CN106156750A - A kind of based on convolutional neural networks to scheme to search car method - Google Patents
A kind of based on convolutional neural networks to scheme to search car method Download PDFInfo
- Publication number
- CN106156750A CN106156750A CN201610590679.3A CN201610590679A CN106156750A CN 106156750 A CN106156750 A CN 106156750A CN 201610590679 A CN201610590679 A CN 201610590679A CN 106156750 A CN106156750 A CN 106156750A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- convolutional neural
- neural networks
- search
- 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
- G06F18/21—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
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- 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
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Medical Informatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of based on convolutional neural networks to scheme to search car method.The present invention collects the image set of different vehicle, identical vehicle is as positive sample pair, different vehicles is as negative sample pair, exemplary convolution neural network model is trained, minimize similarity difference or error in classification, learn one group of vehicle characteristics expression, vehicle image is after this convolutional neural networks model propagated forward, the result of data Layer just can be as the textural characteristics of vehicle, calculate the similarity of vehicle characteristics to be retrieved and retrieved set vehicle by this feature, be ranked up obtaining scheming to search the result of car by similarity.The present invention utilizes the outward appearance method for expressing of convolutional neural networks study vehicle image, higher compared with purposivenesss such as SIFT, HoG features, feature is more directly perceived, need not extra metric learning process, be conducive to increasing substantially the accuracy of search and precision, and the dimension of feature can be controlled in the less order of magnitude, be conducive to realizing fast search in large nuber of images storehouse.
Description
Technical field
The invention belongs to technical field of intelligent traffic, relate to a kind of based on convolutional neural networks to scheme to search car method.
Background technology
Rely on vehicle appearance feature search collection is carried out sequencing of similarity scheming to search car method, it is possible to from video or image set
In find and comprise all images of vehicle that user searches, search procedure does not relies on license board information, to unlicensed, false-trademark and deck
Equally applicable.
Convolutional neural networks is research at present and the focus of commercial Application, compared with traditional intelligent algorithm, such as god
Through network, support vector machine etc., degree of depth learning algorithm can significantly promote the precision of image classification, in face recognition application field
Surmount the accuracy rate of naked eyes identification.
It is shown with a variety of method: " suspected vehicles based on sift feature with the vehicle appearance mark sheet scheming to search in car method
Search method " in 103678558A, tri-Color Channels of R, G, B of vehicle image extract SIFT feature, respectively in conjunction with car
The external appearance characteristic of color description vehicle;" a kind of vehicle retrieval method based on similarity study " 105488099A first extracts
The SIFT feature of vehicle image, utilizes clustering algorithm this SIFT feature discrete, is converted into the vehicle based on neighborhood characteristics and retouches
State feature;" a kind of motor vehicles search method based on image and device " 104361087A is according to vehicle outward appearance in the picture
Vehicle image is divided into multiple subwindow by profile, extracts SIFT or HoG feature, merge the stricture of vagina of all subwindows in subwindow
The external appearance characteristic of reason feature representation vehicle.
In these existing vehicle appearance method for expressing, it is special that the manual method such as commonly used SIFT, HoG calculates texture
Levying, this traditional-handwork feature extraction mode is applicable to most of application of pattern recognition, and versatility is good, but due to computational methods
Fixing, it is impossible to reach the highest precision, be unfavorable for carrying out of vehicle retrieval.
Summary of the invention
The present invention is directed to the deficiencies in the prior art, it is provided that a kind of based on convolutional neural networks to scheme to search car method.
The central scope of the present invention: collecting the image set of different vehicle, identical vehicle is as positive sample pair, different cars
As negative sample pair, at exemplary convolution neural network model, in the network modeies such as TripletLoss, Inception, VGG
Training, minimizes similarity difference or error in classification, learns one group of vehicle characteristics expression, and vehicle image is at this convolutional Neural
After network model's propagated forward, the result of data Layer just can calculate vehicle to be retrieved as the textural characteristics of vehicle by this feature
Feature and the similarity of retrieved set vehicle, be ranked up obtaining scheming to search the result of car by similarity.
Feature representation Learning Scheme therein:
1. first with conventional object detection method location vehicle location.
2. operations such as reasonably being rotated by vehicle region image, scale, be color transformed, increases the quantity of sample.
3. the size of specification vehicle image pattern.
4. being labelled by vehicle image, identical label played by identical vehicle, and different vehicle tag is different.
5., according to the convolutional neural networks framework used and penalty integral data, form training set.
6. in training set, learn network parameter, use GPU accelerator card to improve learning rate.
7. the vehicle image after specification under the network parameter that arrives of study according to convolutional neural networks propagated forward, each
Individual data Layer all can be as the appearance representation of vehicle, and available GPU accelerates.
Based on convolutional neural networks to scheme to search car scheme:
1. utilize conventional object detection method position vehicle to be searched and search for the vehicle location concentrated.
2. the size of specification vehicle image pattern.
3. use GPU to accelerate, by the vehicle image after specification under the network parameter that arrives of study according to convolutional neural networks
Propagated forward, selects the data external appearance characteristic vector as vehicle of a data Layer.
4. calculate vehicle to be searched and the similarity of search collection vehicle according to external appearance characteristic vector, according to similarity descending
Arrangement search collection.
Beneficial effects of the present invention:
1, for scheme to search car problem learning feature representation method, learn to feature extracting method tight with to scheme to search car task
Relevant, improve feature to greatest extent to scheme to search the suitability in car task.
2, metric learning can be attached in convolutional neural networks framework, reduce the complexity of vehicle retrieval task, calculate
Method is more stable controlled.
3, use GPU to accelerate, increase substantially the speed of feature extraction.
Accompanying drawing explanation
Fig. 1 is characterized expression learning process figure.
Fig. 2 is for scheme to search map flow chart.
Detailed description of the invention
Below in conjunction with embodiment and accompanying drawing, the invention will be further described:
In video investigation project, the present embodiment have employed based on convolutional neural networks to scheme to search car scheme.
As it is shown in figure 1, in feature representation learning process: (1) positions vehicle location with conventional object detection method, will
Vehicle region image carries out the rotation in the range of 10 °, scaling in the range of 0.2, carries out color transformed with PCA, every vehicle figure
As disturbance 10 times, collect 3,000,000 images after data set extension altogether and be trained.(2) by all training set image specifications extremely
224x224.(3) by all of image composition tlv triple in data set, (a, p, n), a represents any one vehicle image, p with a is
The different images of same vehicle, n Yu a is different vehicle, and all tlv triple are stored in data base.(4) add on VGG network
Tripletloss penalty composition vehicle characteristics expresses convolutional neural networks structure, training network parameter in training set.
As in figure 2 it is shown, during based on convolutional neural networks to scheme to search car: (1) is fixed with conventional object detection method
The vehicle location that position vehicle to be searched and search are concentrated.(2) specification vehicle image is to 224x224.(3) at GPU accelerator card, will rule
Vehicle image after model according to convolutional neural networks propagated forward, selects Tripletloss layer under the network parameter that study is arrived
Data Layer data above are as the external appearance characteristic vector of vehicle.(4) calculate vehicle to be searched according to external appearance characteristic vector and search
The inner product of rope collection vehicle, arrangement search collection.
To sum up, the present invention utilizes the outward appearance method for expressing of convolutional neural networks study vehicle image, relatively SIFT, HoG feature
Higher Deng purposiveness, feature is more directly perceived, it is not necessary to extra metric learning process, is conducive to increasing substantially the accuracy of search
And precision, and the dimension of feature can be controlled in the less order of magnitude, is conducive to realizing fast search in large nuber of images storehouse.
The above, only presently preferred embodiments of the present invention, it is not intended to limit protection scope of the present invention, should carry
Understanding, the present invention is not limited to implementation as described herein, and the purpose that these implementations describe is to help this area
In technical staff put into practice the present invention.
Claims (3)
1. one kind based on convolutional neural networks to scheme to search car method, it is characterised in that: collect different vehicle image set, identical
Vehicle as positive sample pair, different vehicles as negative sample pair, on exemplary convolution neural network model train, minimize
Similarity difference or error in classification, learn one group of vehicle characteristics expression, and vehicle image is before this convolutional neural networks model
After propagating, the result of data Layer just can calculate vehicle characteristics to be retrieved and retrieval as the textural characteristics of vehicle by this feature
The similarity of collection vehicle, is ranked up obtaining scheming to search the result of car by similarity.
It is the most according to claim 1 a kind of based on convolutional neural networks to scheme to search car method, it is characterised in that:
Feature representation learns specifically:
Step 1. utilizes object detection method to position vehicle location;
Vehicle region image carries out rotating, scales by step 2., color transformed;
The size of step 3. specification vehicle image pattern;
Vehicle image is labelled by step 4., and identical label played by identical vehicle, and different vehicle tag is different;
Step 5., according to the convolutional neural networks framework used and penalty integral data, forms training set;
Step 6. learns network parameter in training set;
Vehicle image after step 7. specification is under the network parameter that study is arrived, according to convolutional neural networks propagated forward, each
Individual data Layer all can be as the appearance representation of vehicle.
It is the most according to claim 2 a kind of based on convolutional neural networks to scheme to search car method, it is characterised in that: step 6
Middle utilization GPU accelerator card improves learning rate.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610590679.3A CN106156750A (en) | 2016-07-26 | 2016-07-26 | A kind of based on convolutional neural networks to scheme to search car method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610590679.3A CN106156750A (en) | 2016-07-26 | 2016-07-26 | A kind of based on convolutional neural networks to scheme to search car method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106156750A true CN106156750A (en) | 2016-11-23 |
Family
ID=58059592
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610590679.3A Pending CN106156750A (en) | 2016-07-26 | 2016-07-26 | A kind of based on convolutional neural networks to scheme to search car method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106156750A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107688819A (en) * | 2017-02-16 | 2018-02-13 | 平安科技(深圳)有限公司 | The recognition methods of vehicle and device |
CN108229276A (en) * | 2017-03-31 | 2018-06-29 | 北京市商汤科技开发有限公司 | Neural metwork training and image processing method, device and electronic equipment |
CN108764018A (en) * | 2018-04-03 | 2018-11-06 | 北京交通大学 | A kind of multitask vehicle based on convolutional neural networks recognition methods and device again |
CN109902732A (en) * | 2019-02-22 | 2019-06-18 | 哈尔滨工业大学(深圳) | Automobile automatic recognition method and relevant apparatus |
CN110009929A (en) * | 2019-03-15 | 2019-07-12 | 北京筑梦园科技有限公司 | A kind of Vehicle berth management method, equipment and system |
CN110222560A (en) * | 2019-04-25 | 2019-09-10 | 西北大学 | A kind of text people search's method being embedded in similitude loss function |
CN110458234A (en) * | 2019-08-14 | 2019-11-15 | 广州广电银通金融电子科技有限公司 | It is a kind of based on deep learning to scheme to search vehicle method |
CN111475666A (en) * | 2020-03-27 | 2020-07-31 | 深圳市墨者安全科技有限公司 | Dense vector-based media accurate matching method and system |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130266214A1 (en) * | 2012-04-06 | 2013-10-10 | Brighham Young University | Training an image processing neural network without human selection of features |
CN104537387A (en) * | 2014-12-16 | 2015-04-22 | 广州中国科学院先进技术研究所 | Method and system for classifying automobile types based on neural network |
CN105354273A (en) * | 2015-10-29 | 2016-02-24 | 浙江高速信息工程技术有限公司 | Method for fast retrieving high-similarity image of highway fee evasion vehicle |
-
2016
- 2016-07-26 CN CN201610590679.3A patent/CN106156750A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130266214A1 (en) * | 2012-04-06 | 2013-10-10 | Brighham Young University | Training an image processing neural network without human selection of features |
CN104537387A (en) * | 2014-12-16 | 2015-04-22 | 广州中国科学院先进技术研究所 | Method and system for classifying automobile types based on neural network |
CN105354273A (en) * | 2015-10-29 | 2016-02-24 | 浙江高速信息工程技术有限公司 | Method for fast retrieving high-similarity image of highway fee evasion vehicle |
Non-Patent Citations (2)
Title |
---|
FLORIAN SCHROFF 等: "FaceNet: A Unified Embedding for Face Recognition and Clustering", 《IN PROCEEDINGS OF THE IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
张琪: "图像检索在公安图像侦查领域的发展", 《中国安防》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107688819A (en) * | 2017-02-16 | 2018-02-13 | 平安科技(深圳)有限公司 | The recognition methods of vehicle and device |
CN108229276B (en) * | 2017-03-31 | 2020-08-11 | 北京市商汤科技开发有限公司 | Neural network training and image processing method and device and electronic equipment |
CN108229276A (en) * | 2017-03-31 | 2018-06-29 | 北京市商汤科技开发有限公司 | Neural metwork training and image processing method, device and electronic equipment |
CN108764018A (en) * | 2018-04-03 | 2018-11-06 | 北京交通大学 | A kind of multitask vehicle based on convolutional neural networks recognition methods and device again |
CN109902732B (en) * | 2019-02-22 | 2021-08-27 | 哈尔滨工业大学(深圳) | Automatic vehicle classification method and related device |
CN109902732A (en) * | 2019-02-22 | 2019-06-18 | 哈尔滨工业大学(深圳) | Automobile automatic recognition method and relevant apparatus |
CN110009929A (en) * | 2019-03-15 | 2019-07-12 | 北京筑梦园科技有限公司 | A kind of Vehicle berth management method, equipment and system |
CN110222560A (en) * | 2019-04-25 | 2019-09-10 | 西北大学 | A kind of text people search's method being embedded in similitude loss function |
CN110222560B (en) * | 2019-04-25 | 2022-12-23 | 西北大学 | Text person searching method embedded with similarity loss function |
CN110458234A (en) * | 2019-08-14 | 2019-11-15 | 广州广电银通金融电子科技有限公司 | It is a kind of based on deep learning to scheme to search vehicle method |
CN110458234B (en) * | 2019-08-14 | 2021-12-03 | 广州广电银通金融电子科技有限公司 | Vehicle searching method with map based on deep learning |
CN111475666A (en) * | 2020-03-27 | 2020-07-31 | 深圳市墨者安全科技有限公司 | Dense vector-based media accurate matching method and system |
CN111475666B (en) * | 2020-03-27 | 2023-10-10 | 深圳市墨者安全科技有限公司 | Dense vector-based media accurate matching method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106156750A (en) | A kind of based on convolutional neural networks to scheme to search car method | |
CN110414368B (en) | Unsupervised pedestrian re-identification method based on knowledge distillation | |
Liu | An improved faster R-CNN for object detection | |
CN106529424B (en) | A kind of logo detection recognition method and system based on selective search algorithm | |
CN110909788B (en) | Statistical clustering-based road intersection position identification method in track data | |
CN103324913B (en) | A kind of pedestrian event detection method of Shape-based interpolation characteristic sum trajectory analysis | |
CN106650731B (en) | Robust license plate and vehicle logo recognition method | |
CN107103326A (en) | The collaboration conspicuousness detection method clustered based on super-pixel | |
CN103886619B (en) | A kind of method for tracking target merging multiple dimensioned super-pixel | |
CN103456013B (en) | A kind of method representing similarity between super-pixel and tolerance super-pixel | |
CN106355602A (en) | Multi-target locating and tracking video monitoring method | |
CN103632146B (en) | A kind of based on head and shoulder away from human body detecting method | |
CN105678297A (en) | Portrait semantic analysis method and system based on label transfer and LSTM model | |
CN105046197A (en) | Multi-template pedestrian detection method based on cluster | |
CN105718912B (en) | A kind of vehicle characteristics object detecting method based on deep learning | |
CN107679512A (en) | A kind of dynamic gesture identification method based on gesture key point | |
CN111340855A (en) | Road moving target detection method based on track prediction | |
CN103473551A (en) | Station logo recognition method and system based on SIFT operators | |
CN104598885A (en) | Method for detecting and locating text sign in street view image | |
CN103020614B (en) | Based on the human motion identification method that space-time interest points detects | |
CN109949593A (en) | A kind of traffic lights recognition methods and system based on crossing priori knowledge | |
CN104200228A (en) | Recognizing method and system for safety belt | |
CN112329559A (en) | Method for detecting homestead target based on deep convolutional neural network | |
CN106886757B (en) | A kind of multiclass traffic lights detection method and system based on prior probability image | |
CN106951829A (en) | A kind of notable method for checking object of video based on minimum spanning tree |
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: 20161123 |
|
RJ01 | Rejection of invention patent application after publication |