CN103295021A - Method and system for detecting and recognizing feature of vehicle in static image - Google Patents
Method and system for detecting and recognizing feature of vehicle in static image Download PDFInfo
- Publication number
- CN103295021A CN103295021A CN2012100428091A CN201210042809A CN103295021A CN 103295021 A CN103295021 A CN 103295021A CN 2012100428091 A CN2012100428091 A CN 2012100428091A CN 201210042809 A CN201210042809 A CN 201210042809A CN 103295021 A CN103295021 A CN 103295021A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- training
- image
- detection
- feature
- 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
Images
Abstract
The invention belongs to the field of recognizing types of vehicles through computer image processing, and relates to a method and system for detecting and recognizing a feature and a brand of a vehicle in a static image by means of a digital picture processing technique. A method for detecting the vehicle in the static image is combined with a vehicle logo detecting method which is based on an AdaBoost framework in the method. The method and system for detecting and recognizing the feature of the vehicle in the static images comprises a training part and a detecting part. The training part includes the following steps of manufacturing a vehicle logo sample, collecting an image containing the vehicle logo from the Internet, positioning the vehicle logo, and extracting the vehicle logo image based on position information; calculating a sample feature, constructing 5 different rectangular features with each rectangular feature corresponding to one Haar feature; training a cascade classifier, inputting the training sample acquired from the last step and conducting training, and finally connecting strong classifiers and multiple corresponding weak classifiers obtained in training in series. The detecting part includes the following steps: loading the image to be detected, converting the image into a grey-scale image and conducting histogram equalization, loading the vehicle logo classifiers which include threshold values of the strong classifiers and the weak classifiers and rectangular feature information corresponding to the selected features, conducting cascade vehicle logo detection with the detected image firstly passing the detection of the former strong classifiers. If the detected image is not the vehicle logo image, the detected image can be excluded at the front end, and only the vehicle logo can finally pass the detection of the strong classifiers at various different levels.
Description
Technical field
The present invention relates to the method and system of vehicle characteristics detection and Identification in a kind of static images, belong to vehicle image identification field, is the identification of carrying out vehicle appearance feature detection and vehicle brand in static images specifically.
Background technology
Along with electronic imaging technology and Internet development, people create picture, it is more and more convenient and various to share picture and obtain the approach of picture, thereby the picture that occurs on the internet is also increasing, except the description of literal to these pictures, computing machine is not also known the content of picture, and such as whether vehicle is arranged in certain pictures, which type of feature vehicle possesses in the picture, be what brand, these all have only and judge by artificial inspection and just can know usually.
Traditional form with keyword is made note to image content and can not be matched well on the corresponding picture, and therefore people just arise at the historic moment according to the technology that certain algorithm directly goes to identify image content by using computing machine.Like this can be so that even if there is not the internet picture of content description in advance, also can go to know by image recognition technology wherein just to comprise wherein content to vehicle detection and identification.
Existing vehicle characteristics detection and Identification technology mainly is used in the video, and such as intelligent transportation monitoring, the multiframe information by video can detect the vehicle of motion with comparalive ease, thereby is located and identify; Yet in static images, be not suitable for adopting these classic methods, just will seek the method for detection and Identification vehicle in static images.Along with Development of E-business, increasing commodity are directly to face the consumer on network, wherein also comprise vehicle.If can allow computing machine know in certain pictures in the internet whether have vehicle, and know the brand of vehicle, will more be conducive to the 0vehicle marketing merchant so to the popularization of vehicle, so the technology of detection and Identification vehicle has broad application prospects in static images.
Summary of the invention
The method that the purpose of this invention is to provide a kind of detection and Identification of static images vehicle characteristics, and provide a kind of system of detection and Identification of the vehicle characteristics at the internet picture, comprise: network chart sheet to be detected is carried out vehicle detection and judges the attitude of vehicle, thereby front region or back region at vehicle are carried out car mark detection identification vehicle, and flow process as shown in Figure 1.
One, the invention process provides vehicle checking method in a kind of static images, comprises training and detects two parts.
The step of training stage is as follows:
(1) make the vehicle sample: the input picture is carried out normalization, and namely each color component to input picture carries out the gamma standardization, crosses dark or the lower situation of contrast to adapt to image, and the operation that the present invention adopts is to the color component computing of taking the logarithm;
(2) sample characteristics calculates: to the image calculation gradient orientation histogram after the normalization (Histogram of Oriented Gradient, HOG) feature pyramid;
(3) training auto model: import the characteristic data set of sample into training classifier implicit expression support vector machine (Latent Support Vector, LSVM) study.Produce the mixture model of root model, partial model and the corresponding variable shape partial model of vehicle by study.
The step of detection-phase is as follows:
(1) load picture to be measured: the input picture is carried out normalization, and namely each color component to input picture carries out the gamma standardization;
(2) feature calculation: network chart sheet to be detected is calculated HOG feature pyramid;
(3) load auto model: the data file of load store auto model;
(4) vehicle detection: the multiple dimensioned detection algorithm by metric space scans the realization vehicle detection location, zone that is complementary with the units of variance model at the feature pyramid.
Two, the invention process provides a kind of static images vehicle identification method, specifically is a kind of car mark detection method based on the AdaBoost framework, comprises training and detects two parts.
The step of training stage is as follows:
(1) makes the car standard specimen originally;
(2) sample characteristics calculates;
(3) training connection level sorter.
The step of detection-phase is as follows:
(1) loads picture to be measured;
(2) load car mark sorter;
(3) connection level car mark detects.
Description of drawings
Fig. 1 is the schematic flow sheet of static images vehicle characteristics detection and Identification.
Fig. 2 is the composition synoptic diagram of training auto model 6 moding shape parts.
Fig. 3 is the synoptic diagram of the target distortion positioning parts during vehicle characteristics detects.
Fig. 4 is the synoptic diagram of the Haar feature of sample characteristics calculating.
Fig. 5 is the schematic flow sheet that connection level car mark detects.
Fig. 6 is the result schematic diagram of level car mark testing process shown in Figure 5.
Embodiment
Describe embodiments of the present invention in detail below with reference to drawings and Examples, how the application technology means solve technical matters to the present invention whereby, and the implementation procedure of reaching technique effect can fully understand and implements according to this.
Existing vehicle checking method is just set up model to the front of vehicle, vehicle in picture is current with side or back, just very possible detection does not go out vehicle, and existing method will address this problem, can only be that model is removed to set up in each side of vehicle, calculated amount will be increased like this, however occlusion issue can not be avoided well.And the method among the present invention is that vehicle front/back, side are set up model (root model) respectively, and each model is divided into different parts, each parts and corresponding distortion thereof are set up model (partial model and deformable component model) respectively, a model permeate at last as final auto model, such processing can not only guarantee that the vehicle of different attitudes obtains detecting, and the problem that can also avoid vehicle sections to be blocked effectively.
Existing vehicle identification method is by the location of vehicle license plate being removed the car mark of positioned vehicle, then by the car brand under the ground difference vehicles such as edge feature of picture of marking on a map, this method depends on the accuracy of car plate location, and a lot of network vehicle pictures do not have car plate, such as the vehicle pictures in the car exhibition; In addition, merely go to judge that by feature the car target is all kinds of, also can reduce accuracy because of the influence of environment, such as illumination and noise.And the method among the present invention be by to a large amount of car standard specimen this, on the basis of extracting the Haar feature, draw a connection level sorter with AdaBoost method training and remove to detect car mark in the picture to be measured, this method can directly be located car mark place in the picture, is not subjected to the influence of environment.
One, the invention process provides vehicle checking method in a kind of static images, comprises training and detects two parts, it is characterized in that:
The step of training stage is as follows:
(1) make the vehicle sample:
Each color component to input picture carries out the gamma standardization, crosses dark or the lower situation of contrast to adapt to image, and the operation that the present invention adopts is to the color component computing of taking the logarithm.Samples pictures is wanted in advance with rectangle wherein vehicle to be marked.
(2) sample characteristics calculates:
Sample image after the normalization is made up image pyramid, on each layer, with a certain size window scanning, in each window, calculate the HOG feature, form the feature pyramid.
(3) training auto model:
In order to adapt to different attitudes and the occlusion issue of vehicle, the present invention has set up 2 template models to auto model, one of them is the model (because visual angle, the front and back difference of vehicle is little, the present invention shares with and represents) at visual angle before and after the vehicle, and another is the vehicular sideview model; To each template model, the present invention is divided into 6 parts with it, as Fig. 2.
Auto model of the present invention contains n=2 subtemplate, with (n+2) first array
Definition, wherein
Expression root wave filter (HOG characteristic model),
Be the model of i subtemplate, b is side-play amount.Each subtemplate comprises m=6 partial model, with m
The tlv triple definition, wherein,
Be the wave filter of j parts, size is
,
Be the fixed position of j parts with respect to root,
The loss function that departs from tram in the auto model for subtemplate.
In the characteristics of image pyramid, the position of each wave filter is
, wherein
Represent that i wave filter is at feature pyramid
To be each wave filter deduct vehicle part in position response summation separately the position of layer, corresponding response changes loss, adds side-play amount:
Wherein, H representation feature pyramid,
Expression with row major order and put among the H with
For the upper left corner
The proper vector of subwindow,
Be the skew of i parts relative fixed position,
Be deformation behaviour.
With model simplification, in order to obtain model parameter with the sorter combination:
With m unit array
Definition contains the mixture model of m group model, wherein
Be the object module of c group, wherein
Simple target model in the mixture model
In the position of each filtering be
, wherein
Be model
In the quantity of parts, the filter location of c group model is reduced to
Model parameter vector is used in the position response of mixture model equally
And vector
The some product representation.Wherein vectorial
Be the series connection of model parameter vector in each simple target model,
And vector
Show as sparse property.Adopt many group model study can obtain the different models of same target, thereby be combined into mixture model.
Adopt LSVM to learn partial model, can take full advantage of hiding Info in the image, be conducive to enrich the study of model.
The step of detection-phase is as follows:
(1) load picture to be measured:
The input picture is carried out normalization, and namely each color component to input picture carries out the gamma standardization.
(2) feature calculation:
Testing image after the normalization is made up image pyramid, on each layer, with a certain size window scanning, in each window, calculate the HOG feature, form the feature pyramid.
(3) load auto model:
The data file of load store auto model.
(4) vehicle detection:
The memory response of calculated characteristics pyramid the feature and the model filtering device,
, the parts wave filter is done the distance conversion:
Utilize near the filter response of the position of distance conversion expansion, and add the part distortion loss, improve accuracy of detection.
Be the ultimate range of i the relative root of parts position, wherein root position filtering response is placed on the l layer of this parts correspondence (x y) locates.
Utilize the summation of respective layer root filter response to add that the parts wave filter through conversion and sampling calculates the response of each layer root position:
Wherein
Be the pyramidal total number of plies of feature.
The optimized migration of calculating unit:
Utilize the root position
Be offset at optimum
Middle corresponding component locations, the location of realizing the target distortion parts, result such as Fig. 3 of seeking.
Two, a kind of car mark detection method based on the AdaBoost framework comprises training and detects two parts, and the training stage step is as follows:
(1) make the car standard specimen originally:
Comprise car target picture from network collection, and calibration vehicle mark position, extract the car picture of marking on a map according to positional information, carry out convergent-divergent according to the intrinsic length breadth ratio of car target, and by histogram equalization elimination illumination effect, as the positive sample of such car target, other do not contain car target position as negative sample to adopt vehicle pictures.
(2) sample characteristics calculates:
Construct 5 kinds of different rectangular characteristic, the corresponding a kind of Haar feature (as Fig. 4) of each rectangular characteristic, this Haar characterizing definition be correspondence the rectangular area pixel value and cum rights value sum, calculate the Haar feature by the mode of integral image.
Integral image is defined as:
Pixel in the expression original image
All pixel value sums of upper left side,
I.e. expression
Certain pixel value in zone, upper left side.
Regulation
, so only need by going or by whole image of row traversal once, can calculating corresponding integral image.And calculate the pixel value sum of certain rectangular area among the former figure, as long as by four vertex positions of rectangle in integral image, inquire about four values, the pixel value sum that certain plus and minus calculation of these four values can this rectangular area of equal value.
The Haar feature calculation:
First kind of feature:
Second kind of feature:
The third feature:
The 4th kind of feature:
The 5th kind of feature:
To n sample calculation feature of input, comprising m positive sample and n-m negative sample, each sample has two attributes
, wherein x represents the Haar proper vector of this sample
, y represents the classification of this sample, for positive sample is to be taken as 1, is taken as-1 during negative sample.
(3) training connection level sorter:
Be sample
The Haar proper vector,
Be sign of inequality direction controller, value is+1 or-1,
Be Weak Classifier training threshold value.
To each feature
, generate its corresponding Weak Classifier
, calculate the error with respect to current weight:
Selection has least error
Weak Classifier
Join in the strong classifier and go, the rectangular characteristic information of record feature correspondence this moment, and upgrade weight:
Wherein,
Under the frequency of training T that sets, training can produce a strong classifier at every turn, and has a plurality of Weak Classifiers to be selected in this process simultaneously, finally then each strong classification and corresponding a plurality of Weak Classifiers thereof is together in series, and forms final connection level sorter.
The detection-phase step is as follows:
(1) loads picture to be measured: transfer gray-scale map to and carry out histogram equalization;
(2) load car mark sorter: training gained classifier data is kept in the corresponding txt file, and wherein data structure is described as:
N strong classifier, the
Individual strong classifier comprises the threshold value of strong classifier
With
Individual Weak Classifier;
Wherein,
jIndividual Weak Classifier comprises the threshold value of Weak Classifier
, direction controller
, coefficient
And the rectangular characteristic information of the selected feature correspondence of this Weak Classifier: the number of rectangle
, type under the rectangular characteristic
, the positional information of each sub-rectangle and weight
(3) connection level car mark detects: in the connection level sorter, the strong classifier one-level is than one-level complexity, detected image is at first passed through the detection of the strong classifier of front, if not the car picture of marking on a map, can be excluded at front end so, having only car to mark picture on a map could final detection by strong classifiers at different levels, and a large amount of non-cars picture of marking on a map will be excluded when preceding what strong classifier detection, process such as Fig. 5, result such as Fig. 6.
Claims (12)
1. the method that the purpose of this invention is to provide a kind of detection and Identification of static images vehicle characteristics, and provide a kind of system of detection and Identification of the vehicle characteristics at the internet picture, comprise: network chart sheet to be detected is carried out vehicle detection and judges the attitude of vehicle, detect the identification vehicle thereby carry out the car mark in the front region of vehicle or back region.
2. one, the invention process provides vehicle checking method in a kind of static images, comprise training and detect two parts.
3. the training stage comprises: by making the vehicle sample, the input picture is carried out normalized; Carry out sample characteristics then and calculate, to the picture construction image pyramid after the normalization; Next, the training auto model imports the characteristic data set of sample into training classifier study; Produce the mixture model of root model, partial model and the corresponding variable shape partial model of vehicle by study; Detection-phase comprises: load picture to be measured: the input picture is carried out normalization, and namely each color component to input picture carries out the gamma standardization; Carry out feature calculation: the testing image after the normalization is made up image pyramid; Load auto model, the data file of load store auto model; Carry out vehicle detection at last, scan the zone that is complementary with the units of variance model at the feature pyramid and realize the vehicle detection location.
4. two, the invention process provides the car mark detection method based on the AdaBoost framework, comprise training and detect two parts.
5. the training stage comprises: make the car standard specimen this, comprise car target picture from network collection, and calibration vehicle mark position, extract the car picture of marking on a map according to positional information; Sample characteristics calculates, structure rectangular characteristic, the corresponding a kind of Haar feature of each rectangular characteristic; The training sample that a training connection level sorter, input are obtained by last step is also trained, and the strong classification and the corresponding a plurality of Weak Classifiers thereof that obtain of training are together in series the most at last; Detection-phase comprises: load picture to be measured: transfer gray-scale map to and carry out histogram equalization; Load car mark sorter, comprise strong, the threshold value of Weak Classifier and the rectangular characteristic information of selected feature correspondence; A connection level car mark detects, and detected image is the detection of the strong classifier by the front at first, if not the car picture of marking on a map, can be excluded at front end so, have only car mark on a map picture could be finally detection by strong classifiers at different levels.
6. training stage making vehicle sample as claimed in claim 1 and detection-phase load the method for picture to be measured, it is characterized in that: each color component to input picture carries out the gamma standardization, cross dark or the lower situation of contrast to adapt to image, the operation of adopting is to the color component computing of taking the logarithm, and samples pictures is wanted in advance with rectangle wherein vehicle to be marked.
7. training stage sample characteristics Calculation Method as claimed in claim 2 is characterized in that: to the sample after the normalization, on each layer, with a certain size window scanning, calculate the HOG feature in each window, form the feature pyramid.
8. the training stage as claimed in claim 3 is trained the method for auto model, it is characterized in that: in order to adapt to different attitudes and the occlusion issue of vehicle, auto model has been set up 2 template models, and one of them is the model at visual angle before and after the vehicle, and another is the vehicular sideview model.
9. the method for detection-phase vehicle detection as claimed in claim 4 is characterized in that: calculated characteristics pyramid
lLayer feature and the
iThe memory response of individual model filtering device is done the distance conversion with the parts wave filter, utilizes the filter response apart near the position conversion expansion, and adds part distortion loss, improves accuracy of detection;
D I, j (x, y)Be
iThe ultimate range of the relative root of individual parts position, wherein root position filtering response is placed on the of this parts correspondence
lLayer (
x,
y) locate, utilize the summation of respective layer root filter response to add that the parts wave filter through conversion and sampling calculates the response of each layer root position; Then, the optimized migration of calculating unit, utilize the root position (
x 0,
y 0,
l 0) in the optimum skew, seek corresponding component locations, realize the location of target distortion parts.
10. the training stage based on the AdaBoost framework as claimed in claim 5 is made car standard specimen method originally, it is characterized in that: extract the car picture of marking on a map according to positional information, carry out convergent-divergent according to the intrinsic length breadth ratio of car target, and by histogram equalization elimination illumination effect, as the positive sample of such car target, other do not contain car target position as negative sample to adopt vehicle pictures.
11. the training stage sample characteristics Calculation Method based on the AdaBoost framework as claimed in claim 6, it is characterized in that: the Haar characterizing definition for corresponding rectangular area pixel value and cum rights value sum, calculate the Haar feature by the mode of integral image; Integral image SAT (
x,
y) expression original image in pixel (
x,
y) all pixel value sums of upper left side, adopt incremental mode to calculate, so only need travel through whole image once by row or by row, can calculate corresponding integral image; And calculate the pixel value sum of certain rectangular area among the former figure, as long as by four vertex positions of rectangle in integral image, inquire about four values, the pixel value sum that certain plus and minus calculation of these four values can this rectangular area of equal value.
12. the method that joins the level sorter based on the training stage training of AdaBoost framework as claimed in claim 7, it is characterized in that: under the frequency of training T that sets, each training can produce a strong classifier, and have a plurality of Weak Classifiers to be selected in this process simultaneously, finally then each strong classification and corresponding a plurality of Weak Classifiers thereof are together in series, form final connection level sorter.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2012100428091A CN103295021A (en) | 2012-02-24 | 2012-02-24 | Method and system for detecting and recognizing feature of vehicle in static image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2012100428091A CN103295021A (en) | 2012-02-24 | 2012-02-24 | Method and system for detecting and recognizing feature of vehicle in static image |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103295021A true CN103295021A (en) | 2013-09-11 |
Family
ID=49095850
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2012100428091A Pending CN103295021A (en) | 2012-02-24 | 2012-02-24 | Method and system for detecting and recognizing feature of vehicle in static image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103295021A (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103559501A (en) * | 2013-10-25 | 2014-02-05 | 公安部第三研究所 | Vehicle sun visor detecting method and device based on image analysis |
CN103984950A (en) * | 2014-04-22 | 2014-08-13 | 北京联合大学 | Moving vehicle stop lamp state recognition method adaptable to day detection |
CN105225523A (en) * | 2015-10-15 | 2016-01-06 | 浙江宇视科技有限公司 | A kind of parking space state detection method and device |
CN105574490A (en) * | 2015-12-10 | 2016-05-11 | 金鹏电子信息机器有限公司 | Vehicle brand identification method and system based on headlight image characteristics |
CN105809184A (en) * | 2015-10-30 | 2016-07-27 | 哈尔滨工程大学 | Vehicle real-time identification tracking and parking space occupancy determining method suitable for gas station |
CN105956632A (en) * | 2016-05-20 | 2016-09-21 | 浙江宇视科技有限公司 | Target detection method and device |
CN106203446A (en) * | 2016-07-05 | 2016-12-07 | 中国人民解放军63908部队 | Three dimensional object recognition positioning method for augmented reality auxiliary maintaining system |
CN106909884A (en) * | 2017-01-17 | 2017-06-30 | 北京航空航天大学 | A kind of hand region detection method and device based on hierarchy and deformable part sub-model |
CN107220650A (en) * | 2017-05-31 | 2017-09-29 | 广州炒米信息科技有限公司 | Food image detection method and device |
CN107229906A (en) * | 2017-05-08 | 2017-10-03 | 上海工程技术大学 | A kind of automobile overtaking's method for early warning based on units of variance model algorithm |
CN107273923A (en) * | 2017-06-02 | 2017-10-20 | 浙江理工大学 | A kind of construction method of weaving face fabric friction sound wave discriminator |
WO2018036277A1 (en) * | 2016-08-22 | 2018-03-01 | 平安科技(深圳)有限公司 | Method, device, server, and storage medium for vehicle detection |
CN108052955A (en) * | 2017-11-03 | 2018-05-18 | 中国科学院计算技术研究所 | A kind of high-precision braille recognition methods and system |
CN109344684A (en) * | 2018-08-03 | 2019-02-15 | 广东工业大学 | Battery-operated motor cycle image detecting method based on units of variance model |
CN110633751A (en) * | 2019-09-17 | 2019-12-31 | 上海眼控科技股份有限公司 | Training method of car logo classification model, car logo identification method, device and equipment |
CN111080612A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Truck bearing damage detection method |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7715597B2 (en) * | 2004-12-29 | 2010-05-11 | Fotonation Ireland Limited | Method and component for image recognition |
CN102419820A (en) * | 2011-08-18 | 2012-04-18 | 电子科技大学 | Method for rapidly detecting car logo in videos and images |
-
2012
- 2012-02-24 CN CN2012100428091A patent/CN103295021A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7715597B2 (en) * | 2004-12-29 | 2010-05-11 | Fotonation Ireland Limited | Method and component for image recognition |
CN102419820A (en) * | 2011-08-18 | 2012-04-18 | 电子科技大学 | Method for rapidly detecting car logo in videos and images |
Non-Patent Citations (2)
Title |
---|
林志阳 等: "基于Adaboost的车标定位方法", 《计算机工程》, vol. 34, no. 11, 30 June 2008 (2008-06-30), pages 206 - 207 * |
胡正平 等: "HOG特征混合模型结合隐SVM的感兴趣目标检测定位算法", 《信号处理》, vol. 27, no. 8, 31 August 2011 (2011-08-31), pages 1206 - 1212 * |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103559501A (en) * | 2013-10-25 | 2014-02-05 | 公安部第三研究所 | Vehicle sun visor detecting method and device based on image analysis |
CN103984950A (en) * | 2014-04-22 | 2014-08-13 | 北京联合大学 | Moving vehicle stop lamp state recognition method adaptable to day detection |
CN103984950B (en) * | 2014-04-22 | 2017-07-14 | 北京联合大学 | A kind of moving vehicle brake light status recognition methods for adapting to detection on daytime |
CN105225523A (en) * | 2015-10-15 | 2016-01-06 | 浙江宇视科技有限公司 | A kind of parking space state detection method and device |
CN105809184A (en) * | 2015-10-30 | 2016-07-27 | 哈尔滨工程大学 | Vehicle real-time identification tracking and parking space occupancy determining method suitable for gas station |
CN105809184B (en) * | 2015-10-30 | 2020-06-16 | 哈尔滨工程大学 | Method for real-time vehicle identification and tracking and parking space occupation judgment suitable for gas station |
CN105574490A (en) * | 2015-12-10 | 2016-05-11 | 金鹏电子信息机器有限公司 | Vehicle brand identification method and system based on headlight image characteristics |
CN105574490B (en) * | 2015-12-10 | 2019-04-09 | 金鹏电子信息机器有限公司 | Vehicle brand recognition methods and system based on headlight characteristics of image |
CN105956632A (en) * | 2016-05-20 | 2016-09-21 | 浙江宇视科技有限公司 | Target detection method and device |
CN105956632B (en) * | 2016-05-20 | 2019-06-11 | 浙江宇视科技有限公司 | A kind of method and apparatus detecting target |
CN106203446A (en) * | 2016-07-05 | 2016-12-07 | 中国人民解放军63908部队 | Three dimensional object recognition positioning method for augmented reality auxiliary maintaining system |
CN106203446B (en) * | 2016-07-05 | 2019-03-12 | 中国人民解放军63908部队 | Three dimensional object recognition positioning method for augmented reality auxiliary maintaining system |
WO2018036277A1 (en) * | 2016-08-22 | 2018-03-01 | 平安科技(深圳)有限公司 | Method, device, server, and storage medium for vehicle detection |
CN106909884A (en) * | 2017-01-17 | 2017-06-30 | 北京航空航天大学 | A kind of hand region detection method and device based on hierarchy and deformable part sub-model |
CN106909884B (en) * | 2017-01-17 | 2020-08-04 | 北京航空航天大学 | Hand region detection method and device based on layered structure and deformable part model |
CN107229906A (en) * | 2017-05-08 | 2017-10-03 | 上海工程技术大学 | A kind of automobile overtaking's method for early warning based on units of variance model algorithm |
CN107220650A (en) * | 2017-05-31 | 2017-09-29 | 广州炒米信息科技有限公司 | Food image detection method and device |
CN107220650B (en) * | 2017-05-31 | 2020-07-17 | 广州炒米信息科技有限公司 | Food image detection method and device |
CN107273923A (en) * | 2017-06-02 | 2017-10-20 | 浙江理工大学 | A kind of construction method of weaving face fabric friction sound wave discriminator |
CN107273923B (en) * | 2017-06-02 | 2020-09-29 | 浙江理工大学 | Construction method of textile fabric friction sound wave discriminator |
CN108052955A (en) * | 2017-11-03 | 2018-05-18 | 中国科学院计算技术研究所 | A kind of high-precision braille recognition methods and system |
CN108052955B (en) * | 2017-11-03 | 2021-03-05 | 中国科学院计算技术研究所 | High-precision Braille identification method and system |
CN109344684A (en) * | 2018-08-03 | 2019-02-15 | 广东工业大学 | Battery-operated motor cycle image detecting method based on units of variance model |
CN110633751A (en) * | 2019-09-17 | 2019-12-31 | 上海眼控科技股份有限公司 | Training method of car logo classification model, car logo identification method, device and equipment |
CN111080612A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Truck bearing damage detection method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103295021A (en) | Method and system for detecting and recognizing feature of vehicle in static image | |
CN105046196B (en) | Front truck information of vehicles structuring output method based on concatenated convolutional neutral net | |
CN109657632B (en) | Lane line detection and identification method | |
CN102509091B (en) | Airplane tail number recognition method | |
CN109165623B (en) | Rice disease spot detection method and system based on deep learning | |
CN108154102B (en) | Road traffic sign identification method | |
CN102096821B (en) | Number plate identification method under strong interference environment on basis of complex network theory | |
CN102968637B (en) | Complicated background image and character division method | |
CN104077577A (en) | Trademark detection method based on convolutional neural network | |
CN105574550A (en) | Vehicle identification method and device | |
CN104166841A (en) | Rapid detection identification method for specified pedestrian or vehicle in video monitoring network | |
CN105913040A (en) | Real time double cameras provided pedestrian detection system for use under scotopic vision conditions | |
CN104036284A (en) | Adaboost algorithm based multi-scale pedestrian detection method | |
CN111325146B (en) | Truck type and axle type identification method and system | |
CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
CN109978872B (en) | White matter microstructure characteristic screening system and method based on white matter fiber tracts | |
Merabet et al. | Building roof segmentation from aerial images using a line-and region-based watershed segmentation technique | |
CN104766344B (en) | Vehicle checking method based on movement edge extractor | |
CN108090459B (en) | Traffic sign detection and identification method suitable for vehicle-mounted vision system | |
CN110659550A (en) | Traffic sign recognition method, traffic sign recognition device, computer equipment and storage medium | |
CN111967313A (en) | Unmanned aerial vehicle image annotation method assisted by deep learning target detection algorithm | |
CN103065163B (en) | A kind of fast target based on static images detects recognition system and method | |
CN110223310A (en) | A kind of line-structured light center line and cabinet edge detection method based on deep learning | |
CN103914829B (en) | Method for detecting edge of noisy image | |
CN110659637A (en) | Electric energy meter number and label automatic identification method combining deep neural network and SIFT features |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130911 |