CN108596115A - A kind of vehicle checking method, apparatus and system based on convolutional neural networks - Google Patents
A kind of vehicle checking method, apparatus and system based on convolutional neural networks Download PDFInfo
- Publication number
- CN108596115A CN108596115A CN201810392105.4A CN201810392105A CN108596115A CN 108596115 A CN108596115 A CN 108596115A CN 201810392105 A CN201810392105 A CN 201810392105A CN 108596115 A CN108596115 A CN 108596115A
- Authority
- CN
- China
- Prior art keywords
- vehicle image
- image
- convolutional neural
- neural networks
- target 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
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of vehicle checking method, apparatus and system based on convolutional neural networks, method include：Divide at least one vehicle image from image to be detected by full convolutional neural networks, and marks the location information of each vehicle image；Extract vehicle image；The pixel number for detecting vehicle image, the vehicle image that pixel number is less than to the first setting numerical value are determined as Small object vehicle image；Small object vehicle image amplification the first setting multiple is formed into amplification target vehicle image；Amplification target vehicle image is identified by sub- convolutional neural networks to divide the first accurate vehicle image；First accurate vehicle image is reduced into the first setting multiple and forms target vehicle image；According to the location information of each vehicle image, detection image is integrated using each target vehicle image, the vehicle image for being not determined to Small object vehicle image and image to be detected.The accuracy of detection for the detection image to be formed can be improved in technical solution through the invention.
Description
Technical field
The present invention relates to field of computer technology, more particularly to a kind of vehicle checking method based on convolutional neural networks,
Apparatus and system.
Background technology
With in vehicle-mounted automatic Pilot, it is often necessary to the realtime graphic of acquisition road in real time, to the realtime graphic of acquisition into
Row identification is to be split the vehicle image in image and mark to form detection image, to realize to the vehicle in image
It is detected.
Currently, being mainly detected to the vehicle in image by full convolutional neural networks.
The remote vehicle of the distance of range image collecting device relatively far away from corresponds to the vehicle image in realtime graphic
Usually relatively small, when being detected to the vehicle in image by convolutional neural networks, convolutional neural networks are not easy to opposite
Smaller vehicle image is split and marks, and the accuracy of detection of the detection image of formation is relatively low.
Invention content
An embodiment of the present invention provides a kind of vehicle checking method, apparatus and system based on convolutional neural networks, can carry
The accuracy of detection for the detection image that height is formed.
In a first aspect, the present invention provides a kind of vehicle checking methods based on convolutional neural networks, including：
Image to be detected is inputted into full convolutional neural networks so that the full convolutional neural networks are to described image to be detected
It is identified, to be partitioned at least one vehicle image from described image to be detected, and each vehicle image is marked to exist
Location information in described image to be detected；
Extract each vehicle image that the full convolutional neural networks are divided；
The corresponding pixel number of each vehicle image institute of Detection and Extraction, and corresponding pixel points number is small
It is determined as Small object vehicle image in each vehicle image of the first setting numerical value；
Be directed to each described Small object vehicle image, by the Small object vehicle image amplification first setting multiple with
Form amplification target vehicle image；
It is directed to each described amplification target vehicle image, it will be described according to the size of the amplification target vehicle image
Amplification target vehicle image inputs corresponding sub- convolutional neural networks so that the sub- convolutional neural networks are to the amplification target
Vehicle image is identified, to be partitioned into the first accurate vehicle image from the amplification target vehicle image；
Each first accurate vehicle image is reduced into the first setting multiple to form target vehicle image；
According to location information of each vehicle image in described image to be detected, each target vehicle is utilized
Image, each vehicle image for being not determined to Small object vehicle image and described image to be detected integrate detection image.
Preferably,
Further include：
Each vehicle image that corresponding pixel points number is more than to the second setting numerical value is determined as big target vehicle figure
Picture；
Be directed to each described big target vehicle image, by the big target vehicle image down second set multiple with
It is formed and reduces target vehicle image；
It is directed to each described diminution target vehicle image, according to the size of the big target vehicle image by the contracting
Small object vehicle image inputs corresponding sub- convolutional neural networks so that the sub- convolutional neural networks are to the diminution target carriage
Image is identified, to be partitioned into the second accurate vehicle image from the diminution target vehicle image；
By each second accurate the second setting of vehicle image amplification multiple to form target vehicle image.
Preferably,
Each vehicle image that the extraction full convolutional neural networks are divided, including：
It is directed to each described vehicle image that the full convolutional neural networks are divided, is executed：
Generate the gray level image corresponding to described image to be detected；
The gray value that the vehicle image is corresponded to each current pixel point in the gray level image is labeled as 255,
And by the gray value of each pixel other than the current pixel point marks and is unless each in the gray level image, with
Form binaryzation vehicle image；
Then, the corresponding pixel number of each vehicle image institute of Detection and Extraction, and by respective pixel
Each vehicle image that point number is less than the first setting numerical value is determined as Small object vehicle image, including：Detect each institute
The corresponding pixel number of binaryzation vehicle image institute is stated, and corresponding pixel points number is less than each of the first setting numerical value
A binaryzation vehicle image is determined as Small object vehicle image.
Preferably,
The value range of the first setting multiple, including：More than 1 and it is less than 4；
And/or
The value range of the second setting multiple, including：More than 1 and it is less than 4.
Second aspect, an embodiment of the present invention provides a kind of vehicle detection apparatus based on convolutional neural networks, including：
First interactive module, for image to be detected to be inputted full convolutional neural networks so that the full convolutional Neural net
Described image to be detected is identified in network, to be partitioned at least one vehicle image from described image to be detected, and marks
Location information of each vehicle image in described image to be detected；
Image zooming-out module, each vehicle image divided for extracting the full convolutional neural networks；
Detection process module, for the corresponding pixel number of each vehicle image institute of Detection and Extraction, and
Each vehicle image that corresponding pixel points number is less than to the first setting numerical value is determined as Small object vehicle image；
Scaling processing module, for being directed to each described Small object vehicle image, by the Small object vehicle image
Amplification the first setting multiple is to form amplification target vehicle image；Each first accurate vehicle image is reduced into the first setting
Multiple is to form target vehicle image；
Second interactive module, for being directed to each described amplification target vehicle image, according to the amplification target vehicle
The amplification target vehicle image is inputted corresponding sub- convolutional neural networks by the size of image so that the sub- convolutional Neural
The amplification target vehicle image is identified in network, to be partitioned into the first accurate vehicle from the amplification target vehicle image
Image；
Integrated processing module, for the location information according to each vehicle image in described image to be detected, profit
With each target vehicle image, be not determined to each vehicle image of Small object vehicle image with it is described to be detected
Image integrates detection image.
Preferably,
The detection process module is further used for corresponding pixel points number being more than each described of the second setting numerical value
Vehicle image is determined as big target vehicle image；
The scaling processing module is further used for being directed to each described big target vehicle image, by the big mesh
Mark vehicle image reduces the second setting multiple and reduces target vehicle image to be formed；By each second accurate vehicle image amplification the
Two setting multiples are to form target vehicle image；
Second interactive module is further used for being directed to each described diminution target vehicle image, according to described
The diminution target vehicle image is inputted corresponding sub- convolutional neural networks by the size of big target vehicle image so that the son
The diminution target vehicle image is identified in convolutional neural networks, to be partitioned into from the diminution target vehicle image
Two accurate vehicle images.
Preferably,
Described image extraction module, for being directed to each described vehicle figure that the full convolutional neural networks are divided
Picture executes：
Generate the gray level image corresponding to described image to be detected；
The gray value that the vehicle image is corresponded to each current pixel point in the gray level image is labeled as 255,
And by the gray value of each pixel other than the current pixel point marks and is unless each in the gray level image, with
Form binaryzation vehicle image；
Then, the detection process module, for detecting the corresponding pixel of each binaryzation vehicle image institute
Number, and corresponding pixel points number is less than each binaryzation vehicle image that first sets numerical value and is determined as Small object vehicle
Image.
Preferably,
The value range of the first setting multiple, including：More than 1 and it is less than 4；
And/or
The value range of the second setting multiple, including：More than 1 and it is less than 4..
The third aspect, an embodiment of the present invention provides a kind of vehicle detecting systems based on convolutional neural networks, including：
Full convolutional neural networks, at least one sub- convolutional neural networks, and as described in any in above-mentioned second aspect
Vehicle detection apparatus based on convolutional neural networks；Wherein,
The full convolutional neural networks are connected with the vehicle detection apparatus based on convolutional neural networks, described in each
Sub- convolutional neural networks are connected with the vehicle detection apparatus based on convolutional neural networks, and each described sub- convolution god
Picture size range there are one being corresponded to respectively through network.
An embodiment of the present invention provides a kind of vehicle checking method, apparatus and system based on convolutional neural networks, the party
In method, image to be detected of input is identified by full convolutional neural networks, to be partitioned at least from image to be detected
One vehicle image, and location information of each vehicle image in image to be detected is marked, then extract convolutional neural networks
Each vehicle image of segmentation is then said when the number of the pixel corresponding to the vehicle image is less than the first setting numerical value
The bright vehicle image is relatively small, and full convolutional neural networks fail accurately to divide the vehicle image in image to be detected
It cuts, i.e., the segmentation precision corresponding to the vehicle image is relatively low；It is subsequent, then it can be determined not from each vehicle image of extraction
The each Small object vehicle image that can be accurately divided by full convolutional neural networks, then by each Small object vehicle image amplification the
One setting multiple is to form corresponding amplification target vehicle image, at this point, amplification target vehicle image has relatively large ruler
It is very little, it can be according to the size of each amplification target vehicle image, respectively by corresponding sub- convolutional neural networks to each amplification mesh
Mark vehicle image is identified again, and each sub- convolutional neural networks can be then partitioned into from the amplification target vehicle image of input
Each the first of segmentation accurate vehicle image is reduced the first setting multiple to be formed by the higher first accurate vehicle image of precision
After target vehicle image with degree of precision, then location information that can be according to each vehicle image in image to be detected,
Using each target vehicle image, it is not determined to each vehicle image and the integrated inspection of image to be detected of Small object vehicle image
Altimetric image, to improve formation detection image accuracy of detection.
Description of the drawings
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is the present invention
Some embodiments for those of ordinary skill in the art without creative efforts, can also basis
These attached drawings obtain other attached drawings.
Fig. 1 is a kind of flow chart for vehicle checking method based on convolutional neural networks that one embodiment of the invention provides；
Fig. 2 is a kind of structural representation for vehicle detection apparatus based on convolutional neural networks that one embodiment of the invention provides
Figure；
Fig. 3 is a kind of structural representation for vehicle detecting system based on convolutional neural networks that one embodiment of the invention provides
Figure；
Fig. 4 is the flow for another vehicle checking method based on convolutional neural networks that one embodiment of the invention provides
Figure.
Specific implementation mode
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments, based on the embodiments of the present invention, those of ordinary skill in the art
The every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
As shown in Figure 1, an embodiment of the present invention provides a kind of vehicle checking methods based on convolutional neural networks, including：
Step 101, image to be detected is inputted into full convolutional neural networks so that the full convolutional neural networks are waited for described
Detection image is identified, and to be partitioned at least one vehicle image from described image to be detected, and marks each vehicle
Location information of the image in described image to be detected；
Step 102, each vehicle image that the full convolutional neural networks are divided is extracted；
Step 103, the corresponding pixel number of each vehicle image institute of Detection and Extraction, and by respective pixel
Each vehicle image that point number is less than the first setting numerical value is determined as Small object vehicle image；
Step 104, it is directed to each described Small object vehicle image, Small object vehicle image amplification first is set
Multiple is determined to form amplification target vehicle image；
Step 105, it is directed to each described amplification target vehicle image, according to the ruler of the amplification target vehicle image
It is very little that the amplification target vehicle image is inputted into corresponding sub- convolutional neural networks so that the sub- convolutional neural networks are to described
Amplification target vehicle image is identified, to be partitioned into the first accurate vehicle image from the amplification target vehicle image；
Step 106, each first accurate vehicle image is reduced into the first setting multiple to form target vehicle image；
Step 107, the location information according to each vehicle image in described image to be detected, using each described
Target vehicle image, each vehicle image for being not determined to Small object vehicle image and the integrated inspection of described image to be detected
Altimetric image.
Embodiment as shown in Figure 1 is identified image to be detected of input by full convolutional neural networks, with from waiting for
It is partitioned at least one vehicle image in detection image, and marks location information of each vehicle image in image to be detected,
Then each vehicle image of extraction convolutional neural networks segmentation, the number of the pixel corresponding to the vehicle image are less than
When the first setting numerical value, then illustrate that the vehicle image is relatively small, full convolutional neural networks fail in image to be detected to this
Vehicle image is accurately divided, i.e., the segmentation precision corresponding to the vehicle image is relatively low；It is subsequent, then it can be from each of extraction
The each Small object vehicle image for failing accurately to be divided by full convolutional neural networks is determined in vehicle image, it then will be each small
Target vehicle image magnification first sets multiple to form corresponding amplification target vehicle image, at this point, amplification target vehicle figure
As having relatively large size, corresponding sub- convolution god can be passed through respectively according to the size of each amplification target vehicle image
Each amplification target vehicle image is identified again through network, each sub- convolutional neural networks then can be from the amplification mesh of input
It is partitioned into the higher first accurate vehicle image of precision in mark vehicle image, each the first of segmentation the accurate vehicle image is reduced
It, then can be according to each vehicle image to be detected after first setting multiple is to form the target vehicle image with degree of precision
Location information in image using each target vehicle image, is not determined to each vehicle image of Small object vehicle image
Integrate detection image with image to be detected, to improve formation detection image accuracy of detection.
In above-described embodiment, various sizes of amplification target image can be inputted respectively with different images recognition capability
In sub- convolutional neural networks, i.e. each sub- convolutional neural networks are corresponding, and there are one picture size ranges, input a son volume
The size of the amplification target image of product neural network should be located within the size range corresponding to the sub- convolutional neural networks, this
When sub- convolutional neural networks then preferably the amplification target image of input can be identified, with from the amplification target vehicle of input
It is partitioned into the first accurate vehicle image with higher precision in image.
In one embodiment of the invention, further include：
Each vehicle image that corresponding pixel points number is more than to the second setting numerical value is determined as big target vehicle figure
Picture；
Be directed to each described big target vehicle image, by the big target vehicle image down second set multiple with
It is formed and reduces target vehicle image；
It is directed to each described diminution target vehicle image, according to the size of the big target vehicle image by the contracting
Small object vehicle image inputs corresponding sub- convolutional neural networks so that the sub- convolutional neural networks are to the diminution target carriage
Image is identified, to be partitioned into the second accurate vehicle image from the diminution target vehicle image；
By each second accurate the second setting of vehicle image amplification multiple to form target vehicle image.
In above-described embodiment, it is detected by the corresponding pixel number of each vehicle image institute to extraction,
It realizes and is classified to vehicle image according to the corresponding pixel number of each vehicle image institute, by corresponding pixel points number
Each vehicle image more than the second setting numerical value is determined as big target vehicle image, then by big target vehicle image down the
Two setting multiples reduce target vehicle image to be formed, and according to each size for reducing target vehicle image by each diminution mesh
Mark vehicle image inputs different convolutional neural networks and is identified to be partitioned into the second accurate vehicle image, by segmentation respectively
After second accurate the second setting of vehicle image amplification multiple is to form the target vehicle image with degree of precision, according to each
Location information of the vehicle image in image to be detected using each target vehicle image, is not determined to Small object vehicle figure
When each vehicle image of picture integrates detection image with image to be detected, relatively large vehicle image also has in detection image
Higher segmentation precision can further improve the accuracy of detection for the detection image to be formed.
It is understandable, each vehicle of the first setting numerical value and the second setting numerical value is located at for corresponding pixel points number
Image directly can input corresponding sub- convolutional neural networks according to the size of current vehicle image and be identified accordingly.
In one embodiment of the invention, each vehicle figure for extracting the full convolutional neural networks and being divided
Picture, including：
It is directed to each described vehicle image that the full convolutional neural networks are divided, is executed：
Generate the gray level image corresponding to described image to be detected；
The gray value that the vehicle image is corresponded to each current pixel point in the gray level image is labeled as 255,
And by the gray value of each pixel other than the current pixel point marks and is unless each in the gray level image, with
Form binaryzation vehicle image；
Then, the corresponding pixel number of each vehicle image institute of Detection and Extraction, and by respective pixel
Each vehicle image that point number is less than the first setting numerical value is determined as Small object vehicle image, including：Detect each institute
The corresponding pixel number of binaryzation vehicle image institute is stated, and corresponding pixel points number is less than each of the first setting numerical value
A binaryzation vehicle image is determined as Small object vehicle image.
In one embodiment of the invention, the value range of the first setting multiple, including：More than 1 and it is less than 4；With/
Or, the value range of the second setting multiple, including：More than 1 and it is less than 4.For example, the first setting multiple and second is set
Determine multiple and may each be 2 or 3.
As shown in Fig. 2, an embodiment of the present invention provides a kind of vehicle detection apparatus based on convolutional neural networks, including：
First interactive module 201, for image to be detected to be inputted full convolutional neural networks so that the full convolutional Neural
Described image to be detected is identified in network, to be partitioned at least one vehicle image from described image to be detected, and marks
Remember location information of each vehicle image in described image to be detected；
Image zooming-out module 202, each vehicle image divided for extracting the full convolutional neural networks；
Detection process module 203, for the corresponding pixel number of each vehicle image institute of Detection and Extraction,
And each vehicle image that corresponding pixel points number is less than to the first setting numerical value is determined as Small object vehicle image；
Scaling processing module 204, for being directed to each described Small object vehicle image, by the Small object vehicle figure
As amplification the first setting multiple is to form amplification target vehicle image；Each first accurate vehicle image is reduced first to set
Multiple is determined to form target vehicle image；
Second interactive module 205, for being directed to each described amplification target vehicle image, according to the amplification target
The amplification target vehicle image is inputted corresponding sub- convolutional neural networks by the size of vehicle image so that the sub- convolution god
The amplification target vehicle image is identified through network, to be partitioned into first from the amplification target vehicle image precisely
Vehicle image；
Integrated processing module 206, is used for the location information according to each vehicle image in described image to be detected,
Using each target vehicle image, be not determined to each vehicle image of Small object vehicle image with it is described to be checked
Altimetric image integrates detection image.
In one embodiment of the invention, the detection process module 203 is further used for corresponding pixel points number being more than
Each vehicle image of second setting numerical value is determined as big target vehicle image；
The scaling processing module 204 is further used for being directed to each described big target vehicle image, will be described big
Target vehicle image down second sets multiple and reduces target vehicle image to be formed；By each second accurate vehicle image amplification
Second setting multiple is to form target vehicle image；
Second interactive module 205 is further used for being directed to each described diminution target vehicle image, according to institute
The diminution target vehicle image is inputted corresponding sub- convolutional neural networks by the size for stating big target vehicle image so that described
The diminution target vehicle image is identified in sub- convolutional neural networks, to be partitioned into from the diminution target vehicle image
Second accurate vehicle image.
In one embodiment of the invention, described image extraction module 202, for being directed to the full convolutional neural networks institute
Each described vehicle image of segmentation executes：
Generate the gray level image corresponding to described image to be detected；
The gray value that the vehicle image is corresponded to each current pixel point in the gray level image is labeled as 255,
And by the gray value of each pixel other than the current pixel point marks and is unless each in the gray level image, with
Form binaryzation vehicle image；
Then, the detection process module 203, for detecting the corresponding pixel of each binaryzation vehicle image institute
Point number, and corresponding pixel points number is less than each binaryzation vehicle image that first sets numerical value and is determined as Small object
Vehicle image.
In one embodiment of the invention, the value range of the first setting multiple, including：More than 1 and it is less than 4；With/
Or, the value range of the second setting multiple, including：More than 1 and it is less than 4.
As shown in figure 3, an embodiment of the present invention provides a kind of vehicle detecting systems based on convolutional neural networks, including：
Full convolutional neural networks 301, at least one sub- convolutional neural networks 302, and any one embodiment of the invention
The vehicle detection apparatus 303 based on convolutional neural networks of middle offer；Wherein,
The full convolutional neural networks 301 are connected with the vehicle detection apparatus 303 based on convolutional neural networks, each
A sub- convolutional neural networks 302 are connected with the vehicle detection apparatus 303 based on convolutional neural networks, and each
The sub- convolutional neural networks 302 are corresponding respectively, and there are one picture size ranges.
The contents such as the information exchange between each unit, implementation procedure in above-mentioned apparatus, due to implementing with the method for the present invention
Example is based on same design, and particular content can be found in the narration in the method for the present invention embodiment, and details are not described herein again.
In order to more clearly illustrate technical scheme of the present invention and advantage, with reference to system provided in an embodiment of the present invention
System, for being realized based on the vehicle checking method of convolutional neural networks by the system, as shown in figure 4, can specifically include
Following each step：
Step 401, full convolutional neural networks and at least one sub- convolutional neural networks are trained in advance.
Wherein, each sub- convolutional neural networks correspond to a picture size range respectively；As the current sub- convolution god of input
When the size of image through network is within the scope of the picture size corresponding to current sub- convolutional neural networks, current sub- convolution
Neural network can carry out the image of input preferable identification to be partitioned into corresponding vehicle image.
In the embodiment of the present invention only for training sub- convolutional neural networks X1, X2, X3, X4.
Step 402, image to be detected is inputted by full convolutional neural networks based on the vehicle detection apparatus of convolutional neural networks.
Step 403, image to be detected of input is identified in full convolutional neural networks, to divide from image to be detected
Go out at least one vehicle image, and marks location information of each vehicle image in image to be detected.
In following each steps of the embodiment of the present invention, only for being partitioned into vehicle image A, B, C.
Step 404, the vehicle detection apparatus based on convolutional neural networks is directed to each vehicle image of segmentation, generates
Vehicle image is corresponded to the gray value of each current pixel point in gray level image by the gray level image corresponding to image to be detected
Labeled as 255, and by the gray value of each pixel other than current pixel point marks and is unless each in gray level image,
To form binaryzation vehicle image.
In following each steps of the embodiment of the present invention, only to form the corresponding binaryzation vehicle image of A, B, C institute
For A1, B1, C1.
Step 405, the corresponding pixel of vehicle detection apparatus detection A1, B1, C1 institutes based on convolutional neural networks
Number, and corresponding pixel points number is less than each binaryzation vehicle image that first sets numerical value and is determined as Small object vehicle figure
Picture, each binaryzation vehicle image that corresponding pixel points number is more than to the first setting numerical value are determined as big target vehicle image,
Corresponding pixel points number is located at each binaryzation vehicle image between the first setting numerical value and the second setting numerical value to be determined as
Middle target vehicle image.
In following each steps of the embodiment of the present invention, only A1 is determined as Small object vehicle image, is determined as B1
C1 is determined as big target vehicle image by middle target vehicle image.
Step 406, the first setting multiple of A1 amplifications is amplified with being formed based on the vehicle detection apparatus of convolutional neural networks
Target vehicle image and C1 reduce the second setting multiple and reduce target vehicle image to be formed.
Here, the first setting multiple and the second setting multiple can be configured in conjunction with practical business scene, the first setting
Multiple and the second setting multiple should usually be more than 1 and be less than 4, for example can be 2 or 3.
In following each steps of the embodiment of the present invention, only with by the first setting multiple of Small object vehicle image A1 amplification with
It forms amplification target vehicle image A2, big target vehicle image C1 is reduced to the second setting multiple to form diminution target vehicle figure
As for C2.
Step 407, based on the vehicle detection apparatus of convolutional neural networks according to the ruler of the size of A2, the size of B1 and C2
It is very little, the corresponding sub- convolutional neural networks of A2, B1 and C2 institute are determined from each trained sub- convolutional neural networks.
In the embodiment of the present invention, only to determine that the sub- convolutional neural networks corresponding to amplification target vehicle image A2 are
Sub- convolutional neural networks corresponding to X1, middle target vehicle image B1 are X2, reduce the son volume corresponding to target vehicle image C2
For product neural network is X3；It is the figure corresponding to X1 that i.e. the size of amplification target vehicle image A2, which is located at sub- convolutional neural networks,
As within size range, it is the picture size model corresponding to X2 that the size of middle target vehicle image B1, which is located at sub- convolutional neural networks,
Within enclosing, reduce target vehicle image C2 size be located at sub- convolutional neural networks be picture size range corresponding to X3 it
It is interior.
Step 408, A2 is identified in sub- convolutional neural networks X1, to be partitioned into the first accurate vehicle image from A2
A3；B1 is identified in sub- convolutional neural networks X2, to be partitioned into the accurate vehicle image B2 of third from B1；Sub- convolutional Neural net
C2 is identified in network X3, to be partitioned into the second accurate vehicle image C3 from C2.
Step 409, A3 is reduced to form target by the first setting multiple based on the vehicle detection apparatus of convolutional neural networks
Vehicle image A4, by C3 amplifications the second setting multiple to form target vehicle image C4.
Step 410, believed according to the position of A, B, C in image to be detected based on the vehicle detection apparatus of convolutional neural networks
Breath integrates detection image using A4, B2, C4 and image to be detected.
In the embodiment of the present invention, i.e., A4 is integrated into image to be detected by the location information according to A in image to be detected,
B2 is integrated into image to be detected according to location informations of the B in image to be detected, according to positions of the C in image to be detected
C4 is integrated into image to be detected by information.
It can specifically be integrated by maximum ballot method.
In a kind of mode in the cards, can according to the location information of A, B, C in image to be detected, using A4, B2,
The profile of C4 is split label in image to be detected to A, B, C.
In conclusion each embodiment of the present invention at least has the advantages that：
1, in one embodiment of the invention, image to be detected of input is identified by full convolutional neural networks, with from
It is partitioned at least one vehicle image in image to be detected, and marks position letter of each vehicle image in image to be detected
Then breath extracts each vehicle image of convolutional neural networks segmentation, the number of the pixel corresponding to the vehicle image
When less than the first setting numerical value, then illustrate that the vehicle image is relatively small, full convolutional neural networks fail in image to be detected
The vehicle image is accurately divided, i.e., the segmentation precision corresponding to the vehicle image is relatively low；It is subsequent, then it can be from extraction
The each Small object vehicle image for failing accurately to be divided by full convolutional neural networks is determined in each vehicle image, it then will be each
A the first setting of Small object vehicle image amplification multiple is to form corresponding amplification target vehicle image, at this point, amplification target vehicle
Image has relatively large size, can pass through corresponding son volume respectively according to the size of each amplification target vehicle image
Each amplification target vehicle image is identified in product neural network again, and each sub- convolutional neural networks then can putting from input
It is partitioned into the higher first accurate vehicle image of precision in big target vehicle image, by each first accurate vehicle image of segmentation
After reducing the first setting multiple to form the target vehicle image with degree of precision, then it can waited for according to each vehicle image
Location information in detection image using each target vehicle image, is not determined to each vehicle of Small object vehicle image
Image and image to be detected integrate detection image, to improve formation detection image accuracy of detection.
2, it in one embodiment of the invention, is carried out by the corresponding pixel number of each vehicle image institute to extraction
Detection is realized and is classified to vehicle image according to the corresponding pixel number of each vehicle image institute, by respective pixel
Each vehicle image that point number is more than the second setting numerical value is determined as big target vehicle image, then by big target vehicle image
It reduces the second setting multiple and reduces target vehicle image to be formed, and will be each according to each size for reducing target vehicle image
Diminution target vehicle image inputs different convolutional neural networks and is identified to be partitioned into the second accurate vehicle image respectively, will
After second accurate the second setting of vehicle image amplification multiple of segmentation is to form the target vehicle image with degree of precision, root
According to location information of each vehicle image in image to be detected, using each target vehicle image, it is not determined to Small object
When each vehicle image of vehicle image integrates detection image with image to be detected, relatively large vehicle image in detection image
Also there is higher segmentation precision, can further improve the accuracy of detection for the detection image to be formed.
It should be noted that herein, such as first and second etc relational terms are used merely to an entity
Or operation is distinguished with another entity or operation, is existed without necessarily requiring or implying between these entities or operation
Any actual relationship or order.Moreover, the terms "include", "comprise" or its any other variant be intended to it is non-
It is exclusive to include, so that the process, method, article or equipment including a series of elements includes not only those elements,
But also include other elements that are not explicitly listed, or further include solid by this process, method, article or equipment
Some elements.In the absence of more restrictions, the element limited by sentence " including one ", is not arranged
Except there is also other identical factors in the process, method, article or apparatus that includes the element.
Finally, it should be noted that：The foregoing is merely presently preferred embodiments of the present invention, is merely to illustrate the skill of the present invention
Art scheme, is not intended to limit the scope of the present invention.Any modification for being made all within the spirits and principles of the present invention,
Equivalent replacement, improvement etc., are included within the scope of protection of the present invention.
Claims (9)
1. a kind of vehicle checking method based on convolutional neural networks, which is characterized in that including：
Image to be detected is inputted into full convolutional neural networks so that the full convolutional neural networks carry out described image to be detected
Identification, to be partitioned at least one vehicle image from described image to be detected, and marks each vehicle image described
Location information in image to be detected；
Extract each vehicle image that the full convolutional neural networks are divided；
The corresponding pixel number of each vehicle image institute of Detection and Extraction, and corresponding pixel points number is less than the
Each vehicle image of one setting numerical value is determined as Small object vehicle image；
It is directed to each described Small object vehicle image, by Small object vehicle image amplification the first setting multiple to be formed
Amplification target vehicle image；
It is directed to each described amplification target vehicle image, according to the size of the amplification target vehicle image by the amplification
Target vehicle image inputs corresponding sub- convolutional neural networks so that the sub- convolutional neural networks are to the amplification target vehicle
Image is identified, to be partitioned into the first accurate vehicle image from the amplification target vehicle image；
Each first accurate vehicle image is reduced into the first setting multiple to form target vehicle image；
According to location information of each vehicle image in described image to be detected, each target vehicle figure is utilized
Picture, each vehicle image for being not determined to Small object vehicle image integrate detection image with described image to be detected.
2. according to the method described in claim 1, it is characterized in that, further including：
Each vehicle image that corresponding pixel points number is more than to the second setting numerical value is determined as big target vehicle image；
It is directed to each described big target vehicle image, the big target vehicle image down second is set into multiple to be formed
Reduce target vehicle image；
It is directed to each described diminution target vehicle image, according to the size of the big target vehicle image by the diminution mesh
It marks vehicle image and inputs corresponding sub- convolutional neural networks so that the sub- convolutional neural networks are to the diminution target vehicle figure
As being identified, to be partitioned into the second accurate vehicle image from the diminution target vehicle image；
By each second accurate the second setting of vehicle image amplification multiple to form target vehicle image.
3. according to the method described in claim 1, it is characterized in that,
Each vehicle image that the extraction full convolutional neural networks are divided, including：
It is directed to each described vehicle image that the full convolutional neural networks are divided, is executed：
Generate the gray level image corresponding to described image to be detected；
The gray value that the vehicle image is corresponded to each current pixel point in the gray level image is labeled as 255, and will
The gray value of each pixel other than the current pixel point marks and is unless each in the gray level image, to be formed
Binaryzation vehicle image；
Then, the corresponding pixel number of each vehicle image institute of Detection and Extraction, and by corresponding pixel points
Each vehicle image that number is less than the first setting numerical value is determined as Small object vehicle image, including：Detect each described two
The corresponding pixel number of value vehicle image institute, and corresponding pixel points number is less than the first each institute for setting numerical value
It states binaryzation vehicle image and is determined as Small object vehicle image.
4. according to any method in claims 1 to 3, which is characterized in that
The value range of the first setting multiple, including：More than 1 and it is less than 4；
And/or
The value range of the second setting multiple, including：More than 1 and it is less than 4.
5. a kind of vehicle detection apparatus based on convolutional neural networks, which is characterized in that including：
First interactive module, for image to be detected to be inputted full convolutional neural networks so that the full convolutional neural networks pair
Described image to be detected is identified, and to be partitioned at least one vehicle image from described image to be detected, and marks each
Location information of the vehicle image in described image to be detected；
Image zooming-out module, each vehicle image divided for extracting the full convolutional neural networks；
Detection process module, for the corresponding pixel number of each vehicle image institute of Detection and Extraction, and will be right
Each vehicle image that pixel number is less than the first setting numerical value is answered to be determined as Small object vehicle image；
Scaling processing module amplifies the Small object vehicle image for being directed to each described Small object vehicle image
First setting multiple is to form amplification target vehicle image；Each first accurate vehicle image is reduced into the first setting multiple
To form target vehicle image；
Second interactive module, for being directed to each described amplification target vehicle image, according to the amplification target vehicle figure
The amplification target vehicle image is inputted corresponding sub- convolutional neural networks by the size of picture so that the sub- convolutional neural networks
The amplification target vehicle image is identified, to be partitioned into the first accurate vehicle figure from the amplification target vehicle image
Picture；
Integrated processing module, for the location information according to each vehicle image in described image to be detected, using each
A target vehicle image, each vehicle image for being not determined to Small object vehicle image and described image to be detected
Integrated detection image.
6. device according to claim 5, which is characterized in that
The detection process module is further used for for corresponding pixel points number being more than each vehicle of the second setting numerical value
Image is determined as big target vehicle image；
The scaling processing module is further used for being directed to each described big target vehicle image, by the big target carriage
Image down second sets multiple and reduces target vehicle image to be formed；Each second accurate vehicle image amplification second is set
Multiple is determined to form target vehicle image；
Second interactive module is further used for being directed to each described diminution target vehicle image, according to the big mesh
The diminution target vehicle image is inputted corresponding sub- convolutional neural networks by the size for marking vehicle image so that the sub- convolution
The diminution target vehicle image is identified in neural network, to be partitioned into the second essence from the diminution target vehicle image
Quasi- vehicle image.
7. device according to claim 5, which is characterized in that
Described image extraction module, each the described vehicle image divided for being directed to the full convolutional neural networks,
It executes：
Generate the gray level image corresponding to described image to be detected；
The gray value that the vehicle image is corresponded to each current pixel point in the gray level image is labeled as 255, and will
The gray value of each pixel other than the current pixel point marks and is unless each in the gray level image, to be formed
Binaryzation vehicle image；
Then, the detection process module, for detecting the corresponding pixel number of each binaryzation vehicle image institute,
And each binaryzation vehicle image that corresponding pixel points number is less than to the first setting numerical value is determined as Small object vehicle figure
Picture.
8. according to any device in claim 5 to 7, which is characterized in that
The value range of the first setting multiple, including：More than 1 and it is less than 4；
And/or
The value range of the second setting multiple, including：More than 1 and it is less than 4.
9. a kind of vehicle detecting system based on convolutional neural networks, which is characterized in that including：
Full convolutional neural networks, at least one sub- convolutional neural networks, and as described in any in the claims 5 to 8
Vehicle detection apparatus based on convolutional neural networks；Wherein,
Each described sub- convolutional neural networks is connected with the vehicle detection apparatus based on convolutional neural networks, each
The sub- convolutional neural networks are corresponding respectively, and there are one picture size ranges.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810392105.4A CN108596115A (en) | 2018-04-27 | 2018-04-27 | A kind of vehicle checking method, apparatus and system based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810392105.4A CN108596115A (en) | 2018-04-27 | 2018-04-27 | A kind of vehicle checking method, apparatus and system based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108596115A true CN108596115A (en) | 2018-09-28 |
Family
ID=63610127
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810392105.4A Pending CN108596115A (en) | 2018-04-27 | 2018-04-27 | A kind of vehicle checking method, apparatus and system based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108596115A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109528230A (en) * | 2018-11-21 | 2019-03-29 | 济南浪潮高新科技投资发展有限公司 | A kind of tumor of breast dividing method and device based on multi-stage transformation network |
| CN109800637A (en) * | 2018-12-14 | 2019-05-24 | 中国科学院深圳先进技术研究院 | A kind of remote sensing image small target detecting method |
| CN111539962A (en) * | 2020-01-10 | 2020-08-14 | 济南浪潮高新科技投资发展有限公司 | Target image classification method, device and medium |
| CN111539961A (en) * | 2019-12-13 | 2020-08-14 | 山东浪潮人工智能研究院有限公司 | Target segmentation method, device and equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103279759A (en) * | 2013-06-09 | 2013-09-04 | 大连理工大学 | Vehicle front trafficability analyzing method based on convolution nerve network |
| CN106503710A (en) * | 2016-10-26 | 2017-03-15 | 北京邮电大学 | A kind of automobile logo identification method and device |
| CN107358177A (en) * | 2017-06-27 | 2017-11-17 | 维拓智能科技（深圳）有限公司 | A kind of medium and long distance pedestrian detection method and terminal device based on graphical analysis |
| CN107944354A (en) * | 2017-11-10 | 2018-04-20 | 南京航空航天大学 | A kind of vehicle checking method based on deep learning |
-
2018
- 2018-04-27 CN CN201810392105.4A patent/CN108596115A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103279759A (en) * | 2013-06-09 | 2013-09-04 | 大连理工大学 | Vehicle front trafficability analyzing method based on convolution nerve network |
| CN106503710A (en) * | 2016-10-26 | 2017-03-15 | 北京邮电大学 | A kind of automobile logo identification method and device |
| CN107358177A (en) * | 2017-06-27 | 2017-11-17 | 维拓智能科技（深圳）有限公司 | A kind of medium and long distance pedestrian detection method and terminal device based on graphical analysis |
| CN107944354A (en) * | 2017-11-10 | 2018-04-20 | 南京航空航天大学 | A kind of vehicle checking method based on deep learning |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109528230A (en) * | 2018-11-21 | 2019-03-29 | 济南浪潮高新科技投资发展有限公司 | A kind of tumor of breast dividing method and device based on multi-stage transformation network |
| CN109528230B (en) * | 2018-11-21 | 2021-08-17 | 山东浪潮科学研究院有限公司 | Method and device for segmenting breast tumor based on multistage transformation network |
| CN109800637A (en) * | 2018-12-14 | 2019-05-24 | 中国科学院深圳先进技术研究院 | A kind of remote sensing image small target detecting method |
| CN111539961A (en) * | 2019-12-13 | 2020-08-14 | 山东浪潮人工智能研究院有限公司 | Target segmentation method, device and equipment |
| CN111539962A (en) * | 2020-01-10 | 2020-08-14 | 济南浪潮高新科技投资发展有限公司 | Target image classification method, device and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108596115A (en) | A kind of vehicle checking method, apparatus and system based on convolutional neural networks | |
| US20190019055A1 (en) | Word segmentation system, method and device | |
| JP3799408B1 (en) | Image processing apparatus and image processing method | |
| CN113344857B (en) | Defect detection network training method, defect detection method and storage medium | |
| CN109858476B (en) | Tag expansion method and electronic equipment | |
| CN103824373B (en) | A kind of bill images amount of money sorting technique and system | |
| CN105160668A (en) | Image segmentation method and system, and cell image segmentation method and system | |
| CN113569863B (en) | Document checking method, system, electronic equipment and storage medium | |
| CN107818321A (en) | A kind of watermark date recognition method for vehicle annual test | |
| CN113191348B (en) | Template-based text structured extraction method and tool | |
| CN108573244B (en) | Vehicle detection method, device and system | |
| CN110569774A (en) | Automatic line graph image digitalization method based on image processing and pattern recognition | |
| CN110414318A (en) | Container number recognition methods under large scene | |
| CN114549993A (en) | Method, system and device for scoring line segment image in experiment and readable storage medium | |
| CN112016481A (en) | Financial statement information detection and identification method based on OCR | |
| CN103699876A (en) | Method and device for identifying vehicle number based on linear array CCD (Charge Coupled Device) images | |
| CN113505781A (en) | Target detection method and device, electronic equipment and readable storage medium | |
| CN108053409B (en) | Automatic construction method and system for remote sensing image segmentation reference library | |
| CN112434582A (en) | Lane line color identification method and system, electronic device and storage medium | |
| CN111460198B (en) | Picture timestamp auditing method and device | |
| CN111914706A (en) | Method and device for detecting and controlling quality of character detection output result | |
| CN111160374A (en) | Color identification method, system and device based on machine learning | |
| CN108734158B (en) | Real-time train number identification method and device | |
| CN115631197A (en) | Image processing method, device, medium, equipment and system | |
| CN112308062B (en) | Medical image access number identification method in complex background image |
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: 20180928 |