CN108376406A - A kind of Dynamic Recurrent modeling and fusion tracking method for channel blockage differentiation - Google Patents
A kind of Dynamic Recurrent modeling and fusion tracking method for channel blockage differentiation Download PDFInfo
- Publication number
- CN108376406A CN108376406A CN201810017676.XA CN201810017676A CN108376406A CN 108376406 A CN108376406 A CN 108376406A CN 201810017676 A CN201810017676 A CN 201810017676A CN 108376406 A CN108376406 A CN 108376406A
- Authority
- CN
- China
- Prior art keywords
- channel
- background
- model
- pixel
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/215—Motion-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to technical field of computer vision, specially a kind of Dynamic Recurrent modeling differentiated for channel blockage and fusion tracking method.The present invention uses Dynamic Recurrent background modeling algorithm, sense channel background and foreground information to channel background model first, is partitioned into the dynamic such as channel pedestrian, deposit and stationary body；Then, binding object color and position feature timely and accurately track object using Fusion Features track algorithm, extract object space characteristic information, and judgment object is to take background local condition more new strategy for different conditions in movement or stationary state.Finally, the residence time according to the object of tracking in channel and width accounting determine whether to block object.Channel solid blockade can be effectively detected out in the method for the present invention, improve the safety assurance ability of the multiple unit of fire, solve the occupied security hidden trouble such as fire escape, emergency exit in the high-risk unit of fire.
Description
Technical field
The invention belongs to technical field of computer vision, and in particular to a kind of Dynamic Recurrent for channel blockage differentiation is built
Mould and fusion tracking method.
Background technology
With the development of the society, people constantly pursue more comfortable, well-to-do house and working space, cause more and more
Fire escape, extra exit are occupied, once there is fire, seriously affect evacuating personnel and fire-fighting rescue, cause a large amount of people
Member's injures and deaths and property loss, cause the concern of more and more people.However, channel blockage detects, depends merely on artificial investigation and at all can not
It realizes, although relevant department arranges camera head monitor at corresponding critical passage crossing at present, finally still needs artificial basis
Monitoring video goes to judge whether channel blocks, without fundamentally solving channel blockage test problems.
Currently, people achieve certain achievement in channel blockage detection.A kind of passageway for fire apparatus based on images match
Background image need to only be matched with real-time monitoring image in specified region, judged specified by obstacle detection method technology
Whether there are obstacles block in region, implement simply, but background image can not constantly update, and cannot adapt to disappear very well
The variation of anti-channel complex environment；A kind of passageway for fire apparatus safety detection method based on adaptive background study, this method use
The foreground target detection of HOG features and adaptive RTS threshold adjustment strategy, can adapt to the variation of channel complex environment very well, improve object
The accuracy rate and robustness that physical examination is surveyed, but not accounting for the short time appears in channel object, and the case where be not belonging to block；Also
There is one kind being based on obstacle detection method technology combined of multi-sensor information, the technology combination millimetre-wave radar and computer regard
Feel technology substantially increases the reliability of passage barrier detection, but party's law technology implementation cost is big, is unfavorable for extensive
Popularization.
Invention content
The purpose of the invention is to overcome the deficiencies in the prior art, propose a kind of dynamic can be used for channel blockage differentiation
Recusive modeling and fusion tracking method take a kind of dynamic local condition to detect and update method channel Gaussian Background model,
The track algorithm of fusion color and position feature tracks foreground target, residence time and width in conjunction with tracking object in channel
Whether degree accounting judgment object results in blockage, and realizes that channel blockage detects automatically.
The Dynamic Recurrent modeling proposed by the present invention that can be used for channel blockage differentiation and fusion tracking method, including following step
Suddenly：
Step 1, to channel background N frame image Dynamic Recurrent background modelings, by real-time monitoring images pixel and background model
Comparison is partitioned into the dynamic such as channel pedestrian, deposit and stationary body；
Step 2, the tracked object space characteristic information of extraction judge that it is to be in movement or stationary state to track object,
Background model local condition more new strategy is taken for different states；
Step 3, the channel object to being partitioned into, using the track algorithm of binding object color and position feature, in real time with
Track channel object；
Step 4 calculates residence time and width accounting of the step 2 tracking object in channel, determines whether to block object,
And according to the degree of blocking, corresponding alarm warning is sent out.
In step (1), takes a kind of dynamic local condition to detect and update method channel Gaussian Background model, detect
Channel background model and foreground model are partitioned into channel foreground to meeting the pixel value binaryzation of foreground and background model respectively
Object.
In step (2), the distance of the centroid position of two frames before and after tracking object is calculated, it is right if being less than preset threshold value
Pixel in object range stops the update of step (1) background model, conversely, then normal update.
In step (3), the track algorithm using binding object color and position feature, i.e., CamShift algorithms and
Kalman filter track algorithm is linearly combined, and CamShift algorithms are for predicting mass center position of object, Kalman filters
Centroid position of the wave track algorithm for constantly correcting object prediction.
In step (4), the residence time refers to object from channel is entered to the time of leaving channel, and width accounting refers to object
The width of contact channel ground surface accounts for the ratio of channel width；Only at the appointed time, it is stifled to detect that object all meets channel
The width accounting of plug can be considered as just channel blockage.
The operation of each step is further described below.
Step 1：Using Dynamic Recurrent background modeling algorithm, sense channel background and foreground information, be partitioned into channel pedestrian,
The dynamic such as deposit and stationary body, step 2：The tracked object space characteristic information of extraction judges that it is in fortune to track object
Dynamic or stationary state, background model local condition more new strategy is taken for different states；Detailed process is as follows：
(1) assume that sampled value of certain pixel before t moment is x in videot=[x(t),x(t-1),...,x(t-T)], to sample
This collection xtGauss hybrid models are established, i.e.,：
Wherein,Respectively represent the weights of m-th of Gauss model, mean value, variance and height
This function；B, F respectively represent background and foreground model, and M represents Gauss model number；
(2) to newly there is pixel xiCompare successively with current M model, comparative approach is：If there is | xi,t-μj i,t-1| ＜
2.5σj i,t-1, then it is assumed that pixel xiMeet current background model, is just considered as background pixel, is otherwise considered as foreground pixel；Wherein,
xi,tIndicate the pixel value of the pixel i in t frame images, μj i,t-1, σj i,t-1It is illustrated respectively in pixel after t-1 frame images are trained
The mean value and standard deviation of j-th of Gauss model in the mixed Gauss model of point i；
(3) once some Gauss model successful match, then the Gauss model is using pixel value xiUpdate, other Gauss models
It remains unchanged, more new formula is：
Wherein,It is the learning rate of variation, value range isI.e. when the static or slow object of movement
When, background model learning rate in object rangePart stops update, other range background models normally update；
(4) weights of M Gauss model are normalized and is ranked sequentially by size, B Gauss model is made before therefrom choosing
For background model, it is left (M-B) a Gauss model as foreground model, wherein background discriminateB
Indicate the smallest positive integral value for meeting b-th of Gaussian Background model of background discriminate, Ψ indicates the ratio that background accounts for, in this implementation
Middle value is 0.9；
(5) according to foreground and background model, binary conversion treatment is carried out to image, that is, meets the pixel of foreground and background model
Value is respectively set to 1 and 0, carries out connected domain analysis to foreground bianry image, by the region merger being linked to be, is divided into one kind, i.e.,
For subject Oi(i=1,2 ... N), and its boundary position information is extracted, obtain each subject OiProfile Ri。
Step 3, to obtaining channel foreground object Oi, the track algorithm of fusion color and position feature is taken, is tracked
Foreground target Oi, extraction object OiPosition feature information, detailed process are as follows：
(1) initialization search box is chosen, to OiProfile RiBackprojection operations are carried out, if (x, y) is RiIn a bit, I
(x, y) is the probability value at color probability figure (x, y), calculates separately RiZeroth order square M00With first moment M10、M01：
It can thus be concluded that RiCentroid position：Mi(xi,yi)=(M01/M00,M10/M00)；
(2) RiCentroid position Mi(xi,yi) it is input to Kalman filter state equation Mk,i=AMk-1,i+ W is measured
Equation Zk,i=HMk,i+ V, update equation M'k,i=Mk,i+Kk(Zk,i-HMk,i), obtain more accurate target location.
Wherein, Mk-1,iAnd Mk,iI-th of object is indicated respectively, in k-1 and k moment center-of-mass coordinate quantity of states, Zk,iAnd M'k,i
I-th of object is respectively represented, in k moment center-of-mass coordinate measured values and correction value, A is Kalman filter quantity of state transfer matrix, H
It is transformed matrix of the Kalman filter state variable to measurand, KkIt is Kalman filter gain matrix, stochastic variable W, V point
It Biao Shi not system noise matrix and measurement noise matrix；
(3) according to two frame O of video moment k-1 and kiCenter-of-mass coordinate Mk-1,i(xk-1,i,yk-1,i), Mk,i(xk,i,yk,i) judge
Channel object space state；If 2 barycenter Euclidean distance di：
If di＜ Dset, think that i-th of object is static or mobile slow at this time, then use background local condition more
Newly, i.e., in RiBackground stops update in range, other range backgrounds normally update, wherein DsetIndicate preset 2 barycenter Europe
Formula distance value, the value preferably between 1~10 pixel；
(4) size of search box is adjusted according to barycenter, if displacement distance is more than preset fixed value, is constantly searched for, directly
It is less than preset fixed value to the distance of the center of search box and center-of-mass coordinate or operation times reaches maximum value and stop meter
It calculates.
Step 4：The residence time in channel and width accounting according to the object of tracking determine whether to block object, stagnant
It refers to time of the object from entrance channel to leaving channel to stay the time, and width accounting refers to the width of object contact channel ground surface
The ratio of channel width is accounted for, detailed process is as follows：
(1) channel width w is setA, i-th of object OiIn the patch ground width w at τ momenti(τ)；Every T in time TsIt extracts
Object considerately width sample data calculates its average patch ground widthN number of object overall average patch ground is calculated again
Width accounts for the ratio V of channel widthOOPR：
(2) according to T and VOOPRBig wisp stopping state falls into three classes：In time T, tracking is consistently present in channel
Object, once object leaving channel, then exclude influence of this object to channel blockage, work as VOOPR50%~70%, it is considered as
Yellow blocks；VOOPR70%~90%, it is considered as orange blocking；VOOPRMore than 90% or more, it is considered as red blocking.
Channel blockage provided by the invention sentences method for distinguishing compared with prior art, has following advantage：
The present invention is taken a kind of dynamic local condition to detect and update method, overcome to channel Gaussian mixture model-universal background model
Background model is sensitive to external conditions such as illumination and foreground melts problem；Take a kind of fusion color and position feature with
Track algorithm keeps track foreground target improves the success rate and robustness of tracking；Residence time by calculating tracking object in channel
With width accounting, whether judgment object causes channel blockage, exclusion short time to appear in clogging caused by the object of channel, accurate
Really reasonably realize the automatic detection of channel blockage.In addition to this, technical method of the invention also has development cost low, is conducive to
It is large-scale to promote and apply.
Description of the drawings
Fig. 1 is the flow diagram that the present invention realizes.
Fig. 2 is characterized fusion tracking schematic diagram.
Fig. 3 is the unobstructed design sketch in channel.
Fig. 4 is that channel slightly blocks design sketch.
Fig. 5 is channel blockage design sketch.
Fig. 6 is channel Severe blockage design sketch.
Specific implementation mode
With reference to implementation steps and attached drawing, the present invention is described in further detail, but embodiments of the present invention are not
It is limited to this.
The Dynamic Recurrent modeling proposed by the present invention that can be used for channel blockage differentiation and fusion tracking method, as shown in Figure 1,
Specific implementation step is as follows：
Step 1：Using Dynamic Recurrent background modeling algorithm, sense channel background and foreground information, be partitioned into channel pedestrian,
The dynamic such as deposit and stationary body, detailed process are as follows：
(1) assume that sampled value of certain pixel before t moment is x in videot=[x(t),x(t-1),...,x(t-T)], to sample
This collection xtGauss hybrid models are established, i.e.,：
Wherein,Respectively represent the weights of m-th of Gauss model, mean value, variance and height
This function；B, F respectively represent background and foreground model, and M represents Gauss model number；
(2) to newly there is pixel xiCompare successively with current M model, comparative approach is：If there is | xi,t-μj i,t-1| ＜
2.5σj i,t-1, then it is assumed that pixel xiMeet current background model, be just considered as background pixel, be otherwise considered as foreground pixel, wherein
xi,tIndicate the pixel value of the pixel i in t frame images, μj i,t-1, σj i,t-1It is illustrated respectively in pixel after t-1 frame images are trained
The mean value and standard deviation of j-th of Gauss model in the mixed Gauss model of point i；
(3) once some Gauss model successful match, then the Gauss model is using pixel value xiUpdate, other Gauss models
It remains unchanged, more new formula is：
Wherein,It is the learning rate of variation, value range isI.e. when the static or slow object of movement,
Background model learning rate in object rangePart stops update, other range background models normally update；
(4) weights of M Gauss model are normalized and is ranked sequentially by size, B Gauss model is made before therefrom choosing
For background model, it is left (M-B) a Gauss model as foreground model, wherein background discriminateB
Indicate the smallest positive integral value for meeting b-th of Gaussian Background model of background discriminate, Ψ indicates the ratio that background accounts for, in this implementation
Middle value is 0.9；
(5) according to foreground and background model, binary conversion treatment is carried out to image, that is, meets the pixel of foreground and background model
Value is respectively set to 1 and 0, carries out connected domain analysis to foreground bianry image, by the region merger being linked to be, is divided into one kind, i.e.,
For subject Oi(i=1,2 ... N), and its boundary position information is extracted, obtain each subject OiProfile Ri；
Step 2：Channel foreground object O is obtained by step 1i, take the track algorithm of fusion color and position feature
Track foreground target Oi, extraction object OiPosition feature information, detailed process are as follows：
(1) initialization search box is chosen, to OiProfile RiBackprojection operations are carried out, if (x, y) is RiIn a bit, I
(x, y) is the probability value at color probability figure (x, y), calculates separately RiZeroth order square M00With first moment M10、M01：
It can thus be concluded that RiCentroid position：Mi(xi,yi)=(M01/M00,M10/M00)；
(2) RiCentroid position Mi(xi,yi) it is input to Kalman filter state equation Mk,i=AMk-1,i+ W is measured
Equation Zk,i=HMk,i+ V, update equation M'k,i=Mk,i+Kk(Zk,i-HMk,i), obtain more accurate target location.
Wherein, Mk-1,iAnd Mk,iI-th of object is indicated respectively, in k-1 and k moment center-of-mass coordinate quantity of states, Zk,iAnd M'k,i
I-th of object is respectively represented, in k moment center-of-mass coordinate measured values and correction value, A is Kalman filter quantity of state transfer matrix, H
It is transformed matrix of the Kalman filter state variable to measurand, KkIt is Kalman filter gain matrix, stochastic variable W, V point
It Biao Shi not system noise matrix and measurement noise matrix；
(3) according to two frame O of video moment k-1 and kiCenter-of-mass coordinate Mk-1,i(xk-1,i,yk-1,i), Mk,i(xk,i,yk,i) judge
Channel object space state；If 2 barycenter Euclidean distance di：
If di＜ Dset, think that i-th of object is static or mobile slow at this time, then use background local condition more
Newly, i.e., in RiBackground stops update in range, other range backgrounds normally update, wherein DsetIndicate preset 2 barycenter Europe
Formula distance value, the value preferably between 1~10 pixel, value is 5 pixels in the present embodiment；
(4) size of search box is adjusted according to barycenter, if displacement distance is more than preset fixed value, is constantly searched for, directly
It is less than preset fixed value to the distance of the center of search box and center-of-mass coordinate or operation times reaches maximum value and stop meter
It calculates；
Step 3：The residence time in channel and width accounting according to the object of tracking determine whether to block object, stagnant
It refers to time of the object from entrance channel to leaving channel to stay the time, and width accounting refers to the width of object contact channel ground surface
The ratio of channel width is accounted for, detailed process is as follows：
(1) channel width w is setA, i-th of object OiIn the patch ground width w at τ momenti(τ)；Every T in time TsIt extracts
Object considerately width sample data calculates its average patch ground widthN number of object overall average patch ground is calculated again
Width accounts for the ratio V of channel widthOOPR：
(2) according to T and VOOPRBig wisp stopping state falls into three classes：In time T, tracking is consistently present in channel
Object, once object leaving channel, then exclude influence of this object to channel blockage, work as VOOPR50%~70%, it is considered as
Yellow blocks；VOOPR70%~90%, it is considered as orange blocking；VOOPRMore than 90% or more, it is considered as red blocking.
According to above-mentioned specific implementation step carry out channel blockage experiment test, it is unobstructed for channel, it is slight block, block and
Four kinds of situations of Severe blockage, as listed in table 1, experimental result is as seen in figures 3-6 for detection data：
Table 1 gives, and within 6 minutes time, was detected, can calculate in channel width to channel object every 1 minute
Spend wAUnder=600 (pixel), work as VOOPRBe 58.30%, channel occur it is slight block, when object width constantly increases, VOOPRIt reaches
To 72.80%, channel blockage, when object width continues growing, VOOPRReach 92.92%, channel Severe blockage.
1 channel blockage experimental result of table is tested
In order to which further image illustrates the effect of the method provided by the present invention, as seen in figures 3-6：
Fig. 3 indicates that channel is unobstructed, and first is classified as PASS VIDEO monitoring schematic diagram, and second, which is classified as passage obstruction body foreground, shows
It is intended to, third is classified as channels track object schematic diagram；
Fig. 4 indicates that channel slightly blocks, and first is classified as PASS VIDEO monitoring schematic diagram, before second is classified as passage obstruction body
Scape schematic diagram, third are classified as channels track object schematic diagram；
Fig. 5 indicates channel blockage, and first is classified as PASS VIDEO monitoring schematic diagram, and second, which is classified as passage obstruction body foreground, shows
It is intended to, third is classified as channels track object schematic diagram, and pedestrian passes through difficulty at this time.
Fig. 6 indicates channel Severe blockage, and first is classified as PASS VIDEO monitoring schematic diagram, before second is classified as passage obstruction body
Scape schematic diagram, third are classified as channels track object schematic diagram, and pedestrian can not pass through at all at this time.
Claims (4)
1. a kind of Dynamic Recurrent modeling differentiated for channel blockage and fusion tracking method, which is characterized in that including following step
Suddenly：
Step 1, to channel background N frame image Dynamic Recurrent background modelings, real-time monitoring images pixel and background model are compared,
It is partitioned into the dynamic such as channel pedestrian, deposit and stationary body；
Step 2, the tracked object space characteristic information of extraction judge that it is to be in movement or stationary state to track object, for
Different states takes background model local condition more new strategy；
Step 3, the channel object to being partitioned into, using the track algorithm of binding object color and position feature, real-time tracking is logical
Road object；
Step 4 calculates residence time and width accounting of the step 2 tracking object in channel, determines whether to block object, and root
According to the degree of blocking, corresponding alarm warning is sent out.
2. the Dynamic Recurrent modeling according to claim 1 differentiated for channel blockage and fusion tracking method, feature
It is, the detailed process of the step 1 and step 2 is as follows：
(1) assume that sampled value of certain pixel before t moment is x in videot=[x(t),x(t-1),...,x(t-T)], to sample set
xtGauss hybrid models are established, i.e.,：
Wherein,Respectively represent the weights of m-th of Gauss model, mean value, variance and Gaussian function
Number；B, F respectively represent background and foreground model, and M represents Gauss model number；
(2) to newly there is pixel xiCompare successively with current M model, comparative approach is：If there is | xi,t-μj i,t-1| ＜ 2.5
σj i,t-1, then it is assumed that pixel xiMeet current background model, is just considered as background pixel, is otherwise considered as foreground pixel, wherein xi,t
Indicate the pixel value of the pixel i in t frame images, μj i,t-1, σj i,t-1It is illustrated respectively in pixel i after t-1 frame images are trained
Mixed Gauss model in j-th of Gauss model mean value and standard deviation；
(3) once some Gauss model successful match, then the Gauss model is using pixel value xiUpdate, other Gauss models are kept
Constant, more new formula is：
Wherein,It is the learning rate of variation, value range isI.e. when the static or slow object of movement, object
Background model learning rate in rangePart stops update, other range background models normally update；
(4) weights of M Gauss model are normalized and is ranked sequentially by size, B Gauss model is as the back of the body before therefrom choosing
Scape model is left (M-B) a Gauss model as foreground model, wherein background discriminateB tables
Show that the smallest positive integral value for meeting b-th of Gaussian Background model of background discriminate, Ψ indicate the ratio that background accounts for；
(5) according to foreground and background model, binary conversion treatment is carried out to image, that is, meets the pixel value point of foreground and background model
It is not set as 1 and 0, connected domain analysis is carried out to foreground bianry image, by the region merger being linked to be, is divided into one kind, as object
Body object Oi(i=1,2 ... N), and its boundary position information is extracted, obtain each subject OiProfile Ri。
3. the Dynamic Recurrent modeling according to claim 2 differentiated for channel blockage and fusion tracking method, feature
It is, the detailed process of the step 3 is as follows：
(1) to obtaining channel foreground object Oi, initialization search box is chosen, to OiProfile RiBackprojection operations are carried out, if
(x, y) is RiIn a bit, I (x, y) is the probability value at color probability figure (x, y), calculates separately RiZeroth order square M00With one
Rank square M10、M01：
It can thus be concluded that RiCentroid position：Mi(xi,yi)=(M01/M00,M10/M00)；
(2) RiCentroid position Mi(xi,yi) it is input to Kalman filter state equation Mk,i=AMk-1,i+ W measures equation
Zk,i=HMk,i+ V, update equation M'k,i=Mk,i+Kk(Zk,i-HMk,i), obtain more accurate target location；
Wherein, Mk-1,iAnd Mk,iI-th of object is indicated respectively, in k-1 and k moment center-of-mass coordinate quantity of states, Zk,iAnd M'k,iRespectively
I-th of object is represented, in k moment center-of-mass coordinate measured values and correction value, A is Kalman filter quantity of state transfer matrix, and H is
Kalman filter state variable is to the transformed matrix of measurand, KkIt is Kalman filter gain matrix, stochastic variable W, V difference
Indicate system noise matrix and measurement noise matrix；
(3) according to two frame O of video moment k-1 and kiCenter-of-mass coordinate Mk-1,i(xk-1,i,yk-1,i), Mk,i(xk,i,yk,i) judge channel object
Body position state；If 2 barycenter Euclidean distance di：
If di＜ Dset, think that i-th of object is static or mobile slow at this time, then use background local condition to update, that is, exist
RiBackground stops update in range, other range backgrounds normally update, wherein DsetIndicate preset 2 barycenter Euclidean distances
Value；
(4) size of search box is adjusted according to barycenter, if displacement distance is more than preset fixed value, is constantly searched for, until searching
The center of rope frame and the distance of center-of-mass coordinate are less than preset fixed value or operation times reach maximum value and stop calculating.
4. the Dynamic Recurrent modeling according to claim 3 differentiated for channel blockage and fusion tracking method, feature
It is, the detailed process of the step 3 is as follows：
(1) channel width w is setA, i-th of object OiIn the patch ground width w at τ momenti(τ)；Every T in time TsExtract object patch
Ground width sample data calculates its average patch ground widthN number of object overall average patch ground width is calculated again to account for
The ratio V of channel widthOOPR：
(2) according to T and VOOPRBig wisp stopping state falls into three classes：In time T, tracking is consistently present in the object in channel
Body, once object leaving channel, then exclude influence of this object to channel blockage, work as VOOPR50%~70%, it is considered as yellow
It blocks；VOOPR70%~90%, it is considered as orange blocking；VOOPRMore than 90% or more, it is considered as red blocking.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810017676.XA CN108376406A (en) | 2018-01-09 | 2018-01-09 | A kind of Dynamic Recurrent modeling and fusion tracking method for channel blockage differentiation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810017676.XA CN108376406A (en) | 2018-01-09 | 2018-01-09 | A kind of Dynamic Recurrent modeling and fusion tracking method for channel blockage differentiation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108376406A true CN108376406A (en) | 2018-08-07 |
Family
ID=63016481
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810017676.XA Pending CN108376406A (en) | 2018-01-09 | 2018-01-09 | A kind of Dynamic Recurrent modeling and fusion tracking method for channel blockage differentiation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108376406A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108989768A (en) * | 2018-08-31 | 2018-12-11 | 安徽恒杰保安服务有限公司 | A kind of passageway for fire apparatus monitoring system based on estate management |
| CN109409238A (en) * | 2018-09-28 | 2019-03-01 | 深圳市中电数通智慧安全科技股份有限公司 | A kind of obstacle detection method, device and terminal device |
| CN109580619A (en) * | 2018-12-03 | 2019-04-05 | 余姚市腾翔电子科技有限公司 | Bathtub blocks alarm mechanism |
| CN110720929A (en) * | 2019-09-23 | 2020-01-24 | 浙江工业大学 | Blood oxygen content estimation method based on binary sensor bounded recursive optimization fusion |
| CN110766915A (en) * | 2019-09-19 | 2020-02-07 | 重庆特斯联智慧科技股份有限公司 | Alarm method and system for identifying fire fighting access state |
| CN110910355A (en) * | 2019-11-07 | 2020-03-24 | 浙江大华技术股份有限公司 | Package blocking detection method and device and computer storage medium |
| CN110909633A (en) * | 2019-11-07 | 2020-03-24 | 浙江大华技术股份有限公司 | Method and device for determining accumulation degree, storage medium, and electronic device |
| CN111091535A (en) * | 2019-11-22 | 2020-05-01 | 三一重工股份有限公司 | Factory management method and system based on deep learning image semantic segmentation |
| CN111681424A (en) * | 2020-06-06 | 2020-09-18 | 浙江长元科技有限公司 | Method for processing illegal parking occupation event of fire fighting truck channel |
| CN112132043A (en) * | 2020-09-24 | 2020-12-25 | 南京安元科技有限公司 | Fire fighting channel occupation self-adaptive detection method based on monitoring video |
| CN113239832A (en) * | 2021-05-20 | 2021-08-10 | 河南中全科技有限公司 | Hidden danger intelligent identification method and system based on image identification |
| CN113651061A (en) * | 2021-08-18 | 2021-11-16 | 上海德启信息科技有限公司 | Monitoring method and monitoring system for logistics circulation state |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101957997A (en) * | 2009-12-22 | 2011-01-26 | 北京航空航天大学 | Regional average value kernel density estimation-based moving target detecting method in dynamic scene |
| CN103366374A (en) * | 2013-07-12 | 2013-10-23 | 重庆大学 | Fire fighting access obstacle detection method based on image matching |
| CN103914698A (en) * | 2014-03-27 | 2014-07-09 | 北京科技大学 | Method for recognizing and classifying road barriers based on video |
| CN104240222A (en) * | 2013-06-19 | 2014-12-24 | 贺江涛 | Intelligent detecting method and device for firefighting access blockage |
| CN104463253A (en) * | 2015-01-06 | 2015-03-25 | 电子科技大学 | Fire fighting access safety detection method based on self-adaptation background study |
| CN104866823A (en) * | 2015-05-11 | 2015-08-26 | 重庆邮电大学 | Vehicle detection and tracking method based on monocular vision |
| CN105261037A (en) * | 2015-10-08 | 2016-01-20 | 重庆理工大学 | Moving object detection method capable of automatically adapting to complex scenes |
| JP2017033506A (en) * | 2015-08-06 | 2017-02-09 | Ｋｙｂ株式会社 | Road surface detection device and road surface detection method |
| KR101716973B1 (en) * | 2015-11-23 | 2017-03-15 | (주)베라시스 | Apparatus and method for detecting vehicle on road using stereo image |
-
2018
- 2018-01-09 CN CN201810017676.XA patent/CN108376406A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101957997A (en) * | 2009-12-22 | 2011-01-26 | 北京航空航天大学 | Regional average value kernel density estimation-based moving target detecting method in dynamic scene |
| CN104240222A (en) * | 2013-06-19 | 2014-12-24 | 贺江涛 | Intelligent detecting method and device for firefighting access blockage |
| CN103366374A (en) * | 2013-07-12 | 2013-10-23 | 重庆大学 | Fire fighting access obstacle detection method based on image matching |
| CN103914698A (en) * | 2014-03-27 | 2014-07-09 | 北京科技大学 | Method for recognizing and classifying road barriers based on video |
| CN104463253A (en) * | 2015-01-06 | 2015-03-25 | 电子科技大学 | Fire fighting access safety detection method based on self-adaptation background study |
| CN104866823A (en) * | 2015-05-11 | 2015-08-26 | 重庆邮电大学 | Vehicle detection and tracking method based on monocular vision |
| JP2017033506A (en) * | 2015-08-06 | 2017-02-09 | Ｋｙｂ株式会社 | Road surface detection device and road surface detection method |
| CN105261037A (en) * | 2015-10-08 | 2016-01-20 | 重庆理工大学 | Moving object detection method capable of automatically adapting to complex scenes |
| KR101716973B1 (en) * | 2015-11-23 | 2017-03-15 | (주)베라시스 | Apparatus and method for detecting vehicle on road using stereo image |
Non-Patent Citations (2)
| Title |
|---|
| 华媛蕾: "改进混合高斯模型的运动目标检测算法", 《计算机应用》 * |
| 邱联奎: "混合高斯背景模型目标检测的一种改进算法", 《计算机仿真》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108989768A (en) * | 2018-08-31 | 2018-12-11 | 安徽恒杰保安服务有限公司 | A kind of passageway for fire apparatus monitoring system based on estate management |
| CN109409238A (en) * | 2018-09-28 | 2019-03-01 | 深圳市中电数通智慧安全科技股份有限公司 | A kind of obstacle detection method, device and terminal device |
| CN109580619A (en) * | 2018-12-03 | 2019-04-05 | 余姚市腾翔电子科技有限公司 | Bathtub blocks alarm mechanism |
| CN110766915A (en) * | 2019-09-19 | 2020-02-07 | 重庆特斯联智慧科技股份有限公司 | Alarm method and system for identifying fire fighting access state |
| CN110720929A (en) * | 2019-09-23 | 2020-01-24 | 浙江工业大学 | Blood oxygen content estimation method based on binary sensor bounded recursive optimization fusion |
| CN110909633A (en) * | 2019-11-07 | 2020-03-24 | 浙江大华技术股份有限公司 | Method and device for determining accumulation degree, storage medium, and electronic device |
| CN110910355A (en) * | 2019-11-07 | 2020-03-24 | 浙江大华技术股份有限公司 | Package blocking detection method and device and computer storage medium |
| CN110909633B (en) * | 2019-11-07 | 2022-08-16 | 浙江大华技术股份有限公司 | Method and device for determining accumulation degree, storage medium, and electronic device |
| CN111091535A (en) * | 2019-11-22 | 2020-05-01 | 三一重工股份有限公司 | Factory management method and system based on deep learning image semantic segmentation |
| CN111681424A (en) * | 2020-06-06 | 2020-09-18 | 浙江长元科技有限公司 | Method for processing illegal parking occupation event of fire fighting truck channel |
| CN112132043A (en) * | 2020-09-24 | 2020-12-25 | 南京安元科技有限公司 | Fire fighting channel occupation self-adaptive detection method based on monitoring video |
| CN113239832A (en) * | 2021-05-20 | 2021-08-10 | 河南中全科技有限公司 | Hidden danger intelligent identification method and system based on image identification |
| CN113239832B (en) * | 2021-05-20 | 2023-02-17 | 众芯汉创（北京）科技有限公司 | Hidden danger intelligent identification method and system based on image identification |
| CN113651061A (en) * | 2021-08-18 | 2021-11-16 | 上海德启信息科技有限公司 | Monitoring method and monitoring system for logistics circulation state |
| CN113651061B (en) * | 2021-08-18 | 2023-04-14 | 上海德启信息科技有限公司 | Monitoring method and monitoring system for logistics circulation state |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108376406A (en) | A kind of Dynamic Recurrent modeling and fusion tracking method for channel blockage differentiation | |
| CN110425005B (en) | Safety monitoring and early warning method for man-machine interaction behavior of belt transport personnel under mine | |
| CN103914688B (en) | A kind of urban road differentiating obstacle | |
| CN103632158B (en) | Forest fire prevention monitor method and forest fire prevention monitor system | |
| CN106571014A (en) | Method for identifying abnormal motion in video and system thereof | |
| CN106203274A (en) | Pedestrian's real-time detecting system and method in a kind of video monitoring | |
| CN108596129A (en) | A kind of vehicle based on intelligent video analysis technology gets over line detecting method | |
| CN106204640A (en) | A kind of moving object detection system and method | |
| CN108053427A (en) | A kind of modified multi-object tracking method, system and device based on KCF and Kalman | |
| CN105303191A (en) | Method and apparatus for counting pedestrians in foresight monitoring scene | |
| CN106778655B (en) | Human body skeleton-based entrance trailing entry detection method | |
| CN109255298A (en) | Safety cap detection method and system in a kind of dynamic background | |
| Yu et al. | Railway obstacle detection algorithm using neural network | |
| CN105184258A (en) | Target tracking method and system and staff behavior analyzing method and system | |
| CN106210634A (en) | A kind of wisdom gold eyeball identification personnel fall down to the ground alarm method and device | |
| CN109101888A (en) | A kind of tourist's flow of the people monitoring and early warning method | |
| CN104378582A (en) | Intelligent video analysis system and method based on PTZ video camera cruising | |
| CN100565557C (en) | System for tracking infrared human body target based on corpuscle dynamic sampling model | |
| CN107133569A (en) | The many granularity mask methods of monitor video based on extensive Multi-label learning | |
| US20150071492A1 (en) | Abnormal behaviour detection | |
| CN108564069A (en) | A kind of industry safe wearing cap video detecting method | |
| CN104318578A (en) | Video image analyzing method and system | |
| CN112071084A (en) | Method and system for judging illegal parking by utilizing deep learning | |
| CN108197575A (en) | A kind of abnormal behaviour recognition methods detected based on target detection and bone point and device | |
| CN110633643A (en) | Abnormal behavior detection method and system for smart community |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180807 |