the predeterminable area using artificial intelligence approach
It obtains, the extraction feature of the Network Recognition model includes the feature of at least one mark；
Early warning is carried out based on the recognition result.
9. a kind of device for detecting position in storehouse, which is characterized in that including：
Acquisition module, the first video image of the predeterminable area for obtaining each warehouse at least one warehouse, wherein institute
State at least one mark for the position in storehouse that predeterminable area is provided with for distinguishing each warehouse；
Identification module, for first video image to be identified using first network identification model, to obtain identification knot
Fruit, wherein the recognition result includes the position in storehouse information in each warehouse, and the first network identification model is using artificial
Intelligent method is trained multiple first sample images of the predeterminable area, the first network identification model
Extraction feature includes the feature of at least one mark.
10. a kind of device of analysis position in storehouse, which is characterized in that including：
Receiving module, for receiving recognition result from server, wherein the recognition result is using Network Recognition model to extremely
What the video image of the predeterminable area in each warehouse in a few warehouse was identified, the recognition result includes described
The position in storehouse information in each warehouse, the Network Recognition model are multiple samples to the predeterminable area using artificial intelligence approach
What image was trained, the extraction feature of the Network Recognition model includes the feature of at least one mark；
Alarm module, for carrying out early warning based on the recognition result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810306459.2A CN108537272A (en) | 2018-04-08 | 2018-04-08 | Method and apparatus for detection and analysis position in storehouse |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810306459.2A CN108537272A (en) | 2018-04-08 | 2018-04-08 | Method and apparatus for detection and analysis position in storehouse |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108537272A true CN108537272A (en) | 2018-09-14 |
Family
ID=63483001
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810306459.2A Pending CN108537272A (en) | 2018-04-08 | 2018-04-08 | Method and apparatus for detection and analysis position in storehouse |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108537272A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109766854A (en) * | 2019-01-15 | 2019-05-17 | 济南浪潮高新科技投资发展有限公司 | A kind of robust human face recognizer based on two stages complementary networks |
| CN111460909A (en) * | 2020-03-09 | 2020-07-28 | 兰剑智能科技股份有限公司 | Vision-based goods location management method and device |
| CN111612398A (en) * | 2020-05-21 | 2020-09-01 | 北京每日优鲜电子商务有限公司 | Warehouse goods distribution method and device, computer equipment and storage medium |
| CN111931762A (en) * | 2020-09-25 | 2020-11-13 | 广州佰锐网络科技有限公司 | AI-based image recognition solution method, device and readable storage medium |
| CN114842332A (en) * | 2022-04-12 | 2022-08-02 | 苏州艾吉威机器人有限公司 | Stock position detection method and system |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103116766A (en) * | 2013-03-20 | 2013-05-22 | 南京大学 | Increment neural network and sub-graph code based image classification method |
| CN105488456A (en) * | 2015-11-23 | 2016-04-13 | 中国科学院自动化研究所 | Adaptive rejection threshold adjustment subspace learning based human face detection method |
| CN106372666A (en) * | 2016-08-31 | 2017-02-01 | 同观科技（深圳）有限公司 | Target identification method and device |
| CN107016356A (en) * | 2017-03-21 | 2017-08-04 | 乐蜜科技有限公司 | Certain content recognition methods, device and electronic equipment |
| CN107122375A (en) * | 2016-12-12 | 2017-09-01 | 南京理工大学 | The recognition methods of image subject based on characteristics of image |
| CN107423719A (en) * | 2017-08-03 | 2017-12-01 | 上海物联网有限公司 | A kind of railway goods yard based on video image protects famous detection method |
| CN107578017A (en) * | 2017-09-08 | 2018-01-12 | 百度在线网络技术（北京）有限公司 | Method and apparatus for generating image |
| CN107748858A (en) * | 2017-06-15 | 2018-03-02 | 华南理工大学 | A kind of multi-pose eye locating method based on concatenated convolutional neutral net |
-
2018
- 2018-04-08 CN CN201810306459.2A patent/CN108537272A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103116766A (en) * | 2013-03-20 | 2013-05-22 | 南京大学 | Increment neural network and sub-graph code based image classification method |
| CN105488456A (en) * | 2015-11-23 | 2016-04-13 | 中国科学院自动化研究所 | Adaptive rejection threshold adjustment subspace learning based human face detection method |
| CN106372666A (en) * | 2016-08-31 | 2017-02-01 | 同观科技（深圳）有限公司 | Target identification method and device |
| CN107122375A (en) * | 2016-12-12 | 2017-09-01 | 南京理工大学 | The recognition methods of image subject based on characteristics of image |
| CN107016356A (en) * | 2017-03-21 | 2017-08-04 | 乐蜜科技有限公司 | Certain content recognition methods, device and electronic equipment |
| CN107748858A (en) * | 2017-06-15 | 2018-03-02 | 华南理工大学 | A kind of multi-pose eye locating method based on concatenated convolutional neutral net |
| CN107423719A (en) * | 2017-08-03 | 2017-12-01 | 上海物联网有限公司 | A kind of railway goods yard based on video image protects famous detection method |
| CN107578017A (en) * | 2017-09-08 | 2018-01-12 | 百度在线网络技术（北京）有限公司 | Method and apparatus for generating image |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109766854A (en) * | 2019-01-15 | 2019-05-17 | 济南浪潮高新科技投资发展有限公司 | A kind of robust human face recognizer based on two stages complementary networks |
| CN111460909A (en) * | 2020-03-09 | 2020-07-28 | 兰剑智能科技股份有限公司 | Vision-based goods location management method and device |
| CN111612398A (en) * | 2020-05-21 | 2020-09-01 | 北京每日优鲜电子商务有限公司 | Warehouse goods distribution method and device, computer equipment and storage medium |
| CN111931762A (en) * | 2020-09-25 | 2020-11-13 | 广州佰锐网络科技有限公司 | AI-based image recognition solution method, device and readable storage medium |
| CN114842332A (en) * | 2022-04-12 | 2022-08-02 | 苏州艾吉威机器人有限公司 | Stock position detection method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108537272A (en) | Method and apparatus for detection and analysis position in storehouse | |
| US11151427B2 (en) | Method and apparatus for checkout based on image identification technique of convolutional neural network | |
| CN106845890B (en) | Storage monitoring method and device based on video monitoring | |
| CN108985359A (en) | A kind of commodity recognition method, self-service machine and computer readable storage medium | |
| CN109087510A (en) | traffic monitoring method and device | |
| CN109543542A (en) | A kind of determination method whether particular place personnel dressing standardizes | |
| CN108961547A (en) | A kind of commodity recognition method, self-service machine and computer readable storage medium | |
| CN105510333B (en) | The localization method and device of defect in a kind of plastic film flexible package printing | |
| CN104966304A (en) | Kalman filtering and nonparametric background model-based multi-target detection tracking method | |
| CN109345876A (en) | A kind of control and monitoring system for ships and method | |
| CN113411542A (en) | Intelligent working condition monitoring equipment | |
| CN111063144A (en) | Abnormal behavior monitoring method, device, equipment and computer readable storage medium | |
| CN112183472A (en) | Method for detecting whether test field personnel wear work clothes or not based on improved RetinaNet | |
| CN111951258A (en) | Goods shelf out-of-stock early warning analysis system and method based on edge calculation | |
| CN113034624A (en) | Temperature early warning image identification method, system, equipment and storage medium based on temperature sensing color-changing adhesive tape | |
| CN109684976A (en) | Door state monitoring method, device, equipment and system and storage medium | |
| CN112949546A (en) | Off-duty detection method and device based on artificial intelligence perception of station service state | |
| EP3901818A1 (en) | Method and system for analyzing commodities on supermarket shelf | |
| CN108510487A (en) | A kind of photovoltaic module remote monitoring device | |
| CN112687022A (en) | Intelligent building inspection method and system based on video | |
| CN111339927A (en) | Intelligent work state identification system for personnel in electric power business hall | |
| CN108629310B (en) | Engineering management supervision method and device | |
| CN110287878A (en) | The determination method and device of marketing strategy, storage medium, electronic device | |
| CN109492573A (en) | A kind of pointer read method and device | |
| CN114612735A (en) | Detection method of hook anti-falling device based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180914 |