CN105844286A - Newly added vehicle logo identification method and apparatus - Google Patents
Newly added vehicle logo identification method and apparatus Download PDFInfo
- Publication number
- CN105844286A CN105844286A CN201610137614.3A CN201610137614A CN105844286A CN 105844286 A CN105844286 A CN 105844286A CN 201610137614 A CN201610137614 A CN 201610137614A CN 105844286 A CN105844286 A CN 105844286A
- Authority
- CN
- China
- Prior art keywords
- car
- newly
- sample
- increased
- car mark
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/09—Recognition of logos
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a newly added vehicle logo identification method and apparatus. The method comprises the steps of extracting vehicle logo features from an image containing a target vehicle, and comparing the vehicle logo features with a vehicle logo model to determine whether the identification confidence coefficient of a vehicle logo to be identified is greater than a preset threshold; if not, making a feature comparison between the vehicle logo to be identified and vehicle logo samples in each newly added vehicle logo storage folder, and updating the sample size in the newly added vehicle log storage folder according to the comparison result; when the stored sample size in the newly added storage folder reaches a set size, carrying out newly added sample cluster detection to obtain a newly added sample set; and storing the newly added sample set into a sample set of the existing vehicle logo model, and updating the vehicle logo model. The invention saves the manpower consumption and improves the vehicle logo identification technology to a large extent. The invention solves the problem that newly added vehicle logos result in a lower vehicle logo identification rate; and also solves the problems of being time-consuming and low in efficiency due to that currently a newly added vehicle brand mainly relies on manual confirmation.
Description
Technical field
The invention belongs to technical field of intelligent traffic, be specifically related to a kind of without supervising adaptive newly-increased car target identification
Method and apparatus.
Background technology
Vehicle-logo recognition technology is an important research content of intelligent transportation research field, and the research of this technology is for traffic
The contents such as control, highway are deployed to ensure effective monitoring and control of illegal activities, highway toll have profound significance.Conventional treatment increases the method for car mark sample set newly, whole
Process includes: picture collection, photo finishing, classification confirmation, classification are added and model replacement, are required to manually participate in, the longest
And efficiency is low.And, along with developing rapidly of automobile industry, domestic car, the new style cumulative year after year of imported car, the most efficiently
The vehicle brand that reply newly increases, it is ensured that the discrimination of car absolute altitude accuracy, it has also become vehicle-logo recognition technical research and later stage car
Identify a new challenge of other system development.
Summary of the invention
For the deficiencies in the prior art, the present invention proposes a kind of newly-increased car target recognition methods and device, to improve newly
Increase car target recognition accuracy and recognition efficiency, promote car target Weigh sensor level.
For achieving the above object, the newly-increased car target recognition methods that the present invention provides, including:
The picture comprising vehicle target is carried out car mark feature extraction, and compares with car mark model, it is judged that be to be identified
Whether car target recognition confidence is more than the threshold value preset；
If it is not, the car standard specimen in then being pressed from both sides with each newly-increased car mark store files by car mark to be identified originally carries out aspect ratio pair,
The sample size in newly-increased car mark store files folder is updated according to comparison result；
When storage sample size in newly-increased storage folder reaches the quantity set, carry out newly-increased sample clustering detection,
Obtain the sample set increased newly；
The sample set that this is newly-increased is stored in the sample set of existing car mark model, more new car mark model.
In some embodiments of the invention, described according to the sample in comparison result renewal newly-increased car mark store files folder
Amount, step include:
It is pre-created multiple for depositing newly-increased car standard specimen newly-increased car mark storage folder originally, is put by car mark feature
The comparative result of reliability and threshold value, originally puts into newly-increased car standard specimen in some file, or again increases a car mark newly and deposit
Storage file.
In some embodiments of the invention, the described comparative result by car mark feature confidence level Yu threshold value, will be newly-increased
Car standard specimen is originally put in some file, or again increases a car mark storage folder newly, step include:
Traversal searches car standard specimen this storage folder identical with newly-increased car mark to be increased output confidence level；
If this document folder exists, then will comprise this car target vehicle pictures and be stored in this sample storage folder as sample
In, the sample total that file is corresponding adds " 1 "；
If this sample file folder does not exists, the most newly-built newly-increased car mark storage folder named with this confidence level, will
Comprising this car target vehicle pictures to be stored in this sample storage folder as sample, the sample total that file is corresponding adds " 1 ".
In some embodiments of the invention, described in carry out the detection of newly-increased sample clustering, obtain the sample set increased newly, step
Suddenly include:
1) sheet of marking the sample car in newly-increased car standard specimen basis on a map extracts car mark feature successively, it is thus achieved that K car mark feature；
2) i-th car mark feature is randomly selectedAs cluster sample class, calculate remaining j=
K-i car mark feature PjWith feature PiBetween feature fitting degree；
3) willGroup fitting result sorts according to order from small to large respectively, and statistics often group degree of fitting is less than ρ0's
Number of samples, wherein ρ0Empirical value for degree of fitting；
4) takeGroup degree of fitting is less than ρ0The maximum number of sample number, as effective sample number M in this sample set；
5) filter outIn group ranking results, often in group sample before in M sample the car standard specimen of repetition this as having
The car mark model sample of effect.
In some embodiments of the invention, the computing formula of described feature fitting degree is as follows:
dij=| | Pj-Pi| |, (i, j ∈ (1,2 ..., K), and i ≠ j).
Invention also proposed a kind of newly-increased car target identification device, including:
Extraction module, for the picture comprising vehicle target carries out car mark feature extraction, and compares with car mark model
Right, it is judged that whether car target recognition confidence to be identified is more than the threshold value preset；
Sample size more new module, if for car target recognition confidence to be identified less than or equal to the threshold value preset, then will treat
Identify that car mark originally carries out aspect ratio pair with the car standard specimen in each newly-increased car mark store files folder, update according to comparison result newly-increased
Sample size in car mark store files folder；
Sample set more new module, when the storage sample size in newly-increased storage folder reaches the quantity set, enters
Row newly-increased sample clustering detection, obtains the sample set increased newly；
Car mark model modification module, for being stored in the sample set of existing car mark model, more new car by the sample set that this is newly-increased
Mark model.
In some embodiments of the invention, described sample size more new module includes:
Newly-increased car mark memory module, multiple for depositing the newly-increased car mark storage originally of newly-increased car standard specimen for being pre-created
File；
Sample size distribution module, is used for the comparative result by car mark feature confidence level Yu threshold value, by newly-increased car standard specimen originally
Put in some file, or again increase a car mark storage folder newly.
In some embodiments of the invention, described sample size distribution module is searched and newly-increased car to be increased for traveling through
The car mark storage folder that mark output confidence level is identical；If this document folder exists, then will comprise this car target vehicle pictures conduct
Sample is stored in this sample storage folder, and the sample total that file is corresponding adds " 1 "；If this sample file folder does not exists, then
A newly-built newly-increased car mark storage folder named with this confidence level, will comprise this car target vehicle pictures and be stored in as sample
In this sample storage folder, the sample total that file is corresponding adds " 1 ".
In some embodiments of the invention, described sample set more new module, it is used for
1) sheet of marking the sample car in newly-increased car standard specimen basis on a map extracts car mark feature successively, it is thus achieved that K car mark feature；
2) i-th car mark feature is randomly selectedAs cluster sample class, calculate remaining j=
K-i car mark feature PjWith PiFeature fitting degree between feature；
3) willGroup fitting result sorts according to order from small to large respectively, and statistics often group degree of fitting is less than ρ0's
Number of samples, wherein ρ0Empirical value for degree of fitting；
4) takeGroup degree of fitting is less than ρ0The maximum number of sample number, as effective sample number M in this sample set；
5) filter outIn group ranking results, often in group sample before in M sample the car standard specimen of repetition this as having
The car mark model sample of effect.
In some embodiments of the invention, the computing formula of described feature fitting degree is as follows:
dij=| | Pj-Pi| |, (i, j ∈ (1,2 ..., K), and i ≠ j).
Newly-increased car target recognition methods and device that the present invention provides combine car mark confidence level so that newly-increased car mark early stage
Collect housekeeping operation, be completely dependent on algorithm and realize, largely save manpower and expend, improve vehicle-logo recognition technology.This
Bright solve the problem that newly-increased car mark drags down vehicle-logo recognition rate；Also solve the most newly-increased vehicle brand too much to rely on manually simultaneously
Confirm, time-consuming long, the inefficient problem brought.
Accompanying drawing explanation
For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with specific embodiment, and reference
Accompanying drawing, the present invention is described in more detail, wherein:
Fig. 1 is the schematic flow sheet of the newly-increased car target recognition methods of one embodiment of the invention；
Fig. 2 is the schematic flow sheet of the mark recognition methods of the newly-increased car of another embodiment of the present invention；
Fig. 3 is the high-level schematic functional block diagram of the newly-increased car target identification device of one embodiment of the invention.
Detailed description of the invention
Below in conjunction with the drawings and specific embodiments, the present invention is described in further detail.
With reference to Fig. 1, it is shown that the schematic flow sheet of the newly-increased car target recognition methods of one embodiment of the invention, specifically may be used
To comprise the following steps:
Step 101, carries out car mark feature extraction, and compares with car mark model, sentence the picture comprising vehicle target
Whether disconnected car target recognition confidence to be identified is more than the threshold value preset.
In this step, by high-definition camera Real-time Collection video flowing, obtain the current frame image comprising vehicle target,
Use multi-scale sliding window mouth detection method, in same window, detect the feature of vehicle region, and enter with sample motor vehicle model
Row coupling, filters out the motor vehicles target in picture, then this motor vehicles target is carried out car mark feature extraction, such that it is able to accurately
Obtain the car mark feature in picture.
Preferably for the vehicle region in picture, it is marked on the position of vehicle region in conjunction with car, car test region can be chosen
The region of lower about 2/3 carries out the extraction of feature.As an alternative embodiment of the invention, it is also possible to choose under car test region 3/
The region of about 4 carries out the extraction of feature.Another embodiment as the present invention, it is also possible to choose 1/3 left side under car test region
Right region carries out the extraction of feature.Car mark feature extraction is mainly based upon the detection of HOG, extract the texture in detection region,
Linearly, gray scale, the feature such as direction extraction, and combine PCA dimensionality reduction, it is achieved to the conversion from higher-dimension to low-dimensional of the car mark characteristic information,
Therefore, it is possible on the premise of ensureing to extract car mark principal character, improve the efficiency of algorithm.Aspect ratio to during, will extract
Car mark feature be fitted with the feature in existing car mark model, digital simulation similarity (i.e. confidence level).
Step 102, if it is not, the car standard specimen in then being pressed from both sides with each newly-increased car mark store files by car mark to be identified originally carries out spy
Levy comparison, update the sample size in newly-increased car mark store files folder according to comparison result.
Certainly, if car target recognition confidence to be identified is more than the threshold value preset, then according to car target recognition result in advance
The car mark model deposited is searched the car mark corresponding with car mark to be identified.Preferably, existing car mark model is prestored, according to car
Identify other accuracy requirement and the empirical value of vehicle-logo recognition result can be set, by car mark to be identified and pre-bicycle parking target aspect ratio
Right, it is judged that whether described car mark to be identified is newly-increased car mark.If described car target recognition confidence to be identified is less than or equal to presetting
Threshold value, then judge that described car to be identified is designated as newly-increased car mark.Accuracy rate requirement according to vehicle-logo recognition result, can be arranged not
Same empirical value, empirical value is the highest, then the accuracy rate of judged result is the highest, otherwise, the lowest.
It is pre-created multiple newly-increased car mark storage folder, for depositing newly-increased car standard specimen originally, according to vehicle-logo recognition
Precision needs, and arranges the empirical value of the other recognition result of car, by the comparative result of car mark feature confidence level Yu threshold value, and will be newly-increased
Car standard specimen is originally put in some file, or again increases a car mark storage folder newly.Specifically, if car mark to be identified
Recognition confidence less than or equal to preset threshold value, category car is designated as doubtful newly-increased car mark.In newly-increased car mark memory element,
Traversal searches the car mark storage folder identical with newly-increased car mark to be increased output confidence level；If this document folder exists, then will
Comprising this car target vehicle pictures to be stored in this sample storage folder as sample, the sample total that file is corresponding adds " 1 "；
If this sample file folder does not exists, the most newly-built newly-increased car mark storage folder named with this confidence level will comprise this car
Target vehicle pictures is stored in this sample storage folder as sample, and the sample total that file is corresponding adds " 1 ".
It should be noted that when initial, by sample size reset in each storage folder, often increase a sample, corresponding
It is " N+1 " that sample size in file then adds " 1 ", i.e. sample size.If the confidence level of car mark feature is less than or equal to empirical value, then recognize
Insincere for recognition result, this car target picture is plucked out, is stored in newly-increased car mark file, and by the sample size in file
Add " 1 ".
Step 103, when the storage sample size in newly-increased storage folder reaches the quantity set, carries out newly-increased sample and gathers
Class detects, and obtains the sample set increased newly.
N the car standard specimen reaching carried out cluster detection and the sample training set is carried out sample clustering inspection the most successively
Surveying, supervision clustering convergence effect, wherein, n is integer.Implement step as follows:
1) sheet (being assumed to be K sample) of marking the sample car in newly-increased car standard specimen basis on a map extracts car mark feature successively, it is thus achieved that K
Individual car mark feature；
2) i-th car mark feature is randomly selectedAs cluster sample class, calculate remaining j=
K-i car mark feature PjWith PiFeature fitting degree between feature, specific formula for calculation is as follows:
dij=| | Pj-Pi| |, (i, j ∈ (1,2 ..., K), and i ≠ j)
For preventing the problem of Local Clustering, this clustering method can randomly selectIndividual sample calculates cluster matching
Degree.
3) willGroup fitting result sorts according to order from small to large respectively, and statistics often group degree of fitting is less than ρ0's
Number of samples, wherein ρ0For the empirical value of degree of fitting, determining according to fitting precision, generally this value is the least, then fitting precision is more
Height, otherwise, fitting precision is relatively low；
4) takeGroup degree of fitting is less than ρ0The maximum number of sample number, as effective sample number M in this sample set；
5) filter outIn group ranking results, often organize the car standard specimen repeated in front M sample in sample originally, as having
The car mark model sample of effect.
Step 104, is stored in the sample set of existing car mark model, more new car mark model by the sample set that this is newly-increased.
In this step, the car mark sample set of newly-increased convergence is appended in existing car mark model sample set, and more
New car mark model, so that detectable car mark style sum increases.
Below by way of a preferred embodiment, the method for the basis that the present invention provides is described in further detail, such as Fig. 2
Shown in, the present embodiment mainly comprises the steps that
Step 201, carries out car mark feature extraction, and compares with car mark model the picture comprising vehicle target.
Step 202, it is judged that car target recognition confidence to be identified whether more than the threshold value preset, the most then performs step
208, illustrate that described car mark to be identified is not newly-increased car mark, searches in the car mark model prestored according to car target recognition result and
The car mark that car mark to be identified is corresponding；If it is not, then perform step 203.
Step 203, the car standard specimen in being pressed from both sides with each newly-increased car mark store files by car mark to be identified originally carries out aspect ratio pair.
Step 204, according to comparison result, originally puts into newly-increased car standard specimen in some file, or again increases one newly
Individual car mark storage folder.
Step 205, it is judged that whether the storage sample size in newly-increased storage folder reaches the quantity set, if it is not, then hold
Row step 204, the most then perform step 206.
Step 206, carries out newly-increased sample clustering detection, obtains the sample set increased newly.
Step 207, is stored in the sample set of existing car mark model, more new car mark model by the sample set that this is newly-increased.
The present invention also provides for a kind of vehicle-logo recognition device, sees Fig. 3, it is shown that the newly-increased car mark of one embodiment of the invention
Identification device, including extraction module 301, sample size more new module 302, sample set more new module 303 and car mark model modification
Module 304.Wherein:
Extraction module 301, for the picture comprising vehicle target carries out car mark feature extraction, and is carried out with car mark model
Comparison, it is judged that whether car target recognition confidence to be identified is more than the threshold value preset.Specifically as described in above-described embodiment.
Sample size more new module 302, if for car target recognition confidence to be identified less than or equal to the threshold value preset, then will
Car standard specimen during car mark to be identified presss from both sides with each newly-increased car mark store files originally carries out aspect ratio pair, updates new according to comparison result
Increase the sample size in car mark store files folder.Specifically as described in above-described embodiment.
Sample set more new module 303, when the storage sample size in newly-increased storage folder reaches the quantity set,
Carry out newly-increased sample clustering detection, obtain the sample set increased newly.Specifically as described in above-described embodiment.
Car mark model modification module 304, for the sample set that this is newly-increased is stored in the sample set of existing car mark model, updates
Car mark model.Specifically as described in above-described embodiment.
As another embodiment of the present invention, described sample size more new module 303 includes:
Newly-increased car mark memory module, multiple for depositing the newly-increased car mark storage originally of newly-increased car standard specimen for being pre-created
File.Specifically as described in above-described embodiment.
Sample size distribution module, is used for the comparative result by car mark feature confidence level Yu threshold value, by newly-increased car standard specimen originally
Put in some file, or again increase a car mark storage folder newly.Specifically as described in above-described embodiment.
In a preferred embodiment of the present invention, described sample size distribution module is in newly-increased car mark memory element
In, traversal searches the car mark storage folder identical with newly-increased car mark to be increased output confidence level；If this document folder exists, then
To comprise this car target vehicle pictures to be stored in this sample storage folder as sample, the sample total that file is corresponding adds
“1”；If this sample file folder does not exists, the most newly-built newly-increased car mark storage folder named with this confidence level will comprise
This car target vehicle pictures is stored in this sample storage folder as sample, and the sample total that file is corresponding adds " 1 ".
In yet another embodiment of the present invention, described sample set more new module 303, it is used for
1) sheet of marking the sample car in newly-increased car standard specimen basis on a map extracts car mark feature successively, it is thus achieved that K car mark feature；
2) i-th car mark feature is randomly selectedAs cluster sample class, calculate remaining j=
K-i car mark feature PjWith PiFeature fitting degree between feature；
3) willGroup fitting result sorts according to order from small to large respectively, and statistics often group degree of fitting is less than ρ0's
Number of samples, wherein ρ0Empirical value for degree of fitting；
4) takeGroup degree of fitting is less than ρ0The maximum number of sample number, as effective sample number M in this sample set；
5) filter outIn group ranking results, often in group sample before in M sample the car standard specimen of repetition this as having
The car mark model sample of effect.
Preferably, the computing formula of described feature fitting degree is as follows:
dij=| | Pj-Pi| |, (i, j ∈ (1,2 ..., K), and i ≠ j).
Therefore, the present invention provides newly-increased car target recognition methods and device combine car mark confidence level so that newly-increased car mark
The collection housekeeping operation of early stage, is completely dependent on algorithm and realizes, and largely saves manpower and expends, improves vehicle-logo recognition skill
Art.The present invention solves the problem that newly-increased car mark drags down vehicle-logo recognition rate；Also solve the most newly-increased vehicle brand too much simultaneously
Rely on manual confirmation, time-consuming long, the inefficient problem brought.
Obviously, above-described embodiment is only for clearly demonstrating example, and not restriction to embodiment.Right
For those of ordinary skill in the field, can also make on the basis of the above description other multi-form change or
Variation.Here without also cannot all of embodiment be given exhaustive.And the obvious change thus extended out or
Change among still in the protection domain of the invention.
Claims (10)
1. a newly-increased car target recognition methods, it is characterised in that including:
The picture comprising vehicle target is carried out car mark feature extraction, and compares with car mark model, it is judged that car mark to be identified
Recognition confidence whether more than the threshold value preset；
If it is not, the car standard specimen in then being pressed from both sides with each newly-increased car mark store files by car mark to be identified originally carries out aspect ratio pair, according to
Comparison result updates the sample size in newly-increased car mark store files folder；
When storage sample size in newly-increased storage folder reaches the quantity set, carry out newly-increased sample clustering detection, obtain
Newly-increased sample set；
The sample set that this is newly-increased is stored in the sample set of existing car mark model, more new car mark model.
Newly-increased car target recognition methods the most according to claim 1, it is characterised in that described new according to comparison result renewal
Increase car mark store files folder in sample size, step include:
It is pre-created multiple for depositing newly-increased car standard specimen newly-increased car mark storage folder originally, by car mark feature confidence level
With the comparative result of threshold value, newly-increased car standard specimen is originally put in some file, or again increase a car mark storage literary composition newly
Part presss from both sides.
Newly-increased car target recognition methods the most according to claim 2, it is characterised in that described by car mark feature confidence level
With the comparative result of threshold value, newly-increased car standard specimen is originally put in some file, or again increase a car mark storage literary composition newly
Part press from both sides, step include:
In newly-increased car mark memory element, traversal searches the car mark storage literary composition identical with newly-increased car mark to be increased output confidence level
Part presss from both sides；If this document folder exists, then will comprise this car target vehicle pictures and be stored in this sample storage folder as sample, literary composition
The sample total that part folder is corresponding adds " 1 "；If this sample file folder does not exists, the most newly-built newly-increased car named with this confidence level
Mark storage folder, will comprise this car target vehicle pictures and be stored in this sample storage folder as sample, and file is corresponding
Sample total add " 1 ".
Newly-increased car target recognition methods the most according to claim 1, it is characterised in that described in carry out the inspection of newly-increased sample clustering
Survey, obtain the sample set increased newly, step include:
1) sheet of marking the sample car in newly-increased car standard specimen basis on a map extracts car mark feature successively, it is thus achieved that K car mark feature；
2) i-th car mark feature is randomly selectedAs cluster sample class, calculate remaining j=K-i
Individual car mark feature PjWith PiFeature fitting degree between feature；
3) willGroup fitting result sorts according to order from small to large respectively, and statistics often group degree of fitting is less than ρ0Sample
Number, wherein ρ0Empirical value for degree of fitting；
4) takeGroup degree of fitting is less than ρ0The maximum number of sample number, as effective sample number M in this sample set；
5) filter outIn group ranking results, often in group sample before in M sample the car standard specimen of repetition this as effective car
Mark model sample.
Newly-increased car target recognition methods the most according to claim 4, it is characterised in that the calculating of described feature fitting degree is public
Formula is as follows:
dij=| | Pj-Pi| |, (i, j ∈ (1,2 ..., K), and i ≠ j).
6. a newly-increased car target identification device, it is characterised in that including:
Extraction module, for the picture comprising vehicle target carries out car mark feature extraction, and compares with car mark model, sentences
Whether disconnected car target recognition confidence to be identified is more than the threshold value preset；
Sample size more new module, if for car target recognition confidence to be identified less than or equal to the threshold value preset, then by be identified
Car standard specimen during car mark presss from both sides with each newly-increased car mark store files originally carries out aspect ratio pair, updates newly-increased car mark according to comparison result
Sample size in store files folder；
Sample set more new module, when the storage sample size in newly-increased storage folder reaches the quantity set, carries out new
Increase sample clustering detection, obtain the sample set increased newly；
Car mark model modification module, for being stored in the sample set of existing car mark model, more new car mark mould by the sample set that this is newly-increased
Type.
Newly-increased car target identification device the most according to claim 6, it is characterised in that described sample size more new module bag
Include:
Newly-increased car mark memory module, multiple for depositing newly-increased car standard specimen newly-increased car mark storage file originally for being pre-created
Folder；
Sample size distribution module, for by the comparative result of car mark feature confidence level Yu threshold value, originally putting into newly-increased car standard specimen
In some file, or again increase a car mark storage folder newly.
Newly-increased car target identification device the most according to claim 7, it is characterised in that described sample size distribution module is used for
In newly-increased car mark memory element, traversal searches the car mark storage file identical with newly-increased car mark to be increased output confidence level
Folder；If this document folder exists, then will comprise this car target vehicle pictures and be stored in this sample storage folder as sample, file
The sample total that folder is corresponding adds " 1 "；If this sample file folder does not exists, the most newly-built newly-increased car mark named with this confidence level
Storage folder, will comprise this car target vehicle pictures and be stored in this sample storage folder as sample, and file is corresponding
Sample total adds " 1 ".
Newly-increased car target identification device the most according to claim 6, it is characterised in that described sample set more new module, uses
In
1) sheet of marking the sample car in newly-increased car standard specimen basis on a map extracts car mark feature successively, it is thus achieved that K car mark feature；
2) i-th car mark feature is randomly selectedAs cluster sample class, calculate remaining j=K-i
Individual car mark feature PjWith PiFeature fitting degree between feature；
3) willGroup fitting result sorts according to order from small to large respectively, and statistics often group degree of fitting is less than ρ0Sample
Number, wherein ρ0Empirical value for degree of fitting；
4) takeGroup degree of fitting is less than ρ0The maximum number of sample number, as effective sample number M in this sample set；
5) filter outIn group ranking results, often in group sample before in M sample the car standard specimen of repetition this as effective car
Mark model sample.
Newly-increased car target identification device the most according to claim 9, it is characterised in that the calculating of described feature fitting degree
Formula is as follows:
dij=| | Pj-Pi| |, (i, j ∈ (1,2 ..., K), and i ≠ j).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610137614.3A CN105844286A (en) | 2016-03-11 | 2016-03-11 | Newly added vehicle logo identification method and apparatus |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610137614.3A CN105844286A (en) | 2016-03-11 | 2016-03-11 | Newly added vehicle logo identification method and apparatus |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105844286A true CN105844286A (en) | 2016-08-10 |
Family
ID=56588059
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610137614.3A Pending CN105844286A (en) | 2016-03-11 | 2016-03-11 | Newly added vehicle logo identification method and apparatus |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105844286A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107153844A (en) * | 2017-05-12 | 2017-09-12 | 上海斐讯数据通信技术有限公司 | The accessory system being improved to flowers identifying system and the method being improved |
| CN109214411A (en) * | 2018-07-12 | 2019-01-15 | 上海斐讯数据通信技术有限公司 | It is a kind of to identify typical picture to the verification method and system of newly-increased entity based on training pattern |
| CN110619341A (en) * | 2018-06-19 | 2019-12-27 | 佛山市顺德区美的电热电器制造有限公司 | Image recognition model training system and method |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6253179B1 (en) * | 1999-01-29 | 2001-06-26 | International Business Machines Corporation | Method and apparatus for multi-environment speaker verification |
| CN101315670A (en) * | 2007-06-01 | 2008-12-03 | 清华大学 | Specific shot body detection device, learning device and method thereof |
| CN101604394A (en) * | 2008-12-30 | 2009-12-16 | 华中科技大学 | Increment study classification method under a kind of limited storage resources |
| CN101630368A (en) * | 2009-08-25 | 2010-01-20 | 华南理工大学 | Self-adaptive method of user writing style for recognizing handwritten Chinese characters |
| CN102521623A (en) * | 2011-12-09 | 2012-06-27 | 南京大学 | Subspace-based incremental learning face recognition method |
| CN103077407A (en) * | 2013-01-21 | 2013-05-01 | 信帧电子技术（北京）有限公司 | Car logo positioning and recognition method and car logo positioning and recognition system |
| CN104182728A (en) * | 2014-07-26 | 2014-12-03 | 佳都新太科技股份有限公司 | Vehicle logo automatic location and recognition method based on pattern recognition |
| CN104331691A (en) * | 2014-11-28 | 2015-02-04 | 深圳市捷顺科技实业股份有限公司 | Vehicle logo classifier training method, vehicle logo recognition method and device |
| CN104504384A (en) * | 2015-01-15 | 2015-04-08 | 博康智能网络科技股份有限公司 | Car logo identifying method and car logo identifying system |
-
2016
- 2016-03-11 CN CN201610137614.3A patent/CN105844286A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6253179B1 (en) * | 1999-01-29 | 2001-06-26 | International Business Machines Corporation | Method and apparatus for multi-environment speaker verification |
| CN101315670A (en) * | 2007-06-01 | 2008-12-03 | 清华大学 | Specific shot body detection device, learning device and method thereof |
| CN101604394A (en) * | 2008-12-30 | 2009-12-16 | 华中科技大学 | Increment study classification method under a kind of limited storage resources |
| CN101630368A (en) * | 2009-08-25 | 2010-01-20 | 华南理工大学 | Self-adaptive method of user writing style for recognizing handwritten Chinese characters |
| CN102521623A (en) * | 2011-12-09 | 2012-06-27 | 南京大学 | Subspace-based incremental learning face recognition method |
| CN103077407A (en) * | 2013-01-21 | 2013-05-01 | 信帧电子技术（北京）有限公司 | Car logo positioning and recognition method and car logo positioning and recognition system |
| CN104182728A (en) * | 2014-07-26 | 2014-12-03 | 佳都新太科技股份有限公司 | Vehicle logo automatic location and recognition method based on pattern recognition |
| CN104331691A (en) * | 2014-11-28 | 2015-02-04 | 深圳市捷顺科技实业股份有限公司 | Vehicle logo classifier training method, vehicle logo recognition method and device |
| CN104504384A (en) * | 2015-01-15 | 2015-04-08 | 博康智能网络科技股份有限公司 | Car logo identifying method and car logo identifying system |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107153844A (en) * | 2017-05-12 | 2017-09-12 | 上海斐讯数据通信技术有限公司 | The accessory system being improved to flowers identifying system and the method being improved |
| CN110619341A (en) * | 2018-06-19 | 2019-12-27 | 佛山市顺德区美的电热电器制造有限公司 | Image recognition model training system and method |
| CN110619341B (en) * | 2018-06-19 | 2024-02-02 | 佛山市顺德区美的电热电器制造有限公司 | Image recognition model training system and method |
| CN109214411A (en) * | 2018-07-12 | 2019-01-15 | 上海斐讯数据通信技术有限公司 | It is a kind of to identify typical picture to the verification method and system of newly-increased entity based on training pattern |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105373794B (en) | A kind of licence plate recognition method | |
| CN108090429B (en) | Vehicle type recognition method for graded front face bayonet | |
| CN105321350B (en) | Fake-licensed car detection method and device | |
| CN104731879B (en) | A kind of highway fee evasion vehicles behavioral data analysis method | |
| CN107122776A (en) | A kind of road traffic sign detection and recognition methods based on convolutional neural networks | |
| CN109816024A (en) | A kind of real-time automobile logo detection method based on multi-scale feature fusion and DCNN | |
| CN112016605B (en) | Target detection method based on corner alignment and boundary matching of bounding box | |
| CN101719142B (en) | Method for detecting picture characters by sparse representation based on classifying dictionary | |
| CN103034848B (en) | A kind of recognition methods of form types | |
| CN108491797A (en) | A kind of vehicle image precise search method based on big data | |
| CN106355188A (en) | Image detection method and device | |
| CN110969160A (en) | License plate image correction and recognition method and system based on deep learning | |
| CN103324955A (en) | Pedestrian detection method based on video processing | |
| CN106529532A (en) | License plate identification system based on integral feature channels and gray projection | |
| CN102819733B (en) | Rapid detection fuzzy method of face in street view image | |
| CN101661559A (en) | Digital image training and detecting methods | |
| CN102332092A (en) | Flame detection method based on video analysis | |
| CN109344886B (en) | Occlusion number plate distinguishing method based on convolutional neural network | |
| CN111078946A (en) | Bayonet vehicle retrieval method and system based on multi-target regional characteristic aggregation | |
| CN108509950B (en) | Railway contact net support number plate detection and identification method based on probability feature weighted fusion | |
| CN106326916A (en) | Multi-scale feature estimation and high-order BING feature-based target detection method | |
| CN101930549A (en) | Second generation curvelet transform-based static human detection method | |
| CN111898491A (en) | Method and device for identifying reverse driving of vehicle and electronic equipment | |
| Sikirić et al. | Classifying traffic scenes using the GIST image descriptor | |
| CN103679214A (en) | Vehicle detection method based on online area estimation and multi-feature decision fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160810 |
|
| RJ01 | Rejection of invention patent application after publication |