CN109657689A - A kind of extracting method, device and the equipment of the vehicle key point based on deep learning - Google Patents
A kind of extracting method, device and the equipment of the vehicle key point based on deep learning Download PDFInfo
- Publication number
- CN109657689A CN109657689A CN201811496187.3A CN201811496187A CN109657689A CN 109657689 A CN109657689 A CN 109657689A CN 201811496187 A CN201811496187 A CN 201811496187A CN 109657689 A CN109657689 A CN 109657689A
- Authority
- CN
- China
- Prior art keywords
- key point
- vehicle pictures
- sample
- preset
- vehicle
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
- G06V10/464—Salient features, e.g. scale invariant feature transforms [SIFT] using a plurality of salient features, e.g. bag-of-words [BoW] representations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The application belongs to technical field of computer vision more particularly to a kind of extracting method, device and the equipment of the vehicle key point based on deep learning.Extracting method, device and the equipment for the vehicle key point based on deep learning that this application discloses a kind of, wherein method includes: to obtain vehicle pictures to be detected；The vehicle pictures to be detected are input to the preset key point based on preset convolutional neural networks and extract model, multiple key point coordinates of the vehicle pictures to be detected are obtained, the preset key point extracts the association relation model that model is vehicle pictures and multiple key point coordinates；The preset convolutional neural networks are as follows: one data layers, five conv layers, two LRN layers, three pooling layers and two fc layers.Solve the extracting method poor robustness of existing vehicle key point and use by limitation the technical issues of.
Description
Technical field
The application belongs to mentioning for technical field of computer vision more particularly to a kind of vehicle key point based on deep learning
Take method, apparatus and equipment.
Background technique
With the development of artificial intelligence and computer vision technique, vehicle recongnition technique has been widely used in many images
In investigation and public security system, such as in bayonet system, electronic police system, intelligent transportation and automatic Pilot field.
Vehicle key point is extracted as the basic technology in vehicle identification, accuracy for vehicle identification to Guan Chong
It wants.The prior art is extracted key point and is carried out using classifier, and detection method is to be obtained using classifier by sliding window
Feature in certain area, and the comparison for carrying out each feature obtains testing result.But this detection method detects block
Compare effective (such as Car license recognition), is wanted when being easy the interference put by other, poor robustness to the detection of key point, and detecting
Image clearly is sought, so that the detection method use is limited to.
Therefore it provides a kind of robustness is good and the extracting method of vehicle key point without limitations is used to become this field skill
Art personnel technical problem urgently to be resolved.
Summary of the invention
Extracting method, device and the equipment for the vehicle key point based on deep learning that this application provides a kind of are used for vehicle
The extraction of key point solves the extracting method poor robustness of existing vehicle key point and the technical issues of use is by limitation.
In view of this, the application first aspect provides a kind of extracting method of vehicle key point based on deep learning,
Include:
Obtain vehicle pictures to be detected；
The vehicle pictures to be detected are input to the preset key point based on preset convolutional neural networks and extract model, are obtained
To multiple key point coordinates of the vehicle pictures to be detected, it is vehicle pictures and multiple passes that the preset key point, which extracts model,
The association relation model of key point coordinate；
The preset convolutional neural networks are as follows: one data layers, five conv layers, two LRN layers, three pooling layers
With two fc layers.
Preferably, the method also includes:
Multiple sample vehicle pictures are obtained, there is multiple key points mark, and described in every sample vehicle pictures
Multiple key point marks are identical with the quantity of the multiple key point coordinate；
According to each key point mark in sample vehicle pictures described in every, determine that each key point mark corresponds to
Sample key point coordinate, obtain the sample key point coordinate of every sample vehicle pictures；
With the preset convolutional neural networks be the first training network, multiple described sample vehicle pictures are first instruction
Practice the input pointer of network, the output that all sample key point coordinates are the first training network is as a result, obtain described
Preset key point extracts model.
Preferably, described to train network, multiple described sample vehicle pictures with the preset convolutional neural networks for first
Input pointer, all sample key point coordinates for the first training network are the output knot of the first training network
Fruit obtains before the preset key point extracts model further include:
Referred to the input that VGGNet is the second training network, multiple described sample vehicle pictures are the second training network
Mark, all sample key point coordinates are the output of the second training network as a result, obtaining the key point based on VGGNet
Extract model；
Described be first training network, multiple described sample vehicle pictures with the preset convolutional neural networks is described the
Input pointer, all sample key point coordinates of one training network are the output of the first training network as a result, obtaining
The preset key point extracts model specifically:
With the preset convolutional neural networks be the first training network, multiple described sample vehicle pictures are first instruction
Input pointer, all sample key point coordinates of white silk network are the output of the first training network as a result, according to described
Key point based on VGGNet extracts model and distills algorithm by knowledge, obtains the preset key point and extracts model.
Preferably, after described multiple sample vehicle pictures of acquisition further include:
Image change is carried out to sample vehicle pictures described in every, the sample vehicle pictures after obtaining multiple image changes,
Described image variation includes: plus makes an uproar and/or selective erasing and/or rotation and/or brightness change and/or affine transformation；
Then each key point according in sample vehicle pictures described in every marks, and determines each key point mark
Corresponding sample key point coordinate is infused, the sample key point coordinate of every sample vehicle pictures is obtained specifically:
Each key point mark in sample vehicle pictures after being changed according to every described image, determines each pass
Key point marks corresponding sample key point coordinate, the sample key point of the sample vehicle pictures after obtaining every described image variation
Coordinate.
Preferably, the method also includes:
The spin matrix of the vehicle pictures to be detected is determined according to the multiple key point coordinate, and according to the rotation
The vehicle pictures to be detected are aligned by matrix.
Preferably, the width of the data layers of input channel of the preset convolutional neural networks and high size phase
Deng.
Preferably, each of described preset convolutional neural networks conv layers of activation primitive is equal are as follows: relu.
The application second aspect provides a kind of extraction element of vehicle key point based on deep learning, comprising:
Picture obtains module, for obtaining vehicle pictures to be detected；
Key point determining module, for being input to the vehicle pictures to be detected based on the pre- of preset convolutional neural networks
It sets key point and extracts model, obtain multiple key point coordinates of the vehicle pictures to be detected, the preset key point extracts mould
Type is the association relation model of vehicle pictures and multiple key point coordinates；
The preset convolutional neural networks are as follows: one data layers, five conv layers, two LRN layers, three pooling layers
With two fc layers.
The application third aspect provides a kind of extract equipment of vehicle key point based on deep learning, and the equipment includes
Processor and memory:
Said program code is transferred to the processor for storing program code by the memory；
The processor is used for the method according to the above-mentioned first aspect of the instruction execution in said program code.
The application fourth aspect provides a kind of computer readable storage medium, and the computer readable storage medium is for depositing
Program code is stored up, said program code is for executing method described in above-mentioned first aspect.
As can be seen from the above technical solutions, the embodiment of the present application has the advantage that
The extracting method for the vehicle key point based on deep learning that this application provides a kind of, comprising: obtain measuring car to be checked
Picture；The vehicle pictures to be detected are input to the preset key point based on preset convolutional neural networks and extract model, are obtained
To multiple key point coordinates of the vehicle pictures to be detected, it is vehicle pictures and multiple passes that the preset key point, which extracts model,
The association relation model of key point coordinate；The preset convolutional neural networks are as follows: one data layers, five conv layers, two LRN
Layer, three pooling layers and two fc layers.
In the application, the vehicle pictures to be detected for needing to extract key point are input to, the depth of neural network is passed through
Acquistion to preset key point extract model in, the key point coordinate of vehicle pictures to be detected is obtained, because neural network has
The advantages of deep learning, thus it is good by the robustness that the preset key point that neural network obtains extracts model, and use is without limitations,
The key point coordinate of the vehicle pictures to be detected obtained from is correct, and simultaneously to the image definition of vehicle pictures to be detected without
It is required that solve the extracting method poor robustness of existing vehicle key point and use by limitation the technical issues of.
Detailed description of the invention
Fig. 1 is a kind of first embodiment of the extracting method of the vehicle key point based on deep learning in the embodiment of the present application
Flow diagram；
Fig. 2 is a kind of second embodiment of the extracting method of the vehicle key point based on deep learning in the embodiment of the present application
Flow diagram；
Fig. 3 is a kind of 3rd embodiment of the extracting method of the vehicle key point based on deep learning in the embodiment of the present application
Flow diagram；
Fig. 4 is a kind of fourth embodiment of the extracting method of the vehicle key point based on deep learning in the embodiment of the present application
Flow diagram；
Fig. 5 is a kind of structural schematic diagram of the extraction element of the vehicle key point based on deep learning in the embodiment of the present application
The structural schematic diagram of Fig. 6 neural network structure provided by the present application；
Fig. 7 is the schematic diagram of the key point mark provided in the application second embodiment.
Specific embodiment
The embodiment of the present application provides extracting method, device and the equipment of a kind of vehicle key point based on deep learning,
For the extraction of vehicle key point, solves the extracting method poor robustness of existing vehicle key point and using the technology limited to
Problem.
In order to make those skilled in the art more fully understand application scheme, below in conjunction in the embodiment of the present application
Attached drawing, the technical scheme in the embodiment of the application is clearly and completely described, it is clear that described embodiment is only this
Apply for a part of the embodiment, instead of all the embodiments.Based on the embodiment in the application, those of ordinary skill in the art exist
Every other embodiment obtained under the premise of creative work is not made, shall fall in the protection scope of this application.
Referring to Fig. 1, in the embodiment of the present application a kind of extracting method of the vehicle key point based on deep learning first
The flow diagram of embodiment, comprising:
Step 101 obtains vehicle pictures to be detected.
Vehicle pictures to be detected are input to the extraction mould of the preset key point based on preset convolutional neural networks by step 102
Type obtains multiple key point coordinates of vehicle pictures to be detected.
It should be noted that preset key point extracts the incidence relation mould of model vehicle pictures and multiple key point coordinates
Type.It should be noted simultaneously that as shown in fig. 6, preset convolutional neural networks are as follows: one data layers, five conv layers, two
LRN layers, three pooling layers and two fc layers.
Further, the width of the input channel of the data layer of preset convolutional neural networks is equal with high size.It can
With understanding, because conventional images are all R, G, B triple channel, therefore the quantity for designing the data layers of input channel is three
It is a.
Further, each conv layers of activation primitive of preset convolutional neural networks is equal are as follows: relu.
In the present embodiment, the vehicle pictures to be detected for needing to extract key point are input to, the depth of neural network is passed through
Learn obtained preset key point to extract in model, obtain the key point coordinate of vehicle pictures to be detected, because neural network has
There is the advantages of deep learning, thus it is good by the robustness that the preset key point that neural network obtains extracts model, using not by office
The key point coordinate of limit, the vehicle pictures to be detected obtained from is correct, and simultaneously to the image clearly of vehicle pictures to be detected
Spend no requirement (NR), solve the extracting method poor robustness of existing vehicle key point and use by limit to the technical issues of.
The above are the first of a kind of extracting method of the vehicle key point based on deep learning provided by the embodiments of the present application
Embodiment, the following are a kind of second realities of the extracting method of the vehicle key point based on deep learning provided by the embodiments of the present application
Apply example.
Referring to Fig. 2, in the embodiment of the present application a kind of extracting method of the vehicle key point based on deep learning second
The flow diagram of embodiment, comprising:
Step 201 obtains multiple sample vehicle pictures.
It should be noted that in every sample vehicle pictures have multiple key points mark, and multiple key points mark and
The quantity of multiple key point coordinates is identical.It is understood that key point mark is then existed in order to which key point to be marked
It only needs to calculate these coordinates with markd point when calculating the coordinate of key point and is considered key point coordinate.Simultaneously
It is understood that the setting of key point mark, can according to need progress, is not specifically limited herein.In the present embodiment, close
Key point mark to vehicle as shown in fig. 7, carried out the mark of 19 key points, wherein 1,2,18, No. 19 point marks vehicle window in total
Outer profile；3,4,6,7,13,14,16, No. 17 points mark automobile front lamp；5, No. 15 points mark vehicle lower edge；8,9,11, No. 12
Point mark license plate area；No. 10 points mark vehicle bottom profiled midpoint.
Step 202 is marked according to each key point in every sample vehicle pictures, determines that each key point mark corresponds to
Sample key point coordinate, obtain the sample key point coordinate of every sample vehicle pictures.
Step 203 take preset convolutional neural networks as first training network, multiple sample vehicle pictures is the first training net
The input pointer of network, all sample key point coordinates are that the output of the first training network extracts mould as a result, obtaining preset key point
Type.
Step 204 obtains vehicle pictures to be detected.
Vehicle pictures to be detected are input to the extraction mould of the preset key point based on preset convolutional neural networks by step 205
Type obtains multiple key point coordinates of vehicle pictures to be detected.
It should be noted that in the present embodiment, there are 19 key points above every sample vehicle pictures when model training, therefore
It is also 19 in the quantity of the key point coordinate of obtained vehicle pictures to be detected.
In the present embodiment, the vehicle pictures to be detected for needing to extract key point are input to, the depth of neural network is passed through
Learn obtained preset key point to extract in model, obtain the key point coordinate of vehicle pictures to be detected, because neural network has
There is the advantages of deep learning, thus it is good by the robustness that the preset key point that neural network obtains extracts model, using not by office
The key point coordinate of limit, the vehicle pictures to be detected obtained from is correct, and simultaneously to the image clearly of vehicle pictures to be detected
Spend no requirement (NR), solve the extracting method poor robustness of existing vehicle key point and use by limit to the technical issues of.
The above are the second of a kind of extracting method of the vehicle key point based on deep learning provided by the embodiments of the present application
Embodiment, the following are a kind of third of the extracting method of the vehicle key point based on deep learning provided by the embodiments of the present application is real
Apply example.
Referring to Fig. 3, a kind of third of the extracting method of the vehicle key point based on deep learning in the embodiment of the present application
The flow diagram of embodiment, comprising:
Step 301 obtains multiple sample vehicle pictures.
Step 301 is identical as the content of step 201 in the application second embodiment, and specific descriptions may refer to the second implementation
The content of example step 201, details are not described herein.
Step 302 is marked according to each key point in every sample vehicle pictures, determines that each key point mark corresponds to
Sample key point coordinate, obtain the sample key point coordinate of every sample vehicle pictures.
Step 303 is that the input that second training network, multiple sample vehicle pictures are the second training network refers to VGGNet
Mark, all sample key point coordinates are that the output of the second training network extracts model as a result, obtaining the key point based on VGGNet.
Step 304 take preset convolutional neural networks as first training network, multiple sample vehicle pictures is the first training net
The input pointer of network, all sample key point coordinates are the output of the first training network as a result, according to the key based on VGGNet
Point extracts model and distills algorithm by knowledge, obtains preset key point and extracts model.
It should be noted that the preset key point extraction model structure provided in the present embodiment is simple, model may be made
Performance is bad, although while VGGNet there is more outstanding ability in feature extraction, therefore this implementation to all kinds of image processing tasks
In example, the characteristic information of the model learning typical structure model of simple structure is made using the method that knowledge is distilled, reaches typical
Performance possessed by structural model, so that model parameter and calculation amount be greatly decreased under the premise of guaranteeing model performance.It can be with
Understand, knowledge distillation algorithm belongs to the prior art, and details are not described herein.
Step 305 obtains vehicle pictures to be detected.
Vehicle pictures to be detected are input to the extraction mould of the preset key point based on preset convolutional neural networks by step 306
Type obtains multiple key point coordinates of vehicle pictures to be detected.
It should be noted that step 306 is identical as the content of step 102 in the application first embodiment, specific descriptions can
With referring to the content of first embodiment step 102, details are not described herein.
In the present embodiment, the vehicle pictures to be detected for needing to extract key point are input to, the depth of neural network is passed through
Learn obtained preset key point to extract in model, obtain the key point coordinate of vehicle pictures to be detected, because neural network has
There is the advantages of deep learning, thus it is good by the robustness that the preset key point that neural network obtains extracts model, using not by office
The key point coordinate of limit, the vehicle pictures to be detected obtained from is correct, and simultaneously to the image clearly of vehicle pictures to be detected
Spend no requirement (NR), solve the extracting method poor robustness of existing vehicle key point and use by limit to the technical issues of.
The above are a kind of thirds of the extracting method of the vehicle key point based on deep learning provided by the embodiments of the present application
Embodiment, the following are a kind of 4th realities of the extracting method of the vehicle key point based on deep learning provided by the embodiments of the present application
Apply example.
Referring to Fig. 4, in the embodiment of the present application a kind of extracting method of the vehicle key point based on deep learning the 4th
The flow diagram of embodiment, comprising:
Step 401 obtains multiple sample vehicle pictures.
Step 401 is identical as the content of step 201 in the application second embodiment, and specific descriptions may refer to the second implementation
The content of example step 201, details are not described herein.
Step 402 carries out image change to every sample vehicle pictures, the sample vehicle figure after obtaining multiple image changes
Piece.
It should be noted that because true vehicle pictures to be detected are likely to occur a variety of situations, in order to enable preset pass
The service condition that key point extracts model is true, and sample vehicle pictures can be carried out with image change, and image change includes: plus makes an uproar
And/or selective erasing and/or rotation and/or brightness change and/or affine transformation.
It is understood that plus the mode of making an uproar can include but is not limited at random plus make an uproar, the spiced salt add make an uproar, Gauss adds and makes an uproar；At random
The mode of erasing: to carry out random areas selection to original image, and carrying out area filling, realizes the selective erasing to picture,
Enhance preset key point and extracts model to the extractability for blocking vehicle pictures to be detected；The mode of rotation are as follows: with original image
Central point be dot, all pixels point on image is rotated, rotating range include rotate clockwise 3~5 °, counterclockwise
3~5 ° of rotation enhances preset key point and extracts model to the adaptability for the vehicle pictures to be detected for having certain angle.Brightness becomes
Change and can be the adjustment for carrying out brightness to image using Adaptive contrast enhancement algorithm.
Step 403 is marked according to each key point in the sample vehicle pictures after every image change, determines each pass
Key point marks corresponding sample key point coordinate, and the sample key point of the sample vehicle pictures after obtaining every image change is sat
Mark.
Step 404 take preset convolutional neural networks as the first training network, the sample vehicle pictures after multiple image changes
Input pointer, all sample key point coordinates for the first training network are the output of the first training network as a result, obtaining preset
Key point extracts model.
Step 405 obtains vehicle pictures to be detected.
Vehicle pictures to be detected are input to the extraction mould of the preset key point based on preset convolutional neural networks by step 406
Type obtains multiple key point coordinates of vehicle pictures to be detected.
It should be noted that step 406 is identical as the content of step 102 in the application first embodiment, specific descriptions can
With referring to the content of first embodiment step 102, details are not described herein.
Step 407, multiple key point coordinates according to vehicle pictures to be detected, determine the spin moment of vehicle pictures to be detected
Battle array, and be aligned vehicle pictures to be detected according to spin matrix.
It should be noted that being aligned to vehicle pictures to be detected, be conducive to the discrimination of subsequent classification recognizer
It is promoted.It is understood that solving spin matrix in the present embodiment using Pu Shi (procrustes) method and carrying out shape normalizing
Change.Belong to now it is understood that vehicle pictures to be detected are carried out alignment using spin matrix after obtaining spin matrix simultaneously
There is technology, details are not described herein.
In the present embodiment, the vehicle pictures to be detected for needing to extract key point are input to, the depth of neural network is passed through
Learn obtained preset key point to extract in model, obtain the key point coordinate of vehicle pictures to be detected, because neural network has
There is the advantages of deep learning, thus it is good by the robustness that the preset key point that neural network obtains extracts model, using not by office
The key point coordinate of limit, the vehicle pictures to be detected obtained from is correct, and simultaneously to the image clearly of vehicle pictures to be detected
Spend no requirement (NR), solve the extracting method poor robustness of existing vehicle key point and use by limit to the technical issues of.
The above are the 4th of a kind of extracting method of the vehicle key point based on deep learning provided by the embodiments of the present application the
Embodiment, the following are a kind of implementations of the extraction element of the vehicle key point based on deep learning provided by the embodiments of the present application
Example, please refers to Fig. 5.
A kind of extraction element of the vehicle key point based on deep learning provided in the embodiment of the present application, comprising:
Picture obtains module 501, for obtaining vehicle pictures to be detected.
Key point determining module 502, for being input to vehicle pictures to be detected based on the pre- of preset convolutional neural networks
It sets key point and extracts model, obtain multiple key point coordinates of vehicle pictures to be detected, it is vehicle that preset key point, which extracts model,
The association relation model of picture and multiple key point coordinates.
Preset convolutional neural networks are as follows: one data layers, five conv layers, two LRN layers, three pooling layers and two
It is fc layers a.
The extract equipment for the vehicle key point based on deep learning that the embodiment of the present application also provides a kind of, equipment include place
Reason device and memory: memory is transferred to processor for storing program code, and by program code, and processor is used for basis
Instruction execution previous embodiment one in program code is to any vehicle key point based on deep learning of example IV
Extracting method, thereby executing various function application and data processing.
The embodiment of the present application also provides a kind of computer readable storage mediums, for storing program code, the program generation
Code is for executing the extracting method of previous embodiment one to any vehicle key point based on deep learning of example IV
In any one embodiment.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description,
The specific work process of device and unit, can refer to corresponding processes in the foregoing method embodiment, and 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 apparatus embodiments described above are merely exemplary, for example, the division of the unit, only
Only a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components can be tied
Another system is closed or is desirably integrated into, or some features can be ignored or not executed.Another point, it is shown or discussed
Mutual coupling, direct-coupling or communication connection can be through some interfaces, the INDIRECT COUPLING or logical of device or unit
Letter connection can be electrical property, mechanical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
It, can also be in addition, each functional unit in each embodiment of the application can integrate in one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of software functional units.
If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product
When, it can store in a computer readable storage medium.Based on this understanding, the technical solution of the application is substantially
The all or part of the part that contributes to existing technology or the technical solution can be in the form of software products in other words
It embodies, which is stored in a storage medium, including some instructions are used so that a computer
Equipment (can be personal computer, server or the network equipment etc.) executes the complete of each embodiment the method for the application
Portion or part steps.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only
Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. are various can store journey
The medium of sequence code.
The above, above embodiments are only to illustrate the technical solution of the application, rather than its limitations；Although referring to before
Embodiment is stated the application is described in detail, those skilled in the art should understand that: it still can be to preceding
Technical solution documented by each embodiment is stated to modify or equivalent replacement of some of the technical features；And these
It modifies or replaces, the spirit and scope of each embodiment technical solution of the application that it does not separate the essence of the corresponding technical solution.
Claims (10)
1. a kind of extracting method of the vehicle key point based on deep learning characterized by comprising
Obtain vehicle pictures to be detected；
The vehicle pictures to be detected are input to the preset key point based on preset convolutional neural networks and extract model, obtain institute
Multiple key point coordinates of vehicle pictures to be detected are stated, it is vehicle pictures and multiple key points that the preset key point, which extracts model,
The association relation model of coordinate；
The preset convolutional neural networks are as follows: one data layers, five conv layers, two LRN layers, three pooling layers and two
It is fc layers a.
2. the method according to claim 1, wherein the method also includes:
Multiple sample vehicle pictures are obtained, there is multiple key points mark, and the multiple in every sample vehicle pictures
Key point mark is identical with the quantity of the multiple key point coordinate；
According to each key point mark in sample vehicle pictures described in every, determine that each key point marks corresponding sample
This key point coordinate obtains the sample key point coordinate of every sample vehicle pictures；
With the preset convolutional neural networks be the first training network, multiple described sample vehicle pictures are first training net
The input pointer of network, all sample key point coordinates are the output of the first training network as a result, obtaining described preset
Key point extracts model.
3. according to the method described in claim 2, it is characterized in that, described trained with the preset convolutional neural networks for first
Network, multiple described sample vehicle pictures are input pointer, all sample key point coordinates of the first training network
Before the output for training network for described first is as a result, obtain the preset key point extraction model further include:
With VGGNet be input pointer that second training network, multiple described sample vehicle pictures are the second training network,
All sample key point coordinates are that the output of the second training network is extracted as a result, obtaining the key point based on VGGNet
Model；
Described be first training network, multiple described sample vehicle pictures with the preset convolutional neural networks is first instruction
Practice the input pointer of network, the output that all sample key point coordinates are the first training network is as a result, obtain described
Preset key point extracts model specifically:
With the preset convolutional neural networks be the first training network, multiple described sample vehicle pictures are first training net
The input pointer of network, all sample key point coordinates are the output of the first training network as a result, being based on according to described
The key point of VGGNet extracts model and distills algorithm by knowledge, obtains the preset key point and extracts model.
4. according to the method described in claim 2, it is characterized in that, after described multiple sample vehicle pictures of acquisition further include:
Image change is carried out to sample vehicle pictures described in every, the sample vehicle pictures after obtaining multiple image changes are described
Image change includes: plus makes an uproar and/or selective erasing and/or rotation and/or brightness change and/or affine transformation；
Then each key point according in sample vehicle pictures described in every marks, and determines each key point mark pair
The sample key point coordinate answered obtains the sample key point coordinate of every sample vehicle pictures specifically:
Each key point mark in sample vehicle pictures after being changed according to every described image, determines each key point
Corresponding sample key point coordinate is marked, the sample key point of the sample vehicle pictures after obtaining every described image variation is sat
Mark.
5. method according to claim 1 to 4, which is characterized in that the method also includes:
The spin matrix of the vehicle pictures to be detected is determined according to the multiple key point coordinate, and according to the spin matrix
The vehicle pictures to be detected are aligned.
6. method according to claim 1 to 4, which is characterized in that the institute of the preset convolutional neural networks
The width for stating data layers of input channel is equal with high size.
7. method according to claim 1 to 4, which is characterized in that the preset convolutional neural networks it is every
A conv layers of activation primitive is equal are as follows: relu.
8. a kind of extraction element of the vehicle key point based on deep learning characterized by comprising
Picture obtains module, for obtaining vehicle pictures to be detected；
Key point determining module, for the vehicle pictures to be detected to be input to the preset pass based on preset convolutional neural networks
Key point extracts model, obtains multiple key point coordinates of the vehicle pictures to be detected, and the preset key point extracts model and is
The association relation model of vehicle pictures and multiple key point coordinates；
The preset convolutional neural networks are as follows: one data layers, five conv layers, two LRN layers, three pooling layers and two
It is fc layers a.
9. a kind of extract equipment of the vehicle key point based on deep learning, which is characterized in that the equipment include processor with
And memory:
Said program code is transferred to the processor for storing program code by the memory；
The processor is used for the method according to instruction execution any one of claims 1 to 7 in said program code.
10. a kind of computer readable storage medium, which is characterized in that the computer readable storage medium is for storing program generation
Code, said program code is for method described in any one of perform claim requirement 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811496187.3A CN109657689A (en) | 2018-12-07 | 2018-12-07 | A kind of extracting method, device and the equipment of the vehicle key point based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811496187.3A CN109657689A (en) | 2018-12-07 | 2018-12-07 | A kind of extracting method, device and the equipment of the vehicle key point based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109657689A true CN109657689A (en) | 2019-04-19 |
Family
ID=66113582
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811496187.3A Pending CN109657689A (en) | 2018-12-07 | 2018-12-07 | A kind of extracting method, device and the equipment of the vehicle key point based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109657689A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111444972A (en) * | 2020-03-31 | 2020-07-24 | 高新兴科技集团股份有限公司 | Hierarchical vehicle target detection method based on deep learning |
| CN112348035A (en) * | 2020-11-11 | 2021-02-09 | 东软睿驰汽车技术（沈阳）有限公司 | Vehicle key point detection method and device and electronic equipment |
| CN112905824A (en) * | 2021-02-08 | 2021-06-04 | 智慧眼科技股份有限公司 | Target vehicle tracking method and device, computer equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107247989A (en) * | 2017-06-15 | 2017-10-13 | 北京图森未来科技有限公司 | A kind of neural network training method and device |
| CN107679078A (en) * | 2017-08-29 | 2018-02-09 | 银江股份有限公司 | A kind of bayonet socket image vehicle method for quickly retrieving and system based on deep learning |
| CN108491827A (en) * | 2018-04-13 | 2018-09-04 | 腾讯科技（深圳）有限公司 | A kind of vehicle checking method, device and storage medium |
| CN108733719A (en) * | 2017-04-24 | 2018-11-02 | 优信拍（北京）信息科技有限公司 | A kind of recognition methods of vehicle position, device, equipment and computer-readable medium |
-
2018
- 2018-12-07 CN CN201811496187.3A patent/CN109657689A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108733719A (en) * | 2017-04-24 | 2018-11-02 | 优信拍（北京）信息科技有限公司 | A kind of recognition methods of vehicle position, device, equipment and computer-readable medium |
| CN107247989A (en) * | 2017-06-15 | 2017-10-13 | 北京图森未来科技有限公司 | A kind of neural network training method and device |
| CN107679078A (en) * | 2017-08-29 | 2018-02-09 | 银江股份有限公司 | A kind of bayonet socket image vehicle method for quickly retrieving and system based on deep learning |
| CN108491827A (en) * | 2018-04-13 | 2018-09-04 | 腾讯科技（深圳）有限公司 | A kind of vehicle checking method, device and storage medium |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111444972A (en) * | 2020-03-31 | 2020-07-24 | 高新兴科技集团股份有限公司 | Hierarchical vehicle target detection method based on deep learning |
| CN112348035A (en) * | 2020-11-11 | 2021-02-09 | 东软睿驰汽车技术（沈阳）有限公司 | Vehicle key point detection method and device and electronic equipment |
| CN112905824A (en) * | 2021-02-08 | 2021-06-04 | 智慧眼科技股份有限公司 | Target vehicle tracking method and device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Oh et al. | Approaching the computational color constancy as a classification problem through deep learning | |
| CN108509859B (en) | Non-overlapping area pedestrian tracking method based on deep neural network | |
| CN106683046A (en) | Real-time image splicing method for police unmanned aerial vehicle investigation and evidence obtaining | |
| CN112200111A (en) | Global and local feature fused occlusion robust pedestrian re-identification method | |
| CN107169954B (en) | Image significance detection method based on parallel convolutional neural network | |
| CN105335722A (en) | Detection system and detection method based on depth image information | |
| Qu et al. | A pedestrian detection method based on yolov3 model and image enhanced by retinex | |
| CN111814860A (en) | Multi-target detection method for garbage classification | |
| CN108918536A (en) | Tire-mold face character defect inspection method, device, equipment and storage medium | |
| CN109657689A (en) | A kind of extracting method, device and the equipment of the vehicle key point based on deep learning | |
| CN112949633B (en) | Improved YOLOv 3-based infrared target detection method | |
| CN106446862A (en) | Face detection method and system | |
| CN109784278A (en) | The small and weak moving ship real-time detection method in sea based on deep learning | |
| CN110414439A (en) | Anti- based on multi-peak detection blocks pedestrian tracting method | |
| CN108647695A (en) | Soft image conspicuousness detection method based on covariance convolutional neural networks | |
| CN109816001A (en) | A kind of more attribute recognition approaches of vehicle based on deep learning, device and equipment | |
| CN103839042A (en) | Human face recognition method and human face recognition system | |
| CN110232326A (en) | A kind of D object recognition method, device and storage medium | |
| CN106709500A (en) | Image feature matching method | |
| CN110490232A (en) | Method, apparatus, the equipment, medium of training literal line direction prediction model | |
| CN111582337A (en) | Strawberry malformation state detection method based on small sample fine-grained image analysis | |
| CN109325407B (en) | Optical remote sensing video target detection method based on F-SSD network filtering | |
| CN113468996A (en) | Camouflage object detection method based on edge refinement | |
| CN109993806A (en) | A kind of color identification method, device and electronic equipment | |
| CN112001219A (en) | Multi-angle multi-face recognition attendance checking method and system |
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: 20190419 |