CN102169631A - Manifold-learning-based traffic jam event cooperative detecting method - Google Patents
Manifold-learning-based traffic jam event cooperative detecting method Download PDFInfo
- Publication number
- CN102169631A CN102169631A CN2011101009103A CN201110100910A CN102169631A CN 102169631 A CN102169631 A CN 102169631A CN 2011101009103 A CN2011101009103 A CN 2011101009103A CN 201110100910 A CN201110100910 A CN 201110100910A CN 102169631 A CN102169631 A CN 102169631A
- Authority
- CN
- China
- Prior art keywords
- prototype
- manifold
- traffic
- texture
- classification
- 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 relates to the technical field of intelligent identification of traffic state, in particular to a manifold-learning-based traffic jam event cooperative detecting method. The method is characterized by comprising a training and learning process, an event detecting process and a feedback process, wherein the training and learning process is to acquire typical images of each traffic state for forming a traffic jam event training image collection, perform texture conversion and manifold dimensionality reduction treatment, select a model prototype pattern and obtain an event classification prototype pattern collection and a manifold cooperative detection model; the event detecting process is to acquire traffic video frame image in real time, perform texture conversion and manifold dimensionality reduction treatment, transmit to a manifold cooperative detecting module, perform traffic jam event classification detection and obtain the result of the detection; and the feedback process is to compare the detection result of adjacent frame images and update or correct corresponding event classification prototype, The method helps to improve the detection accuracy rate of traffic jam events and has high robustness and good use effect.
Description
Technical field
The present invention relates to traffic behavior intelligent identification technology field, particularly a kind of traffic congestion incident collaborative detection method based on manifold learning.
Background technology
The traffic congestion incident may be defined as the traffic behavior that the average stroke speed of a motor vehicle has been lower than certain limit.This moment, road traffic flow played pendulum, and traffic flow inside has little disturbance just will produce big operation problem, even interruption of communication takes place.Vehicle often standed in a long line when this incident took place, and whole fleet can loiter.
Traditional traffic congestion event detector has coil, microwave, infrared ray etc. several, these detecting devices are earlier by detecting traffic parameters such as the speed of a motor vehicle, vehicle flowrate, queue length basically, then, detect the traffic congestion incident by some predictions or indirect conversion algorithm again.Existing relatively more classical detection method comprises California algorithm, McMaster algorithm, exponential smoothing etc.Development along with technology, differentiating algorithm is also improving, the Video Detection algorithm has also obtained development fast, but they all respectively have quality, the performance that does not have a kind of algorithm is better than other algorithms fully, because various algorithms all are based on certain thought or theory, its qualifications and applicability are arranged all also.
Along with the continuous development of computing machine and image processing techniques, utilizing video detector to carry out the traffic events detection has become a kind of potential especially method, is expected to replace the important component part that conventional detector becomes intelligent transportation system in recent years.Video detector is compared with traditional detecting device, have that processing speed is fast, installation and maintenance are convenient and but the lower monitoring range of expense is wide, can obtain plurality of advantages such as a greater variety of traffic parameters, in intelligent transportation system, obtained application more and more widely.
The at present widely used Indirect Detecting Method that is based on the video frequency vehicle tracking in countries in the world: at first, automatically extract traffic parameter (speed of a motor vehicle, vehicle flowrate, queue length etc.) by image processing techniques, then, according to relevant traffic parameter indirect detection traffic congestion incident.This kind method is compared with traditional detection method based on coil, just the extraction of traffic parameter has certain difference, but similar algorithms is still adopted in the automatic detection of the traffic congestion incident of core, we can say the raising that does not have essence in the automatic context of detection of traffic congestion incident.Because there are a lot of Unpredictabilities in the diversity of road traffic environment and traffic congestion incident when taking place, existing video detection technology is subject to the interference of external environment conditions such as vehicle running state or weather conditions, and the error of background modeling, target extraction and each step of vehicle tracking all can directly influence the automatic extraction of traffic parameter.In addition, indirectly video detecting method relatively relies on the extraction precision of traffic parameter, and the error of parameter extraction indirect detection algorithm often is difficult to remedy.Therefore, the block up method of incident of this indirect detection based on video technique still exists bigger limitation in utilization.
Video detecting method is meant the method for using image processing technique and intellectual technology to discern traffic parameter and state, this method utilizes video camera to obtain vision signal, by the video signal detection road traffic condition, convert vision signal to digital picture by image processing equipment, again by computing machine traffic parameter and the status detection identification of being correlated with.It is the most various detection method of a class, and existing method commonly used has: method one is to utilize information such as the shape of vehicle, color, symmetry, carries out coherent detection in conjunction with general knowledge information such as road and shades.This method is simple, directly perceived, is easy to programming and realizes, but need to estimate a plurality of empirical values, as the gray difference threshold value of minimum length, vehicle shadow and the road of the experience ratio of vehicle length and width, vehicle edge etc.Accurately whether empirical value is directly connected to the quality that detects performance.Method two is to utilize the great deal of related information that exists between the sequence image to carry out vehicle detection, and this method is not known any information of scene in advance, and is applicable to the video camera situation of movement.This method shortcoming is consuming time big, and is bad to too complicated, too fast or slow excessively motion detection effect.Method three is directly to detect the method that interframe changes, and this method is quick and easy, and real-time is better, is fit to the fast and bigger moving target of deformation of motion, but is not suitable for the scene of global motion, as the more road of unevenness or detour etc.
Summary of the invention
The objective of the invention is to overcome the deficiencies in the prior art, a kind of traffic congestion incident collaborative detection method based on manifold learning is provided, this method not only helps improving the detection accuracy rate of traffic congestion incident, and has stronger robustness, and result of use is good.
For achieving the above object, the technical solution used in the present invention is: a kind of traffic congestion incident collaborative detection method based on manifold learning, it is characterized in that: this method comprises training study process, event detection procedure and feedback procedure, and described training study process is carried out as follows:
Step 1.1: the traffic jam level classification is defined, and the typical image of gathering corresponding to each traffic jam level classification forms traffic congestion incident training atlas;
Step 1.2: described each traffic congestion incident training atlas is carried out texture transformation, generate traffic congestion incident texture atlas;
Step 1.3: described traffic congestion incident texture atlas is carried out manifold dimension-reducing handle, obtain dimensionality reduction texture atlas;
Step 1.4: corresponding to each traffic jam level classification, carry out prototype pattern according to dimensionality reduction texture atlas and select, obtain event classification prototype collection and based on the collaborative detection model of stream shape of described event classification prototype collection;
Described event detection procedure is carried out as follows:
Step 2.1: gather the traffic video two field picture in real time;
Step 2.2: each two field picture is carried out texture transformation, obtain real-time texture maps;
Step 2.3: described real-time texture maps is carried out manifold dimension-reducing handle, obtain real-time dimensionality reduction texture maps;
Step 2.4: described real-time dimensionality reduction texture maps is sent into the collaborative detection model of described stream shape carry out the detection of traffic congestion event classification, testing result that obtains being correlated with and correlation parameter;
Described feedback procedure carries out as follows:
The testing result that compares the consecutive frame image, if sudden change does not appear in testing result, the event classification prototype of the classification under it of then real-time dimensionality reduction texture maps information being superposeed into, upgrading prototype pattern, and discern next the collection in real time and the sample to be detected of process texture transformation and manifold dimension-reducing with the prototype pattern after upgrading; Otherwise the real-time dimensionality reduction texture maps of feedback false retrieval is revised events corresponding classification prototype.
The invention has the beneficial effects as follows synergetic neural network technology and the manifold learning method utilized, adopt according to blocking up and the direct video method of the vehicle distribution configuration feature of the non-road conditions of blocking up detects the incident of blocking up, this method has stronger robustness, be not subject to the interference of local conditions, environmental change is little to the influence of testing result comparatively speaking.Simultaneously, this method can improve the verification and measurement ratio of traffic congestion incident, reduces misclassification rate, accelerates detection speed, is expected to further improve the Video Detection level of traffic congestion incident, has very strong practicality, has broad application prospects.
Description of drawings
Fig. 1 is the implementation structure synoptic diagram of the inventive method.
Embodiment
The present invention is based on the traffic congestion incident collaborative detection method of manifold learning, it is characterized in that: this method comprises training study process, event detection procedure and feedback procedure, and described training study process is carried out as follows:
Step 1.1: the traffic jam level classification is defined, and the typical image of gathering corresponding to each traffic jam level classification forms traffic congestion incident training atlas;
Step 1.2: described each traffic congestion incident training atlas is carried out texture transformation, generate traffic congestion incident texture atlas;
Step 1.3: described traffic congestion incident texture atlas is carried out manifold dimension-reducing handle, obtain dimensionality reduction texture atlas;
Step 1.4: corresponding to each traffic jam level classification, carry out prototype pattern according to dimensionality reduction texture atlas and select, obtain event classification prototype collection and based on the collaborative detection model of stream shape of described event classification prototype collection;
Described event detection procedure is carried out as follows:
Step 2.1: gather the traffic video two field picture in real time;
Step 2.2: each two field picture is carried out texture transformation, obtain real-time texture maps;
Step 2.3: described real-time texture maps is carried out manifold dimension-reducing handle, obtain real-time dimensionality reduction texture maps;
Step 2.4: described real-time dimensionality reduction texture maps is sent into the collaborative detection model of described stream shape carry out the detection of traffic congestion event classification, testing result that obtains being correlated with and correlation parameter;
Described feedback procedure carries out as follows:
The testing result that compares the consecutive frame image, if sudden change does not appear in testing result, the event classification prototype of the classification under it of then real-time dimensionality reduction texture maps information being superposeed into, upgrading prototype pattern, and discern next the collection in real time and the sample to be detected of process texture transformation and manifold dimension-reducing with the prototype pattern after upgrading; Otherwise the real-time dimensionality reduction texture maps of feedback false retrieval is revised events corresponding classification prototype.
Specific embodiments to this method committed step is described further below.
1, texture transformation algorithm:The method of texture analysis is a lot, as co-occurrence matrix, and GAbor small echo, FRACTAL DIMENSION, texture spectrum etc.The texture properties that co-occurrence matrix extracts lacks visual similarity; Though and the GAbor small echo meets human visual signature most, its calculated amount is too big, is difficult to satisfy real-time processing requirements; The fractal dimension threshold value is determined difficulty.And texture spectrum algorithm is simple, and texture cell is particularly suitable for the representing grain roughness, and therefore, the present invention selects texture cell to show the traffic texture image.
Texture image can be counted as being made of one group of texture cell, and texture cell has been represented a pixel and the local grain feature that neighborhood territory pixel constituted thereof, and then can represent the textural characteristics of entire image to the statistics of the texture cell of entire image.The texture cell representation is to consider the gray-scale relation of center pixel and 8 neighborhood territory pixels in 3 * 3 windows.The definition texture cell is a set that comprises 8 pixels:
TU={E 1 , E 2 ..., E 8 ,
Wherein
I=1,2 ..., 8,
V 0 Be processed pixel,
V i Be its neighborhood territory pixel,
ΔBe the grey scale change threshold value.
Definition texture cell mark value
N TU For:
2, manifold learning dimensionality reduction algorithm:Manifold learning is intended to find the inherent law of high dimensional data collection distribution, its basic thought is: the point in the higher-dimension observation space grows up to a stream shape by the acting in conjunction of minority independent variable at observation space, if can launch stream shape or the inherent primary variables of discovery that observation space curls effectively, just can carry out dimensionality reduction to this data set.Manifold learning definition: order
Be
The space
The dimension territory, the present is a smooth embedding, wherein.Data point is generated by certain stochastic process, warp
Mapping forms the data of observation space
The target of manifold learning is will be from observation data
Middle reconstruct
With
The general title
Be latent space,
Be latent data.
Manifold learning has become a hot research field of machine learning at present, mainly contains four class manifold learning methods: neural network, main flow shape, core principle component analysis and embedding grammar.Wherein, embedding grammar has characteristics such as thought is deep, algorithm simple, global optimum with it, becomes the focus direction of manifold learning.Since published on the Science magazine in 2000 use algorithm that embedding grammar LLE and Isomap carry out manifold learning and test after, LLE and Isomap method become the outstanding representative of manifold learning, the former can be seen as the representative of partial approach, and the latter can regard the representative of global approach as, and the concern that has caused numerous theories and used with its advantage separately.The embedding grammar that occurs is of a great variety at present, and it is different that the difference between the various embedding grammars is that it makes hypothesis to embedding, and perhaps the part between the data of description is different with the mode of holotopy.But most local embedding grammar all exist in various degree owe the match phenomenon, be that reconstruct stream shape can not be approached data acquisition effectively, this phenomenon is to cause owing to same neighborhood that the sample of zones of different that current embedding grammar will be arranged in stream shape is mapped to latent space, its profound reason then is the global information that current many embedding grammars have been ignored sample set, and global approach too strictly limits the manifold distance that keeps between the sample, thus the effect that some can't be embedded into data set in the low-dimensional Euclidean space under the condition that keeps overall manifold distance on, is difficult to obtain.Therefore, how guaranteeing geometric relationship and distance measure unchangeability between data when the stream shape of higher dimensional space maps to the stream shape of lower dimensional space, is the key that embeds the manifold learning method.
The present invention utilizes known classification information that the selection of neighborhood point is exercised supervision, and simultaneously according to the correlativity between the classification information, adaptively selected each sample neighborhood of a point yardstick merges local and global information realizes embedding manifold learning arithmetic.In addition, stream shape algorithm all compares sensitivity to noise and algorithm parameter, and the existence of noise makes that input parameter is difficult to select more, and the less variation of parameter can cause the learning outcome of significant difference.At this problem, the present invention adopts pretreated scheme that source images is carried out the texture image conversion, with the input sample of the data after the conversion as manifold learning, so not only can reduce sample information amount, and minimizing interference of noise, simultaneously, can also obtain the roughness classification information with the supervision manifold learning.
3, the collaborative model of cognition of stream shape:The basic ideas that the present invention proposes based on the collaborative model of cognition of manifold learning are at first to use non-linear embedding manifold learning arithmetic traffic texture image data are carried out learning to the mapping of lower dimensional space stream shape from the stream shape of higher dimensional space, obtain the input sample mode of the collaborative model of cognition of stream shape matrix conduct in the corresponding lower dimensional space.Therefore, pattern that occurs in the below collaborative model of cognition or sample are the low-dimensional stream shape matrix that obtains through manifold learning.
According to the basic thought of synergetics, mode identification procedure can be understood as some preface parameter process of competition.For treating test model
q, can construct a dynamic process, it can " draw "
q, make it through intermediateness
Q (t)Enter into a prototype pattern of all prototype patterns
v k , promptly
v k With
Q (0)The most close, that is to say and draw
qIt is in
v k The attraction the lowest point, this process can be described as:
In image model identification, image array is converted into one-dimensional vector earlier, and it is normalized to the column vector with zero-mean and unit length.The kinetics equation of image model identification is:
(1)
Wherein,
qBe test model vector, λ
kBe attention parameters, have only the λ of working as
k0 o'clock, pattern just can be identified,
v k Be the prototype pattern vector,
v k +For
v k Adjoint vector, it must satisfy:
q + With
qAlso must satisfy same relation.
B Kk ' , C is constant coefficient, comprise
B Kk ' Be used for distinguishing between a plurality of patterns, the item that comprises C is used for restriction
qExponential increase, F (t) is a fluctuating force, can ignore generally speaking.
In order to reduce dimension, can introduce the preface parameter
x k Redescribe equation (1).The preface parameter of system
x k Be described as
qUnder the least square meaning in
v k On projection,
Be residual vector:
, (2),
The kinetics equation of pattern-recognition (4) can be expressed as 3 layer networks, and therefore, the Synergetic Pattern Recognition model is also referred to as synergetic neural network, and the output of network is projected to the 3rd layer, according to formula: (5)
Realize the association function of network, finally finish the identifying of synergetic neural network.
The structure of the selection of prototype pattern vector, preface parameter and the setting of attention parameters are the keys of Synergetic Pattern Recognition.
1) selection of prototype pattern: the selection of prototype has determined the recognition capability of cooperation model.Prototype selects main method to have---based on the experience back-and-forth method, based on the genetic algorithm back-and-forth method, based on the clustering procedure back-and-forth method etc.Experience back-and-forth method usable range is limited, and requirement must be understood object of classification is deep comprehensively; The genetic algorithm training time is very long and training effectiveness is low; Clustering procedure can be carried out autopolymerization to given sample, and time complexity is also low than genetic algorithm, and merges multisample information in the prototype stream shape pattern.The present invention utilizes dynamically stack real-time information to carry out the prototype integrated mode and selects according to the characteristics of traffic image diversity, real-time.Owing to a kind of traffic congestion incident multiple vehicle combination distribution texture form is arranged, a conventional kind selects the method for a prototype not to be suitable for the identification of polymorphic similar sample.To this, the present invention selects a plurality of prototypes of different shape in generic sample, as such other prototype collection, at first concentrate prototype to carry out similar coupling competition in sample to be detected to prototype, getting like the maximal phase prototype exports as optimum prototype, and then sample to be detected carried out similar coupling to optimum prototype of all categories respectively, with maximum analog result as classification results.Simultaneously, test sample book is upgraded the prototype pattern of classification under it as feedback quantity, discern the test sample book of next collection in real time and process texture transformation and manifold dimension-reducing with the prototype pattern after upgrading, so continuous immediate transport information of time is added in the prototype, information that superposes in the prototype and sample acquisition time to be tested differed within several seconds, have stronger consistance, can effectively solve interference such as light, shake.
2) setting of preface parameter structure and attention parameters: the Synergetic Pattern Recognition process can be considered some preface parameter process of competition, and classical construction algorithm is the M-P generalized inverse matrix of directly asking by prototype pattern matrix that vector is formed, and obtains the preface parameter then.This preface parameter of asking based on pseudoinverse technique can reflect the similarity degree between input pattern and the prototype pattern well, but required operand is very big, and adopt and to compete optimum prototype method at the prototype collection and ask the types of tissue object prototype, to different test sample books, optimum prototype is also inequality, need recomputate the M-P generalized inverse matrix, calculated amount is bigger, and is impracticable.For fear of finding the inverse matrix, introduce the measure of geodesic line distance between the manifold learning sample, as the distance evaluation function, directly close series structure preface parameter according to manifold distance between test sample book and each prototype pattern, solve quick problem concerning study.
What attention parameters was provided with employing in the cooperation model is static set-up mode, promptly only attention parameters is set according to prototype pattern, and traffic image test sample book and prototype sample have certain gap, attention parameters will be provided with according to the relation between test sample book and the prototype pattern, allow those prototype patterns more close of contract network " concern ", further improve the recognition efficiency of contract network with test sample book.Simultaneously, allow residual vector come balance " concern " degree, promptly allow the size of residual vector, affect the size of attention parameters " concern " degree.
3) feedback modifiers, adjustment study: the pattern that misclassification rate is the highest in the learning sample is revised prototype pattern as feedback quantity, further improve the study extended capability of system.Simultaneously, have the characteristic of time remaining,, and recognition result is assessed constantly to the judging of the congestion status argument sequence of real-time continuous frame according to the traffic congestion incident.If assessment result is inconsistent, the real-time traffic dimensionality reduction texture maps that the feedback mistake is known is revised the prototype classification in real time.
More than be preferred embodiment of the present invention, all changes of doing according to technical solution of the present invention, the function that is produced does not exceed
During the scope of technical solution of the present invention, all belong to protection scope of the present invention.
Claims (2)
1. traffic congestion incident collaborative detection method based on manifold learning, it is characterized in that: this method comprises training study process, event detection procedure and feedback procedure, described training study process is carried out as follows:
Step 1.1: the traffic jam level classification is defined, and the typical image of gathering corresponding to each traffic jam level classification forms traffic congestion incident training atlas;
Step 1.2: described each traffic congestion incident training atlas is carried out texture transformation, generate traffic congestion incident texture atlas;
Step 1.3: described traffic congestion incident texture atlas is carried out manifold dimension-reducing handle, obtain dimensionality reduction texture atlas;
Step 1.4: corresponding to each traffic jam level classification, carry out prototype pattern according to dimensionality reduction texture atlas and select, obtain event classification prototype collection and based on the collaborative detection model of stream shape of described event classification prototype collection;
Described event detection procedure is carried out as follows:
Step 2.1: gather the traffic video two field picture in real time;
Step 2.2: each two field picture is carried out texture transformation, obtain real-time texture maps;
Step 2.3: described real-time texture maps is carried out manifold dimension-reducing handle, obtain real-time dimensionality reduction texture maps;
Step 2.4: described real-time dimensionality reduction texture maps is sent into the collaborative detection model of described stream shape carry out the detection of traffic congestion event classification, testing result that obtains being correlated with and correlation parameter;
Described feedback procedure carries out as follows:
The testing result that compares the consecutive frame image, if sudden change does not appear in testing result, the event classification prototype of the classification under it of then real-time dimensionality reduction texture maps information being superposeed into, upgrading prototype pattern, and discern next the collection in real time and the sample to be detected of process texture transformation and manifold dimension-reducing with the prototype pattern after upgrading; Otherwise the real-time dimensionality reduction texture maps of feedback false retrieval is revised events corresponding classification prototype.
2. the traffic congestion incident collaborative detection method based on manifold learning according to claim 1, it is characterized in that: in the traffic congestion event classification of step 2.4 detects, the system of selection of prototype pattern is: a plurality of prototypes of selecting different shape in generic sample, as such other prototype collection, at first concentrate each prototype to carry out similar coupling competition in sample to be detected to prototype, getting like the maximal phase prototype exports as optimum prototype, and then sample to be detected carried out similar coupling to optimum prototype of all categories respectively, with maximum analog result as classification results.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2011101009103A CN102169631A (en) | 2011-04-21 | 2011-04-21 | Manifold-learning-based traffic jam event cooperative detecting method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2011101009103A CN102169631A (en) | 2011-04-21 | 2011-04-21 | Manifold-learning-based traffic jam event cooperative detecting method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102169631A true CN102169631A (en) | 2011-08-31 |
Family
ID=44490779
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2011101009103A Pending CN102169631A (en) | 2011-04-21 | 2011-04-21 | Manifold-learning-based traffic jam event cooperative detecting method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102169631A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102768801A (en) * | 2012-07-12 | 2012-11-07 | 复旦大学 | Method for detecting motor vehicle green light follow-up traffic violation based on video |
CN102800113A (en) * | 2012-07-18 | 2012-11-28 | 合肥工业大学 | Digital image analysis method based on fractal dimension |
CN103258429A (en) * | 2013-04-26 | 2013-08-21 | 青岛海信网络科技股份有限公司 | Video detecting method aims at vehicles which enter into jammed intersection by force |
CN104834977A (en) * | 2015-05-15 | 2015-08-12 | 浙江银江研究院有限公司 | Traffic alarm condition level predication method based on distance metric learning |
CN105788272A (en) * | 2016-05-16 | 2016-07-20 | 杭州智诚惠通科技有限公司 | Alarming method and system for road flow congestion |
CN108921200A (en) * | 2018-06-11 | 2018-11-30 | 百度在线网络技术(北京)有限公司 | Method, apparatus, equipment and medium for classifying to Driving Scene data |
CN109410582A (en) * | 2018-11-27 | 2019-03-01 | 易念科技(深圳)有限公司 | Traffic condition analysis method and terminal device |
CN112365112A (en) * | 2019-07-26 | 2021-02-12 | 通用电气公司 | System and method for manifold learning of aeronautical network data |
CN112733734A (en) * | 2021-01-13 | 2021-04-30 | 中南大学 | Traffic abnormal event detection method based on combination of Riemann manifold characteristics and LSTM network |
CN113487890A (en) * | 2021-08-13 | 2021-10-08 | 广州市迪声音响有限公司 | Control system and method for dynamically adjusting traffic light time |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101128858A (en) * | 2005-05-18 | 2008-02-20 | Lg电子株式会社 | Method and apparatus for providing transportation status information and using it |
CN101188002A (en) * | 2007-12-24 | 2008-05-28 | 北京大学 | A city traffic dynamic prediction system and method with real time and continuous feature |
CN101615347A (en) * | 2009-07-20 | 2009-12-30 | 交通部公路科学研究所 | Processing method of traffic control data and device |
CN101739820A (en) * | 2009-11-19 | 2010-06-16 | 北京世纪高通科技有限公司 | Road condition predicting method and device |
CN102024326A (en) * | 2009-09-22 | 2011-04-20 | 上海遥薇实业有限公司 | Vehicle-type-recognition flow detection system based on video identification |
-
2011
- 2011-04-21 CN CN2011101009103A patent/CN102169631A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101128858A (en) * | 2005-05-18 | 2008-02-20 | Lg电子株式会社 | Method and apparatus for providing transportation status information and using it |
CN101188002A (en) * | 2007-12-24 | 2008-05-28 | 北京大学 | A city traffic dynamic prediction system and method with real time and continuous feature |
CN101615347A (en) * | 2009-07-20 | 2009-12-30 | 交通部公路科学研究所 | Processing method of traffic control data and device |
CN102024326A (en) * | 2009-09-22 | 2011-04-20 | 上海遥薇实业有限公司 | Vehicle-type-recognition flow detection system based on video identification |
CN101739820A (en) * | 2009-11-19 | 2010-06-16 | 北京世纪高通科技有限公司 | Road condition predicting method and device |
Non-Patent Citations (1)
Title |
---|
王伟智等: "基于协同方法交通状态识别", 《中国体视学与图像分析》, vol. 12, no. 1, 31 March 2007 (2007-03-31), pages 37 - 42 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102768801A (en) * | 2012-07-12 | 2012-11-07 | 复旦大学 | Method for detecting motor vehicle green light follow-up traffic violation based on video |
CN102768801B (en) * | 2012-07-12 | 2014-08-06 | 复旦大学 | Method for detecting motor vehicle green light follow-up traffic violation based on video |
CN102800113A (en) * | 2012-07-18 | 2012-11-28 | 合肥工业大学 | Digital image analysis method based on fractal dimension |
CN102800113B (en) * | 2012-07-18 | 2014-12-03 | 合肥工业大学 | Digital image analysis method based on fractal dimension |
CN103258429A (en) * | 2013-04-26 | 2013-08-21 | 青岛海信网络科技股份有限公司 | Video detecting method aims at vehicles which enter into jammed intersection by force |
CN103258429B (en) * | 2013-04-26 | 2015-06-03 | 青岛海信网络科技股份有限公司 | Video detecting method aims at vehicles which enter into jammed intersection by force |
CN104834977A (en) * | 2015-05-15 | 2015-08-12 | 浙江银江研究院有限公司 | Traffic alarm condition level predication method based on distance metric learning |
CN104834977B (en) * | 2015-05-15 | 2018-02-27 | 浙江银江研究院有限公司 | Traffic alert grade prediction technique based on learning distance metric |
CN105788272A (en) * | 2016-05-16 | 2016-07-20 | 杭州智诚惠通科技有限公司 | Alarming method and system for road flow congestion |
CN108921200A (en) * | 2018-06-11 | 2018-11-30 | 百度在线网络技术(北京)有限公司 | Method, apparatus, equipment and medium for classifying to Driving Scene data |
US11783590B2 (en) | 2018-06-11 | 2023-10-10 | Apollo Intelligent Driving Technology (Beijing) Co., Ltd. | Method, apparatus, device and medium for classifying driving scenario data |
CN109410582A (en) * | 2018-11-27 | 2019-03-01 | 易念科技(深圳)有限公司 | Traffic condition analysis method and terminal device |
CN112365112A (en) * | 2019-07-26 | 2021-02-12 | 通用电气公司 | System and method for manifold learning of aeronautical network data |
CN112733734A (en) * | 2021-01-13 | 2021-04-30 | 中南大学 | Traffic abnormal event detection method based on combination of Riemann manifold characteristics and LSTM network |
CN113487890A (en) * | 2021-08-13 | 2021-10-08 | 广州市迪声音响有限公司 | Control system and method for dynamically adjusting traffic light time |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102169631A (en) | Manifold-learning-based traffic jam event cooperative detecting method | |
CN112380952B (en) | Power equipment infrared image real-time detection and identification method based on artificial intelligence | |
CN106909902B (en) | Remote sensing target detection method based on improved hierarchical significant model | |
CN101236608B (en) | Human face detection method based on picture geometry | |
CN105718866B (en) | A kind of detection of sensation target and recognition methods | |
CN109871902B (en) | SAR small sample identification method based on super-resolution countermeasure generation cascade network | |
CN111898736B (en) | Efficient pedestrian re-identification method based on attribute perception | |
CN106446895A (en) | License plate recognition method based on deep convolutional neural network | |
CN105528794A (en) | Moving object detection method based on Gaussian mixture model and superpixel segmentation | |
CN103942557B (en) | A kind of underground coal mine image pre-processing method | |
CN106557740B (en) | The recognition methods of oil depot target in a kind of remote sensing images | |
CN106128121B (en) | Vehicle queue length fast algorithm of detecting based on Local Features Analysis | |
CN104166841A (en) | Rapid detection identification method for specified pedestrian or vehicle in video monitoring network | |
CN104103033A (en) | Image real-time processing method | |
CN109635634A (en) | A kind of pedestrian based on stochastic linear interpolation identifies data enhancement methods again | |
CN101996308A (en) | Human face identification method and system and human face model training method and system | |
CN102542293A (en) | Class-I extraction and classification method aiming at high-resolution SAR (Synthetic Aperture Radar) image scene interpretation | |
CN101916379A (en) | Target search and recognition method based on object accumulation visual attention mechanism | |
CN110334656A (en) | Multi-source Remote Sensing Images Clean water withdraw method and device based on information source probability weight | |
CN105405138A (en) | Water surface target tracking method based on saliency detection | |
CN113780242A (en) | Cross-scene underwater sound target classification method based on model transfer learning | |
CN104835142B (en) | A kind of vehicle queue length detection method based on textural characteristics | |
CN106056078A (en) | Crowd density estimation method based on multi-feature regression ensemble learning | |
CN103577804A (en) | Abnormal human behavior identification method based on SIFT flow and hidden conditional random fields | |
CN114359702A (en) | Method and system for identifying building violation of remote sensing image of homestead based on Transformer |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C12 | Rejection of a patent application after its publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20110831 |