CN105894815B - Traffic congestion method for early warning based on semantic region segmentation - Google Patents
Traffic congestion method for early warning based on semantic region segmentation Download PDFInfo
- Publication number
- CN105894815B CN105894815B CN201610363423.9A CN201610363423A CN105894815B CN 105894815 B CN105894815 B CN 105894815B CN 201610363423 A CN201610363423 A CN 201610363423A CN 105894815 B CN105894815 B CN 105894815B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- frame
- semantic region
- region
- video
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/2163—Partitioning the feature space
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Traffic Control Systems (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of traffic congestion method for early warning based on semantic region segmentation, and it comprises the following steps：Using the normal monitor video of the traffic flow for treating prewarning area as training sample, the movement locus of vehicle is obtained, monitoring scene is divided into multiple semantic regions；The semantic region that each movement locus passes through successively sequentially in time is converted into a directed rooted tree, merges into multiple directed trees according to whether root node is identical, the vehicle shunting point branch point of out-degree >=2 in every directed tree being defined as under the monitoring scene；Obtain the movement locus of vehicle in video segment to be analyzed；The vehicle shunting point in training sample and video segment to be analyzed is respectively obtained, is such as different from training sample, then judges the possibility that traffic congestion be present.The present invention can not obtain and alarm in the case of for vehicle supervision department provide real-time early warning, improve the promptness of traffic violation and traffic accident treatment, promote to have a good transport and communication network, greatly relieve the congestion of traffic situation.
Description
Technical field
The invention belongs to Digital Video Processing field, and in particular to a kind of traffic congestion early warning based on semantic region segmentation
Method.
Background technology
With the development of the social economy, vehicle guaranteeding organic quantity is continuously increased, urban traffic congestion problem getting worse.Traffic
Congestion not only results in all deterioration of economic society, but also city living environment will be triggered continuous worsening, hampers city
The development in city, reduce the quality of life of people.Trigger traffic congestion the reason for except road capacity deficiency or design it is improper in addition to,
Traffic violation and traffic accident are to cause another main cause of traffic congestion.Traffic violation and traffic accident are such as
Fruit can not rapidly be handled, it will caused the delay of road vehicles to a certain extent, or even can be caused serious traffic
Congestion phenomenon.
What 2012-07-25 was authorized, Authorization Notice No. is that CN102024325B Chinese invention patent discloses a kind of base
In the traffic jam point identification method of floating car technology, this method using Floating Car, compile in the process of moving by taken at regular intervals vehicle
Number, Position, Velocity and Time information, and the data collected are sent to data center；Data center is in the data
Velocity information carry out data and filter out pretreatment, obtain effective floating car data, then extract what all Ultra-Low Speeds travelled
Data acquisition system；Data center by the inspection to the frequent Ultra-Low Speed running region of more cars, can automatic identification traffic jam point, from
And realize the dynamic renewal of traffic jam point information.What 2013-05-01 was authorized, Authorization Notice No. is in CN102254428B
State's patent of invention discloses a kind of traffic jam detection method based on Video Analysis Technology of knowing clearly, and this method is in Video segmentation and pass
By obtaining three average dissimilarity, crucial frame number, average light flow field energy congestions of video lens on the basis of the extraction of key frame
Characteristic quantity, realize that traffic congestion detects using more classification SVM methods.However, only to having occurred and that traffic is blocked up in existing method
The state of plug is judged, does not consider the influence of traffic violation and traffic accident to traffic congestion, it is impossible to by handing over
Logical illegal activities and traffic accident may trigger the situation of traffic jam to carry out early warning.
The content of the invention
It is an object of the invention to provide a kind of traffic congestion method for early warning based on semantic region segmentation, to by traffic offence
Behavior and traffic accident may trigger the situation of traffic jam to carry out early warning.
To reach above-mentioned purpose, present invention research finds that in specific monitoring scene, the motion of vehicle is by scene structure
Influence can show certain regularity, i.e. vehicle can move along certain path.The regularity of vehicle movement can be by
Similar track (similar direction and similar position) is embodied in the concurrency of area of space, of the invention by such area of space
It is defined as semantic region.When topic model is used to model track, theme represents the semantic region shared between track, i.e., many tools
Same semantic region is passed through in the track for having similar movement direction；Semantic region is modeled as the locus and side of path segment
To discrete distribution.
The movement locus of vehicle is the directly perceived form of expression of the microscopic motion behavior of vehicle on time-space domain, and any vehicle exists
Motion path in monitoring scene is a semantic region combination.In the case where there is traffic violation and traffic accident,
The traffic diverging point of abnormal semantic region combination and exception be present in the vehicle movement path in monitoring scene.If it can know
Not abnormal semantic region combination and the traffic diverging point of exception, can not obtain and alarm in the case of be traffic administration
Department provides real-time early warning, can improve the promptness of traffic violation and traffic accident treatment, promote to have a good transport and communication network, greatly
Ground relieves the congestion of traffic situation.
The technical scheme is that：
Based on the traffic congestion method for early warning of semantic region segmentation, it comprises the following steps：
Step 1：Using the normal monitor video of the traffic flow for treating prewarning area as training sample, and according to the training
The movement locus of vehicle in sample acquisition training sample, the monitoring scene for treating prewarning area is divided into multiple semantic regions；
Step 2：The semantic region that each movement locus passes through successively sequentially in time is converted to one oriented
Tree, merges into multiple directed trees, wherein having by the directed rooted tree corresponding to all movement locus according to whether root node is identical
More directed rooted trees of identical root node merge into a directed tree, and the branch point of out-degree >=2 in every directed tree is defined as
Vehicle shunting point under the monitoring scene；
Step 3：Obtain the movement locus of vehicle in video segment to be analyzed；Using the method for step 2, instruction is respectively obtained
Practice the vehicle shunting point in the vehicle shunting point and video segment to be analyzed in sample, such as the vehicle point in video segment to be analyzed
Flow point is different from the vehicle shunting point in training sample, then judges the possibility that traffic congestion be present, carry out early warning.
Preferably, in the traffic congestion method for early warning based on semantic region segmentation, the prison of prewarning area will be treated
Control scene cut is multiple semantic regions, including：
Motion dictionary is built according to monitoring scene, movement locus is divided into path segment sequence while path segment is entered
Row coding, afterwards using the path segment after coding as a document, is modeled, using gibbs using LDA topic models
Parameter in method of sampling estimation model, obtains " theme-word " matrix of K × V dimensionsThe number that K is the theme, V are motion
The number of word, the probability for each motion word that often row expression theme Z is included；" theme-word " square tieed up according to K × V
Battle arrayThe monitoring scene for treating prewarning area is divided into K semantic region and is numbered for each semantic region.
Preferably, in the traffic congestion method for early warning based on semantic region segmentation, built according to monitoring scene
Dictionary is moved, including：
It is several rectangular areas that size is P (pixel) × Q (pixel) by monitoring scene cutting；
By vehicle in monitoring scene can movable direction be divided into four angular intervals：
Assuming that monitoring scene size is M (pixel) × N (pixel), the size for moving dictionary is (M/P) × (N/Q) × 4,
It is expressed as V, the motion word in dictionary is expressed asWherein a ∈ { 1,2,3,4 }, four angular intervals are corresponded to respectively： And
c∈{1,2,…,N/Q}。
Preferably, in the traffic congestion method for early warning based on semantic region segmentation, the movement locus of vehicle rises
The frame of video for detecting this vehicle is started from, this frame of video is the start frame of first path segment of movement locus；Terminate at not
The frame of video of this vehicle is detected, this frame of video is the end frame of last path segment of movement locus.
Preferably, will fortune according to following rule in the traffic congestion method for early warning based on semantic region segmentation
Dynamic rail mark is divided into path segment sequence, it is assumed that vehicle A is the vehicle detected in traffic video stream：
Rule 1：When having detected that other vehicles enter or leave monitoring scene in current video frame, previous video is set
Frame is the end frame of vehicle A current track fragment, and present frame is the start frame of vehicle A next path segment；
Rule 2：Vehicle in continuous two frame of video is matched.If the vehicle A in present frame is in former frame
The vehicle of matching is can not find, then sets start frame of the present frame as vehicle A first path segment；If the car in present frame
A can not find the vehicle of matching in a later frame, then sets end frame of the present frame as vehicle A last path segment；
Rule 3：When vehicle A enters the rectangular area in any monitoring scene, previous frame of video working as vehicle A is set
The end frame of preceding path segment, present frame are the start frame of vehicle A next path segment；
Rule 4：Compare tracing point direction of the vehicle A movement locus in current video frame to regard previous with vehicle A
Tracing point direction in frequency frame.If the direction of two tracing points is in different angular intervals, previous frame of video is set as car
The end frame of A current track fragment, present frame are the start frame of vehicle A next path segment.
Preferably, in the traffic congestion method for early warning based on semantic region segmentation,
Rectangular area according to residing for path segment and the angular interval residing for the direction of motion of path segment, by track piece
Section is mapped to corresponding motion word in motion dictionary；The direction of motion of path segment is by first tracing point in path segment
The direction of motion represent.
Preferably, in the traffic congestion method for early warning based on semantic region segmentation, the prison of prewarning area will be treated
Scene cut is controlled into K semantic region, including：
1) the motion word of each self-contained most probable value in K theme is selected, by the square corresponding to this motion word
Seed region of the shape region as K initial semantic region, the corresponding semantic region of a theme；Theme is expressed as Zi,
Theme ZiCorresponding semantic region is expressed as Ri, i=1,2 ..., K.
2) centered on seed region, 4, the upper and lower, left and right of the respective seed region in K semantic region are chosen successively
The rectangular area not being merged in rectangular area；If meeting merging criterion, by the rectangular area of selection and K semantic space
The respective seed region in domain is merged into new semantic region, while the rectangular area being merged is pressed into K semantic region each
Storehouse.
3) it is treated as seed region not to take out a rectangular area respectively in empty storehouse from K semantic region and holds
Row step 2), until the respective storehouse in K semantic region is sky；
4) rectangular area of remaining connection in monitoring scene is merged into new semantic region respectively；
5) semantic region segmentation terminates.
Preferably, in the traffic congestion method for early warning based on semantic region segmentation, rectangular area and seed zone
The merging criterion in domain is as follows：
A. if rectangular area is only by a semantic region RiIt is chosen for rectangular area to be combined, i=1,2 ..., K, than
Compared with probability of four motion words in K theme corresponding to this rectangular area；If there is a motion word in theme
ZiIn probability be the probability in K theme peak, then this rectangular area merge with seed region；
B. if rectangular area is chosen for rectangular area to be combined, 2≤n≤K, comparing this rectangle by n semantic region
Probability of four motion words in the theme corresponding to n semantic region corresponding to region；By this rectangular area and motion
Seed region of the word in the theme corresponding to n semantic region corresponding to probability highest theme merges.
The present invention provides a kind of traffic congestion method for early warning based on semantic region segmentation, and it comprises the following steps：It will treat
The normal monitor video of traffic flow of prewarning area obtains the movement locus of vehicle, monitoring scene is split as training sample
For multiple semantic regions；By the semantic region that each movement locus passes through successively sequentially in time be converted to one it is oriented
Root tree, multiple directed trees are merged into according to whether root node is identical, the branch point of out-degree >=2 in every directed tree is defined as this
Vehicle shunting point under monitoring scene；Obtain the movement locus of vehicle in video segment to be analyzed；Respectively obtain training sample and
Vehicle shunting point in video segment to be analyzed, as being different from the vehicle shunting point in training sample, then judge that traffic be present gathers around
The possibility of plug, carry out early warning.The present invention can not obtain and alarm in the case of provide in real time for vehicle supervision department
Early warning, improve the promptness of traffic violation and traffic accident treatment, promote to have a good transport and communication network, greatly relieve the congestion of traffic
Situation.
Further advantage, target and the feature of the present invention embodies part by following explanation, and part will also be by this
The research and practice of invention and be understood by the person skilled in the art.
Brief description of the drawings
Fig. 1 is the overview flow chart of the traffic congestion method for early warning provided by the invention based on semantic region segmentation；
Fig. 2 is segmentation flow in semantic region in the traffic congestion method for early warning provided by the invention based on semantic region segmentation
Figure.
Embodiment
The present invention is described in further detail below in conjunction with the accompanying drawings, to make those skilled in the art with reference to specification text
Word can be implemented according to this.
It should be appreciated that such as " having ", "comprising" and " comprising " term used herein do not allot one or more
The presence or addition of individual other elements or its combination.
As shown in figure 1, a kind of traffic congestion method for early warning based on semantic region segmentation, comprises the following steps：
First, traffic video stream to be analyzed is obtained, motion dictionary is built according to the monitoring scene of traffic video
It is several rectangular areas that size is P (pixel) × Q (pixel) first by monitoring scene cutting；Secondly will monitoring
In scene vehicle can movable direction be divided into four angular intervals：And
Assuming that monitoring scene size is M (pixel) × N (pixel), the size for moving dictionary is (M/P) × (N/Q) × 4,
It is expressed as V.Motion word in dictionary is expressed asWherein a ∈ { 1,2,3,4 }, four angular intervals are corresponded to respectively： And
c∈{1,2,…,N/Q}。
2nd, by being tracked to the vehicle in traffic video stream, the movement locus of vehicle is obtained.
3rd, the movement locus of vehicle is divided into path segment sequence
The movement locus of vehicle originates in the frame of video for detecting this vehicle, and this frame of video is first rail of movement locus
The start frame of mark fragment；The frame of video for being not detected by this vehicle is terminated at, this frame of video is last track of movement locus
The end frame of fragment.Movement locus is divided into by path segment sequence according to following rule in the process, it is assumed that vehicle A is to hand over
The vehicle detected in logical video flowing：
Rule 1：When having detected that other vehicles enter or leave monitoring scene in current video frame, previous video is set
Frame is the end frame of vehicle A current track fragment, and present frame is the start frame of vehicle A next path segment.
Rule 2：Vehicle in continuous two frame of video is matched.If the vehicle A in present frame is in former frame
The vehicle of matching is can not find, then sets start frame of the present frame as vehicle A first path segment；If the car in present frame
A can not find the vehicle of matching in a later frame, then sets end frame of the present frame as vehicle A last path segment.
Rule 3：When vehicle A enters the rectangular area in any monitoring scene, previous frame of video working as vehicle A is set
The end frame of preceding path segment, present frame are the start frame of vehicle A next path segment.
Rule 4：Compare tracing point direction of the vehicle A movement locus in current video frame to regard previous with vehicle A
Tracing point direction in frequency frame.If the direction of two tracing points is in different angular intervals, previous frame of video is set as car
The end frame of A current track fragment, present frame are the start frame of vehicle A next path segment.
4th, the movement locus of vehicle is encoded
Rectangular area according to residing for path segment and the angular interval residing for the direction of motion of path segment, path segment
It is mapped to corresponding motion word in motion dictionary.The direction of motion of path segment is by first tracing point in path segment
The direction of motion represents.Now, the movement locus of vehicle is encoded as moving sequence of terms, is regarded as a document.
5th, encoded movement locus is modeled using LDA topic models, mould is estimated using Gibbs sampling method
Parameter in type, obtain " theme-word " matrix of K × V dimensionsThe number that K is the theme, V are the number of motion word, are often gone
Represent the probability for each motion word that theme Z is included.
6th, monitoring scene is divided into multiple semantic regions.As shown in Fig. 2 comprise the following steps that：
(1) the motion word of each self-contained most probable value in K theme is selected, by the square corresponding to this motion word
Seed region of the shape region as K initial semantic region, the corresponding semantic region of a theme.Theme is expressed as Zi,
Theme ZiCorresponding semantic region is expressed as Ri, i=1,2 ..., K.
(2) centered on seed region, 4, the upper and lower, left and right of the respective seed region in K semantic region are chosen successively
The rectangular area not being merged in rectangular area.If meeting merging criterion, by the rectangular area of selection and K semantic space
The respective seed region in domain is merged into new semantic region, while the rectangular area being merged is pressed into K semantic region each
Storehouse.
(3) it is treated as seed region not to take out a rectangular area respectively in empty storehouse from K semantic region and holds
Row step (2), until the respective storehouse in K semantic region is sky.
(4) rectangular area of remaining connection in monitoring scene is merged into new semantic region respectively.
(5) semantic region segmentation terminates.
The merging criterion of rectangular area and seed region is as follows：
(1) if rectangular area is only by a semantic region RiIt is chosen for rectangular area to be combined, i=1,2 ..., K,
Compare probability of four motion words in K theme corresponding to this rectangular area.If there is a motion word in master
Inscribe ZiIn probability be the probability in K theme peak, then this rectangular area merge with seed region.
(2) if rectangular area is chosen for rectangular area to be combined, 2≤n≤K, comparing this rectangle by n semantic region
Probability of four motion words in the theme corresponding to n semantic region corresponding to region.By this rectangular area and motion
Seed region of the word in the theme corresponding to n semantic region corresponding to probability highest theme merges.
7th, the semantic region split is numbered.
8th, each movement locus is converted to a directed rooted tree according to the semantic region passed through sequentially in time.Have
Root node to root tree is first semantic region that movement locus enters monitoring scene, and leaf node is that movement locus leaves monitoring
Last semantic region of scene, remaining node are the semantic regions that movement locus passes through in monitoring scene.
9th, the directed rooted tree with identical root node is merged into directed tree, the number of directed tree is exactly different root sections
The number of point.Vehicle shunting point branch point of the out-degree in directed tree more than or equal to 2 being defined as in monitoring scene, this vehicle point
Flow point is the semantic region that vehicle produces shunting.
The tenth, all tracks in video segment to be judged are converted into the collection of directed tree according to step 8 and step 9
Close.The vehicle shunting point in the set of this directed tree is calculated, if vehicle shunting point is different from the car of step 9 training gained
Split point, then judge that this video segment has the possibility for triggering traffic congestion, carry out real-time early warning.
In order to verify the validity of the traffic congestion method for early warning based on semantic region segmentation of the invention provided, Soviet Union is chosen
The traffic video for adding up to 24 hours length of state city disk South Road-southern loop Northern Cross crossing CAM9 collections is tested, and tests number
AVI format according to source for single fixed ccd video camera collection, resolution ratio is 320 × 240, and the acquisition rate of image is
25fps。
320 × 240 monitoring scene be divided into size be 10 × 10 unit, each unit can movable direction from
Dispersion is orthogonal four direction：Upper and lower, left and right.Therefore, the size for moving dictionary is 32 × 24 × 4.
Setting model parameter alpha=K/50, β=0.01, number of topics K are 20.
Training sample have chosen different time sections in experimental data, traffic flow is in the total 300 minutes of normal condition
Traffic video.The semantic region split and vehicle shunting point are trained through above-mentioned steps one to step 9.
The monitor video of remaining 1140 minutes is divided into upper nonoverlapping 380 of time according to the equal length of 3 minutes and regarded
Frequency fragment.57 sections of video segments are marked out by traffic professional person and different degrees of traffic violation and traffic accident be present,
And the traffic congestion phenomenon in subsequent video fragment is triggered；35 sections are marked out to lead because vehicle flowrate is big, road capacity is insufficient
Cause the video segment of traffic congestion.
380 sections of video segments in this example pass through the execution of step 8 to step 10,76 sections of video segments presence of early warning
Trigger the possibility of traffic congestion, wherein meet traffic professional person mark for 52 sections of video segments.With the accuracy rate of early warning and
Recall rate evaluates the quality of early warning result.The piece of video hop count of the accurate piece of video hop count/early warning of accuracy rate=early warning, is recalled
The piece of video hop count that traffic professional person marks out in the accurate piece of video hop count/sample of rate=early warning.The accuracy rate of early warning is
68.42%, recall rate 91.27%.
Although embodiment of the present invention is disclosed as above, it is not restricted in specification and embodiment listed
With it can be applied to various suitable the field of the invention completely, can be easily for those skilled in the art
Other modification is realized, therefore under the universal limited without departing substantially from claim and equivalency range, it is of the invention and unlimited
In specific details and shown here as the legend with description.
Claims (8)
1. the traffic congestion method for early warning based on semantic region segmentation, it is characterised in that comprise the following steps：
Step 1：Using the normal monitor video of the traffic flow for treating prewarning area as training sample, and according to the training sample
The movement locus of vehicle in training sample is obtained, the monitoring scene for treating prewarning area is divided into multiple semantic regions；
Step 2：The semantic region that each movement locus passes through successively sequentially in time is converted into a directed rooted tree,
Directed rooted tree corresponding to all movement locus is merged into multiple directed trees according to whether root node is identical, wherein with identical
More directed rooted trees of root node merge into a directed tree, and the branch point of out-degree >=2 in every directed tree is defined as into the prison
Control the vehicle shunting point under scene；
Step 3：Obtain the movement locus of vehicle in video segment to be analyzed；Using the method for step 2, training sample is respectively obtained
The vehicle shunting point in vehicle shunting point and video segment to be analyzed in this, such as the vehicle shunting point in video segment to be analyzed
Different from the vehicle shunting point in training sample, then judge the possibility that traffic congestion be present, carry out early warning.
2. the traffic congestion method for early warning as claimed in claim 1 based on semantic region segmentation, it is characterised in that early warning will be treated
The monitoring scene in region is divided into multiple semantic regions, including：
Motion dictionary is built according to monitoring scene, movement locus is divided into path segment sequence while path segment is compiled
Code, afterwards using the path segment after coding as a document, is modeled, using gibbs sampler using LDA topic models
Parameter in method estimation model, obtains " theme-word " matrix of K × V dimensionsThe number that K is the theme, V are motion word
Number, often row represent the probability of each motion word that theme Z is included；" theme-word " matrix tieed up according to K × V
The monitoring scene for treating prewarning area is divided into K semantic region and is numbered for each semantic region.
3. the traffic congestion method for early warning as claimed in claim 2 based on semantic region segmentation, it is characterised in that according to monitoring
Scenario building moves dictionary, including：
It is several rectangular areas that size is the pixel of P pixel × Q by monitoring scene cutting；
The direction of vehicle movement in monitoring scene is divided into four angular intervals： And
Assuming that monitoring scene size is M pixel × N number of pixel, number as (M/P) × (N/ that word is moved in dictionary is moved
Q) × 4, it is expressed as V, the motion word in dictionary is expressed asWherein a ∈ { 1,2,3,4 }, four angular intervals are corresponded to respectivelyAndb∈{1,2,…,M/
P}；c∈{1,2,…,N/Q}.
4. the traffic congestion method for early warning as claimed in claim 2 based on semantic region segmentation, it is characterised in that the fortune of vehicle
Dynamic rail mark originates in the frame of video for detecting this vehicle, and this frame of video is the start frame of first path segment of movement locus；
The frame of video for being not detected by this vehicle is terminated at, this frame of video is the end frame of last path segment of movement locus.
5. the traffic congestion method for early warning as claimed in claim 4 based on semantic region segmentation, it is characterised in that according to following
Movement locus is divided into path segment sequence by rule, it is assumed that vehicle A is the vehicle detected in traffic video stream：
Rule 1：When having detected that other vehicles enter or leave monitoring scene in current video frame, set previous frame of video as
The end frame of vehicle A current track fragment, present frame are the start frame of vehicle A next path segment；
Rule 2：Vehicle in continuous two frame of video is matched, if the vehicle A in present frame is looked for not in former frame
To the vehicle of matching, then start frame of the present frame as vehicle A first path segment is set；If the vehicle A in present frame
The vehicle of matching is can not find in a later frame, then sets end frame of the present frame as vehicle A last path segment；
Rule 3：When vehicle A enters the rectangular area in any monitoring scene, set previous frame of video and work as front rail as vehicle A
The end frame of mark fragment, present frame are the start frame of vehicle A next path segment；
Rule 4：Compare tracing point direction of the vehicle A movement locus in current video frame with vehicle A in previous video frame
In tracing point direction, if the direction of two tracing points is in different angular intervals, set previous frame of video as vehicle A's
The end frame of current track fragment, present frame are the start frame of vehicle A next path segment.
6. the traffic congestion method for early warning as claimed in claim 5 based on semantic region segmentation, it is characterised in that
Rectangular area according to residing for path segment and the angular interval residing for the direction of motion of path segment, path segment is reflected
It is mapped to corresponding motion word in motion dictionary；The direction of motion of path segment by first tracing point in path segment fortune
Dynamic direction represents.
7. the traffic congestion method for early warning as claimed in claim 6 based on semantic region segmentation, it is characterised in that early warning will be treated
The monitoring scene in region is divided into K semantic region, including：
1) the motion word of each self-contained most probable value in K theme is selected, by the rectangle region corresponding to this motion word
Seed region of the domain as K initial semantic region, the corresponding semantic region of a theme, theme are expressed as Zi, theme Zi
Corresponding semantic region is expressed as Ri, i=1,2 ..., K；
2) centered on seed region, 4, the upper and lower, left and right rectangle of the respective seed region in K semantic region is chosen successively
The rectangular area not being merged in region；It is if meeting merging criterion, the rectangular area of selection and K semantic region is each
From seed region be merged into new semantic region, while the rectangular area being merged is pressed into the respective heap in K semantic region
Stack；
3) it is treated as seed region not to take out a rectangular area respectively in empty storehouse from K semantic region and performs step
It is rapid 2), until the respective storehouse in K semantic region is empty；
4) rectangular area of remaining connection in monitoring scene is merged into new semantic region respectively；
5) semantic region segmentation terminates.
8. the traffic congestion method for early warning as claimed in claim 7 based on semantic region segmentation, it is characterised in that rectangular area
It is as follows with the merging criterion of seed region：
A. if rectangular area is only by a semantic region RiRectangular area to be combined is chosen for, i=1,2 ..., K, compares this
Probability of four motion words in K theme corresponding to rectangular area；If there is a motion word in theme ZiIn
Probability be the probability in K theme peak, then this rectangular area merge with seed region；
B. if rectangular area is chosen for rectangular area to be combined, 2≤n≤K, comparing this rectangular area by n semantic region
Probability of the four corresponding motion words in the theme corresponding to n semantic region；By this rectangular area and motion word
Seed region in the theme corresponding to n semantic region corresponding to probability highest theme merges.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610363423.9A CN105894815B (en) | 2016-05-27 | 2016-05-27 | Traffic congestion method for early warning based on semantic region segmentation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610363423.9A CN105894815B (en) | 2016-05-27 | 2016-05-27 | Traffic congestion method for early warning based on semantic region segmentation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105894815A CN105894815A (en) | 2016-08-24 |
| CN105894815B true CN105894815B (en) | 2018-03-02 |
Family
ID=56717971
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610363423.9A Expired - Fee Related CN105894815B (en) | 2016-05-27 | 2016-05-27 | Traffic congestion method for early warning based on semantic region segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105894815B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086794B (en) * | 2018-06-27 | 2022-03-01 | 武汉理工大学 | Driving behavior pattern recognition method based on T-LDA topic model |
| CN110211404B (en) * | 2019-06-30 | 2021-11-05 | 湖南华咨工程咨询有限公司 | Intelligent transportation system based on wifi is wireless |
| CN112699942A (en) * | 2020-12-30 | 2021-04-23 | 东软睿驰汽车技术（沈阳）有限公司 | Operating vehicle identification method, device, equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101706996A (en) * | 2009-11-12 | 2010-05-12 | 北京交通大学 | Method for identifying traffic status of express way based on information fusion |
| CN102289933A (en) * | 2011-08-08 | 2011-12-21 | 上海电科智能系统股份有限公司 | Method for predicting grades of spatial effect ranges of traffic event effects on city expressways |
| CN204087487U (en) * | 2014-07-24 | 2015-01-07 | 兰州交通大学 | A kind of road traffic acquiring video information and intelligent processing system |
| CN105608918A (en) * | 2016-03-31 | 2016-05-25 | 宇龙计算机通信科技(深圳)有限公司 | Road condition information monitoring method and system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7813870B2 (en) * | 2006-03-03 | 2010-10-12 | Inrix, Inc. | Dynamic time series prediction of future traffic conditions |
-
2016
- 2016-05-27 CN CN201610363423.9A patent/CN105894815B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101706996A (en) * | 2009-11-12 | 2010-05-12 | 北京交通大学 | Method for identifying traffic status of express way based on information fusion |
| CN102289933A (en) * | 2011-08-08 | 2011-12-21 | 上海电科智能系统股份有限公司 | Method for predicting grades of spatial effect ranges of traffic event effects on city expressways |
| CN204087487U (en) * | 2014-07-24 | 2015-01-07 | 兰州交通大学 | A kind of road traffic acquiring video information and intelligent processing system |
| CN105608918A (en) * | 2016-03-31 | 2016-05-25 | 宇龙计算机通信科技(深圳)有限公司 | Road condition information monitoring method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105894815A (en) | 2016-08-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110570658B (en) | Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering | |
| CN105070053B (en) | A kind of intelligent traffic monitoring video camera for recognizing rule-breaking vehicle motor pattern | |
| CN105894815B (en) | Traffic congestion method for early warning based on semantic region segmentation | |
| CN107204114A (en) | A kind of recognition methods of vehicle abnormality behavior and device | |
| CN102880692A (en) | Retrieval-oriented monitoring video semantic description and inspection modeling method | |
| CN104504897A (en) | Intersection traffic flow characteristic analysis and vehicle moving prediction method based on trajectory data | |
| CN103456175A (en) | Accompanying vehicle real-time detection method based on vehicle registration plate recognition and meshing monitoring | |
| CN103400391A (en) | Multiple-target tracking method and device based on improved random forest | |
| CN110738857A (en) | vehicle violation evidence obtaining method, device and equipment | |
| CN113936465B (en) | Traffic event detection method and device | |
| CN106778688A (en) | The detection method of crowd's throat floater event in a kind of crowd scene monitor video | |
| CN111768619A (en) | Express way vehicle OD point determining method based on checkpoint data | |
| Xiong et al. | Vehicle re-identification with image processing and car-following model using multiple surveillance cameras from urban arterials | |
| CN108156406A (en) | The information processing method and device of automobile data recorder | |
| CN114494998A (en) | Intelligent analysis method and system for vehicle data | |
| Wang et al. | Vision-based highway traffic accident detection | |
| CN110610118A (en) | Traffic parameter acquisition method and device | |
| Lochrane et al. | Modeling driver behavior in work and nonwork zones: Multidimensional psychophysical car-following framework | |
| CN115311458B (en) | Real-time expressway pedestrian intrusion event detection method based on multi-task learning | |
| Hussein et al. | Analysis of road user behavior and safety during New York City’s summer streets program | |
| Grigoropoulos et al. | Detection and Classification of Bicyclist Group Behavior for Automated Vehicle Applications | |
| Katariya et al. | A pov-based highway vehicle trajectory dataset and prediction architecture | |
| Sayed et al. | A novel approach for diagnosing road safety issues using automated computer vision techniques | |
| CN113408333B (en) | Method for distinguishing pedestrian traffic behaviors in subway station based on video data | |
| CN114814825B (en) | Vehicle track sensing and state extraction method based on radar and video fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180302 Termination date: 20210527 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |