CN104298967B - A kind of vehicle image comparison method of view-based access control model feature - Google Patents
A kind of vehicle image comparison method of view-based access control model feature Download PDFInfo
- Publication number
- CN104298967B CN104298967B CN201410482824.7A CN201410482824A CN104298967B CN 104298967 B CN104298967 B CN 104298967B CN 201410482824 A CN201410482824 A CN 201410482824A CN 104298967 B CN104298967 B CN 104298967B
- Authority
- CN
- China
- Prior art keywords
- vehicle image
- image
- feature
- access control
- control model
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Library & Information Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
- Processing Or Creating Images (AREA)
Abstract
The present invention relates to a kind of vehicle image comparison methods of view-based access control model feature, include the following steps：1) a large amount of vehicle image data are acquired and forms image data base, the inverted list concordance list of vehicle image set is established according to image data base；2) it according to the inverted list concordance list, the n most like with the Query image of input image of retrieval, and exports.Compared with prior art, speed of the present invention is fast, and average 1 Query image only needs 1 second time to compare completion, and on the database of million ranks, accuracy can achieve 8 one-tenth.
Description
Technical field
The present invention relates to a kind of image-recognizing methods, compare other side more particularly, to a kind of vehicle image of view-based access control model feature
Method.
Background technique
In recent years, intelligent transportation system development is intelligent with the development of computer vision and mode identification technology quickly
Traffic system more effectively application provides opportunity.Computer vision is that the visual performance of people is simulated using computer, from visitor
It sees in the image of things and extracts information, handled and understood, eventually for actually detected, measurement and control.
The prior art, which is not directed to Vehicular system and specially does, to be optimized, therefore performance is unable to satisfy demand.There is presently no energy
The technology of similar vehicle is found on the database of million ranks.
Summary of the invention
Fast, search essence that it is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide a kind of speed
Spend the vehicle image comparison method of high view-based access control model feature.
The purpose of the present invention can be achieved through the following technical solutions：
A kind of vehicle image comparison method of view-based access control model feature, includes the following steps：
1) a large amount of vehicle image data are acquired and forms image data base, vehicle image set is established according to image data base
Inverted list concordance list；
2) it according to the inverted list concordance list, the n most like with the Query image of input image of retrieval, and exports.
The inverted list concordance list for establishing vehicle image set is specially：
(a) car plate detection is carried out to every vehicle image in image data base and brand recognition is handled；
(b) vehicle image is extracted according to the license plate position detected, and be normalized；
(c) feature point extraction is carried out using every vehicle image of the various features extracting method to extraction, to distinct methods
The characteristic point of acquisition carries out non-maxima suppression processing, only retains a key point in same area；
(d) feature, including tri- kinds of features of SIFT, SURF and LSSD are extracted to the key point of acquisition, finally obtains all figures
The characteristic point of picture；
(e) clustering is carried out to the characteristic point obtained in step (d), generates vocabulary；
(f) each characteristic point in every vehicle image is compared with vocabulary, by every vehicle image with feature to
The form of amount is expressed, and described eigenvector is (s1, s2, s3 ..., sN), wherein sK indicates the of in present image vocabulary
The number that K word occurs, K=1,2 ..., N, N are the sum of word in vocabulary；
(g) TF-IDF normalization is carried out to feature vector, and is established under different brands according to the result that brand recognition is handled
Inverted list concordance list.
The car plate detection is executed using AdaBoost detector.
Feature extracting method in the step (c) includes SIFT, MSER and Harris Laplace.
In the step (c), when carrying out feature point extraction using various features extracting method, any two methods are extracted
Characteristic point between repetitive rate be greater than 50%.
In the step (e), method that clustering uses is the Kmeans after utilizing KD-Forests technology to accelerate
Clustering algorithm.
The described retrieval n images most like with the Query image of input the specific steps are：
(aa) it is detected, is obtained opposite with Query image according to characteristic area of step (a)-(f) to Query image
Feature vector after just TF-IDF is normalized；
(bb) brand recognition is carried out to Query image；
(cc) calculate separately the feature vector of Query image and all features in the inverted list concordance list under corresponding brand to
The cosine similarity of amount, the smallest n of cosine similarity images of output.
The n is 15-25.
Compared with prior art, the present invention has the following advantages that：
1, speed is fast, and average 1 Query image only needs 1 second time；
2, search precision is high, and on the database of million ranks, accuracy can achieve 8 one-tenth；
3, strong robustness can use under different scenes.
Specific embodiment
The present invention is described in detail combined with specific embodiments below.The present embodiment is based on the technical solution of the present invention
Implemented, the detailed implementation method and specific operation process are given, but protection scope of the present invention be not limited to it is following
Embodiment.
A kind of vehicle image comparison method of view-based access control model feature, this method, input include Query image, Query figure
As upper characteristic area and image data base, the n vehicle most like for image data base is exported, specifically includes and builds library and retrieval two
A step, wherein build that library is only related with image data base, and process can handle any Query image after the completion.This method is specific
It is described as follows：
1) library is built：It acquires a large amount of vehicle image data and forms image data base, vehicle image is established according to image data base
The inverted list concordance list of set, specially：
(a) car plate detection is carried out to every vehicle image in image data base and brand recognition is handled, car plate detection is adopted
It is executed with AdaBoost detector, brand recognition processing is using technical solution disclosed in Chinese patent CN103488973A.
(b) vehicle image is extracted according to the license plate position detected, and be normalized.
(c) feature point extraction is carried out using the every vehicle image of various features extracting method (detector) to extraction,
Feature extracting method includes the figure of SIFT (also referred to as DoG), the acquisition of MSER and Harris Laplace different vehicle monitoring system
Image quality amount has notable difference, therefore cannot use identical parameter to all images, and the mode of hill climbing can be used to adjust ginseng
Number, final goal are that 3 kinds of detector extract the characteristic point of 2K or so (vision that point can not accurately describe vehicle very little is special
Point), and there is 5 one-tenth of repetitive rate (to avoid some detector due to parameter between the characteristic point of any two detector extraction
Reason performance is too poor, all detects the useless regions such as ground).The characteristic point that distinct methods obtain is carried out at non-maxima suppression
Reason, only retains a key point in same area.
(d) feature, including tri- kinds of features of SIFT, SURF and LSSD are extracted to the key point of acquisition, finally obtains all figures
The characteristic point of picture.After all vehicle images all extraction features, every image all extracts the characteristic point of certain amount, finally receives altogether
Collect about 100M characteristic point.
(e) clustering is carried out to the characteristic point obtained in step (d), generates vocabulary, there are about 1M words in vocabulary.It is common
Clustering method is Kmeans, but its speed is too slow, and the lower common machines of normal condition need one-month period, so this
Invention uses KD-Forests technology and is accelerated, and completes it in 3~5 hours.
(f) each characteristic point in every vehicle image is compared with vocabulary, by every vehicle image with feature to
The form of amount is expressed, and described eigenvector is (s1, s2, s3 ..., sN), wherein sK indicates the K of the vocabulary in present image
The number that a word occurs, K=1,2 ..., N, N are the sum of word in vocabulary.
(g) TF-IDF normalization is carried out to feature vector, and is established under different brands according to the result that brand recognition is handled
Inverted list concordance list, to accelerate subsequent Index process.
2) it retrieves：According to the inverted list concordance list, retrieval opens image with the most like n of the Query image of input, and defeated
Out, n can be set as 15-25 as needed, and similar definitions here are the Cos similitudes of two final feature vectors of image.Retrieval
Step is specially：
(aa) it is detected, is obtained opposite with Query image according to characteristic area of step (a)-(f) to Query image
Feature vector after just TF-IDF is normalized；
(bb) brand recognition is carried out to Query image；
(cc) calculate separately the feature vector of Query image and all features in the inverted list concordance list under corresponding brand to
The cosine similarity of amount, the smallest n of cosine similarity images of output.
Claims (5)
1. a kind of vehicle image comparison method of view-based access control model feature, which is characterized in that include the following steps：
1) a large amount of vehicle image data are acquired and forms image data base, the row of falling of vehicle image set is established according to image data base
Table index table；
2) it according to the inverted list concordance list, the n most like with the Query image of input image of retrieval, and exports；
The inverted list concordance list for establishing vehicle image set is specially：
(a) car plate detection is carried out to every vehicle image in image data base and brand recognition is handled；
(b) vehicle image is extracted according to the license plate position detected, and be normalized；
(c) feature point extraction is carried out using every vehicle image of the various features extracting method to extraction, distinct methods is obtained
Characteristic point carry out non-maxima suppression processing, only retain a key point in same area；
(d) feature, including tri- kinds of features of SIFT, SURF and LSSD are extracted to the key point of acquisition, finally obtains all images
Characteristic point；
(e) clustering is carried out to the characteristic point obtained in step (d), generates vocabulary；
(f) each characteristic point in every vehicle image is compared with vocabulary, by every vehicle image with feature vector
Form expression, described eigenvector are (s1, s2, s3 ..., sN), wherein sK indicates the k-th of the vocabulary in present image
The number that word occurs, K=1,2 ..., N, N are the sum of word in vocabulary；
(g) TF-IDF normalization is carried out to feature vector, and falling under different brands is established according to the result that brand recognition is handled
Table index table is arranged,
Wherein, in the step (c), when carrying out feature point extraction using various features extracting method, any two methods are mentioned
Between the characteristic point taken repetitive rate be greater than 50%, and
Wherein, the described retrieval n images most like with the Query image of input the specific steps are：
(aa) it is detected, is obtained corresponding with Query image according to characteristic area of step (a)-(f) to Query image
Feature vector, and TF-IDF normalization is carried out to this feature vector, the feature vector after obtaining TF-IDF normalization；
(bb) brand recognition is carried out to Query image；
(cc) feature vector of Query image obtained in step (aa) and the inverted list concordance list under corresponding brand are calculated separately
In all feature vectors cosine similarity, the output the smallest n of cosine similarity images.
2. a kind of vehicle image comparison method of view-based access control model feature according to claim 1, which is characterized in that described
Car plate detection is executed using AdaBoost detector.
3. a kind of vehicle image comparison method of view-based access control model feature according to claim 1, which is characterized in that described
Feature extracting method in step (c) includes SIFT, MSER and Harris Laplace.
4. a kind of vehicle image comparison method of view-based access control model feature according to claim 1, which is characterized in that described
In step (e), method that clustering uses is the Kmeans clustering algorithm after utilizing KD-Forests technology to accelerate.
5. a kind of vehicle image comparison method of view-based access control model feature according to claim 1, which is characterized in that described
N is 15-25.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410482824.7A CN104298967B (en) | 2014-09-19 | 2014-09-19 | A kind of vehicle image comparison method of view-based access control model feature |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410482824.7A CN104298967B (en) | 2014-09-19 | 2014-09-19 | A kind of vehicle image comparison method of view-based access control model feature |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104298967A CN104298967A (en) | 2015-01-21 |
| CN104298967B true CN104298967B (en) | 2018-11-20 |
Family
ID=52318688
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410482824.7A Active CN104298967B (en) | 2014-09-19 | 2014-09-19 | A kind of vehicle image comparison method of view-based access control model feature |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104298967B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113868500A (en) * | 2020-06-30 | 2021-12-31 | 华为技术有限公司 | Retrieval method, index construction method and related equipment |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7920716B2 (en) * | 2003-12-25 | 2011-04-05 | Fujitsu Limited | Vehicle search system and vehicle search method |
| CN103810505B (en) * | 2014-02-19 | 2017-11-24 | 北京大学 | Vehicles identifications method and system based on multiple layer description |
-
2014
- 2014-09-19 CN CN201410482824.7A patent/CN104298967B/en active Active
Non-Patent Citations (3)
| Title |
|---|
| 基于车标识别的车型细分类技术研究;潘祥;《中国硕士学位论文全文数据库信息科技辑》;20110415;全文 * |
| 汽车车型自动识别系统;张红星;《中国硕士学位论文全文数据库信息科技辑》;20080815;全文 * |
| 运动车辆识别技术;李丹丹;《中国硕士学位论文全文数据库信息科技辑》;20060115;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104298967A (en) | 2015-01-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105005772B (en) | A kind of video scene detection method | |
| CN102663411B (en) | Recognition method for target human body | |
| Eweiwi et al. | Temporal key poses for human action recognition | |
| CN102663401B (en) | Image characteristic extracting and describing method | |
| CN106355154B (en) | Method for detecting frequent passing of people in surveillance video | |
| CN103279738B (en) | Automatic identification method and system for vehicle logo | |
| CN103279768A (en) | Method for identifying faces in videos based on incremental learning of face partitioning visual representations | |
| CN103679142A (en) | Target human body identification method based on spatial constraint | |
| CN103226713A (en) | Multi-view behavior recognition method | |
| CN103020658A (en) | Recognition method for objects in two-dimensional images | |
| CN107369162B (en) | Method and system for generating insulator candidate target area | |
| CN105825215B (en) | It is a kind of that the instrument localization method of kernel function is embedded in based on local neighbor and uses carrier | |
| CN104298967B (en) | A kind of vehicle image comparison method of view-based access control model feature | |
| Zhang et al. | An on-road vehicle detection method for high-resolution aerial images based on local and global structure learning | |
| Li et al. | Downsampling sparse representation and discriminant information aided occluded face recognition | |
| CN103235955A (en) | Extraction method of visual word in image retrieval | |
| Huang et al. | Person re-identification across multi-camera system based on local descriptors | |
| CN104008095A (en) | Object recognition method based on semantic feature extraction and matching | |
| Pei et al. | Pedestrian detection based on hog and lbp | |
| CN103390150A (en) | Human body part detection method and device | |
| CN102279925B (en) | Chain processing face recognition method and system | |
| Wirdiani et al. | Medicinal plant recognition of leaf shape using Localized Arc Pattern Method | |
| Wang et al. | Fast loop closure detection via binary content | |
| Rashwan et al. | Gait representation and recognition from temporal co-occurrence of flow fields | |
| Yunsheng et al. | Auto-recognition pedestrians research based on hog feature and svm classifier for vehicle images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |