CN107808126A - Vehicle retrieval method and device - Google Patents
Vehicle retrieval method and device Download PDFInfo
- Publication number
- CN107808126A CN107808126A CN201710940974.1A CN201710940974A CN107808126A CN 107808126 A CN107808126 A CN 107808126A CN 201710940974 A CN201710940974 A CN 201710940974A CN 107808126 A CN107808126 A CN 107808126A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- characteristic vector
- score value
- detected
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- 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
-
- 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
Abstract
The invention discloses a kind of vehicle retrieval method and device, wherein, the vehicle retrieval method includes：Extract N number of detection characteristic vector of vehicle to be detected；The similarity score sequence of each detection characteristic vector and all characteristic vectors in characteristic vector storehouse of vehicle to be detected is calculated successively, i-th of detection characteristic vector is subjected to Similarity Measure with all characteristic vectors in ith feature vector storehouse, obtain i-th of similarity score value sequence, wherein, ith feature vector storehouse is the characteristic vector storehouse of the identical vehicle part corresponding with i-th of detection characteristic vector pre-established；The similarity score value sequence calculated is ranked up, from similarity score value sequence in characteristic vector corresponding to maximum preceding M similarity score value, filters out target vehicle.The vehicle retrieval method carries out vehicle retrieval by extracting and integrating the multiple target feature of vehicle, and the vehicle retrieval degree of accuracy is higher.
Description
Technical field
The present invention relates to technical field of image processing, and in particular to a kind of vehicle retrieval method and device.
Background technology
It is an important problem in computer vision field to scheme to search figure, main task is being schemed by the picture of input
Technology as retrieving the image similar to its in storehouse, the technology of associated picture retrieval is provided for the mankind.It relate to computer and regards
Feel, image procossing, many technical fields such as pattern-recognition and information processing, the face retrieval, network either for maturation
Picture retrieval, or the car plate vehicle retrieval of monitoring field are required for putting into substantial amounts of manpower and go to handle.
In recent years, have extensively in fields such as intelligent video monitoring, Vehicular automatic driving, robot environment's perception to scheme to search figure
General application.For example, in public security big data system to scheme to search figure, be the vehicle image that user is provided as mesh
The image of vehicle is marked, the traveling record of target vehicle is searched in the alert data record of bayonet socket or electricity of magnanimity.
Vehicle retrieval method of the prior art is that the figure of view picture target vehicle is extracted by Sift, Surf, Hog scheduling algorithm
As feature, then the entire image feature by each vehicle in same algorithm extraction database, two groups of features are matched,
Target vehicle is retrieved from database.
However, be in above-mentioned technical proposal by being matched to the entire image feature of vehicle, once entire image
In difference of some position caused by artificial origin, can all influence matching effect.
On the other hand, inventor has found the field of video monitoring in traffic, due to the shadow by natural causes such as illumination, brightness
Ring, very big similitude be present in information such as the color and veins plus vehicle so that the vehicle retrieval degree of accuracy in the above method
Substantially reduce.
The content of the invention
In view of this, the embodiments of the invention provide a kind of vehicle retrieval method and device, to solve car in the prior art
Retrieval the degree of accuracy it is low the problem of.
First aspect present invention provides a kind of vehicle retrieval method, comprises the following steps：
Obtain the image of vehicle to be detected；
N number of detection characteristic vector of the vehicle to be detected is extracted from the image of the vehicle to be detected, wherein, N is big
In equal to 2, each detection characteristic vector corresponds to a part of the vehicle to be detected；
Following steps are performed respectively for N number of detection characteristic vector, obtain N number of similarity score value sequence：By i-th
Individual detection characteristic vector carries out Similarity Measure with all characteristic vectors in ith feature vector storehouse, obtains i-th of similarity
Score value sequence, wherein, the ith feature vector storehouse be pre-establish with described i-th detection characteristic vector it is corresponding
The characteristic vector storehouse of identical vehicle part, i take 1 to N natural number；Wherein, in different characteristic vector storehouse same vehicle spy
Being established between sign vector has corresponding relation；
Choose respectively feature corresponding to preceding M similarity score value maximum in N number of similarity score value sequence to
Amount；
Target vehicle is judged whether according to the corresponding relation between the characteristic vector of selection, the target vehicle corresponds to
N number of characteristic vector be respectively present in the characteristic vector of the selection；
If it is present determine that the target vehicle is the vehicle to be detected.
Alternatively, if target vehicle exists multiple, the vehicle retrieval method also includes：
Obtain all characteristic vectors of multiple target vehicles；
For the same target vehicle, respectively according to the phase of its characteristic vector and the characteristic vector of the vehicle to be detected
The similarity score value of the target vehicle is calculated like degree score value, and the result of calculation is ranked up；
The image of multiple target vehicles and its corresponding similarity score value are shown according to the ranking results.
Alternatively, the vehicle to be detected is extracted from the image of the vehicle to be detected using convolutional neural networks model
N number of detection characteristic vector.
Alternatively, before the image of vehicle to be detected is obtained, in addition to：
Obtain multiple images for including vehicle；
Using convolutional neural networks model vehicle image is extracted from the multiple image for including vehicle；
Vehicle characteristics are extracted from the vehicle image of extraction using convolutional neural networks model, and to the difference of same vehicle
Feature establishes corresponding relation, obtains characteristic vector storehouse.
Alternatively, after the step of image of the acquisition vehicle to be detected, in addition to：By the figure of the vehicle to be detected
As being converted into the second form by the first form.
Alternatively, the first form rgb format, second form are yuv format.
Second aspect of the present invention provides a kind of vehicle retrieval device, including：
First acquisition unit, for obtaining the image of vehicle to be detected；
Extraction unit, for extracting N number of detection feature of the vehicle to be detected from the image of the vehicle to be detected
Vector, wherein, N is more than or equal to 2, each part for detecting characteristic vector and corresponding to the vehicle to be detected；
First computing unit, for performing following steps respectively for N number of detection characteristic vector, obtain N number of similar
Spend score value sequence：I-th of detection characteristic vector is subjected to similarity meter with all characteristic vectors in ith feature vector storehouse
Calculate, obtain i-th of similarity score value sequence, wherein, the ith feature vector storehouse be pre-establish with described i-th
The characteristic vector storehouse of identical vehicle part corresponding to characteristic vector is detected, i takes 1 to N natural number；Wherein, different characteristic to
Being established in amount storehouse between the characteristic vector of same vehicle has corresponding relation；
Unit is chosen, for choosing preceding M similarity score value maximum in N number of similarity score value sequence respectively
Corresponding characteristic vector；
Judging unit, target vehicle is judged whether for the corresponding relation between the characteristic vector according to selection, institute
N number of characteristic vector corresponding to target vehicle is stated to be respectively present in the characteristic vector of the selection；
Determining unit, for determining that the target vehicle is the vehicle to be detected.
Alternatively, the vehicle retrieval device also includes：
Second acquisition unit, for obtaining all characteristic vectors of multiple target vehicles；
Second computing unit, for for the same target vehicle, respectively according to its characteristic vector with it is described to be detected
The similarity score value of the characteristic vector of vehicle calculates the similarity score value of the target vehicle, and the result of calculation is entered
Row sequence；
Display unit, for the image that multiple target vehicles are shown according to the ranking results and its corresponding similarity
Score value.
Third aspect present invention provides a kind of image processing apparatus, including at least one processor；And with it is described extremely
The memory of few processor communication connection；Wherein, have can be by the finger of one computing device for the memory storage
Order, the instruction is by least one computing device, so that at least one computing device first aspect present invention
Or the vehicle retrieval method described in any one optional mode of first aspect.
Fourth aspect present invention provides a kind of non-transient computer readable storage medium storing program for executing, and the non-transient computer is readable
Storage medium stores computer instruction, and the computer instruction is used to make computer perform first aspect present invention or first party
Vehicle retrieval method described in the optional mode of any one of face.
Fifth aspect present invention provides a kind of computer program product, and the computer program product is non-including being stored in
Calculation procedure in transitory computer readable storage medium, the computer program include programmed instruction, when described program instructs
When being computer-executed, the computer is set to perform in any one optional mode of first aspect present invention or first aspect
Described vehicle retrieval method.
Technical scheme provided by the invention, has the following advantages that：
1. vehicle retrieval method provided in an embodiment of the present invention, first, extract N number of detection feature of vehicle to be detected to
Amount；Secondly, calculate successively each detection characteristic vector of vehicle to be detected in characteristic vector storehouse all characteristic vectors it is similar
Scoring sequence is spent, i.e., i-th of detection characteristic vector is subjected to similarity meter with all characteristic vectors in ith feature vector storehouse
Calculate, obtain i-th of similarity score value sequence, wherein, ith feature vector storehouse be pre-establish with i-th detection feature
The characteristic vector storehouse of identical vehicle part corresponding to vector；Finally, the similarity score value sequence calculated is ranked up,
From similarity score value sequence in characteristic vector corresponding to maximum preceding M similarity score value, target vehicle is filtered out.Should
Vehicle retrieval method carries out vehicle retrieval by extracting and integrating the multiple target feature of vehicle, and the vehicle retrieval degree of accuracy is higher.
2. vehicle retrieval method provided in an embodiment of the present invention, target vehicle exist it is multiple in the case of, vehicle inspection
Suo Fangfa also includes：Obtain all characteristic vectors of multiple target vehicles；For same target vehicle, respectively according to its feature to
The similarity score value of amount and the characteristic vector of vehicle to be detected calculates the similarity score value of target vehicle, and to result of calculation
It is ranked up；The image of multiple target vehicles and its corresponding similarity score value are shown according to ranking results.It is same by calculating
The similarity score value of target vehicle, and the score value and corresponding multiple target vehicle images are shown, it can be visualized
As a result, additionally it is possible to according to visualization result, vehicle to be detected is filtered out from some target vehicles, accuracy in detection is higher.
3. vehicle retrieval method provided in an embodiment of the present invention, by using convolutional neural networks model from vehicle to be detected
Image in extract N number of detection characteristic vector of vehicle to be detected, i.e., deep learning training is carried out to the image of vehicle to be detected
Go out model, so as to extract N number of detection characteristic vector of vehicle to be detected, loss is low, and the degree of accuracy is high.
4. vehicle retrieval method provided in an embodiment of the present invention, in addition to：Open with N number of detection characteristic vector one by one
Corresponding multiple threads, i-th of detection characteristic vector is performed respectively and is carried out with all characteristic vectors in ith feature vector storehouse
Similarity Measure, obtain i-th of similarity score value sequence.By carrying out multithreading operation, speed on image processing apparatus
It hurry up, the degree of accuracy is high.
5. vehicle retrieval device provided in an embodiment of the present invention, driving is entered by the multiple target feature for extracting and integrating vehicle
Retrieval, the vehicle retrieval degree of accuracy is higher.
Brief description of the drawings
The features and advantages of the present invention can be more clearly understood by reference to accompanying drawing, accompanying drawing is schematically without that should manage
Solve to carry out any restrictions to the present invention, in the accompanying drawings：
Fig. 1 shows a flow chart specifically illustrated of vehicle retrieval method in the embodiment of the present invention 1；
Fig. 2 shows a flow chart specifically illustrated of vehicle retrieval method in the embodiment of the present invention 2；
Fig. 3 shows the visualization result of vehicle retrieval in the embodiment of the present invention 2；
Fig. 4 shows a flow chart specifically illustrated of vehicle retrieval method in the embodiment of the present invention 3；
Fig. 5 shows a structure chart specifically illustrated of vehicle retrieval device in the embodiment of the present invention 4；
Fig. 6 shows another structure chart specifically illustrated of vehicle retrieval device in the embodiment of the present invention 4；
Fig. 7 shows a structure chart specifically illustrated of image processing apparatus in the embodiment of the present invention 5；
Fig. 8 shows the application scenarios specifically illustrated in the embodiment of the present invention 5；
Fig. 9 shows the step of multiple target feature that vehicle is extracted in the embodiment of the present invention 5；
The step of Figure 10 shows vehicle extraction feature to be retrieved in the embodiment of the present invention 5 and fits into planting modes on sink characteristic.
Embodiment
To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below in conjunction with the embodiment of the present invention
In accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
Part of the embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, those skilled in the art are not having
There is the every other embodiment made and obtained under the premise of creative work, belong to the scope of protection of the invention.
Embodiment 1
Originally apply example and a kind of vehicle retrieval method is provided, in vehicle retrieval device.As shown in figure 1, the vehicle retrieval side
Method comprises the following steps：
Step S11, obtain the image of vehicle to be detected.
The image of vehicle to be detected in the present embodiment can be previously stored in vehicle retrieval device, or vehicle inspection
The vehicle figure that the vehicle image or vehicle retrieval device that rope device obtains from the external world in real time extract from an image
Picture.
Step S12, N number of detection characteristic vector of vehicle to be detected is extracted from the image of vehicle to be detected, wherein, N is big
In equal to 2, each detection characteristic vector corresponds to a part of vehicle to be detected.
By vehicle retrieval device in this example, N number of detection of vehicle to be detected is extracted from the image of vehicle to be detected
Characteristic vector, wherein, N is more than or equal to 2, each part for detecting characteristic vector and corresponding to vehicle to be detected；That is vehicle retrieval
Device extracts a detection characteristic vector from a vehicle part of vehicle to be detected.The vehicle part can be logo, car
Window, annual inspection label, in-car ornament or paper towel box etc., the present invention in above-mentioned vehicle part at least two, extraction it is to be checked
Characteristic vector is detected corresponding to measuring car.In the following description, the vehicle part of selected vehicle to be detected be logo and
Vehicle window.
Step S13, following steps are performed respectively for N number of detection characteristic vector, obtain N number of similarity score value sequence：
I-th of detection characteristic vector is subjected to Similarity Measure with all characteristic vectors in ith feature vector storehouse, obtained i-th
Similarity score value sequence, wherein, the ith feature vector storehouse be pre-establish with i-th detection characteristic vector it is corresponding
Identical vehicle part characteristic vector storehouse, i takes 1 to N natural number；Wherein, same vehicle in different characteristic vector storehouse
Being established between characteristic vector has corresponding relation.
In the present embodiment, the vehicle part identical vehicle with vehicle to be detected is pre-established with vehicle retrieval device
The characteristic vector storehouse of part.The corresponding characteristic vector storehouse of one vehicle part, this feature vector storehouse includes some vehicles
The characteristic vector of same vehicle part；For vehicle to be detected, the corresponding detection characteristic vector of a vehicle part.
In addition, having corresponding relation in different characteristic vector storehouse between the characteristic vector of same vehicle, i.e., vehicle retrieval fills
Put and can be sorted out the characteristic vector corresponding to same vehicle according to the corresponding relation between each characteristic vector.This implementation
Corresponding relation in example can be realized by mark or classification chart, only need to ensure that same vehicle has identical identification information i.e.
Can.
Wherein, the corresponding relation between characteristic vector and detection characteristic vector is as shown in table 1.
The characteristic vector of table 1 and detection characteristic vector
As shown in table 1, vehicle part 1 corresponds to vehicle 1 to vehicle P (P=20000), respectively with characteristic vector 11 to
Characteristic vector 1P, wherein, characteristic vector 11 to characteristic vector 1P forms the characteristic vector storehouse corresponding to vehicle part 1, i.e. this reality
The 1st characteristic vector storehouse in example is applied, includes P characteristic vector in the 1st characteristic vector storehouse；Vehicle part 1, which corresponds to, to be treated
Detect vehicle, there is detection characteristic vector 1, i.e., the 1st detection characteristic vector in the present embodiment.The like, vehicle part
I, corresponding to vehicle 1 to vehicle P, there is ith feature vector storehouse；Corresponding to vehicle to be detected, have i-th of detection feature to
Amount.
In addition, vehicle retrieval device can be returned each characteristic vector of same vehicle according to the identification information of vehicle
Class.For example, corresponding to vehicle 1, vehicle retrieval device can sort out characteristic vector 11, characteristic vector 21 ..., characteristic vector N1
To belong to the characteristic vector of vehicle 1.
In the present embodiment, calculated respectively for N number of detection characteristic vector, obtain N number of similarity score value sequence, i.e., by the
All characteristic vectors in i detection characteristic vector and ith feature vector storehouse carry out Similarity Measure, obtain i-th it is similar
Score value sequence is spent, wherein, i takes 1 to N natural number.For example, by the 1st detection characteristic vector respectively at the 1st characteristic vector
Characteristic vector 11 in storehouse carries out Similarity Measure to characteristic vector 1P, obtains the 1st similarity score value sequence.1st
Similarity score value sequence, including, similarity score value 11, similarity score value 12 ..., similarity score value 1P；Wherein, phase
It is special for detection like similarity ... of the degree score value 11 between detection characteristic vector 1 and characteristic vector 11, similarity score value 1j
The similarity between 1 and characteristic vector 1j of vector is levied, j is 1 to P natural number.In addition, similarity score value 1j is detection feature
Vectorial the distance between 1 and characteristic vector 1j, apart from smaller, corresponding similarity score value is bigger；The distance can pass through Europe
Formula distance, it can also be calculated by COS distance.
Step S14, choose respectively special corresponding to preceding M similarity score value maximum in N number of similarity score value sequence
Sign vector.
In the present embodiment, by being ranked up to the similarity score value in each similarity score value sequence, respectively from
The preceding M similarity score value of maximum is chosen in each similarity score value sequence, and preceding M similarity that should be maximum is obtained
Score value selected characteristic vector.
For example, include P similarity score value in the 1st similarity score value sequence, to the P similarity score value
Carry out descending sort, characteristic vector corresponding to 5 similarity score values before selection, i.e., spy corresponding to preceding 5 similarity score values
Sign vector and detection the distance between characteristic vector 1 minimum, so as to vehicle part corresponding to this feature vector and detection feature to
Vehicle part identical probability corresponding to amount 1 is larger.
Step S15, target vehicle, target vehicle are judged whether according to the corresponding relation between the characteristic vector of selection
Corresponding N number of characteristic vector is respectively present in the characteristic vector of selection；If in the presence of execution step S16；Otherwise, it is performed
He operates.
In the present embodiment, vehicle retrieval device carries out target vehicle according to the corresponding relation between the characteristic vector of selection
Judgement, i.e., by judging whether there is identical identification information between selected characteristic vector, so as to selected by judging
With the presence or absence of all vehicle parts for being used to detect that can range same vehicle in characteristic vector, you can judge whether deposit
In target vehicle.
In addition, other operations can be drawn in the absence of target vehicle or change vehicle part in the present embodiment,
Target vehicle judgement is carried out, i.e., other detection characteristic vectors of vehicle to be detected are extracted from the image of vehicle to be detected, again
Perform step S13 to step S15.
Step S16, it is vehicle to be detected to determine target vehicle.
In the present embodiment, judging that N number of characteristic vector corresponding to target vehicle is respectively present in the characteristic vector of selection
In, you can it is vehicle to be detected to determine the target vehicle.
Vehicle retrieval method provided in an embodiment of the present invention, first, extract N number of detection characteristic vector of vehicle to be detected；
Secondly, the similarity for calculating each detection characteristic vector and all characteristic vectors in characteristic vector storehouse of vehicle to be detected successively obtains
Sub-sequence；Finally, the similarity score value sequence calculated is ranked up, the maximum preceding M from similarity score value sequence
In characteristic vector corresponding to individual similarity score value, target vehicle is filtered out.The vehicle retrieval method is by extracting and integrating car
Multiple target feature carry out vehicle retrieval, the vehicle retrieval degree of accuracy is higher.
Embodiment 2
Originally apply example and a kind of vehicle retrieval method is provided, in vehicle retrieval device.As shown in Fig. 2 the vehicle retrieval side
Method comprises the following steps：
Step S20, obtain the image of vehicle to be detected.
In the present embodiment, after the image for obtaining vehicle to be detected, the image of vehicle to be detected is entered into row format conversion, will
The image of first form is converted to the image of the second form, is easy to follow-up calculating to handle, and improves the efficiency of vehicle retrieval.For example,
The image of the vehicle to be detected obtained is rgb format, then is converted into yuv format.
Remaining is identical with the step S11 of embodiment 1, will not be repeated here.
Step S21, N number of detection characteristic vector of vehicle to be detected is extracted from the image of vehicle to be detected, wherein, N is big
In equal to 2, each detection characteristic vector corresponds to a part of vehicle to be detected.
Inventor is had found in research process, and the depth learning technology of big data is applied in vehicle retrieval, can be significantly
Improve the degree of accuracy of retrieval.The attribute classification of target is represented by extracting the abstract characteristics of image of bottom, to find to scheme
As the distributed nature expression of data, complicated high performance model is constructed.
In the present embodiment, entered using convolutional neural networks (Convolutional Neural Network, referred to as CNN)
The feature extraction of row image, i.e., N number of detection feature of vehicle to be detected is extracted from the image of vehicle to be detected using CNN models
Vector, different vehicle parts can carry out detecting the extraction of characteristic vector using different CNN models.Wherein, CNN models bag
Include RCNN, Fast RCNN, Faster RCNN, YOLO, SSD, Resnet, Squeezenet, Alexnet, VGGnet.
Remaining is identical with the step S12 of embodiment 1, will not be repeated here.
Step S22, calculated respectively for N number of detection characteristic vector, obtain N number of similarity score value sequence：By i-th of inspection
Survey characteristic vector and carry out Similarity Measure with all characteristic vectors in ith feature vector storehouse, obtain i-th of similarity score
Value sequence, wherein, the ith feature vector storehouse is the identical car corresponding with i-th of detection characteristic vector pre-established
The characteristic vector storehouse of part, i take 1 to N natural number；Wherein, in different characteristic vector storehouse same vehicle characteristic vector it
Between establish have corresponding relation.
In the present embodiment, opened in image processing apparatus and detect characteristic vector multiple threads correspondingly with N number of, point
All characteristic vectors that Zhi Hang do not detect for i-th in characteristic vector and ith feature vector storehouse carry out Similarity Measure, obtain the
I similarity score value sequence；I.e. correspond to one detection characteristic vector open a thread, for calculate the detection feature to
Amount and the similarity between all characteristic vectors in corresponding characteristic vector storehouse, so as to obtain similarity score value sequence.It is logical
Cross and multithreading operation is carried out on image processing apparatus, speed is fast, and the degree of accuracy is high.
Remaining is identical with the step S13 of embodiment 1, will not be repeated here.
Step S23, choose respectively special corresponding to preceding M similarity score value maximum in N number of similarity score value sequence
Sign vector.It is identical with the step S14 of embodiment 1, it will not be repeated here.
Step S24, target vehicle, target vehicle are judged whether according to the corresponding relation between the characteristic vector of selection
Corresponding N number of characteristic vector is respectively present in the characteristic vector of selection；If in the presence of execution step S25；Otherwise, step is performed
Rapid S26.It is identical with the step S15 of embodiment 1, it will not be repeated here.
Step S26, detection different in N number of S21 from step of vehicle to be detected is extracted from the image of vehicle to be detected
Characteristic vector.
By being extracted again to vehicle part in vehicle to be detected, the vehicle part in the step is different from step S21's
Vehicle part.In extraction completion again and then secondary execution step S22.
Step S25, judge target vehicle with the presence or absence of multiple.If in the presence of execution step S27；Otherwise, it determines target carriage
It is vehicle to be detected.
In the present embodiment, vehicle retrieval device can be by judging whether there is identical between selected characteristic vector
Identification information, judge in selected characteristic vector with the presence or absence of all vehicles for being used to detect that can range same vehicle
Part, therefore, vehicle retrieval device is while judging whether there is target vehicle, additionally it is possible to judges of target vehicle
Number.
Vehicle retrieval device to selected characteristic vector by sorting out, if categorization results draw selected feature
Presence can range all vehicle parts for being used to detect of multiple vehicles in vector, you can judge multiple target carriages be present
.
Step S27, obtain all characteristic vectors of multiple target vehicles.
In the case where judging multiple target vehicles to be present, vehicle retrieval device has identical mark according to same vehicle
Know information, obtain all characteristic vectors of multiple target vehicles.For example, as shown in table 1, target vehicle is vehicle 1, vehicle 2, with
And vehicle 3, then obtain：Corresponding to the characteristic vector 11 of vehicle 1, characteristic vector 21 ..., characteristic vector N1；Corresponding to vehicle 2
Characteristic vector 12, characteristic vector 22 ..., characteristic vector N2；It is special corresponding to the characteristic vector 13 of vehicle 3, characteristic vector 23 ...
Levy vectorial N3.
Step S28, for same target vehicle, respectively according to the phase of its characteristic vector and the characteristic vector of vehicle to be detected
The similarity score value of target vehicle is calculated like degree score value, and result of calculation is ranked up.
In the present embodiment, to same target vehicle, pass through the characteristic vector to its all characteristic vector and vehicle to be detected
Similarity score value be weighted averaging, you can obtain the similarity score value of same target vehicle.
For example, the similarity score value point of all characteristic vectors and the characteristic vector of vehicle to be detected corresponding to vehicle 1
It is not：Similarity score value 11, similarity score value 21 ..., similarity score value N1；The similarity that can obtain vehicle 1 obtains
Score value 1 is：
It is respectively corresponding to all characteristic vectors of vehicle 2 and the similarity score value of the characteristic vector of vehicle to be detected：
Similarity score value 12, similarity score value 22 ..., similarity score value N2；It can obtain the similarity score value 2 of vehicle 2
For：
It is respectively corresponding to all characteristic vectors of vehicle 3 and the similarity score value of the characteristic vector of vehicle to be detected：
Similarity score value 13, similarity score value 23 ..., similarity score value N3；It can obtain the similarity score value 3 of vehicle 3
For：
Descending sort is carried out to the similarity score value of all target vehicles, i.e. the similarity score value of vehicle is bigger, then
Represent target vehicle closer to vehicle to be detected.
Step S29, the image of multiple target vehicles and its corresponding similarity score value are shown according to ranking results.
In the present embodiment, as shown in figure 3, corresponding to the similarity score value of the vehicle after sequence, corresponding display target car
Image and the target vehicle calculated similarity score value.
The present embodiment using CNN models extracted from the image of vehicle to be detected N number of detection feature of vehicle to be detected to
Amount, i.e., deep learning is carried out to the image of vehicle to be detected, should so as to extract N number of detection characteristic vector of vehicle to be detected
CNN models are a kind of deep neural network models for handling mass image data, and loss is low, and the degree of accuracy is high；In addition, pass through meter
The similarity score value of same target vehicle is calculated, and shows the score value and corresponding multiple target vehicle images, can be obtained
Visualization result, additionally it is possible to according to visualization result, filter out vehicle to be detected from some target vehicles, accuracy in detection compared with
It is high.
Embodiment 3
Originally apply example and a kind of vehicle retrieval method is provided, in vehicle retrieval device.As shown in figure 4, the vehicle retrieval side
Method comprises the following steps：
Step S311, obtain multiple images for including vehicle.
In the present embodiment, vehicle retrieval device in the electric police of input or bayonet socket image to obtaining multiple figures for including vehicle
Picture.
Step S312, vehicle image is extracted from multiple images for including vehicle using convolutional neural networks model.
Wherein, the CNN models for extracting vehicle image are：RCNN、Fast RCNN、Faster RCNN、YOLO、SSD
One of them.
Step S313, vehicle characteristics are extracted from the vehicle image of extraction using convolutional neural networks model, and to same
The different characteristic of vehicle establishes corresponding relation, obtains characteristic vector storehouse.
Wherein, for from the vehicle image of extraction extract vehicle characteristics CNN models for Resnet, Squeezenet,
One of Alexnet, VGGnet, and identical mark is assigned to the different characteristic of same vehicle, to establish characteristic vector storehouse.
Step S314, obtain the image of vehicle to be detected.It is identical with the step S20 of embodiment 2, it will not be repeated here.
Step S315, N number of detection characteristic vector of vehicle to be detected is extracted from the image of vehicle to be detected, wherein, N is big
In equal to 2, each detection characteristic vector corresponds to a part of vehicle to be detected.It is identical with the step S21 of embodiment 2, herein
Repeat no more.
Step S316, calculated respectively for N number of detection characteristic vector, obtain N number of similarity score value sequence：By i-th
Detect characteristic vector and carry out Similarity Measure with all characteristic vectors in ith feature vector storehouse, obtain i-th of similarity and obtain
Score value sequence, wherein, the ith feature vector storehouse is the identical corresponding with i-th of detection characteristic vector pre-established
The characteristic vector storehouse of vehicle part, i take 1 to N natural number；Wherein, in different characteristic vector storehouse same vehicle characteristic vector
Between establish have corresponding relation.It is identical with the step S22 of embodiment 2, it will not be repeated here.
Step S317, choose respectively special corresponding to preceding M similarity score value maximum in N number of similarity score value sequence
Sign vector.It is identical with the step S23 of embodiment 2, it will not be repeated here.
Step S318, target vehicle, target carriage are judged whether according to the corresponding relation between the characteristic vector of selection
N number of characteristic vector is respectively present in the characteristic vector of selection corresponding to；If in the presence of execution step S319；Otherwise, perform
Step S320.It is identical with the step S24 of embodiment 2, it will not be repeated here.
Step S319, judge target vehicle with the presence or absence of multiple.If in the presence of execution step S321；Otherwise, it determines target
Vehicle is vehicle to be detected.It is identical with the step S25 of embodiment 2, it will not be repeated here.
Step S321, obtain all characteristic vectors of multiple target vehicles.It is identical with the step S27 of embodiment 2, herein not
Repeat again.
Step S322, for same target vehicle, respectively according to its characteristic vector and the characteristic vector of vehicle to be detected
Similarity score value calculates the similarity score value of target vehicle, and result of calculation is ranked up.The step of with embodiment 2
S28 is identical, will not be repeated here.
Step S323, the image of multiple target vehicles and its corresponding similarity score value are shown according to ranking results.With reality
It is identical to apply the step S29 of example 2, will not be repeated here.
The present embodiment extracts vehicle image by using CNN models from the picture for include vehicle, and reuses CNN
Model extracts vehicle characteristics from the vehicle image of extraction, and establishes corresponding relation to the different characteristic of same vehicle, obtains spy
Levy vectorial storehouse；The picture for including vehicle carries out deep learning twice, establishes characteristic vector storehouse, is examined for follow-up vehicle
Rope, the loss of the vehicle retrieval method is low, and the degree of accuracy is high.
As a kind of embodiment of the present embodiment, the vehicle retrieval method includes three processes, when extraction and
Preserve storage vehicle characteristics：Electric police or bayonet socket image to input carry out vehicle detection using car test CNN models, to each car
Based on logo model extraction vehicle CNN features；The vehicle window of vehicle window model inspection each car, extract vehicle window CNN features；Establish feature
Vectorial storehouse, finally preserve the vehicle image feature of all storages；Second, matching retrieval vehicle and storage vehicle characteristics：For input
Vehicle to be retrieved image, vehicle detection and extraction logo and vehicle window CNN features；Read and parse warehouse entering trolley feature；
To be retrieved and storage vehicle characteristics are matched, calculate the similarity score value between them and the score that sorts.Third, to sequence
Score value is corresponded, visual m odeling technique result.Method flow is specially：
1：Using all vehicles of CNN1 model inspections storage image, such CNN1 models include：RCNN、Fast
RCNN、Faster RCNN、YOLO、SSD；
2：For each car, CNN2 logo model extraction vehicle image features are used；
2-1：The vehicle image of RGB types is transformed into YUV types；
2-2：CNN2 logo models are read, such CNN2 network models include：Resnet、Squeezenet、
Alexnet、VGGnet；
2-3：Extract the CNN2 models layer feature second from the bottom of vehicle image；
3：For each car, CNN3 vehicle window model extraction vehicle glazing features are used；
3-1：The vehicle image of RGB types is transformed into YUV types；
3-2：CNN3 vehicle window models are read, such CNN3 network models include：Resnet、Squeezenet、
Alexnet、VGGnet；
3-3：Extract the CNN3 models layer feature second from the bottom of vehicle window image；
4：Characteristic vector storehouse is established, preserves all storage vehicle characteristics；
5：Extract the CNN2 logo features of vehicle to be retrieved；
5-1：The image type of vehicle to be retrieved is transformed into YUV types；
5-2：CNN2 models are read, such CNN2 network models include：Resnet、Squeezenet、Alexnet、
VGGnet；
5-3：The CNN2 models layer feature second from the bottom of extraction retrieval vehicle image；
5-4：Obtain the detection characteristic vector corresponding to logo；
6：CNN3 vehicle window model extraction vehicle glazing features are used to retrieval vehicle, obtain the detection feature corresponding to vehicle window
Vector；
7：Read and parse warehouse entering trolley feature；
8：The calculating of similarity score value is carried out to vehicle to be retrieved and storage vehicle, sort similarity score value；
9：Score value and vehicle image are corresponded, visual m odeling technique result.
Embodiment 4
The present embodiment provides a kind of vehicle retrieval device, for performing the vehicle retrieval side in embodiment 1 to embodiment 3
Method, as shown in figure 5, the vehicle retrieval device includes：
First acquisition unit 41, for obtaining the image of vehicle to be detected；
Extraction unit 42, N number of detection for extracting the vehicle to be detected from the image of the vehicle to be detected are special
Sign vector, wherein, N is more than or equal to 2, each part for detecting characteristic vector and corresponding to the vehicle to be detected；
First computing unit 43, for performing following steps respectively for N number of detection characteristic vector, obtain N number of phase
Like degree score value sequence：I-th of detection characteristic vector is subjected to similarity with all characteristic vectors in ith feature vector storehouse
Calculate, obtain i-th of similarity score value sequence, wherein, the ith feature vector storehouse is to pre-establish with described i-th
The characteristic vector storehouse of identical vehicle part, i take 1 to N natural number corresponding to individual detection characteristic vector；Wherein, different characteristic
Being established in vectorial storehouse between the characteristic vector of same vehicle has corresponding relation；
Unit 44 is chosen, for choosing preceding M similarity score maximum in N number of similarity score value sequence respectively
Characteristic vector corresponding to value；
Judging unit 45, target vehicle is judged whether for the corresponding relation between the characteristic vector according to selection,
N number of characteristic vector is respectively present in the characteristic vector of the selection corresponding to the target vehicle；
Determining unit 46, for determining that the target vehicle is the vehicle to be detected.
Vehicle retrieval device in the present embodiment, vehicle retrieval is carried out by the multiple target feature for extracting and integrating vehicle,
The vehicle retrieval degree of accuracy is higher.
As a kind of optional embodiment of the present embodiment, as shown in fig. 6, the vehicle retrieval device also includes：
Second acquisition unit 47, for obtaining all characteristic vectors of multiple target vehicles；
Second computing unit 48, for for the same target vehicle, respectively according to its characteristic vector with it is described to be checked
The similarity score value of the characteristic vector of measuring car calculates the similarity score value of the target vehicle, and to the result of calculation
It is ranked up；
Display unit 49, for showing the image of multiple target vehicles according to the ranking results and its corresponding to similar
Spend score value.
Embodiment 5
Originally apply example and a kind of image processing apparatus is provided, for performing the vehicle retrieval method in embodiment 1 to embodiment 3.
Fig. 7 is the hardware architecture diagram of image processing apparatus provided in an embodiment of the present invention, as shown in fig. 7, the device
Including one or more processors 51 and memory 52, in Fig. 7 by taking a processor 51 as an example.
The image processing apparatus can also include：Image display (not shown), for showing the result of vehicle retrieval.Place
Reason device 51, memory 52 can be connected with image display by bus or other modes, to be connected as by bus in Fig. 7
Example.
Processor 51 can be central processing unit (Central Processing Unit, CPU).Processor 61 can be with
For other general processors, digital signal processor (Digital Signal Processor, DSP), application specific integrated circuit
(Application Specific Integrated Circuit, ASIC), field programmable gate array (Field-
Programmable Gate Array, FPGA) either other PLDs, discrete gate or transistor logic,
The chips such as discrete hardware components, or the combination of above-mentioned all kinds of chips.General processor can be microprocessor or the processing
Device can also be any conventional processor etc..
Memory 52 is used as a kind of non-transient computer readable storage medium storing program for executing, available for storing non-transient software program, non-
Transient computer executable program and module, programmed instruction/mould as corresponding to the vehicle retrieval method in the embodiment of the present invention
Block.Processor 51 is stored in non-transient software program, instruction and module in memory 52 by operation, so as to perform service
The various function application of device and data processing, that is, realize in above-described embodiment, vehicle retrieval method.
Memory 52 can include storing program area and storage data field, wherein, storing program area can storage program area,
Application program required at least one function；Storage data field can store uses created number according to vehicle retrieval device
According to etc..In addition, memory 52 can include high-speed random access memory, non-transient memory can also be included, for example, at least
One disk memory, flush memory device or other non-transient solid-state memories.In certain embodiments, memory 52 can
Choosing includes by network connection to the colour of skin to strengthen relative to the remotely located memory of processor 51, these remote memories
Processing unit.The example of above-mentioned network include but is not limited to internet, intranet, LAN, mobile radio communication and its
Combination.
One or more of modules are stored in the memory 52, when by one or more of processors 51
During execution, the vehicle retrieval method any one of embodiment 1 to embodiment 3 is performed.
The said goods can perform the method that the embodiment of the present invention is provided, and possesses the corresponding functional module of execution method and has
Beneficial effect.The ins and outs of detailed description, the correlation that for details, reference can be made in embodiment 1 to embodiment 3 are not retouched in the present embodiment
State.
As specific application example, the present embodiment also provides the application example of following scene,
As shown in figure 8, characteristic extraction part includes：Input storage image file, vehicle detection MODEL C NN1, logo feature
Extraction model CNN2, vehicle window feature detection and Feature Selection Model CNN3 and its extraction feature execution file.Characteristic matching portion
Dividing includes：Image file to be retrieved, vehicle detection MODEL C NN1, logo Feature Selection Model CNN2, vehicle window feature detection and
Feature Selection Model CNN3 and the executable file for matching characteristic.Specifically：By the image path generation .set texts of storage
Part, image file, CNN1 models, CNN2 models, CNN3 models and the execution file of feature extraction are placed on a file mesh
Under record, change the file in .bat files and catalogue and graphics process sole duty option be present, can be given birth to after performing feature extraction file
Into .fea tag file, storage tag file copy is put into the file directory of vehicle retrieval, storage CNN1 models, CNN2
Model, CNN3 models and characteristic matching file, perform .bat, you can obtain visual vehicle retrieval result and also have one
Result document, record the retrieval result of each vehicle.
As shown in Figure 9 and Figure 10, including two processes, extract the multiple target feature of vehicle and matching vehicle characteristics and show
As a result.Vehicle retrieval system realizes that detailed process is as follows.
Extract the multiple target feature of vehicle and preserve, as shown in figure 9, concretely comprising the following steps：
1：2w storage images are read, by coordinates measurement lib.set files, are put into the execution file directory for extracting feature
In, each figure is read, uses SSD vehicle detection model inspections all vehicles therein；
2：For each car, Resnet50 logo model extraction vehicle characteristics are used；
2-1：The vehicle image of RGB types is transformed into YUV types；
2-2：Read resnet50 models, including its deploy and average file；
2-3：Resnet50 fc7 layer features are extracted, obtain the detection characteristic vector corresponding to logo；
3：For each car, vehicle window feature is detected and extracted using Resnet18 vehicle windows detection model；
3-1：The vehicle image of RGB types is transformed into YUV types；
3-2：Read resnet18 models, including its deploy and average file；
3-3：Resnet18 fc7 layer features are extracted, obtain the detection characteristic vector corresponding to vehicle window；
4：Preserve all storage vehicles and be characterized as lib.fea files；
Vehicle extraction feature to be retrieved simultaneously fits into planting modes on sink characteristic, visualization result, as shown in Figure 10, concretely comprises the following steps：
5：By under the folder content of lib.fea file copies to characteristic matching；
6：Read vehicle pictures to be retrieved, repeat the above steps 2 and 3 process, obtain retrieve image logo and car
Window feature, integrate and obtain the current query.fea for retrieving vehicle；
7：Read and parse warehouse entering trolley feature lib.fea files；
8：Similarity score value is carried out to retrieval vehicle characteristics query.fea and all vehicle characteristics lib.fea of storage
Calculating, calculated using COS distance, sort score value；
9：Score value and vehicle image are corresponded, obtain visual m odeling technique result.
The advantages of vehicle retrieval system combined based on multiple target CNN features that the present embodiment uses, is as follows：
1. using deep approach of learning, the multiple target feature extracted and integrate vehicle carries out vehicle rope, and the accuracy rate of retrieval is high.
2. by carrying out multithreading operation on image processing apparatus, the whole system speed of service is fast.
3. the CNN models of the detection vehicle used, loss is low, and accuracy rate is high.
4. the CNN models of the extraction feature used can be Resnet, Squeezenet, Alexnet, VGGnet any
Kind.
5. score value is corresponding with vehicle figure, obtain visual result.
Embodiment 6
The embodiment of the present invention additionally provides a kind of non-transient computer storage medium, and the computer-readable storage medium is stored with
Computer executable instructions, the computer executable instructions can perform the vehicle inspection any one of embodiment 1 to embodiment 3
Suo Fangfa.Wherein, the storage medium can be magnetic disc, CD, read-only memory (Read-Only Memory, ROM), with
Machine storage memory (Random Access Memory, RAM), flash memory (Flash Memory), hard disk (Hard
Disk Drive, abbreviation：) or solid state hard disc (Solid-State Drive, SSD) etc. HDD；The storage medium can also wrap
Include the combination of the memory of mentioned kind.
It is to lead to it will be understood by those skilled in the art that realizing all or part of flow in above-described embodiment method
Computer program is crossed to instruct the hardware of correlation to complete, described program can be stored in a computer read/write memory medium
In, the program is upon execution, it may include such as the flow of the embodiment of above-mentioned each method.Wherein, described storage medium can be magnetic
Dish, CD, read-only memory (ROM) or random access memory (RAM) etc..
Although being described in conjunction with the accompanying embodiments of the invention, those skilled in the art can not depart from the present invention
Spirit and scope in the case of various modification can be adapted and modification, such modifications and variations are each fallen within by appended claims institute
Within the scope of restriction.
Claims (10)
- A kind of 1. vehicle retrieval method, it is characterised in that comprise the following steps：Obtain the image of vehicle to be detected；N number of detection characteristic vector of the vehicle to be detected is extracted from the image of the vehicle to be detected, wherein, N is more than etc. In 2, each detection characteristic vector corresponds to a part of the vehicle to be detected；Following steps are performed respectively for N number of detection characteristic vector, obtain N number of similarity score value sequence：By i-th of inspection Survey characteristic vector and carry out Similarity Measure with all characteristic vectors in ith feature vector storehouse, obtain i-th of similarity score Value sequence, wherein, the ith feature vector storehouse be pre-establish with described i-th detection characteristic vector corresponding to it is identical Vehicle part characteristic vector storehouse, i takes 1 to N natural number；Wherein, in different characteristic vector storehouse same vehicle feature to Being established between amount has corresponding relation；Characteristic vector corresponding to preceding M similarity score value maximum in N number of similarity score value sequence is chosen respectively；Target vehicle, N corresponding to the target vehicle are judged whether according to the corresponding relation between the characteristic vector of selection Individual characteristic vector is respectively present in the characteristic vector of the selection；If it is present determine that the target vehicle is the vehicle to be detected.
- 2. vehicle retrieval method according to claim 1, it is characterised in that if target vehicle has multiple, the car Search method also includes：Obtain all characteristic vectors of multiple target vehicles；For the same target vehicle, respectively according to its characteristic vector and the similarity of the characteristic vector of the vehicle to be detected Score value calculates the similarity score value of the target vehicle, and the result of calculation is ranked up；The image of multiple target vehicles and its corresponding similarity score value are shown according to the ranking results.
- 3. vehicle retrieval method according to claim 1, it is characterised in that treated using convolutional neural networks model from described Detect N number of detection characteristic vector that the vehicle to be detected is extracted in the image of vehicle.
- 4. vehicle retrieval method according to claim 1, it is characterised in that before the image of vehicle to be detected is obtained, Also include：Obtain multiple images for including vehicle；Using convolutional neural networks model vehicle image is extracted from the multiple image for including vehicle；Vehicle characteristics are extracted from the vehicle image of extraction using convolutional neural networks model, and to the different characteristic of same vehicle Corresponding relation is established, obtains characteristic vector storehouse.
- 5. vehicle retrieval method according to claim 1, it is characterised in that the vehicle retrieval method, in addition to：Open With N number of detection characteristic vector multiple threads correspondingly, i-th of detection characteristic vector and ith feature are performed respectively All characteristic vectors in vectorial storehouse carry out Similarity Measure, obtain i-th of similarity score value sequence.
- 6. vehicle retrieval method according to claim 1, it is characterised in that the step of the image for obtaining vehicle to be detected After rapid, in addition to：The image of the vehicle to be detected is converted into the second form by the first form.
- 7. vehicle retrieval method according to claim 6, it is characterised in that the first form rgb format, described second Form is yuv format.
- A kind of 8. vehicle retrieval device, it is characterised in that including：First acquisition unit, for obtaining the image of vehicle to be detected；Extraction unit, for extracting N number of detection characteristic vector of the vehicle to be detected from the image of the vehicle to be detected, Wherein, N is more than or equal to 2, each part for detecting characteristic vector and corresponding to the vehicle to be detected；First computing unit, for performing following steps respectively for N number of detection characteristic vector, obtain N number of similarity and obtain Score value sequence：I-th of detection characteristic vector is subjected to Similarity Measure with all characteristic vectors in ith feature vector storehouse, Obtain i-th of similarity score value sequence, wherein, the ith feature vector storehouse be pre-establish with described i-th detection The characteristic vector storehouse of identical vehicle part corresponding to characteristic vector, i take 1 to N natural number；Wherein, different characteristic vector storehouse Being established between the characteristic vector of middle same vehicle has corresponding relation；Unit is chosen, it is corresponding for choosing preceding M similarity score value maximum in N number of similarity score value sequence respectively Characteristic vector；Judging unit, target vehicle, the mesh are judged whether for the corresponding relation between the characteristic vector according to selection N number of characteristic vector corresponding to mark vehicle is respectively present in the characteristic vector of the selection；Determining unit, for determining that the target vehicle is the vehicle to be detected.
- 9. a kind of image processing apparatus, it is characterised in that including at least one processor；And with least one processor The memory of communication connection；Wherein, have can be by the instruction of one computing device, the instruction quilt for the memory storage At least one computing device, so that the car any one of at least one computing device claim 1 to 7 Search method.
- 10. a kind of non-transient computer readable storage medium storing program for executing, it is characterised in that the non-transient computer readable storage medium storing program for executing is deposited Computer instruction is stored up, the computer instruction is used to make the vehicle retrieval any one of computer perform claim requirement 1 to 7 Method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710940974.1A CN107808126A (en) | 2017-10-11 | 2017-10-11 | Vehicle retrieval method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710940974.1A CN107808126A (en) | 2017-10-11 | 2017-10-11 | Vehicle retrieval method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107808126A true CN107808126A (en) | 2018-03-16 |
Family
ID=61592641
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710940974.1A Pending CN107808126A (en) | 2017-10-11 | 2017-10-11 | Vehicle retrieval method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107808126A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108491797A (en) * | 2018-03-22 | 2018-09-04 | 河北省科学院应用数学研究所 | A kind of vehicle image precise search method based on big data |
| CN109902194A (en) * | 2019-01-18 | 2019-06-18 | 平安科技（深圳）有限公司 | Image duplicate checking method neural network based and relevant device |
| CN110110627A (en) * | 2019-04-24 | 2019-08-09 | 江南大学 | A method of the real-time target detection towards computing resource limitation Platform deployment |
| CN110222604A (en) * | 2019-05-23 | 2019-09-10 | 复钧智能科技(苏州)有限公司 | Target identification method and device based on shared convolutional neural networks |
| CN110245577A (en) * | 2019-05-23 | 2019-09-17 | 复钧智能科技(苏州)有限公司 | Target vehicle recognition methods, device and Vehicular real time monitoring system |
| CN110555125A (en) * | 2018-05-14 | 2019-12-10 | 桂林远望智能通信科技有限公司 | Vehicle retrieval method based on local features |
| CN110555777A (en) * | 2018-06-01 | 2019-12-10 | 易保网络技术(上海)有限公司 | Method for simultaneously providing at least two car insurance quotations |
| WO2020047907A1 (en) * | 2018-09-07 | 2020-03-12 | 佳都新太科技股份有限公司 | License plate failed-detection estimation method and processing terminal |
| CN111540203A (en) * | 2020-04-30 | 2020-08-14 | 东华大学 | Method for adjusting green light passing time based on fast-RCNN |
| CN112419751A (en) * | 2020-10-10 | 2021-02-26 | 同济大学 | Signalized intersection lane queuing length estimation method based on single-section electric alarm data |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106295541A (en) * | 2016-08-03 | 2017-01-04 | 乐视控股（北京）有限公司 | Vehicle type recognition method and system |
| CN106649490A (en) * | 2016-10-08 | 2017-05-10 | 中国人民解放军理工大学 | Depth feature-based image retrieval method and apparatus |
| CN107066586A (en) * | 2017-04-17 | 2017-08-18 | 清华大学深圳研究生院 | Footwear model index management method and system |
-
2017
- 2017-10-11 CN CN201710940974.1A patent/CN107808126A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106295541A (en) * | 2016-08-03 | 2017-01-04 | 乐视控股（北京）有限公司 | Vehicle type recognition method and system |
| CN106649490A (en) * | 2016-10-08 | 2017-05-10 | 中国人民解放军理工大学 | Depth feature-based image retrieval method and apparatus |
| CN107066586A (en) * | 2017-04-17 | 2017-08-18 | 清华大学深圳研究生院 | Footwear model index management method and system |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108491797A (en) * | 2018-03-22 | 2018-09-04 | 河北省科学院应用数学研究所 | A kind of vehicle image precise search method based on big data |
| CN110555125A (en) * | 2018-05-14 | 2019-12-10 | 桂林远望智能通信科技有限公司 | Vehicle retrieval method based on local features |
| CN110555777A (en) * | 2018-06-01 | 2019-12-10 | 易保网络技术(上海)有限公司 | Method for simultaneously providing at least two car insurance quotations |
| WO2020047907A1 (en) * | 2018-09-07 | 2020-03-12 | 佳都新太科技股份有限公司 | License plate failed-detection estimation method and processing terminal |
| CN109902194A (en) * | 2019-01-18 | 2019-06-18 | 平安科技（深圳）有限公司 | Image duplicate checking method neural network based and relevant device |
| CN110110627A (en) * | 2019-04-24 | 2019-08-09 | 江南大学 | A method of the real-time target detection towards computing resource limitation Platform deployment |
| CN110222604A (en) * | 2019-05-23 | 2019-09-10 | 复钧智能科技(苏州)有限公司 | Target identification method and device based on shared convolutional neural networks |
| CN110245577A (en) * | 2019-05-23 | 2019-09-17 | 复钧智能科技(苏州)有限公司 | Target vehicle recognition methods, device and Vehicular real time monitoring system |
| CN110222604B (en) * | 2019-05-23 | 2023-07-28 | 复钧智能科技(苏州)有限公司 | Target identification method and device based on shared convolutional neural network |
| CN111540203A (en) * | 2020-04-30 | 2020-08-14 | 东华大学 | Method for adjusting green light passing time based on fast-RCNN |
| CN111540203B (en) * | 2020-04-30 | 2021-09-17 | 东华大学 | Method for adjusting green light passing time based on fast-RCNN |
| CN112419751A (en) * | 2020-10-10 | 2021-02-26 | 同济大学 | Signalized intersection lane queuing length estimation method based on single-section electric alarm data |
| CN112419751B (en) * | 2020-10-10 | 2021-10-08 | 同济大学 | Signalized intersection lane queuing length estimation method based on single-section electric alarm data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107808126A (en) | Vehicle retrieval method and device | |
| Chen et al. | Deep integration: A multi-label architecture for road scene recognition | |
| CN109829467A (en) | Image labeling method, electronic device and non-transient computer-readable storage medium | |
| WO2017190574A1 (en) | Fast pedestrian detection method based on aggregation channel features | |
| CN105144239B (en) | Image processing apparatus, image processing method | |
| CN105574550A (en) | Vehicle identification method and device | |
| CN112801146B (en) | Target detection method and system | |
| CN105512683A (en) | Target positioning method and device based on convolution neural network | |
| CN107273832B (en) | License plate recognition method and system based on integral channel characteristics and convolutional neural network | |
| CN102385592B (en) | Image concept detection method and device | |
| CN110533654A (en) | The method for detecting abnormality and device of components | |
| CN110263712A (en) | A kind of coarse-fine pedestrian detection method based on region candidate | |
| CN104615986A (en) | Method for utilizing multiple detectors to conduct pedestrian detection on video images of scene change | |
| CN109871789A (en) | Vehicle checking method under a kind of complex environment based on lightweight neural network | |
| CN110599453A (en) | Panel defect detection method and device based on image fusion and equipment terminal | |
| CN110020669A (en) | A kind of license plate classification method, system, terminal device and computer program | |
| CN105809181B (en) | Method and apparatus for Logo detection | |
| CN106845458A (en) | A kind of rapid transit label detection method of the learning machine that transfinited based on core | |
| CN109857878B (en) | Article labeling method and device, electronic equipment and storage medium | |
| CN110909656B (en) | Pedestrian detection method and system integrating radar and camera | |
| CN108596244A (en) | A kind of high spectrum image label noise detecting method based on spectrum angle density peaks | |
| CN107315984A (en) | A kind of method and device of pedestrian retrieval | |
| CN115115825A (en) | Method and device for detecting object in image, computer equipment and storage medium | |
| CN112270671B (en) | Image detection method, device, electronic equipment and storage medium | |
| CN116452950A (en) | Multi-target garbage detection method based on improved YOLOv5 model |
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 |
Application publication date: 20180316 |
|
| RJ01 | Rejection of invention patent application after publication |