CN103226712A - Finite-stage machine-based method for detecting abandoned object - Google Patents
Finite-stage machine-based method for detecting abandoned object Download PDFInfo
- Publication number
- CN103226712A CN103226712A CN2013101841955A CN201310184195A CN103226712A CN 103226712 A CN103226712 A CN 103226712A CN 2013101841955 A CN2013101841955 A CN 2013101841955A CN 201310184195 A CN201310184195 A CN 201310184195A CN 103226712 A CN103226712 A CN 103226712A
- Authority
- CN
- China
- Prior art keywords
- state
- input
- transferred
- pixel
- legacy
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
The invention discloses a finite-stage machine-based method for detecting an abandoned object. The finite-stage machine-based method comprises the steps of: firstly, establishing a Gaussian mixture background model, respectively establishing a short-time background model and a long-time background model according to different learning rates; secondly, establishing a finite-stage machine for each pixel; with results obtained by detection of two different background models with different updating speeds as inputs and a pixel classifying result output by each finite-stage machine as one binary image, carrying out communicated region analysis on the obtained binary image to obtain the shape and contour of the abandoned object by using a region growth method, and working out a rectangular box enclosing the abandoned object; and finally, timing the obtained rectangular box of the abandoned object, and detecting the abandoned object and alarming when a threshold is reached. According to the finite-stage machine-based method, under the condition that the abandoned object is absorbed by two backgrounds, the abandoned object can be always detected according to the pixel history, less false-alarms and wrong-alarms exist in an actual scene, and a good detection effect can also be obtained in an occasion with intensive pedestrian flow without depending on tracking information.
Description
Technical field
The present invention relates to a kind of video legacy detection method, thereby relate more specifically to a kind ofly based on double-background model each pixel of image be set up extended finite state machine and classify and detect the method for legacy.
Background technology
In recent years, spread along with terroristic, the situation that domestic and international public safety incident pilosity, particularly public place explosive create greater casualties happens occasionally.The video legacy can send a kind of technological means of warning rapidly when detect being exactly to take advantage of people's placement dangerous material off guard in public places such as boat station, railway stations at the offender.Simultaneously, legacy detects and also can be used to detect the parking incident in traffic scene.
Video legacy detection method mainly detects legacy by background subtraction at present.Each pixel in the static camera video image is set up Gauss model obtain background, background needs constantly study to adapt to illumination and scene changes slowly.Use the background model of two different update speed to detect legacy.Wherein, upgrade fast model and be model in short-term, legacy is dissolved into background model soon; Upgrade slow model when being long, legacy will just incorporate model after a while.Incorporate background model in short-term at legacy, when also not being dissolved into when long background model, the difference of two backgrounds is exactly a legacy.
In the conventional method, legacy also can background model absorb when long over time, and at this moment legacy just can not be detected again.In addition, if there is object from scene, to be removed, can be detected as legacy.Existing method addresses these problems, and needs legacy that selectively use detected and upgrades background model when long, and keep the tracking to legacy.Yet incorrect update strategy can cause incorrect detection, even causes illumination variation to be detected as legacy.In the intensive stream of people's of reality occasion, the stream of people of motion forms legacy and blocks and shade especially, is difficult to initialization and the maintenance tracking to legacy, and the false-alarm and the false dismissed rate of existing method are very high, are difficult to practicality.
In essence, the result that existing method need be followed the tracks of the legacy detection intervenes background model, prevents that legacy from being absorbed by background model.Under simple scenario, since fairly simple to the tracking of object, thereby it is reliable to detect legacy.In case target is more in the scene, block the influence with shade mutually, mistake appears in the tracking of legacy, will driedly scratch background model, and is difficult to recover, and false-alarm occurs and mistake is alert.
Summary of the invention
Technical matters to be solved by this invention is the defective at background technology, a kind of legacy detection method based on state machine is proposed, each pixel is set up finite state machine, be defined under the given state, according to the available result of the background model of two different update speed.This method legacy by the situation of two background absorption under, can still detect legacy according to pixel history.Because do not rely on trace information, the detection effect that under the occasion that solves the intensive stream of people, also can obtain.
The present invention is for solving the problems of the technologies described above by the following technical solutions:
A kind of legacy detection method based on finite state machine may further comprise the steps:
Step 1: set up the Gaussian mixture model-universal background model of video image, set up in short-term background model respectively and background model when long according to different learning rates;
Step 2: gather video flowing in real time, each frame video image for collecting, carry out following processing:
201, the background model of setting up according to step 1 in short-term, background model is extracted the foreground image of this frame video image respectively when long, classify for each pixel in this frame video image, if pixel belongs to prospect, then it is designated as 1, otherwise is designated as 0, obtain the binary classification result (X that each pixel corresponds respectively to background model when background model is with length in short-term, Y), as the input alphabet of finite state machine; The classification results that the X classification results in the background model that is pixel when long wherein, Y are pixel in background model in short-term, X=0 or 1, Y=0 or 1;
202, for each pixel is set up finite state machine (I, Q, a Z, δ, ω), with each pixel respectively background model in short-term and when long the binary classification result of background model as the input I={(0 of finite state machine, 0), (0,1), (1,0), (1,1) }, the state that obtains each pixel present frame correspondence according to the state set Q and the state transition function δ thereof of finite state machine, Z={0,1 ..., | Q|} is an output alphabet, and ω={ 0,1} is an output function;
203, the state of each pixel present frame corresponding states machine number is carried out classification map: Z-as output alphabet〉ω, classification results forms bianry image, and corresponding this image value is that 1 pixel belongs to legacy;
Step 3: the bianry image that step 2 is obtained carries out UNICOM's regional analysis, uses region-growing method to obtain the shape and the profile of legacy, calculates the rectangular box of surrounding legacy;
Step 4: circulation execution in step 2-3 to the legacy rectangular box timing that obtains, detects legacy and warning when reaching thresholding.
Preferably, a kind of legacy detection method of the present invention based on finite state machine, described finite state machine is made of input alphabet, state set, output alphabet, state transition function and output function, and is specific as follows:
Input alphabet: background model and background model collection { (0,0), (0,1), (1,0) (1,1) } that the pixel classification results is constituted in short-term when being long, wherein 0 represents background, 1 expression prospect;
State set and transfer function thereof are specific as follows:
State 0-pixel is a background; When input was (0,0), state remained unchanged; When input is (1,1), be transferred to state 1; When input is (0,1), be transferred to state 3; When input is (1,0), be transferred to state 7;
State 1-pixel is moving object, when input is (0,0), and return state 0; When input is (0,1), be transferred to state 8; When being input as (1,0), be transferred to state 2; When being input as (1,1), state remains unchanged;
State 2-pixel belongs to an object, and by background model absorption in short-term, background model absorbs during also not by length, and this object is current to be of short duration static; When being input as (0,0), be transferred to state 4; When being input as (0,1), be transferred to state 3; When being input as (1,1), be transferred to state 6; When being input as (1,0), state remains unchanged;
State 3-pixel is the background of just being blocked by of short duration static object; When input is (0,0), be transferred to state 0, will determine that this is a scene background; When input is (1,0), be transferred to state 2, pixels illustrated is an object; When input is (1,1), be transferred to state 6; When input was (0,1), state remained unchanged;
State 4-pixel is absorbed by two background models, static object when being long; When input is (1,1), be transferred to state 5; When input is (1,0), be transferred to state 6; When input is (0,1), be transferred to state 10; When input was (0,0), state remained unchanged;
State 5-undistinguishable, chaos state; When being input as (1,1), state remains unchanged; When being input as (0,0), be transferred to state 4; When being input as (1,0) or (0,1), be transferred to state 6;
One of state 6-definition enables function f: the pixel value when remembeing that this pixel is not blocked by object recently, compare with current pixel value, specifically be with the squared difference of rgb space and after extraction of square root again, obtain pixel value, difference 20 with the interior background of thinking, the f value is 1, otherwise is prospect, and the f value is 0; Under the identical situation of input, the different functional values that enable are transferred to different conditions; When being input as (0,1) and f=0, be transferred to state 10; When being input as (0,1) and f=1, be transferred to state 3; When being input as (1,0) and f=0, be transferred to state 9; When being input as (1,0) and f=1, be transferred to state 7; When being input as (0,0) and f=0, be transferred to state 4; When being input as (0,0) and f=1, be transferred to state 0; Other input state remains unchanged;
State 7-model in short-term divides into background, and model is divided into prospect when long, and empirical tests is the background pixel of scene; When input is (0,0), be transferred to state 0, will determine that this is a scene background; When input is (0,1), be transferred to state 8; When input is (1,1), be transferred to state 6; When input was (1,0), state remained unchanged;
When stationary object was taken away when state 8-was long, pixel was the scene background pixel; When input is (0,0), be transferred to state 4, illustrate that legacy has blocked another legacy; When being input as (1,1), be transferred to state 6; When input is (1,0), be transferred to state 7; When input was (0,1), state remained unchanged;
State 9-pixel be one in short-term stationary object block stationary object when long; When being input as (0,0), be transferred to state 4, illustrate that stationary object is removed in short-term, stationary object is revealed again when long; When being input as (0,1) or (1,1), be transferred to state 6; When input was (1,0), state remained unchanged;
State 10-pixel is after stationary object blocks stationary object when long in short-term, to take away stationary object in short-term, pixel is current stationary object when being long; When next one input is (0,0), be transferred to state 4, will determine that this is a legacy; When being input as (1,0) or (1,1), be transferred to state 6; When input was (0,1), state remained unchanged;
Output alphabet Z be 0,1 ..., N}, i.e. the state value of state machine, N is the state number in the state set;
Output function Z is mapped as 0,1}, wherein 4} be mapped as 1}, its residual value is mapped as { 0} among the Z;
Every frame pixel status machine is exported the bianry image that obtains, as the foundation of legacy detection.
Preferably, a kind of legacy detection method of the present invention based on finite state machine, described state set also comprises following state:
The intermediateness of state 11-replication status 8: when being input as (1,0), be transferred to state 12; When being input as (0,0), be transferred to state 4; When being input as (1,1), be transferred to state 13; When being input as (0,1), state remains unchanged;
The intermediateness of state 12-replication status 3: when being input as (0,0), be transferred to state 0; When being input as (1,1), be transferred to state 6; When being input as (0,1), be transferred to state 11; When being input as (1,0), state remains unchanged;
The intermediateness of state 13-replication status 1: when being input as (0,0), be transferred to state 6; When being input as (0,1), be transferred to state 11; When being input as (1,0), be transferred to state 14; When being input as (1,1), state remains unchanged;
The intermediateness of state 14-replication status 2: when being input as (0,1) or (1,1), be transferred to state 6; When being input as (0,0), be transferred to state 4; When being input as (1,0), state remains unchanged;
The intermediateness of state 15-replication status 1: when being input as (0,1), be transferred to state 3; When being input as (1,0), be transferred to state 2; When being input as (0,0), be transferred to state 6; When being input as (1,1), state remains unchanged.
Preferably, a kind of legacy detection method of the present invention based on finite state machine, for the increase system resists the dried ability of scratching to noise, below state transition function need import and keep that continuous 5 frames are constant just to be shifted: state 0 is transferred to state 7, state 0 is transferred to state 3, and state 15 is transferred to state 6, and state 13 is transferred to state 6, state 4 is transferred to state 6, and state 4 is transferred to state 10.
The present invention adopts above technical scheme compared with prior art, has following technique effect:
The present invention is based on double-background model each pixel is set up extended finite state machine, can filter complicated event in the scene, be removed again after being left over as object or scene in object be removed, and arbitrarily the increase state object that adapts to scene block mutually, complex situations such as project objects.
The present invention is directed to complicated scene legacy detection and set up the common treatment framework, when legacy is absorbed by background model, need not to intervene background model, reduced the complicacy of system design, increased the reliability of system, false-alarm and mistake are alert less in actual scene, have reached practical purpose.
Description of drawings
Fig. 1 is the state machine transition diagram.
Fig. 2 is the finite state machine transition diagram of increase state.
Fig. 3 is a process flow diagram of the present invention.
Embodiment
Below in conjunction with accompanying drawing technical scheme of the present invention is described in further detail:
As shown in Figure 3, idiographic flow of the present invention is as follows:
Step 1: set up Gaussian mixture model-universal background model, model when setting up in short-term with length according to different learning rates.Extract the prospect and the background of video image, if pixel belongs to prospect, be designated as 1, otherwise be 0.Obtain two values like this, as the jump condition of finite state machine;
Step 2: each pixel is set up a finite state machine, is used for the type of classified pixels.Two each pixel classification of background model promote state machine state and change as jump condition, and details are as follows for state machine:
State machine is made of input, state set, output, state transition function.
Background model and background model is to the 2 system manifolds { (0,0), (0,1), (1,0), (1,1) } of pixel classification results formation in short-term when input is long, 0 is background, 1 is prospect.
As shown in Figure 1, the state description pixel is in the situation of actual scene, and each pixel must be in wherein a kind of situation, and following state is arranged:
When input is (1,1), be transferred to state 5; When input is (1,0), be transferred to state 6; When input is (0,1), be transferred to state 10; When input was (0,0), state remained unchanged.
State 5, undistinguishable, chaos state is input as (1,1), keeps this state; Be input as (0,0), transfer to state 4; When being input as (1,0) or (0,1), transfer to state 6.
State 6, model is divided into background in short-term, and model is divided into prospect when long.Be that object is removed, but uncertain current pixel is a kind of state of scene background.Block the object that stops when long or just block background as the object of short stay, the object of short stay is removed at this moment.The state transitions of observing state 6 also needs extra information, and this is to have described the background of actual scene better because can not determine which background model this moment.Some state transition function need be known the historical information of state transitions link, and shift direction is can not determine in the input historical by pixel and background model of this state light, also will compare the value of the last background of this pixel; So used the technology of similar extended finite state machine at this state, specific as follows:
Define one and enable function f: the pixel value when remembeing that this pixel is not blocked by object recently, compare with current pixel value, specifically be with the squared difference of rgb space and after extraction of square root again, obtain pixel value, difference 20 with the interior background of thinking, the f value is 1, otherwise is prospect, and the f value is 0; Under the identical situation of input, the different functional values that enable are transferred to different conditions; When being input as (0,1) and f=0, be transferred to state 10; When being input as (0,1) and f=1, be transferred to state 3; When being input as (1,0) and f=0, be transferred to state 9; When being input as (1,0) and f=1, be transferred to state 7; When being input as (0,0) and f=0, be transferred to state 4; When being input as (0,0) and f=1, be transferred to state 0; Other input state remains unchanged.
State 7, model is divided into background in short-term, and model is divided into prospect when long, may be the background pixel of scene.Be a transition state,, will determine that this is a scene background, is transferred to state 0 if input is (0,0); If input is (1,0), state transitions is to state 8; If input is (1,1), state transitions is to state 6; When input was (1,0), state remained unchanged.
State 8, when stationary object was taken away when long, pixel may be the scene background pixel, also may be to leave over object.If input is (0,0), state transitions illustrates that to state 4 legacy has blocked another legacy; If be input as (1,1), transfer to state 6; If input is (1,0), state transitions is to state 7; When input was (0,1), state remained unchanged.
State 9, pixel be one in short-term stationary object block stationary object when long.If be input as (0,0), illustrate that stationary object is removed in short-term, stationary object is revealed again when long, and state transitions is to state 4; If be input as (0,1) or (1,1), state transitions is to state 6; When input was (1,0), state remained unchanged.
Can cause the extra computation cost because enable function in the state 6, can be by increasing intermediateness, state machine gets the hang of 6 under the minimizing certain situation, reaches better calculated performance.As shown in Figure 2, we need expand above-mentioned state machine, increase by five states:
Every frame pixel status is output as 1 for " state 4 ", the expression legacy, all the other states are output as 0, obtain bianry image.
For the increase system resists the dried ability of scratching to noise, below state transition function need import and keep that continuous 5 frames are constant just to be shifted: state 0 is transferred to state 7, state 0 is transferred to state 3, state 15 is transferred to state 6, state 13 is transferred to state 6, state 4 is transferred to state 6, and state 4 is transferred to state 10, and its essence is equivalent to increase the implicit state of 4 serials between these states.
Step 3: the bianry image that step 2 is obtained carries out UNICOM's regional analysis, uses region-growing method to obtain the shape and the profile of legacy, calculates the rectangular box of surrounding legacy;
Step 4:, reach thresholding and promptly detect legacy and warning to the legacy rectangular box timing that obtains.
The present invention legacy by the situation of two background absorption under, can still detect legacy according to pixel history, false-alarm and mistake are alert less in actual scene because do not rely on trace information, the detection effect that under the intensive stream of people's occasion, also can obtain.
Claims (5)
1. the legacy detection method based on finite state machine is characterized in that, may further comprise the steps:
Step 1: set up the Gaussian mixture model-universal background model of video image, set up in short-term background model respectively and background model when long according to different learning rates;
Step 2: gather video flowing in real time,, carry out following processing successively for each frame video image that collects:
201, the background model of setting up according to step 1 in short-term, background model is extracted the foreground image of this frame video image respectively when long, classify for each pixel in this frame video image, if pixel belongs to prospect, then it is designated as 1, otherwise is designated as 0, obtain each pixel and correspond respectively to when long the background model and the binary classification result (X of background model in short-term, Y), as the input alphabet of finite state machine; The classification results that the X classification results in the background model that is pixel when long wherein, Y are pixel in background model in short-term, X=0 or 1, Y=0 or 1;
202, for each pixel is set up finite state machine (I, Q, a Z, δ, ω), with each pixel respectively when long background model and in short-term the binary classification result of background model as the input I={(0 of finite state machine, 0), (0,1), (1,0), (1,1) }, the state that obtains each pixel present frame correspondence according to the state set Q and the state transition function δ thereof of finite state machine, Z={0,1 ..., | Q|} is an output alphabet, and ω={ 0,1} is an output function;
203, the state of each pixel present frame corresponding states machine number is carried out classification map: Z-as output alphabet〉ω, classification results forms bianry image, and corresponding this image value is that 1 pixel belongs to legacy;
Step 3: the bianry image that step 2 is obtained carries out UNICOM's regional analysis, uses region-growing method to obtain the shape and the profile of legacy, calculates the rectangular box of surrounding legacy;
Step 4: circulation execution in step 2-3 to the legacy rectangular box timing that obtains, detects legacy and warning when reaching thresholding.
2. a kind of legacy detection method based on finite state machine according to claim 1 is characterized in that described finite state machine is made of input alphabet, state set, output alphabet, state transition function and output function, and is specific as follows:
Input alphabet: background model and background model collection { (0,0), (0,1), (1,0) (1,1) } that the pixel classification results is constituted in short-term when being long, wherein 0 represents background, 1 expression prospect;
State set and transfer function thereof are specific as follows:
State 0-pixel is a background; When input was (0,0), state remained unchanged; When input is (1,1), be transferred to state 1; When input is (0,1), be transferred to state 3; When input is (1,0), be transferred to state 7;
State 1-pixel is moving object, when input is (0,0), and return state 0; When input is (0,1), be transferred to state 8; When being input as (1,0), be transferred to state 2; When being input as (1,1), state remains unchanged;
State 2-pixel belongs to an object, and by background model absorption in short-term, background model absorbs during also not by length, and this object is current to be of short duration static; When being input as (0,0), be transferred to state 4; When being input as (0,1), be transferred to state 3; When being input as (1,1), be transferred to state 6; When being input as (1,0), state remains unchanged;
State 3-pixel is the background of just being blocked by of short duration static object; When input is (0,0), be transferred to state 0, will determine that this is a scene background; When input is (1,0), be transferred to state 2, pixels illustrated is an object; When input is (1,1), be transferred to state 6; When input was (0,1), state remained unchanged;
State 4-pixel is absorbed by two background models, static object when being long; When input is (1,1), be transferred to state 5; When input is (1,0), be transferred to state 6; When input is (0,1), be transferred to state 10; When input was (0,0), state remained unchanged;
State 5-undistinguishable, chaos state; When being input as (1,1), state remains unchanged; When being input as (0,0), be transferred to state 4; When being input as (1,0) or (0,1), be transferred to state 6;
One of state 6-definition enables function f: the pixel value when remembeing that this pixel is not blocked by object recently, compare with current pixel value, specifically be with the squared difference of rgb space and after extraction of square root again, obtain pixel value, difference 20 with the interior background of thinking, the f value is 1, otherwise is prospect, and the f value is 0; Under the identical situation of input, the different functional values that enable are transferred to different conditions; When being input as (0,1) and f=0, be transferred to state 10; When being input as (0,1) and f=1, be transferred to state 3; When being input as (1,0) and f=0, be transferred to state 9; When being input as (1,0) and f=1, be transferred to state 7; When being input as (0,0) and f=0, be transferred to state 4; When being input as (0,0) and f=1, be transferred to state 0; Other input state remains unchanged;
State 7-model in short-term divides into background, and model is divided into prospect when long, and empirical tests is the background pixel of scene; When input is (0,0), be transferred to state 0, will determine that this is a scene background; When input is (0,1), be transferred to state 8; When input is (1,1), be transferred to state 6; When input was (1,0), state remained unchanged;
When stationary object was taken away when state 8-was long, pixel was the scene background pixel; When input is (0,0), be transferred to state 4, illustrate that legacy has blocked another legacy; When being input as (1,1), be transferred to state 6; When input is (1,0), be transferred to state 7; When input was (0,1), state remained unchanged;
State 9-pixel be one in short-term stationary object block stationary object when long; When being input as (0,0), be transferred to state 4, illustrate that stationary object is removed in short-term, stationary object is revealed again when long; When being input as (0,1) or (1,1), be transferred to state 6; When input was (1,0), state remained unchanged;
State 10-pixel is after stationary object blocks stationary object when long in short-term, to take away stationary object in short-term, pixel is current stationary object when being long; When next one input is (0,0), be transferred to state 4, will determine that this is a legacy; When being input as (1,0) or (1,1), be transferred to state 6; When input was (0,1), state remained unchanged;
Output alphabet Z be 0,1 ..., N}, i.e. the state value of state machine, N is the state number in the state set;
Output function Z is mapped as 0,1}, wherein 4} be mapped as 1}, its residual value is mapped as { 0} among the Z;
Every frame pixel status machine is exported the bianry image that obtains, as the foundation of legacy detection.
3. a kind of legacy detection method based on finite state machine according to claim 2 is characterized in that described state set also comprises following state:
The intermediateness of state 11-replication status 8: when being input as (1,0), be transferred to state 12; When being input as (0,0), be transferred to state 4; When being input as (1,1), be transferred to state 13; When being input as (0,1), state remains unchanged;
The intermediateness of state 12-replication status 3: when being input as (0,0), be transferred to state 0; When being input as (1,1), be transferred to state 6; When being input as (0,1), be transferred to state 11; When being input as (1,0), state remains unchanged;
The intermediateness of state 13-replication status 1: when being input as (0,0), be transferred to state 6; When being input as (0,1), be transferred to state 11; When being input as (1,0), be transferred to state 14; When being input as (1,1), state remains unchanged;
The intermediateness of state 14-replication status 2: when being input as (0,1) or (1,1), be transferred to state 6; When being input as (0,0), be transferred to state 4; When being input as (1,0), state remains unchanged;
The intermediateness of state 15-replication status 1: when being input as (0,1), be transferred to state 3; When being input as (1,0), be transferred to state 2; When being input as (0,0), be transferred to state 6; When being input as (1,1), state remains unchanged.
4. a kind of legacy detection method according to claim 2 based on finite state machine, it is characterized in that, following state transition function need be imported and keep that continuous 5 frames are constant just to be shifted: state 0 is transferred to state 7, state 0 is transferred to state 3, state 4 is transferred to state 6, and state 4 is transferred to state 10.
5. a kind of legacy detection method based on finite state machine according to claim 3 is characterized in that, following state transition function need be imported and keep that continuous 5 frames are constant just to be shifted: state 15 is transferred to state 6, and state 13 is transferred to state 6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310184195.5A CN103226712B (en) | 2013-05-19 | 2013-05-19 | A kind of remnant object detection method based on finite state machine |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310184195.5A CN103226712B (en) | 2013-05-19 | 2013-05-19 | A kind of remnant object detection method based on finite state machine |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103226712A true CN103226712A (en) | 2013-07-31 |
CN103226712B CN103226712B (en) | 2016-01-20 |
Family
ID=48837153
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310184195.5A Active CN103226712B (en) | 2013-05-19 | 2013-05-19 | A kind of remnant object detection method based on finite state machine |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103226712B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104156942A (en) * | 2014-07-02 | 2014-11-19 | 华南理工大学 | Detection method for remnants in complex environment |
CN105141901A (en) * | 2015-08-12 | 2015-12-09 | 青岛中星微电子有限公司 | Video processing method and device |
CN106408554A (en) * | 2015-07-31 | 2017-02-15 | 富士通株式会社 | Remnant detection apparatus, method and system |
CN106560836A (en) * | 2015-10-02 | 2017-04-12 | Lg电子株式会社 | Apparatus, Method And Mobile Terminal For Providing Object Loss Prevention Service In Vehicle |
RU2676028C1 (en) * | 2018-03-14 | 2018-12-25 | Акционерное Общество "Крафтвэй Корпорэйшн Плс" | Method of detecting left object in video stream |
CN109792508A (en) * | 2016-09-29 | 2019-05-21 | 菲力尔系统公司 | It is detected using the failure safe of thermal image analysis method |
CN110232359A (en) * | 2019-06-17 | 2019-09-13 | 中国移动通信集团江苏有限公司 | It is detained object detecting method, device, equipment and computer storage medium |
CN110782425A (en) * | 2018-07-13 | 2020-02-11 | 富士通株式会社 | Image processing method, image processing device and electronic equipment |
CN112070033A (en) * | 2020-09-10 | 2020-12-11 | 天津城建大学 | Video carry-over detection method based on finite-state machine analysis |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080247599A1 (en) * | 2007-04-05 | 2008-10-09 | Porikli Fatih M | Method for Detecting Objects Left-Behind in a Scene |
CN102314695A (en) * | 2011-08-23 | 2012-01-11 | 北京黄金视讯科技有限公司 | Abandoned object detection method based on computer vision |
CN102509075A (en) * | 2011-10-19 | 2012-06-20 | 北京国铁华晨通信信息技术有限公司 | Remnant object detection method and device |
CN102902960A (en) * | 2012-09-25 | 2013-01-30 | 武汉烽火众智数字技术有限责任公司 | Leave-behind object detection method based on Gaussian modelling and target contour |
US20130051613A1 (en) * | 2011-08-29 | 2013-02-28 | International Business Machines Corporation | Modeling of temporarily static objects in surveillance video data |
-
2013
- 2013-05-19 CN CN201310184195.5A patent/CN103226712B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080247599A1 (en) * | 2007-04-05 | 2008-10-09 | Porikli Fatih M | Method for Detecting Objects Left-Behind in a Scene |
CN102314695A (en) * | 2011-08-23 | 2012-01-11 | 北京黄金视讯科技有限公司 | Abandoned object detection method based on computer vision |
US20130051613A1 (en) * | 2011-08-29 | 2013-02-28 | International Business Machines Corporation | Modeling of temporarily static objects in surveillance video data |
CN102509075A (en) * | 2011-10-19 | 2012-06-20 | 北京国铁华晨通信信息技术有限公司 | Remnant object detection method and device |
CN102902960A (en) * | 2012-09-25 | 2013-01-30 | 武汉烽火众智数字技术有限责任公司 | Leave-behind object detection method based on Gaussian modelling and target contour |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104156942B (en) * | 2014-07-02 | 2017-02-15 | 华南理工大学 | Detection method for remnants in complex environment |
CN104156942A (en) * | 2014-07-02 | 2014-11-19 | 华南理工大学 | Detection method for remnants in complex environment |
US10212397B2 (en) | 2015-07-31 | 2019-02-19 | Fujitsu Limited | Abandoned object detection apparatus and method and system |
CN106408554A (en) * | 2015-07-31 | 2017-02-15 | 富士通株式会社 | Remnant detection apparatus, method and system |
CN106408554B (en) * | 2015-07-31 | 2019-07-09 | 富士通株式会社 | Residue detection device, method and system |
CN105141901A (en) * | 2015-08-12 | 2015-12-09 | 青岛中星微电子有限公司 | Video processing method and device |
CN106560836A (en) * | 2015-10-02 | 2017-04-12 | Lg电子株式会社 | Apparatus, Method And Mobile Terminal For Providing Object Loss Prevention Service In Vehicle |
CN106560836B (en) * | 2015-10-02 | 2022-04-05 | Lg电子株式会社 | Apparatus, method and mobile terminal for providing anti-loss of goods service in vehicle |
CN109792508A (en) * | 2016-09-29 | 2019-05-21 | 菲力尔系统公司 | It is detected using the failure safe of thermal image analysis method |
US10937140B2 (en) | 2016-09-29 | 2021-03-02 | Flir Systems, Inc. | Fail-safe detection using thermal imaging analytics |
CN109792508B (en) * | 2016-09-29 | 2021-07-13 | 菲力尔系统公司 | Fail-safe detection using thermal imaging analysis |
RU2676028C1 (en) * | 2018-03-14 | 2018-12-25 | Акционерное Общество "Крафтвэй Корпорэйшн Плс" | Method of detecting left object in video stream |
CN110782425A (en) * | 2018-07-13 | 2020-02-11 | 富士通株式会社 | Image processing method, image processing device and electronic equipment |
CN110232359A (en) * | 2019-06-17 | 2019-09-13 | 中国移动通信集团江苏有限公司 | It is detained object detecting method, device, equipment and computer storage medium |
CN112070033A (en) * | 2020-09-10 | 2020-12-11 | 天津城建大学 | Video carry-over detection method based on finite-state machine analysis |
Also Published As
Publication number | Publication date |
---|---|
CN103226712B (en) | 2016-01-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103226712A (en) | Finite-stage machine-based method for detecting abandoned object | |
CN102184552B (en) | Moving target detecting method based on differential fusion and image edge information | |
CN112016414A (en) | Method and device for detecting high-altitude parabolic event and intelligent floor monitoring system | |
CN102509075B (en) | Remnant object detection method and device | |
CN103246896B (en) | A kind of real-time detection and tracking method of robustness vehicle | |
CN103500324B (en) | Violent behavior recognition methods based on video monitoring | |
CN101266689B (en) | A mobile target detection method and device | |
CN104392468A (en) | Improved visual background extraction based movement target detection method | |
CN102034240A (en) | Method for detecting and tracking static foreground | |
Liu et al. | Anomaly detection in surveillance video using motion direction statistics | |
CN103400117A (en) | Method for positioning and tracking personnel in well on basis of compute vision | |
CN105046948A (en) | System and method of monitoring illegal traffic parking in yellow grid line area | |
CN104766079A (en) | Remote infrared weak object detecting method | |
CN114639075A (en) | Method and system for identifying falling object of high altitude parabola and computer readable medium | |
Su et al. | A new local-main-gradient-orientation HOG and contour differences based algorithm for object classification | |
CN107657260B (en) | Method for detecting ground hidden danger in power transmission line surrounding environment | |
CN103106796A (en) | Vehicle detection method and device of intelligent traffic surveillance and control system | |
CN103793921A (en) | Moving object extraction method and moving object extraction device | |
CN111325292A (en) | Object behavior identification method and device | |
CN107729811B (en) | Night flame detection method based on scene modeling | |
Wang et al. | Real-time vehicle signal lights recognition with HDR camera | |
CN103400395A (en) | Light stream tracking method based on HAAR feature detection | |
CN103516955B (en) | Intrusion detection method in video monitoring | |
Varghese et al. | Background subtraction for vehicle detection | |
Bhaskar et al. | Enhanced and effective parallel optical flow method for vehicle detection and tracking |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |