Summary of the invention
The present invention relates to a kind of that have at least one device that is used to absorb video image and have a kind of video smoke detection system of signal Processing level of determining each pixel of video image therein or respectively organizing the brightness of pixel.
The task that should solve with the present invention is, a kind of Smoke Detection rapidly and reliably of realizing is described, and is particularly suitable for the video smoke detection system that uses in highway and railway tunnel.Should carry out described Smoke Detection in the stage as far as possible early that fire alarm forms, and should in fact get rid of false alarm.
According to video smoke detection system of the present invention, having at least one is used to absorb the device of video image and has a kind of therein by handling the signal processor of determining each pixel of video image or respectively organizing pixel intensity, in described processing, obtain a kind of for the representational value of described brightness, and the variation that occurs at sign smog comes the time course of described value is analyzed and researched, described processing is that a kind of edge extraction that distributes a rim value for each pixel handles, carry out the comparison of described rim value and a mean value for each pixel, and obtain the counter image of an explanation rim value in from then on comparing with respect to the time response of described mean value.
Smoke detection system of the present invention is based on following understanding, i.e. the appearance of smog causes the reduction of contrast.Make described edge fog when determining pixel intensity handling, or these edges will disappear by a kind of edge extraction.At this moment, the advantage of this process is that rim value is insensitive with respect to the illumination change of integral body.
Especially each relatively time the at rim value and described mean value upgrades described counter image, this counter picture specification the brightness of relevant pixel how once on average to be positioned on the described mean value continually on certain time period.
Described counter image and a kind of threshold value are compared, and when surpassing this threshold value, an initialization value is added up to a currency.
The 3rd preferred implementing form of video smoke detection system of the present invention is characterised in that, also carries out a kind of analysis and research hereinafter referred to as motion detection from the motion aspect to described video image except that the edge extracts processing.
At first arranging in the highway and railway tunnel of system of the present invention for its monitoring, almost only is that what to produce the edge by the mobile object between relative edges and the video camera is arranged is not by the caused covering of smog.Because such object is not to embody suddenly, cover on the position at described edge but moved on to them usually, so can be from following situation, promptly under the situation that is not the edges cover that caused by smog, the object that covers described edge must just produce motion before this.Therefore described motion detection provides a kind of and has been used to distinguish those edges of being covered by object and by the reliable criterion at the edge that smog covered.
Especially the counter image that upgrades by a kind of hysteresis algorithm of continuous usefulness had both carried out the described definite of rim value, also carried out described motion detection.Especially a kind of algorithm based on the standardization simple crosscorrelation is used for described motion detection.
Described hysteresis algorithm especially has a minimum value and a maximal value, and two therebetween threshold values, wherein, described counter image just skips on the maximal value when surpassing lower threshold value when upwards counting, and just skips on the minimum value when being lower than upper threshold value when counting downwards.
This hysteresis algorithm has realized noisy image is used for detection algorithm.Under the situation of the hysteresis of suitable parameterization, the edge that causes by noise in the counter image, do not occur, and edge can single there be not noise image and disappear because of one.
The 4th preferred implementing form of video smoke detection system of the present invention is characterised in that, adopt three kinds of data structures, a kind of have data segment about the information that is present in the edge in the image separately, a kind of have be used for eliminating for Smoke Detection do not need the data segment of bit mask of the image range considered and observed image itself, wherein, described edge and image remain unchanged between the subsequent iteration of processing procedure, and reinitialize described bit mask for each iteration.
The 5th preferred implementing form of video smoke detection system of the present invention is characterised in that, analyzes described image and edge by pixel ground, and for the analysis of carrying out bit mask hereinafter referred to as the group of a plurality of pixels of piece.
The 6th preferred implementing form of video smoke detection system of the present invention is characterised in that, on two paths, that promptly be used for calculating the edge that is present in image and be used to upgrade first path of the data on the edge Already at one, and a processing that is used to set up the enterprising line data in second path of bit mask, wherein, this second path comprises described motion detection.
The 7th preferred implementing form by video smoke detection system of the present invention, second path comprises saturated to the described device that is used to absorb video image of described block check in addition, wherein with some pieces of saturated pixel mark of ascertaining the number, and do not consider these pieces for the analysis of the counter image at edge.
Another preferred implementing form of video smoke detection system of the present invention is characterised in that, image segments arbitrarily can be got rid of outside analyzing by a kind of shielding.Especially will be used for by the bit mask that described motion detection and saturated check are set up, be updated in counter image used when eliminating the image range that to consider for Smoke Detection.
Another preferred implementing form of video smoke detection system of the present invention is characterised in that, carries out a kind of check before making about the judgement that exists smog, whether exists the edge of enough numbers for this judgement.
Embodiment
Press accompanying drawing 1, video smoke detection system of the present invention is made up of a plurality of video cameras 1 and a common processor 2 that carries out video camera 1 Signal Processing and analysis basically.Described video camera 1 for example is installed in the vcehicular tunnel, and is used for traffic monitoring, for example is used for observing and being used to detect dust formation, accident and similar thing of monitoring and controlling traffic rule.Described video camera is to link to each other with the people operational centre on duty that has of observing traffic conditions in the tunnel through monitor.Arranged described processor 2 dispersedly, wherein, common processor 2 is distributed to certain number, for example 8 to 10 video cameras respectively.
Be broken down into pixel at the video image described in the processor 2, give the set of dispense brightness value of each pixel and/or these pixels, and by the brightness value of described pixel and relatively carrying out about there being the judgement of smog of a reference value.Importantly, this distribution does not rely on whole brightness and changes, and does not just rely on the variation of entire image illumination when distributing brightness value for each pixel or pixel groups.Can reach this independence as follows, promptly the rim value of a kind of deduction of real expression be distributed to described pixel illumination.The identification of smog on the basis of this assumption, promptly described edge is weakened by smog, or disappears.
Signal Processing in the processor 2 and signal analysis further can be divided into two function square frames of representing with pixel intensity 3 and Smoke Detection 4 in accompanying drawing 1.By this division, the process flow diagram of accompanying drawing 2 showed for the asking for of the representational value of pixel intensity (pixel intensity 3), and that process flow diagram of accompanying drawing 3 has then been showed from smog and existed the aspect that these values are made further to analyze and research (Smoke Detection 4).Accompanying drawing 4 has been showed by the described method of accompanying drawing 2 for the necessary additional step of some purposes (Smoke Detection in the inner space that for example resembles passageway, entrance hall and similarity).
To be decomposed into pixel by the video image of each video camera 1 picked-up, and digitizing, thus for each pixel with coordinate i and j determine this pixel such as can be at the intensity level I between 0 and 255
I, jFrom described intensity level I
I, jIn form mean value M for certain group of 3 * 3 or 5 * 5 pixels for example
I, jOr intermediate value or by value that low-pass filtering obtained.The advantage that described intermediate value has is can be with 8 calculating of carrying out it.
Be parallel to the calculating of mean value or intermediate value, from described intensity I
I, jRim value of middle acquisition, this is by inferring or realizing by frequency analysis (high-pass filtering, such as wavelet transform).For example can obtain the rim value K of each pixel by using Robert arithmetical unit or Sobel arithmetical unit
I, jBut can certainly adopt more complex calculations device for edge calculations, and be applied to such as resemble on the big zone of 5 * 5 or 7 * 7 pixels.
Analysis and research then, described rim value K
I, jWhether be positioned on mean value or the intermediate value.If sure, then to a value Z
I, jAdd meter one number δ
Ob, and by the new old value Z of value replacement
I, j, if negate, then from a value Z
I, jDeduct a number δ
Un, and by the new old value Z of value replacement
I, jDescribed value Z
I, jBe a following number, it has illustrated the brightness of described rim value and relevant pixel is how to be positioned at certain threshold value (mean value or intermediate value M continually on certain time period fifty-fifty
I, j) on.This number Z
I, jHereinafter referred to as the counter image.Z
I, jThe value scope be 0 to 255 for example, Z
I, jInitial value be 0 when system initialization.Described several δ
UnAnd δ
ObCan be identical or different, such as both can equal 1.
In view of the influence of motion to rim value, described counter image Z
I, jHas a kind of special advantage.If an object moves through described image, then at least one edge also moves through this image, and this causes at the edge separately, and the pixel on the place has a higher rim value, described counter image Z
I, jTherefore δ rises.In case described edge has left relevant pixel, described counter image Z
I, jThen reduce δ
Un, make the edge pass video image on the whole to the counter image Z of each pixel
I, jThere is not any influence.
The last counter image Z that is obtained
I, jSo especially a kind of representational value of brightness for relevant pixel.At analysis and research counter image Z
I, jThree kinds of time scales of Shi Caiyong: the frequency of the video image that absorbs, for example 1/25 second, 255 images after per 10 seconds and such as per half an hour.
Press accompanying drawing 3 with described counter image Z
I, jWith a threshold value S
zCompare.If described counter image Z
I, jBe positioned at described threshold value S
zUnder, then do not carry out any running, if it is positioned at described threshold value S
zOn, then carry out addition, promptly be worth ∑
xImprove 1, and new thus value replaces.So obtain described initialization value ∑
x 0, make when initialization with ∑=0 beginning and add up, wherein, a kind of initialization value ∑ that is used as then appearred after certain stabilization sub stage in several seconds
x 0Stationary value.∑
xUnder normal situation, should equal ∑
x 0
If ∑
xObviously greater than ∑
x 0, new edge then appears, and this may following reason cause that a static object is arranged in the image range of video camera.Object such in the tunnel may be a static automobile for example, or may be a kind of object that is parked in this passageway in the passageway; Covered certain image range by described object in both cases, this is called covering in accompanying drawing 3.Under situation about covering, redefine the initialization value ∑
x 0Form merchant's ∑ subsequently
x/ ∑
x 0, and with smog threshold value S
RCompare.If described merchant is positioned under the smog threshold value, and therefore the edge is weakened or is disappeared, and then starts and reports to the police.
As long as edge translation in prospect clearly, this is always this usually situation in the tunnel, then for the accurate and reliable not Smoke Detection of false alarm, merchant's ∑
x/ ∑
x 0With smog threshold value S
RRelatively be absolute enough.For the Smoke Detection in highway and the railway tunnel, thereby adopt a kind of functional system shown in accompanying drawing 2 and 3 that has.
During Smoke Detection in relating to the inner space that the people stays, situation then is very.Confirm, stand on the three unities and mutually the people of talk implement a kind of round-robin or back and forth motion on described place, different with translation motion is that this motion no longer drops on described counter image Z
I, jOutside.The motion of reticulate pattern or pattern also is debatable.These motions cause producing new edge, and this may compensate because the reduction or the minimizing at the edge that smog causes make and can not detect smog reliably more sometimes.In general, motion causes new edge usually and also covers the edge sometimes, and smog is not directed at new edge, but the reduction edge.The smog that may be directed at new edge sometimes on remote is a kind of exception of this rule.Owing to the part that is arranged in the top of video image from video camera scope farthest, so all can get rid of this effect by the image section that shields this top, maybe can take following hypothesis, promptly the edge that is formed by smog only moves very lentamente.
For the disturbing effect that prevents to move, adopt when needed shown in the accompanying drawing 4, be used to eliminate motion and from edge K
I, jThe subroutine that (accompanying drawing 2) sets out.In principle also can be from described intensity I
I, jSet out, but this is with the shortcoming that exists interference direct current composition.Form the difference DELTA K of successive images
I, jAnd with this difference DELTA K
I, jWith a kind of movement threshold S
BCompare.If Δ K
I, jBe positioned under this threshold value, then do not have motion.At Δ K
I, j>S
BThe time, the pixel that satisfies this condition is aggregated into the subregion that therefrom shields described motion.Realize the latter as follows, promptly do not upgrade described counter image Z
I, j, and the last counter image before the described motion is used for described subregion.
By a kind of morphologic wave filter (erosion) erasure signal noise.This means following content: it is a kind of bianry image that the error image that changes number of pixels in the described subregion is provided.Run over this bianry image with a kind of pattern, and give the pixel that covers with described pattern with value " 1 ".So point out the end of described motion, make described subregion from described image, disappear in succession, and reduce the edge.
Accompanying drawing 5 has been showed the process flow diagram of second embodiment of video smoke detection system of the present invention, and the characteristics of this video smoke detection system especially are the height reliability of jamproof height robustness and Smoke Detection.Represent observed image with reference symbol A in the accompanying drawing 5.
Below introduce several general elaborations before the described process flow diagram in explanation: because described edge not only may be covered by smog, and may be by at video camera with there is the object between the relative edges to cover, so additionally from the observed image of motion aspect analysis and research.At this moment, starting point is, an object that covers the edge does not suddenly produce on this position, goes but moved to there.
When Smoke Detection, the on the other hand differing time frames that note, that will consider on the one hand and that point that should differentiate mutually on the other hand.For example as the DE Camera Shake that is caused by the truck that closely crosses, exist the effect very rapidly in the submicrosecond scope, this can eliminate by forming unsteady mean value.Also have for example to resemble and cause by smog, medium effect rapidly in 10 seconds scopes because described smog needs about 10 seconds, arrives its detected place up to it, and effect slowly in 10 minutes scopes, perhaps slower effect arranged for example.The latter for example is the obviously mobile influence that is produced by the sun.A kind of possibility that be used for distinguishing these time frames and that be used to be identified in the effect of correct time frame is the counter image with hysteresis.
A counter image is a series of value, is generally the size of an image that can amplify or dwindle.These values are generally used for for example counting of incident.The not only rim detection of algorithm shown in the process flow diagram, and Smoke Detection all depends on the counter image that upgrades with a kind of hysteresis algorithm.Described hysteresis is characterised in that four values, and is minimum, low, high and the highest, and wherein, " minimum " and " the highest " forms the counter limits that can not be lower than or exceed.Described value " low " is positioned on the described value " minimum ", and described value " height " is positioned between described value " low " and " the highest ".If meter reading is positioned between " minimum " and " low ", or be positioned between " height " and " the highest ", then normal fully counting, promptly meter reading improves 1 to the incident that each detected.When meter reading reached value " low " and detect other a incident, it then jumped to the highest.Similarly, when meter reading reduces when reaching described value " height " from the top down, meter reading then jumps to minimum.
This hysteresis mechanisms makes can use noisy image for detection algorithm.Under the situation of the hysteresis of suitable parameterization, the edge that causes by noise in the counter image, will not occur, and the edge can be owing to a single noisy image disappears.In addition, also be suitable for following relation: the number of the single image that described value " " continues, on these single images, must there be a kind of feature that is used to detect or incident, such as edge, and gone out the number of following single image in succession in described value " ", promptly after these single images, described incident disappears under the situation that Counter Value reduces.Because certain time interval is equivalent to this number of single image respectively, so these time intervals then are yardsticks in the reaction time of described algorithm.
Analysis shown in the process flow diagram is from adopting a kind of such as based on the rim detection 5 of the method for Sobel arithmetical unit.The brightness of each pixel of described each single image of Algorithm Analysis, and follow the tracks of live history by the counter image 6 that the above-mentioned hysteresis mechanisms of usefulness is upgraded.At this moment, the environment for each pixel calculates two values:
With a kind of value of providing q
SobelThe Sobel edge detection filter be applied in the described environment;
Pixel for described environment is calculated a mean value q
Sum
Then by adopting more described two values of two scalar factor (DiffFac and SumFac):
f
DiffFac·q
sobel<?>f
sumFac·q
sum
When the left side of this inequality during greater than the right side, then improve counter 6, when being not more than, then reduce it.Use described hysteresis mechanisms in both cases.
Be parallel to rim detection 5 and carry out a kind of motion detection 7, for example employing is based on the algorithm of standardization simple crosscorrelation to it, and described algorithm is roughly carried out as follows:
Described standardization simple crosscorrelation is:
(formula 1)
At time t, the image that takes out 4 * 4 pixels for example and is seen these pixels as vector among a small circle
Use vector
The same scope of the next image when being illustrated in time t+1.If described scope did not change, then at all
1 Shang Zeyou value 1 by formula.Otherwise the variation in described scope can change described merchant, the feasible yardstick that the degree of this change can be adopted as the change intensity in the described scope.
For adaptive, in formula 1, take advantage of molecule and denominator, and be similar to the standardization simple crosscorrelation of formula 1 and write down the product of formation like this with the factor with the processor that is adopted.If molecule is less, then carried out motion, and the corresponding scope of mark.The unexpected variation of light situation or lighting condition influences the both sides of inequality about equally consumingly, makes that described motion detection is impregnable with respect to the unified variation in the image.Obtain a present image map in this way with 4 * 4 block of pixels.
Next procedure is to calculate the piece that should not consider when analyzing the counter image 6 at edge.Certain number in should detected image, all pieces of the incident that taken place therein of 4 * 4 pixels for example, the described Smoke Detection algorithm of these incident negative effects.These pieces generate a kind of as the complete image size 1/16 counter image and the bit mask 8 that shows.Determine described size by observed piece when the motion detection, the size of this piece can change again.
The next stage is revise video sensor saturated.Thisly saturatedly can bring different problems:
Have only not only unsaturated but also when being not complete black, described standardized simple crosscorrelation just works when the pixel of image;
The boundary of saturated fully image segments occurs as the edge.As if the unexpected variation of illumination can give this impression, moved relevant edge and also can disappear subsequently.
In the scope of saturated pixel, there is not the edge.When relevant scope arrives when saturated, the edge of originally having discerned disappears.
For this reason, in saturated check 9 for each scope by with relatively the checking an of ultimate value, whether the pixel of certain number is saturated.If sure, the then same relevant piece of mark.For fear of the object that moves have " hole "-since described motion detection only detect the part of mobile object-, may be applied in the described bit mask 8 by the porose extended arithmetic device 10 of filling a kind of.
In the described stage, the bit mask 8 of being calculated in motion detection 7, saturated check 9 and the extended arithmetic device 10 is used to the counter image of replacement and obsoleteness process 11 (eliminating the image range that needn't consider for Smoke Detection) now, wherein, use the hysteresis mechanisms that has illustrated again.
On this aspect of algorithm, have two kinds of counter images, also promptly: be detected the counter image 6 that has upgraded of all pixels at edge and the counter image 11 that has upgraded of all pieces that will eliminate.Now the latter's counter image and a parameter are used for so changing described counter image 6 jointly, feasiblely not assessed by any reliable edge that may disturb the effects of Smoke Detection to become possibility.Each piece in the counter image 8 and a threshold value are compared.In the time of on described value is positioned at this threshold value, all pixels in the counter image 6 are set on the minimum value.From the counter image 6 at edge, calculate two kinds of amounts of representing different number of edges constantly now.Number of pixels in the edge that first kind of amount is the present existence on first threshold.Second kind of amount is the number of pixels on second threshold value, wherein, and the number of pixels in the edge that this second threshold value can be interpreted as roughly existing in the moment in a past.
In order to calculate these two kinds a kind of functions of amount definition
This function is counted the pixel C in the image of counter with a value on the threshold value l
I, jNumber.Can calculate described two kinds of amounts now with this function, this moment, estimation like this was positioned at the number of a pixel on the edge now, made l very near-earth is changed to described pixel C
I, jThe maximal value W that can reach
mOn.In order to consider the noise in the image A, for the common selective value (W of l
m-k), wherein, k means the number of frame, and for example being generally on the fixing video camera in the tunnel is roughly 250.
Can add a parameters " picture altitude to the subroutine that pixel counts is used ", this parameter is impelled the upper part of the image of only considering that Smoke Detection is used, such as the first half.Because smog upwards rises usually, this is rational.Additionally can be outside analyzing with a kind of shielding image segments eliminating arbitrarily.
Before judging whether smog exists, whether check exists enough edges in a step 12 now, so that can make this judgement.This check directly is parked in the video camera front for big truck for example, and the situation that described image does not have any edge is necessary.Owing to can not detect fire in this case, indicate that described algorithm can not workable undesired signal under present case so should start one.In order to postpone other action in short-term, and for more insensitive, and adopt one can or to be zero for smog, or greater than zero interrupting value.Before soon, detected in the latter case and do not had enough edges.
Be less than the number of pixels that is equivalent to the edge minimum number if exist, then can carry out two kinds of actions: if described interrupting value has been not equal to zero, then reduce it, and if it reach 1, then start undesired signal.On the contrary, if described interrupting value is zero, then it is brought up to value greater than zero.If exist abundant edge,, and proceed to handle then with described interrupting value set to zero.
If the reliable detection that exists for smog is enough number of edges, then average summation and the difference by the edge carried out about there being the judgement of smog in the stage 13.Described difference and a parameter multiply each other, and compare with described summation.If described summation is bigger, then there is not smog; Report to the police otherwise start.Stop the processing of present image in both cases, and begin next treatment of picture.
Such as the startup of can reporting to the police as follows, promptly connecting relevant video camera, show a kind of corresponding warning in alarm center that someone is on duty or the Surveillance center, this warning impels operating personnel that the image that is provided by relevant video camera more is provided with eyes.Described center for example can be a kind of police service center or the fire-fighting center in the base in city or area, perhaps also can be the command centre of vcehicular tunnel.