CN105809954B - Traffic incidents detection method and system - Google Patents
Traffic incidents detection method and system Download PDFInfo
- Publication number
- CN105809954B CN105809954B CN201410840963.2A CN201410840963A CN105809954B CN 105809954 B CN105809954 B CN 105809954B CN 201410840963 A CN201410840963 A CN 201410840963A CN 105809954 B CN105809954 B CN 105809954B
- Authority
- CN
- China
- Prior art keywords
- state
- characteristic point
- movement
- entropy
- traffic
- 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.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 72
- 238000000034 method Methods 0.000 claims abstract description 26
- 230000033001 locomotion Effects 0.000 claims abstract description 24
- 230000008859 change Effects 0.000 claims abstract description 18
- 238000004364 calculation method Methods 0.000 claims abstract description 14
- 230000003287 optical effect Effects 0.000 claims description 12
- 230000001133 acceleration Effects 0.000 claims description 7
- SAZUGELZHZOXHB-UHFFFAOYSA-N acecarbromal Chemical compound CCC(Br)(CC)C(=O)NC(=O)NC(C)=O SAZUGELZHZOXHB-UHFFFAOYSA-N 0.000 claims 1
- 238000004458 analytical method Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 230000035772 mutation Effects 0.000 description 6
- 239000002245 particle Substances 0.000 description 6
- 230000006399 behavior Effects 0.000 description 5
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 206010039203 Road traffic accident Diseases 0.000 description 1
- 229910000831 Steel Inorganic materials 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000002547 anomalous effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000009017 pursuit movement Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Traffic Control Systems (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to traffic incidents detection method and system.Traffic incidents detection method, including：Movement-state obtaining step, obtain the movement-state of the motion change of each moment characteristic point, the Based on Feature Points vehicle；Entropy calculation procedure, entropy of the wagon flow at each moment is calculated based on the value of the movement-state at acquired each moment；And traffic incidents detection step, the entropy of the wagon flow that calculates at each moment is fitted at predetermined time intervals, traffic events are determined whether according to fitting result.
Description
Technical field
The present invention relates to traffic incidents detection method and system.
Background technology
In the past, in traffic incidents detection, the identification of traffic events is realized often through the processing of complexity.Such as to hand over
Exemplified by interpreter's part video identification, in this process, it is necessary to reference to image procossing, pattern-recognition, machine learning scheduling theory and side
Method, such as following technology be present：Initialization background, context table first against background model reach carries out further investigated with renewal,
Realize the adaptive foreground moving region detection under complex scene；Designed with reference to the morphological character of moving target with kinetic characteristic
Sorting algorithm based on multi-class support vector machine, the classification for realizing mixed traffic moving target judge, it is proposed that based on card
The multiple features Matching pursuitalgorithm of Kalman Filtering and the occlusion handling method based on historical movement information compensation, ensure that complicated screening
Motion state in the case of gear is accurately estimated；Consider the spatial character of track, directional characteristic, category Properties, it is proposed that fortune
The multi-level learning method of dynamic model formula, and thus construct and matched based on Bayes spatial models and based on origin and destination direction mould
The anomaly detection method of formula matching；The status attribute of moving target and the context-related information in traffic scene are mutually tied
Close, define the concrete concepts such as simple event, complicated event, general expression-form is provided for event recognition；It is basic herein
On, construct the elementary event recognition methods that is combined based on Bayes graders with logical constraint and based on hidden Markov mould
The complicated event recognition methods of type.
In addition, in the past, in the method detected for Video Events, it is proposed that the Video Events based on semanteme detect
Analysis method.This method carries out rational description and expression using a kind of general semantic meaning representation form to traffic events, and then
Land use models know effective automatic identification that method for distinguishing realizes event.This method around video do more physical exercises object identification features description
Video complicated event detection with classification, based on semanteme is excavated with analysis, Context event relevance and event level high-level semantic is retouched
State and understand four aspect key technology, it is proposed that adaptive combined invariant moment value do more physical exercises characteristics of objects description with
Sorting technique, the detection of track multi-tag hypergraph model and analysis complicated event method, temporal sequence association rule excavate event semantics and calculated
Method and the description of case grammar network structure framework understand video multithreading event technology.
The content of the invention
But in the apparatus and method detected for traffic events more than, various problems be present.
Firstly, there are cost it is high the problem of.It is diversified in the movement of traffic target of real world, classifies in identification
During, it is necessary to select the feature of magnanimity, and then realize more accurate, careful target category division.The traffic of moving target
Behavior complexity is various, and randomness is relatively strong, it is necessary to gather the track sample of magnanimity, and then establishes rationally effective behavior pattern
Practise model.Need to build Video Events database that is extensive, covering polymorphic type, the definition of event type and the unification of standard are just
In the expansion of video content semantic analysis correlative study work.
It additionally, there are the problem of complicated.Effective identification of traffic events and the context-related information of traffic events
Close association, on the basis of goal behavior analysis, it is also necessary to strengthen the INTELLIGENT IDENTIFICATION of context.Video file includes abundant
Semantic information is, it is necessary to describe and extract more attributive character in video, and find out relevance therein.Video semanteme description standard
Research.After analysis video semanteme, it is also necessary to the research on standard for describing video semanteme comprehensively, accurate and general.
Also, the problem of application is limited be present.Under current video detection technology, under night scenes, sleety weather, know
Other moving object is simultaneously classified, and the stability of system work is not still high, causes its application to be very limited.
Therefore, the present invention proposes that traffic events and traffic incidents detection technology easy to use can be judged rapidly.
The present invention provides a kind of traffic incidents detection method, including：Movement-state obtaining step, obtain each moment feature
The movement-state of the motion change of point, the Based on Feature Points vehicle；Entropy calculation procedure, based on acquired each moment
The value of the movement-state calculates entropy of the wagon flow at each moment；And traffic incidents detection step, at predetermined time intervals
The entropy of the wagon flow that calculates at each moment is fitted, traffic events are determined whether according to fitting result.
According to above-mentioned traffic incidents detection method, in the entropy calculation procedure, the value based on movement-state
Multiple state intervals are divided, calculate the quantity of characteristic point in each state interval with the ratio of number of whole characteristic points as each
The probability of the state interval, based on entropy described in the probability calculation.
According to above-mentioned traffic incidents detection method, the characteristic point be the vehicle in itself, in the movement-state
The movement-state is obtained by vehicle checker in obtaining step.
According to above-mentioned traffic incidents detection method, the characteristic point is the angle point of vehicle described in video image, in institute
State in movement-state obtaining step, the coordinate based on the angle point in video image at different moments, it is anti-to obtain each moment
Reflect the movement-state of the motion change of each angle point.
According to above-mentioned traffic incidents detection method, the movement-state is the size of the speed of characteristic point, characteristic point
The direction of speed, the either multinomial combination of the acceleration of characteristic point, any one of the position coordinates of characteristic point.
According to above-mentioned traffic incidents detection method, the movement-state is size and the institute of the speed of the characteristic point
State the combination in the direction of the speed of characteristic point, in the entropy calculation procedure, the size of the speed based on the characteristic point and
The direction of the speed of the characteristic point carries out the division of state interval, calculates by the size of the speed of the characteristic point and described
Feature in each state interval of the direction division of the speed of the characteristic point ratio counted out relative to all features of counting out is made
For the probability of each state interval.
According to above-mentioned traffic incidents detection method, when dividing state interval, each state interval of decile.
According to above-mentioned traffic incidents detection method, when dividing state interval, the not each state interval of decile.
According to above-mentioned traffic incidents detection method, in addition to：Corner Detection step, from each pixel in video image
In detect angle point；And angle point tracking step, it is tracked using motion of the optical flow tracking to the angle point, wherein, it is described
Characteristic point is the angle point of the vehicle.
The traffic incidents detection method stated according to entropy, in traffic incidents detection step, according to the straight of entropy linear fit
The slope of line calculates alarm level, using the beginning and ending time at the time of sliding window beginning and end of linear fit as alarm.
The present invention relates to a kind of traffic incident detecting system, including：Movement-state acquisition unit, the movement-state obtain
Portion is taken to obtain the movement-state of the motion change of each characteristic point of each moment reflection, the Based on Feature Points vehicle；Entropy meter
Calculation portion, the entropy calculating part calculate wagon flow when each based on the value of the movement-state at acquired each moment
The entropy at quarter；And traffic incidents detection portion, the traffic incidents detection portion is at predetermined time intervals to the wagon flow that calculates each
The entropy at moment is fitted, and traffic events are determined whether according to fitting result.
According to above-mentioned traffic incident detecting system, the entropy calculating part is divided multiple based on the value of movement-state
State interval, the quantity of characteristic point in each state interval is calculated with the ratio of number of whole characteristic points as each state
The probability in section, based on entropy described in the probability calculation.
According to above-mentioned traffic incident detecting system, the characteristic point be the vehicle in itself, in the movement-state
The movement-state is obtained by vehicle checker in obtaining step.
According to above-mentioned traffic incident detecting system, the characteristic point is the angle point of vehicle described in video image, described
Coordinate of the movement-state acquisition unit based on the angle point in video image at different moments, it is each to obtain current time reflection
The movement-state of the motion change of the angle point.
According to above-mentioned traffic incident detecting system, the movement-state is the size of the speed of characteristic point, characteristic point
The direction of speed, the either multinomial combination of the acceleration of characteristic point, any one of the position coordinates of characteristic point.
According to above-mentioned traffic incident detecting system, the movement-state is size and the institute of the speed of the characteristic point
The combination in the direction of the speed of characteristic point is stated, the movement-state is the size of the speed of the characteristic point and the characteristic point
Speed direction combination, the size of the speed of the entropy calculating part based on the characteristic point and the speed of the characteristic point
Direction carry out state interval division, calculate the side by the size of the speed of the characteristic point and the speed of the characteristic point
Feature into each state interval of division is counted out relative to the ratio that all features are counted out as the general of each state interval
Rate.
According to above-mentioned traffic incident detecting system, when dividing state interval, each state interval of decile.
According to above-mentioned traffic incident detecting system, when dividing state interval, the not each state interval of decile.
According to above-mentioned traffic incident detecting system, in addition to：Corner Detection portion, the Corner Detection portion is from video image
In each pixel in detect angle point；And angle point tracking portions, the angle point tracking portions are using optical flow tracking to the angle point
Motion be tracked, wherein, the characteristic point is the angle point of the vehicle.
According to above-mentioned traffic incident detecting system, the traffic incidents detection portion is according to the straight line of entropy linear fit
Slope calculates alarm level, using the beginning and ending time at the time of sliding window beginning and end of linear fit as alarm.
It is not necessary according to the invention that the video features of selection magnanimity, it is not necessary to realize accurate target classification and identification, no
Need the track sample of collection magnanimity, it is not necessary to build the Video Events storehouse of complexity, it is not necessary to which the context of traffic events is believed
Breath carries out INTELLIGENT IDENTIFICATION, it is not necessary to builds abundant semantic base.Using entropy the complexity of system is reduced to detect traffic events
Degree, and and then reduce application cost.It is in addition, in accordance with the invention it is possible to grand to detect by the microstate (i.e. entropy) of system
Event (i.e. traffic events) in sight.
In addition, in the traffic incidents detection using video image, due to using angle point, therefore angle point is hardly by illumination
The influence of condition, under the weather such as night scenes, sleet, it can also preferably detect traffic events.
Brief description of the drawings
Fig. 1 is the flow chart for showing traffic incidents detection method；
Fig. 2 shows the figure divided based on the size and Orientation of speed to state interval；
Fig. 3 is an illustration entropy time history plot；
Fig. 4 is the figure that an illustration is fitted to the curve of entropy；
Fig. 5 is the curve map of an illustration alarm level and time of fire alarming；
Fig. 6 is the flow chart for the concrete mode for showing traffic incidents detection method；
Fig. 7 is the structure chart for showing traffic incident detecting system；
Fig. 8 is the structure chart for the concrete structure for showing traffic incident detecting system；
Fig. 9 A~Fig. 9 C are the figures for showing to detect vehicle congestion situation, and Fig. 9 A are to show detected angle point
Figure, Fig. 9 B are the figures for the change for showing entropy, and Fig. 9 C are to show the figure that alarm level changes over time.
Embodiment
Below, the present invention is described in detail refer to the attached drawing.
Fig. 1 shows the flow chart of traffic incidents detection method.First, the fortune of the motion change of each moment characteristic point is obtained
Dynamic quantity of state, wherein the Based on Feature Points vehicle；Then, the value of the movement-state based on acquired each moment come
Calculate entropy of the wagon flow at each moment；Finally, intend at predetermined time intervals calculating the entropy of the wagon flow at each moment
Close, traffic events are determined whether according to fitting result.Traffic events mentioned here are to instigate traffic in a short time
The anomalous event varied widely.Traffic events, which have, is clearly distinguishable from the feature of normal traffic wagon flow, such as traffic accident,
Wagon flow mutation etc..Specifically, traffic events be, for example, traffic congestion, collide, thrust into, vehicle abnormality travel speed, vehicle doubling,
Car hits railing, car hits roadbed etc..The mutation of entropy, it is meant that traffic behavior is mutated, so as to mean that there occurs once
New traffic events.
First, as shown in Fig. 1 step S1, the movement-state of vehicle is obtained, the movement-state can be speed
Size, the amount relevant with speed such as direction, acceleration, position coordinates or the combination between them of speed.Obtain vehicle
The mode of movement-state be that can be obtained by the internet of things equipment such as vehicle checker terminal or other approach.But as long as can
Movement-state is obtained, is not particularly limited.
As shown in Fig. 1 step S2, after movement-state is obtained, entropy is calculated based on acquired movement-state.
So that movement-state is velocity field as an example, come calculate people in video, the velocity field that car collectively forms entropy.Fig. 2
The figure divided based on the size and Orientation of speed to state interval is shown.As illustrated in fig. 2, it is assumed that total of the vehicle of tracking
Number is N (i.e. total number of particles is N), according to the velocity attitude and size of current time each particle, calculates the entropy at this moment.Meter
It is as follows to calculate flow：
Interval division is carried out to the velocity attitude of particle, from 0-360 degree, is divided into M deciles.To the velocity magnitude of particle
Interval division is carried out, is divided into the by stages such as N.
Calculate total state interval C (M, N)=M*N.
The two-dimentional frequency H (i, j) of calculating speed, that is, fall into state interval Qi*jParticle number, acquisition fall into each state
Deng the particle number of by stages, i is greater than being equal to 1 integer for being less than or equal to M, and j is greater than being equal to 1 integer for being less than or equal to N.
X-Y scheme H (i, j) is drawn, as shown in Fig. 2 abscissa is velocity magnitude, ordinate is velocity attitude.
As shown in formula (1), the probability P of each state interval is calculatedij。
Formula (1)
As shown in formula (2), the entropy S at each moment is calculated
Formula (2)
Entropy is multiplied by the result sum of the log functions of probability for the probability of each state interval.
Here movement-state is not limited to the speed of vehicle, as long as the quantity of state of the motion of vehicle can be reflected i.e.
Can.Such as movement-state can only be the size of the speed of vehicle, it can also only be the direction of the speed of vehicle, can also be
Combinations thereof.In addition, movement-state can also be the acceleration of vehicle, the position coordinates of vehicle can also be.
In addition, when dividing state interval based on movement-state, isloation state can not also be waited with decile state interval
Section.
By repeating above-mentioned each step, the entropy at all moment can be calculated, and draw entropy and change over time
Curve.Fig. 3 shows one of entropy time history plot.Fig. 3 abscissa is the time, and ordinate is entropy.Such as
Low from the 1 to 10th moment entropy shown in Fig. 3, system is in order state；The moment, entropy increased from 11 to 20, and system is in nothing
Sequence state.
As shown in Fig. 1 step S3, traffic incidents detection is carried out using the entropy calculated in above-mentioned steps.According to
The change of entropy, detect the mutation of traffic events and wagon flow.The mutation of entropy, it is meant that traffic behavior is mutated, from
And mean that there occurs once new traffic events.
It is fitted for the entropy at each moment.Fig. 4 shows to carry out entropy one of linear fit.Here with Linear Quasi
Example is combined into illustrate.To the entropy at all moment calculated, (assuming that 2 seconds) are to entropy progress Linear Quasi at regular intervals
Close.As shown in figure 4, the 9 sublinears fitting of the line segment as shown in digital 1-9 has been carried out altogether.
According to the slope of every section of line segment of linear fit illustrated above, alarm level is calculated, alarm level is fit line
Slope divided by 90 degree.At the time of the beginning and ending time of alarm is sliding window beginning and end.As shown in figure 5, abscissa is report
The alert time, ordinate is alarm level.According to the slope of the linear fit of entropy, the alarm level at each moment is calculated.
It is above-mentioned based on entropy come during obtaining alarm level, however it is not limited to entropy is carried out curve fitting, as long as can obtain
Can be the fitting of any other mode to alarm level.
Embodiments of the present invention are illustrated above, the traffic events inspection of video image can be applied the present invention to
In survey.Specific mode is as described below.
Illustrated in case of detecting traffic events by video image.Here highlight using angle point come
Obtain movement-state and then carry out the entropy in video image and calculate.
In the picture, angle point is that two dimensional image brightness changes curvature maximum on violent point or image border curve
Point, is an important local feature, and it determines the shape of key area in image, embodies feature letter important in image
Breath, has decisive action to the contour feature of master goal, and mesh has substantially been grasped once it have found the contour feature of target also
Target shape, it is of great significance so angle point has in terms of target identification, images match, Image Reconstruction.
Further, since angle point has rotational invariance, therefore angle point is hardly influenceed by illumination condition, and it is in three dimensional field
Scape reconstruction, estimation, target following, target identification, image registration with match etc. computer vision field play it is extremely important
Effect.
In video image, vehicle can be represented by the multiple angle points of itself.The angle point of vehicle is, for example, two profiles
The intersecting point of line.The detection that detection tracking to vehicle can be simplified to multiple angle points of vehicle itself is followed the trail of to realize.Base
In the above-mentioned angle point the characteristics of, it can be held by detected angle point in traffic events important local special in video image
Sign, directly vehicle and personnel are identified so as to avoid.Vehicle is represented by angle point, key character letter can retained
Data volume is efficiently reduced while breath so that image procossing hour operation quantity is greatly reduced.Therefore, can for video image
With as shown in fig. 6, first, Corner Detection is carried out in step s 11, so as to obtain the angle point for representing vehicle.
The method of Corner Detection is varied, but can generally be divided into 4 classes：Corner Detection, base based on edge feature
Corner Detection in gray level image, the Corner Detection based on bianry image and the Corner Detection based on mathematical morphology.Angle point is examined
The principle of survey is most intuitively construed to：In any two mutually orthogonal direction, there is the point of large change.
It can be detected using Harris Corner Detection angle steel joints, harris angle points possess rotational invariance.
As shown in Fig. 6 step S12, optical flow tracking is carried out for the above-mentioned angle point being detected.
Light stream is divided into sparse optical flow and dense optical flow, each pixel of dense optical flow and speed or can be described as and displacement
Correlation, the method for being able to pursuit movement using dense optical flow have Horn-Shrunk methods, also block matching method.In the present invention
In illustrated by using exemplified by sparse optical flow.Need to specify one group of point (angle point), Ran Houli before tracked when calculating
With pyramid LK optical flow algorithms, motion is tracked.
As shown in Fig. 6 step S13, the movement-state of each angle point is obtained.In front and rear two frames video, if the time is t
And t+1, angle point i coordinate are respectively pit(x, y) and pi(t+1)(x, y).Coordinate based on time and angle point is calculated on the angle
Point i movement-state.The movement-state is, for example, the speed v of angle pointi=pi(t+1)(x, y)-pit(x, y).Wherein, x directions
Velocity component be vix=pi(t+1)(x)-pit(x), the velocity component in y directions is viy=pi(t+1)(y)-pit(y)。
As shown in Fig. 6 step S14, after movement-state is obtained, entropy is calculated based on acquired movement-state.
So that movement-state is velocity field as an example, come calculate people in video, the velocity field that car collectively forms entropy.Fig. 6 step S14
With S15 compared with step S2 and S3 in Fig. 1, in addition to the vehicle in embodiment 1 is replaced using angle point, other and embodiment
1 is identical, therefore is not repeated to illustrate.
Fig. 7 shows the structure chart of traffic incident detecting system.Traffic incident detecting system 1 shown in Fig. 7 includes motion shape
State amount acquisition unit 11, entropy calculating part 12 and traffic incidents detection portion 13.
Movement-state acquisition unit 11 obtains the movement-state for reflecting the motion change of each characteristic point at each moment, described
Based on Feature Points vehicle；Entropy calculating part 12 calculates wagon flow based on the value of the movement-state at acquired each moment
In the entropy at each moment；And traffic incidents detection portion 13 is at predetermined time intervals to the wagon flow that calculates described in each moment
Entropy is fitted, and traffic events are determined whether according to fitting result.
Entropy calculating part 12 divides multiple state intervals based on the value of movement-state, calculates in each state interval
Probability of the ratio of number of the quantity of characteristic point and whole characteristic points as each state interval, based on the probability calculation institute
State entropy.
The characteristic point can be the vehicle in itself, obtained in the movement-state acquisition unit 11 by vehicle checker
The movement-state.The characteristic point can also be the angle point of vehicle in video image, and movement-state portion 11 is based on described
Coordinate of the angle point in video image at different moments, obtain the motion shape of the motion change of each angle point of each moment reflection
State amount.
Movement-state can be the size of the speed of characteristic point, the direction of the speed of characteristic point, characteristic point acceleration,
The either multinomial combination of any one of position coordinates of characteristic point
Be in movement-state the speed of characteristic point size and characteristic point speed direction combination in the case of, institute
The direction for stating the size of speed of the entropy calculating part 12 based on the characteristic point and the speed of the characteristic point carries out state interval
Division, and be calculated as all characteristic points the speed of the size and the characteristic point by the speed of the characteristic point
The two-dimentional frequency that feature in each state interval of direction division is counted out.
In addition, when dividing state interval, can be with each state interval of decile, can not also each state interval of decile.
The traffic incidents detection portion 25 calculates alarm level according to the slope of the straight line of entropy linear fit, by Linear Quasi
Beginning and ending time at the time of the sliding window beginning and end of conjunction as alarm.
The above-mentioned traffic incident detecting system of the present invention specifically can be with as shown in figure 8, utilize figure for video image
The angle point of vehicle carries out the calculating of the movement-state of vehicle as in.
Traffic incident detecting system shown in Fig. 8 also includes：Corner Detection portion 21, the Corner Detection portion is from video image
In each pixel in detect angle point；And angle point tracking portions 22, the angle point tracking portions are using optical flow tracking to the angle
The motion of point is tracked, wherein, the characteristic point is the angle point of the vehicle.Fig. 8 traffic incident detecting system 20 utilizes
Corner Detection obtains the movement-state of angle point with tracking, such as can utilize the coordinate of angle point and be calculated at the time of correspondence
The speed of angle point.After the movement-state of angle point is obtained, the same processing of Fig. 7, entropy calculating part 24 and traffic events are carried out
Test section 25 has same function with the entropy calculating part of traffic incident detecting system 10 and traffic incidents detection portion 25, here
It is not repeated to describe.
Application examples 1
Reference picture 9A to Fig. 9 C, detection vehicle congestion situation is illustrated.Fig. 9 A are to show detected angle point
Figure, wherein, the circle in Fig. 9 A represents angle point, and Fig. 9 B are the figures for the change for showing entropy, and Fig. 9 C are to show alarm level at any time
Between the figure that changes.
According to the traffic incidents detection method and system of the present invention.Following apply can be carried out.Detection traffic thing in real time
Part, it can detect the traffic events of early warning；Real-time early warning is carried out to traffic events, according to the entropy tendency chart calculated in real time, to not
Carry out traffic tendency and make anticipation, early warning is carried out to the traffic events that will occur；To traffic events Realtime Alerts, according to real-time meter
Traffic events are alarmed by the entropy catastrophe of calculation, and alarm level is directly proportional to the mutation of entropy, and mutation is bigger, alarm
Higher grade.
According to above-mentioned, the present invention, which provides, can judge traffic events and traffic incidents detection side easy to use rapidly
Method and system.
The present invention is described by reference to the specific embodiment of the present invention above, but those skilled in the art are equal
Solution, can to these specific embodiments carry out it is various modification, combination and change, without departing from by appended claims or its etc.
The spirit and scope of the present invention that jljl limits.
Claims (16)
1. a kind of traffic incidents detection method, including：
Movement-state obtaining step, obtain the movement-state of the motion change of each moment characteristic point, the Based on Feature Points
Vehicle；
Entropy calculation procedure, entropy of the wagon flow at each moment is calculated based on the value of the movement-state at acquired each moment
Value；And
Traffic incidents detection step, the entropy of the wagon flow that calculates at each moment is fitted at predetermined time intervals, root
Traffic events are determined whether according to fitting result, in the entropy calculation procedure, are drawn based on the value of movement-state
Divide multiple state intervals, calculate the quantity of characteristic point in each state interval with the ratio of number of whole characteristic points as each institute
The probability of state interval is stated, based on entropy described in the probability calculation.
2. traffic incidents detection method as claimed in claim 1, it is characterised in that
The characteristic point is the angle point of vehicle described in video image,
In the movement-state obtaining step, the coordinate based on the angle point in video image at different moments, obtain
Each moment reflects the movement-state of the motion change of each angle point.
3. traffic incidents detection method as claimed in claim 1 or 2, it is characterised in that
The movement-state is the size of the speed of characteristic point, the direction of the speed of characteristic point, the acceleration of characteristic point, feature
The either multinomial combination of any one of position coordinates of point.
4. traffic incidents detection method as claimed in claim 1 or 2, it is characterised in that
The movement-state be the speed of the characteristic point size and the characteristic point speed direction combination,
In the entropy calculation procedure, the direction of the speed of the size of the speed based on the characteristic point and the characteristic point is entered
The division of row state interval, calculate by the speed of the characteristic point size and the characteristic point speed direction divide
Feature in each state interval is counted out relative to probability of the ratio that all features are counted out as each state interval.
5. traffic incidents detection method as claimed in claim 1, it is characterised in that
When dividing state interval, each state interval of decile.
6. traffic incidents detection method as claimed in claim 1, it is characterised in that
When dividing state interval, the not each state interval of decile.
7. traffic incidents detection method as claimed in claim 1 or 2, it is characterised in that
Also include：
Corner Detection step, angle point is detected from each pixel in video image；And
Angle point tracking step, it is tracked using motion of the optical flow tracking to the angle point,
Wherein, the characteristic point is the angle point of the vehicle.
8. traffic incidents detection method as claimed in claim 1 or 2, it is characterised in that
In traffic incidents detection step, alarm level is calculated according to the slope of the straight line of entropy linear fit, by linear fit
Sliding window beginning and end at the time of beginning and ending time as alarm.
9. a kind of traffic incident detecting system, including：
Movement-state acquisition unit, the movement-state acquisition unit obtain the motion change of each characteristic point of each moment reflection
Movement-state, the Based on Feature Points vehicle；
Entropy calculating part, the entropy calculating part calculate car based on the value of the movement-state at acquired each moment
Flow the entropy at each moment；And
Traffic incidents detection portion, the traffic incidents detection portion is at predetermined time intervals to the wagon flow that calculates described in each moment
Entropy is fitted, and determines whether traffic events according to fitting result, the entropy calculating part is based on movement-state
Value and divide multiple state intervals, calculate the quantity of characteristic point in each state interval and the ratio of number of whole characteristic points
As the probability of each state interval, based on entropy described in the probability calculation.
10. traffic incident detecting system as claimed in claim 9, it is characterised in that
The characteristic point is the angle point of vehicle described in video image,
Coordinate of the movement-state acquisition unit based on the angle point in video image at different moments, it is anti-to obtain each moment
Reflect the movement-state of the motion change of each angle point.
11. the traffic incident detecting system as described in claim 9 or 10, it is characterised in that
The movement-state is the size of the speed of characteristic point, the direction of the speed of characteristic point, the acceleration of characteristic point, feature
The either multinomial combination of any one of position coordinates of point.
12. the traffic incident detecting system as described in claim 9 or 10, it is characterised in that
The movement-state be the speed of the characteristic point size and the characteristic point speed direction combination, it is described
The direction of the size of speed of the entropy calculating part based on the characteristic point and the speed of the characteristic point carries out drawing for state interval
Point, calculate by the speed of the characteristic point size and the characteristic point speed direction divide each state interval in
Feature is counted out relative to probability of the ratio that all features are counted out as each state interval.
13. traffic incident detecting system as claimed in claim 9, it is characterised in that
When dividing state interval, each state interval of decile.
14. traffic incident detecting system as claimed in claim 9, it is characterised in that
When dividing state interval, the not each state interval of decile.
15. traffic incident detecting system as claimed in claim 9, it is characterised in that
Also include：
Corner Detection portion, the Corner Detection portion detect angle point from each pixel in video image；And
Angle point tracking portions, the angle point tracking portions are tracked using motion of the optical flow tracking to the angle point,
Wherein, the characteristic point is the angle point of the vehicle.
16. the traffic incident detecting system as described in claim 9 or 10, it is characterised in that
The traffic incidents detection portion calculates alarm level according to the slope of the straight line of entropy linear fit, by the cunning of linear fit
Beginning and ending time at the time of dynamic window beginning and end as alarm.
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410840963.2A CN105809954B (en) | 2014-12-30 | 2014-12-30 | Traffic incidents detection method and system |
| PCT/CN2015/099526 WO2016107561A1 (en) | 2014-12-30 | 2015-12-29 | Traffic event detection method and system |
| SG11201705390RA SG11201705390RA (en) | 2014-12-30 | 2015-12-29 | Traffic event detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410840963.2A CN105809954B (en) | 2014-12-30 | 2014-12-30 | Traffic incidents detection method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105809954A CN105809954A (en) | 2016-07-27 |
| CN105809954B true CN105809954B (en) | 2018-03-16 |
Family
ID=56284291
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410840963.2A Active CN105809954B (en) | 2014-12-30 | 2014-12-30 | Traffic incidents detection method and system |
Country Status (3)
| Country | Link |
|---|---|
| CN (1) | CN105809954B (en) |
| SG (1) | SG11201705390RA (en) |
| WO (1) | WO2016107561A1 (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108256380A (en) * | 2016-12-28 | 2018-07-06 | 南宁市浩发科技有限公司 | Road traffic exception automatic testing method |
| CN108345894B (en) * | 2017-01-22 | 2019-10-11 | 北京同方软件有限公司 | A kind of traffic incidents detection method based on deep learning and entropy model |
| CN109145732B (en) * | 2018-07-17 | 2022-02-15 | 东南大学 | Black smoke vehicle detection method based on Gabor projection |
| CN109887276B (en) * | 2019-01-30 | 2020-11-03 | 北京同方软件有限公司 | Night traffic jam detection method based on fusion of foreground extraction and deep learning |
| CN113255405B (en) * | 2020-02-12 | 2024-06-25 | 广州汽车集团股份有限公司 | Parking space line identification method and system, parking space line identification equipment and storage medium |
| CN113286194A (en) * | 2020-02-20 | 2021-08-20 | 北京三星通信技术研究有限公司 | Video processing method and device, electronic equipment and readable storage medium |
| CN112053563B (en) * | 2020-09-16 | 2023-01-20 | 阿波罗智联(北京)科技有限公司 | Event detection method and device applicable to edge computing platform and cloud control platform |
| CN112464734B (en) * | 2020-11-04 | 2023-09-15 | 昆明理工大学 | Automatic identification method for walking motion characteristics of quadruped based on vision |
| CN113920728B (en) * | 2021-10-11 | 2022-08-12 | 南京微达电子科技有限公司 | Detection and early warning method and system for obstacles thrown on highway |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2007026301A (en) * | 2005-07-20 | 2007-02-01 | Matsushita Electric Ind Co Ltd | Stopping/low-speed vehicle detector and stopping/low-speed vehicle detection method |
| CN101639983A (en) * | 2009-08-21 | 2010-02-03 | 任雪梅 | Multilane traffic volume detection method based on image information entropy |
| CN102436740A (en) * | 2011-09-29 | 2012-05-02 | 东南大学 | Automatic detection method of traffic incident on highway |
| CN102542797A (en) * | 2010-12-09 | 2012-07-04 | 财团法人工业技术研究院 | Image-based traffic parameter detection system and method and computer program product |
| CN103971521A (en) * | 2014-05-19 | 2014-08-06 | 清华大学 | Method and device for detecting road traffic abnormal events in real time |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2816919B2 (en) * | 1992-11-05 | 1998-10-27 | 松下電器産業株式会社 | Spatial average speed and traffic volume estimation method, point traffic signal control method, traffic volume estimation / traffic signal controller control device |
| CN100573618C (en) * | 2008-07-07 | 2009-12-23 | 山东省计算中心 | A kind of traffic intersection four-phase vehicle flow detection method |
-
2014
- 2014-12-30 CN CN201410840963.2A patent/CN105809954B/en active Active
-
2015
- 2015-12-29 SG SG11201705390RA patent/SG11201705390RA/en unknown
- 2015-12-29 WO PCT/CN2015/099526 patent/WO2016107561A1/en active Application Filing
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2007026301A (en) * | 2005-07-20 | 2007-02-01 | Matsushita Electric Ind Co Ltd | Stopping/low-speed vehicle detector and stopping/low-speed vehicle detection method |
| CN101639983A (en) * | 2009-08-21 | 2010-02-03 | 任雪梅 | Multilane traffic volume detection method based on image information entropy |
| CN102542797A (en) * | 2010-12-09 | 2012-07-04 | 财团法人工业技术研究院 | Image-based traffic parameter detection system and method and computer program product |
| CN102436740A (en) * | 2011-09-29 | 2012-05-02 | 东南大学 | Automatic detection method of traffic incident on highway |
| CN103971521A (en) * | 2014-05-19 | 2014-08-06 | 清华大学 | Method and device for detecting road traffic abnormal events in real time |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2016107561A1 (en) | 2016-07-07 |
| SG11201705390RA (en) | 2017-08-30 |
| CN105809954A (en) | 2016-07-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105809954B (en) | Traffic incidents detection method and system | |
| Atev et al. | A vision-based approach to collision prediction at traffic intersections | |
| CN106980829B (en) | Abnormal behaviour automatic testing method of fighting based on video analysis | |
| Santosh et al. | Tracking multiple moving objects using gaussian mixture model | |
| Qu et al. | Moving vehicle detection with convolutional networks in UAV videos | |
| US20090319560A1 (en) | System and method for multi-agent event detection and recognition | |
| CN101727672A (en) | Method for detecting, tracking and identifying object abandoning/stealing event | |
| CN103942533A (en) | Urban traffic illegal behavior detection method based on video monitoring system | |
| CN100565557C (en) | System for tracking infrared human body target based on corpuscle dynamic sampling model | |
| CN102063613A (en) | People counting method and device based on head recognition | |
| CN106570490A (en) | Pedestrian real-time tracking method based on fast clustering | |
| CN103336947A (en) | Method for identifying infrared movement small target based on significance and structure | |
| KR101472674B1 (en) | Method and apparatus for video surveillance based on detecting abnormal behavior using extraction of trajectories from crowd in images | |
| Malhi et al. | Vision based intelligent traffic management system | |
| Lejmi et al. | Event detection in video sequences: Challenges and perspectives | |
| Kushwaha et al. | Automatic multiple human detection and tracking for visual surveillance system | |
| Wang et al. | Traffic camera anomaly detection | |
| CN110889347A (en) | Density traffic flow counting method and system based on space-time counting characteristics | |
| CN108876807A (en) | A kind of real-time piggyback satellite image motion object detection tracking | |
| Soleh et al. | A real time vehicle counting based on adaptive tracking approach for highway videos | |
| Srilekha et al. | A novel approach for detection and tracking of vehicles using Kalman filter | |
| Xiong et al. | Crowd density estimation based on image potential energy model | |
| Hao et al. | A detection method of abnormal event in crowds based on image entropy | |
| Khan et al. | Multiple moving vehicle speed estimation using Blob analysis | |
| Kulkarni et al. | Managing crowd density and social distancing |
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 |