CN102663491A - Method for counting high density population based on SURF characteristic - Google Patents
Method for counting high density population based on SURF characteristic Download PDFInfo
- Publication number
- CN102663491A CN102663491A CN2012100645430A CN201210064543A CN102663491A CN 102663491 A CN102663491 A CN 102663491A CN 2012100645430 A CN2012100645430 A CN 2012100645430A CN 201210064543 A CN201210064543 A CN 201210064543A CN 102663491 A CN102663491 A CN 102663491A
- Authority
- CN
- China
- Prior art keywords
- crowd
- people
- mst
- characteristic
- vector
- 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.)
- Granted
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
A method for counting high density population based on an SURF characteristi comprises the following steps: 1) collecting real time monitoring data by a camera, and pretreating images; 2) extracting feature points of moving people from the pretreated images; 3) feature points clustering: using an MST-DBSCAN algorithm which is based on a traditional DBSCN algorithm; 4) counting of people flow: according to the result of clustering in step 3), calculating a real time feature vector T of the crowd by (4), obtaining (5) using a support vector regression, and providing a predicted number of people by (5). The invention provides an effective method for counting high density population based on the SURF characteristic, and the method is suitable for scenes that have a large flow of people or a high density population.
Description
Technical field
The present invention relates to the monitoring technique of computer vision field, relate in particular to crowd's method of counting based on computer vision.
Background technology
Crowd's traffic statistics have widely uses, like automobile bus station, subway platform, tourist attractions and gateway, market etc.Utilize the data on flows of statistics, managerial personnel can the rational management man power and material, the reasonable disposition resource; In public arenas such as some squares, passages, realize effectively dredging of effective management, emergency condition etc. through people's stream information, thereby avoid a series of social security problems that management plays an important role stream of people's statistics to social security in addition.Except mode was checked in manual work, common automatic stream of people's statistical method mainly contained following several kinds at present:
(1) the foot-operated approach sensor of machinery.This mode is installed mechanical foot pedals on ground, passway, the sensor that is connected by pedal is delivered to end for process with the information of trampling, and tramples Information Statistics at the information processing end.This mode is handled the distribution situation that depends on pedal, and when turnover stream of people skewness, the accumulation precision of this method reduces.
(2) infrared induction mode.This mode is installed infrared transmitting device and infrared signal induction receiving trap in the both sides of entrance and exit of the passage, block situation through the infrared signal of penetrating via, analyzes the pedestrian and passes through situation.This mode is applicable to minority pedestrian turnover, for the stream of people block, situation such as people's current density is bigger, serious omission situation can appear.
(3) computer vision mode.This method adopts camera to obtain the monitoring scene realtime graphic, realizes that through computer vision technique the stream of people of video adds up, because this method is easy, relative cost is lower, therefore is to use stream of people's statistical method comparatively widely at present.This also is that the present invention is with the mode of taking.
Traditional stream of people's statistical method based on computer vision is mainly through accurate location and follow the tracks of each pedestrian and realize counting, and it is only applicable to video definition height and the little situation of flow of the people, and or scene that crowd density high and inapplicable big to flow of the people.
Summary of the invention
The deficiency of or scene that crowd density high big for the inapplicable flow of the people that overcomes existing stream of people's statistical method, the present invention provides a kind of high density crowd's method of counting based on the SURF characteristic that effectively is suitable for the scene that flow of the people is big or crowd density is high.
The technical solution adopted for the present invention to solve the technical problems is:
A kind of high density crowd's method of counting based on the SURF characteristic, said method of counting comprises the steps:
1) carries out pre-service through the real-time monitor data of camera collection, and to image;
2) pretreated image is extracted the unique point of sport people:
Detect the unique point that obtains through the SURF algorithm and comprise background and sport people two parts simultaneously, the employing block matching algorithm, the ownership of coming judging characteristic point through analysis to the movable information of adjacent two frames:
Wherein, (x y) is used for mark (x to p; Whether the unique point of y) locating belongs to sport people, and 1 expression is, otherwise is not;
is in the piece coupling computation process; Present frame and consecutive frame (x, the motion vector of y) locating, α are discrimination threshold;
3) unique point cluster:
Employing is based on the MST-DBSCAN algorithm of traditional DBSCAN algorithm; With the data conversion that needs cluster is the point of two-dimensional space; And on the basis of these points, make up a MST; Through analysis, provide the minimum territory of search of traditional DBSCAN algorithm, and finally obtain having the adaptive cluster effect of clusters number MST:
MST-DBSCAN(data,λ,minPts) (3)
Wherein, δ
iBe that N be the limit number of this MST according to the length on the i bar limit of the MST of sport people unique point structure, β is a regulatory factor, and the λ that is obtained by (2) is as the search minimum territory of MST-DBSCAN;
4) people stream counting:
According to the crowd characteristic vector T of the cluster result of step 3), adopt support vector regression to obtain (5), provide the prediction number through (5) through (4) calculating real-time;
The expression of crowd characteristic amount vector T and number valuation functions is following:
T=(n
points,S
cluster,d) (4)
n
people=f(T) (5)
Wherein:
n
PointsThe quantity of expression crowd characteristic point, S
ClusterThe crowd area of expression after the contrary perspective transform of IPM; D representes through sport people after the contrary perspective transform of IPM and the distance between the video camera.
Further, adopt convex closure to come approximate expression crowd's area, with the dimension of area as the crowd characteristic vector T.
Further again; Said support vector regression has learning ability; Through the supported vector regression of the training of proper vector, process is: earlier the real-time training image crowd of taking exercises is cut apart, carried out feature point detection on this basis and make up the crowd characteristic vector with this; Train supported vector regression according to proper vector at last, said support vector regression is the crowd's state model with environment self-adaption property.
Technical conceive of the present invention is: the present invention is directed to the deficiency of classic method, the crowd is studied high density crowd traffic statistics as a global feature analysis, through setting up the crowd characteristic model; Carry out cluster analysis then; Carry out the parameter optimization assessment again, finally not only can add up whole flow situation, can also obtain local flow; Have great importance for safety monitoring, resource management, be significantly improved and innovate than classic method.Fig. 1 is the application scenarios of traditional people stream counting method, and Fig. 2 is the application scenarios of high density crowd's method of counting of proposing of the present invention.
Beneficial effect of the present invention mainly shows: it is good effectively to be suitable for flow of the people is big or crowd density is high scene, reliability.
Description of drawings
Fig. 1 is the synoptic diagram of traditional stream of people's technical method.
Fig. 2 is the synoptic diagram of high density crowd's method of counting of the present invention.
Fig. 3 is the realization hierarchy chart of core algorithm of the present invention.
Fig. 4 is the process flow diagram of core algorithm of the present invention.
Embodiment
Below in conjunction with accompanying drawing the present invention is further described.
With reference to Fig. 1~Fig. 4; A kind of high density crowd's method of counting based on the SURF characteristic; Utilize computer vision to dispose the big advantage of dirigibility in understanding, identification and the project used aspect the processing image; Cut apart achievement that obtains in the field and deficiency and the shortcoming that under high density stream of people environment, exists in conjunction with at present traditional safety monitoring, stream of people's statistics aspect in Image Acquisition, pre-service, feature extraction, detection; Study crowd's counting is run in the open environment in high density crowd and visual field new problem, a difficult problem; Through analysis, solution, finally propose a kind ofly practicablely to have high density crowd's method of counting of environment self-adaption property and provide vague generalization implementation framework and the design proposal under this method to these new problems, a difficult problem.
The related algorithm level that the present invention relates to is seen Fig. 2, mainly is made up of three parts such as data acquisition pretreatment layer, core algorithm layer and displaying interbedded formation.Wherein, the data acquisition preprocessing part mainly is responsible for through the real-time monitor data of camera collection and is done the image pre-service work of some bottoms, like signal denoising, filtering, histogram equalization etc., for the core algorithm layer provides high-quality image; The convenience that core algorithm is partly considered to transplant and the facility of management maintenance; Correlator algorithm use module package; Realize crowd's motion segmentation, feature extraction, cluster, the adaptive training of valuation functions type and last real-time assessment, mainly contained compositions such as crowd characteristic processing module, self-adaptive estimation module.Simultaneously, the core algorithm layer provides real-time high density crowd's information for the displaying interbedded formation, and can do corresponding actions according to the alarm threshold value that the user sets, and the user is provided in system's real time execution process the adjustment interactive interface to systematic parameter.
The correlation computer sense of vision factor that has combined actual environment in the core algorithm flow process of the present invention, crowd's state model of taking training earlier to have environment self-adaption carries out effective strength's evaluation work again.Earlier the real-time training image crowd of taking exercises is cut apart in the training process; Carry out feature point detection on this basis and make up the crowd characteristic vector, the crowd's state model that combines some parameters (like the information such as relative position of ambient brightness, visual field and camera) training of actual environment to obtain having environment self-adaption property according to proper vector at last with this.After the training of crowd's state model is accomplished; Get into effective strength's evaluation work stage; This stage was compared with the training stage of crowd's state model, crowd characteristic vector to obtain part be identical, difference is; This stage finally obtains the assessment of crowd characteristic vector through crowd's state model crowd's virtual condition and feeds back to the display alarm system doing further processing.Its flow process is as shown in Figure 4.Wherein, the committed step based on high density crowd's method of counting core algorithm of SURF characteristic is following:
Detect the unique point of sport people: the feature point detection of sport people mainly was divided into for two steps: (i), at first detect all unique points in the image to be analyzed; (ii), through block-matching technique (blob-matching technique), consider crowd's movable information, the unique point that does not belong to the crowd is rejected.
Can be used for the feature point detection algorithm at present has a lot, like the Harris algorithm, and SIFT algorithm, SURF algorithm etc.The present invention finally selects the SURF algorithm for use through a large amount of experiment in early stage.This algorithm is with respect to other detection algorithms, has translation, rotation, scale, brightness are changed, block the unchangeability with noise etc., and stability is stronger; The SURF algorithm is owing to adopted the Hessian matrix to carry out the eigenwert detection simultaneously, and overall detection speed is faster.
But detect the unique point that obtains through the SURF algorithm and comprise background and sport people two parts simultaneously; In order more accurately to reject the static unique point of background parts; Adopt block matching algorithm, the ownership of coming judging characteristic point through analysis to the movable information of adjacent two frames:
Wherein, p (x, y) be used for mark (x, whether the unique point of y) locating belongs to sport people, 1 the expression be, otherwise be not.
is in the piece coupling computation process; Present frame and consecutive frame are at (x; Y) motion vector of locating, α are discrimination threshold (the α empirical value that the present invention obtains is 0.5).
The unique point cluster: the effect of unique point cluster quality directly influence the vectorial structure of crowd characteristic, and then to influence with the crowd characteristic vector be the number evaluator of training data, and finally influences final assessment result.
Because the sport people stochastic distribution, also determined the random partial property of its unique point simultaneously, so some clustering algorithms commonly used (like k-means etc.) situation good performance can not be arranged.Because such algorithm needs to provide in advance artificially the number of cluster usually, and crowd's quantity can't be confirmed in advance.In order to overcome these not enough points, the present invention proposes a kind of MST-DBSCAN algorithm based on traditional DBSCAN algorithm.
The key of MST-DBSCAN algorithm is with the data conversion that needs cluster to be the point of two-dimensional space; And on the basis of these points, make up a MST; Through analysis to MST; Provide the minimum territory of search of traditional DBSCAN algorithm, and finally obtain having the adaptive cluster effect of clusters number:
MST-DBSCAN(data,λ,minPts) (3)
Wherein, δ
iBe that N be the limit number of this MST according to the length on the i bar limit of the MST of sport people unique point structure, β is regulatory factor (the actual empirical value of testing acquisition of the present invention is 2.0).The λ that is obtained by (2) is as the minimum territory of the search of MST-DBSCAN.
The structure of proper vector and training: through after the cluster of unique point, the pedestrian in the visual field is divided into crowd independent of each other, and the structure of crowd characteristic vector also will launch and be used for further training on this basis.
In the building process of crowd characteristic vector; The influence that distance between crowd and the video camera is counted to its characteristic adds consideration; Through IPM (Inverse Perspective Mapping; Contrary perspective) conversion, with one of them dimension of the distance between crowd and the video camera as proper vector, as constraint condition in order to the characteristic that overcomes same people under the different distance different problems of counting.
In view of crowd characteristic count and number between relation be not simple mapping relations, for the training of crowd characteristic vector, the present invention has adopted support vector regression with learning ability (ε-SVR).Through study, be the valuation functions in space with providing a dimension with T to the crowd characteristic vector T.In high density crowd's counting process of reality, import real-time crowd characteristic vector T, provide the prediction number through (5).
The expression of crowd characteristic amount vector T and number valuation functions is following:
T=(n
points,S
cluster,d) (4)
n
people=f(T) (5)
Wherein:
n
PointsThe quantity of expression crowd characteristic point;
S
ClusterThe crowd area of expression after the contrary perspective transform of IPM, suggestion comes approximate expression crowd's area with convex closure, and with respect to boundary rectangle, the fitting degree of convex closure is higher.With the dimension of area, mainly be to consider to increase the constraint of sport people with the different perspective phenomenons that cause of video camera distance through the crowd characteristic dot density as the crowd characteristic vector T;
D representes through sport people after the contrary perspective transform of IPM and the distance between the video camera.
Be the accuracy that the method for checking the present invention proposition is counted the high density crowd under the actual environment, the inventor adopts professional pedestrian's database experiment Analysis (http://www.cvg.rdg.ac.uk/PETS2009/) of PETS2009.This database is divided into 4 parts (being respectively S0, S1, S2, S3), and 4 parts all have the video of taking (being respectively View1, View2......View8) from 8 different angles.It is the S1 of special topic that experimental section is selected for use with " crowd's counting and crowd density are estimated ", and video angle is View1.For the purpose of convenient, the video sequence S1.L1.13-57 that uses in the experimentation, S1.L1.13-59, S1.L2.14-06, S1.L3.14-17 noted by abridging respectively be V1, V2, V3, V4.
In unique point cluster link; The present invention has adopted according to the improved MST-DBSCAN algorithm of traditional DBSCAN algorithm; Effectively having overcome the sport people cluster can't adaptive deficiency; Realize the self-adaption cluster of sport people unique point, the structure vectorial for crowd characteristic provides the segmentation result of sport people accurately.
The good and bad main effective strength through contrast experiment's video sequence of high density crowd's method of counting makes a decision with corresponding prediction number.Two indexs of main consideration in the process, MAE (Mean Absolute Error, average absolute mistake) and MRE (Mean Relative Error, on average wrong relatively):
Wherein, N is experiment video sequence frame number, and G (i) is the prediction number of i frame, and T (i) is the effective strength of i frame.
In the process of sport people proper vector training, training sample accounts for the proportional control of experiment video sequence about 5%.ε-SVR study through to training sample obtains number valuation functions f (T) and is used to remain the prediction of testing video sequence.
Video sequence | MAE | MRE |
V1 | 1.02 | 5.08% |
V2 | 1.16 | 9.8% |
[0068]?
V3 | 4.26 | 18.05% |
V4 | 1.49 | 7.38% |
Each video interpretation of table 1
Experimental analysis through to PETS2009 specialty pedestrian database shows that the method that the present invention proposes has higher accuracy and robustness to the counting of the crowd under the open environment of high density and visual field.
Claims (3)
1. high density crowd's method of counting based on the SURF characteristic, it is characterized in that: said method of counting comprises the steps:
1) carries out pre-service through the real-time monitor data of camera collection, and to image;
2) pretreated image is extracted the unique point of sport people:
Detect the unique point that obtains through the SURF algorithm and comprise background and sport people two parts simultaneously, the employing block matching algorithm, the ownership of coming judging characteristic point through analysis to the movable information of adjacent two frames:
Wherein, (x y) is used for mark (x to p; Whether the unique point of y) locating belongs to sport people, and 1 expression is, otherwise is not;
is in the piece coupling computation process; Present frame and consecutive frame (x, the motion vector of y) locating, α are discrimination threshold;
3) unique point cluster:
Employing is based on the MST-DBSCAN algorithm of traditional DBSCAN algorithm; With the data conversion that needs cluster is the point of two-dimensional space; And on the basis of these points, make up a MST; Through analysis, provide the minimum territory of search of traditional DBSCAN algorithm, and finally obtain having the adaptive cluster effect of clusters number MST:
MST-DBSCAN(data,λ,minPts) (3)
Wherein, δ
iBe that N be the limit number of this MST according to the length on the i bar limit of the MST of sport people unique point structure, β is a regulatory factor, and the λ that is obtained by (2) is as the search minimum territory of MST-DBSCAN;
4) people stream counting:
According to the crowd characteristic vector T of the cluster result of step 3), adopt support vector regression to obtain (5), provide the prediction number through (5) through (4) calculating real-time;
The expression of crowd characteristic amount vector T and number valuation functions is following:
T=(n
points,S
cluster,d) (4)
n
people=f(T) (5)
Wherein:
n
PointsThe quantity of expression crowd characteristic point, S
ClusterThe crowd area of expression after the contrary perspective transform of IPM; D representes through sport people after the contrary perspective transform of IPM and the distance between the video camera.
2. a kind of high density crowd's method of counting based on the SURF characteristic as claimed in claim 1 is characterized in that: adopt convex closure to come approximate expression crowd's area, with the dimension of area as the crowd characteristic vector T.
3. according to claim 1 or claim 2 a kind of high density crowd's method of counting based on the SURF characteristic; It is characterized in that: said support vector regression has learning ability; Through the supported vector regression of the training of proper vector; Process is: earlier the real-time training image crowd of taking exercises is cut apart; Carry out feature point detection on this basis and make up the crowd characteristic vector with this, train supported vector regression according to proper vector at last, said support vector regression is the crowd's state model with environment self-adaption property.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210064543.0A CN102663491B (en) | 2012-03-13 | 2012-03-13 | Method for counting high density population based on SURF characteristic |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210064543.0A CN102663491B (en) | 2012-03-13 | 2012-03-13 | Method for counting high density population based on SURF characteristic |
Publications (2)
Publication Number | Publication Date |
---|---|
CN102663491A true CN102663491A (en) | 2012-09-12 |
CN102663491B CN102663491B (en) | 2014-09-03 |
Family
ID=46772972
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210064543.0A Active CN102663491B (en) | 2012-03-13 | 2012-03-13 | Method for counting high density population based on SURF characteristic |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102663491B (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102982341A (en) * | 2012-11-01 | 2013-03-20 | 南京师范大学 | Self-intended crowd density estimation method for camera capable of straddling |
CN104778468A (en) * | 2014-01-15 | 2015-07-15 | 索尼公司 | Image processing device, image processing method and monitoring equipment |
CN104933412A (en) * | 2015-06-16 | 2015-09-23 | 电子科技大学 | Abnormal state detection method of medium and high density crowd |
CN106570465A (en) * | 2016-10-31 | 2017-04-19 | 深圳云天励飞技术有限公司 | Visitor flow rate statistical method and device based on image recognition |
CN107607205A (en) * | 2017-09-30 | 2018-01-19 | 江苏西格数据科技有限公司 | Wire harness color sequences detecting system and method |
CN107784321A (en) * | 2017-09-28 | 2018-03-09 | 深圳市奇米教育科技有限公司 | Numeral paints this method for quickly identifying, system and computer-readable recording medium |
CN111950491A (en) * | 2020-08-19 | 2020-11-17 | 成都飞英思特科技有限公司 | Personnel density monitoring method and device and computer readable storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101196991A (en) * | 2007-12-14 | 2008-06-11 | 同济大学 | Close passenger traffic counting and passenger walking velocity automatic detection method and system thereof |
CN101325690A (en) * | 2007-06-12 | 2008-12-17 | 上海正电科技发展有限公司 | Method and system for detecting human flow analysis and crowd accumulation process of monitoring video flow |
CN102063613A (en) * | 2010-12-28 | 2011-05-18 | 北京智安邦科技有限公司 | People counting method and device based on head recognition |
CN102184421A (en) * | 2011-04-22 | 2011-09-14 | 北京航空航天大学 | Training method of support vector regression machine |
CN102184409A (en) * | 2011-04-22 | 2011-09-14 | 北京文安科技发展有限公司 | Machine-vision-based passenger flow statistics method and system |
CN101835034B (en) * | 2010-05-27 | 2011-12-14 | 王巍 | Crowd characteristic counting system |
-
2012
- 2012-03-13 CN CN201210064543.0A patent/CN102663491B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101325690A (en) * | 2007-06-12 | 2008-12-17 | 上海正电科技发展有限公司 | Method and system for detecting human flow analysis and crowd accumulation process of monitoring video flow |
CN101196991A (en) * | 2007-12-14 | 2008-06-11 | 同济大学 | Close passenger traffic counting and passenger walking velocity automatic detection method and system thereof |
CN101835034B (en) * | 2010-05-27 | 2011-12-14 | 王巍 | Crowd characteristic counting system |
CN102063613A (en) * | 2010-12-28 | 2011-05-18 | 北京智安邦科技有限公司 | People counting method and device based on head recognition |
CN102184421A (en) * | 2011-04-22 | 2011-09-14 | 北京航空航天大学 | Training method of support vector regression machine |
CN102184409A (en) * | 2011-04-22 | 2011-09-14 | 北京文安科技发展有限公司 | Machine-vision-based passenger flow statistics method and system |
Non-Patent Citations (5)
Title |
---|
何鹏等: "实时人数计数系统", 《中国图象图形学报》 * |
张锐娟等: "基于SURF的图像配准方法研究", 《红外与激光工程》 * |
彭欣等: "基于SURF目标根据算法研究", 《长春理工大学学报(自然科学版)》 * |
朱军军等: "基于OpenCV的视频人流计数系统的设计与开发", 《电脑知识与技术》 * |
荣秋生等: "基于DBSCAN聚类算法的研究与实现", 《计算机应用》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102982341A (en) * | 2012-11-01 | 2013-03-20 | 南京师范大学 | Self-intended crowd density estimation method for camera capable of straddling |
CN102982341B (en) * | 2012-11-01 | 2015-06-24 | 南京师范大学 | Self-intended crowd density estimation method for camera capable of straddling |
CN104778468A (en) * | 2014-01-15 | 2015-07-15 | 索尼公司 | Image processing device, image processing method and monitoring equipment |
CN104933412A (en) * | 2015-06-16 | 2015-09-23 | 电子科技大学 | Abnormal state detection method of medium and high density crowd |
CN104933412B (en) * | 2015-06-16 | 2018-05-04 | 电子科技大学 | The abnormal state detection method of middle-high density crowd |
CN106570465A (en) * | 2016-10-31 | 2017-04-19 | 深圳云天励飞技术有限公司 | Visitor flow rate statistical method and device based on image recognition |
CN106570465B (en) * | 2016-10-31 | 2018-04-20 | 深圳云天励飞技术有限公司 | A kind of people flow rate statistical method and device based on image recognition |
CN107784321A (en) * | 2017-09-28 | 2018-03-09 | 深圳市奇米教育科技有限公司 | Numeral paints this method for quickly identifying, system and computer-readable recording medium |
CN107784321B (en) * | 2017-09-28 | 2021-06-25 | 深圳市快易典教育科技有限公司 | Method and system for quickly identifying digital picture books and computer readable storage medium |
CN107607205A (en) * | 2017-09-30 | 2018-01-19 | 江苏西格数据科技有限公司 | Wire harness color sequences detecting system and method |
CN111950491A (en) * | 2020-08-19 | 2020-11-17 | 成都飞英思特科技有限公司 | Personnel density monitoring method and device and computer readable storage medium |
CN111950491B (en) * | 2020-08-19 | 2024-04-02 | 成都飞英思特科技有限公司 | Personnel density monitoring method and device and computer readable storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN102663491B (en) | 2014-09-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102663491B (en) | Method for counting high density population based on SURF characteristic | |
CN108898085B (en) | Intelligent road disease detection method based on mobile phone video | |
CN110364008B (en) | Road condition determining method and device, computer equipment and storage medium | |
CN103985250B (en) | The holographic road traffic state vision inspection apparatus of lightweight | |
CN101807352B (en) | Method for detecting parking stalls on basis of fuzzy pattern recognition | |
CN105608456B (en) | A kind of multi-direction Method for text detection based on full convolutional network | |
CN102542289B (en) | Pedestrian volume statistical method based on plurality of Gaussian counting models | |
CN103218816B (en) | A kind of crowd density estimation method and people flow rate statistical method based on video analysis | |
CN104504377B (en) | A kind of passenger on public transport degree of crowding identifying system and method | |
CN103310444B (en) | A kind of method of the monitoring people counting based on overhead camera head | |
CN111462488A (en) | Intersection safety risk assessment method based on deep convolutional neural network and intersection behavior characteristic model | |
CN104166841A (en) | Rapid detection identification method for specified pedestrian or vehicle in video monitoring network | |
CN108764167A (en) | A kind of target of space time correlation recognition methods and system again | |
CN109637151A (en) | A kind of recognition methods that highway Emergency Vehicle Lane is driven against traffic regulations | |
CN107948166A (en) | Traffic anomaly detection method and device based on deep learning | |
CN105303191A (en) | Method and apparatus for counting pedestrians in foresight monitoring scene | |
CN103679212A (en) | Method for detecting and counting personnel based on video image | |
CN109766746A (en) | A kind of track foreign matter detecting method of unmanned plane video | |
CN103425967A (en) | Pedestrian flow monitoring method based on pedestrian detection and tracking | |
CN103164711A (en) | Regional people stream density estimation method based on pixels and support vector machine (SVM) | |
CN110232379A (en) | A kind of vehicle attitude detection method and system | |
CN102855508B (en) | Opening type campus anti-following system | |
CN106339687A (en) | Self-adaptive pedestrian street-crossing signal calculating method based on video | |
CN109948690A (en) | A kind of high-speed rail scene perception method based on deep learning and structural information | |
CN112967252B (en) | Rail vehicle machine sense hanger assembly bolt loss detection method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |