CN101021949A - Automatic monitoring method for miner entry and exit of coal mine - Google Patents

Abstract

This invention provides an automatic monitor method for in and out of miners used in coal mines including: 1, input of original video: utilizing a camera head fixed on the passageway of a mine to get a continuous video image sequence, 2, test of movement objects: testing miners when they appear in the video sphere and recording correspondingly to put it into the being tracked sequence, 3, tracking Stat: tracking miners in the sequence one by one to calculate the sum of the in and out persons to get the sum number of in and out miners.

Description

The method that a kind of miner who is used for the colliery comes in and goes out and monitors automatically

Technical field

The invention belongs to automatic monitoring field, the method that particularly a kind of miner who is used for the colliery comes in and goes out and monitors automatically.

Background technology

Coal industry is as the pillar industry of Chinese national economy, and it makes major contribution for national economic development and social development.But because widely distributed, the coal mining complicated condition of coal resources in China, disaster is serious, coal enterprise's level of the productive forces integral body is on the low side, the safety in production basis is relatively weaker, add the constraint of conditions such as numerous little Coal Production mode falls behind, practitioner's quality is lower, make it become an index the most worrying, that be difficult to hold most safely.Now, great serious accident fundamentally is not effectively controlled, and minor accident happens occasionally, and causes heavy losses for people's lives and properties, country and society.Thereby more and more be subject to people's attention for the research of safety, in the Safety of Coal Mine Production system, introduce the video monitoring relevant technologies and obtain real-time security monitoring and detection information, seem particularly urgent.

Existing relevant technologies has the disclosed well head auto inventory people counting apparatus of document " application of well head auto inventory people counting apparatus " in 2006 02 phases " energy technology and management ", this device is realized research purpose by hardware (revolving door), though can tentatively realize checking the into effect of well number, but fail to realize video monitoring, its intelligent degree is not high, can not distinguish identification preferably for fortuitous event (false touch is moved or repeated and touches) in actual use.A kind of " personnel are by the number count device " disclosed in the patent No. is the Chinese utility model patent of 98229686.X, this device is made up of counter, timer, counting pedal, realize counting by counting pedal trigger pip, this device exist equally aforementioned can not video monitoring, intelligent degree is low, the unfavorable shortcoming of action effect.

Summary of the invention

The objective of the invention is to overcome the shortcoming of prior art with not enough, a kind of intelligent degree height is provided, real-time and accuracy are good, can carry out monitoring intuitively in real time and writing down mine turnover miner's number, and then to the miner that borehole operation miner total number of persons the is controlled method of monitoring automatically of coming in and going out.The executive condition of this method is to need strict carry out " the mine safety systems of inquirie " of miner, and turnover must be worn coloured cap during the colliery.

Purpose of the present invention is achieved through the following technical solutions: the method that a kind of miner who is used for the colliery comes in and goes out and monitors automatically comprises the steps:

(1) original video input: utilize the camera that is installed in top, mine gateway, stationkeeping to obtain continuous sequence of video images;

(2) moving object detects: in the time of in the miner appears at the video boundary, it is detected processing, isolate the miner from complicated background, orient initial position, direction of motion that the miner occurs, list in to the corresponding record of its work and with it and follow the trail of sequence fully;

(3) follow the trail of statistics: for miner's number of people of following the trail of fully in the sequence, follow the trail of one by one, if in image sequence frame thereafter, can track continuously, and meet some requirements and (cross counting line such as the number of people, direction forward, the initial position that occurs at certain bar line with inferior), at this moment system can advance at it/go out on the number sum to add up accordingly, and then can be advanced/go out miner's number of mine real-time and accurately.

In the described step (2), the fundamental purpose that detects step is exactly to judge whether two field picture the inside has the existence of the number of people, follows the trail of sequence fully if exist then it is listed in, follows the trail of the back through some frames and participates in the decision-making of coming in and going out.Before detecting, need video pictures is carried out a series of pre-service, mainly comprise following step:

(2.1) background subtraction is handled

The purpose of this step is to extract foreground image with the background subtraction separating method, and also can get rid of the noise of some illumination to a certain extent.Mainly contain for the moving object detection method under the fixed scene: frame-to-frame differences method, optical flow method and background subtraction point-score.What we adopted in this project is that the background subtraction point-score extracts moving target, why adopt this method to be because frame-to-frame differences method instability, because the speed of personnel's walking is different in the colliery, and also there is actionless situation in some cases, so adopt single frame-to-frame differences method effect bad, and adopting m width of cloth image to do many frame-to-frame differencess method, the m here is not easy to determine, adopts fixing m result very undesirable.And the illumination effect that optical flow method is subjected to is too big, and a lot of illumination noises is exactly arranged under the colliery.In the present invention in this case, the brightness of prospect generally is the brightness greater than background, so we adopt the background subtraction point-score, when nobody passes through below the colliery, utilizes manual extraction one width of cloth background as the reference image.

The background difference is chosen a background frames exactly as the reference image, does difference with present frame and reference picture, if reference picture is chosen suitably, then can be partitioned into moving object more accurately.When background image is static, promptly do not change in time, and the image with moving target changes except the pixel value of motion target area, all the other belong to background parts, remain unchanged, such situation is the ideal situation of background difference, but this situation is non-existent in actual conditions, so generally will take a real-time context update.Overall process in this system's operation all has a real-time context update process, just with present frame background is not once upgraded when we detect when present frame has moving object, and is better to realize, more accurately differential effect.

Background difference formula commonly used is:

D _ij＝|I _ij-B _ij|

D wherein _IjRepresent background difference recoil mark i, the pixel value at j place, I _Ij, B _IjRepresent original image and background image i respectively, the pixel value at j place.Hour can introduce a lot of unnecessary information (because we mainly consider be foreground image information) when the brightness of the brightness ratio prospect of background in this case, and for our coal mine environment, because camera is mounted in the top, and the colliery worker wears coloured cap, the brightness of prospect in this case is often greater than the brightness of background, therefore we adopt another kind of background subtraction divisional processing method, just

D _ij＝I _ij-B _ij

Can well under the situation of accurate extraction prospect, contain the noise that some are unnecessary again in this case.Remaining image comprised the head part of motion and illumination effect at random after the background image that utilization extracts was done the background difference, can utilize following morphological operation to remove the influence of illumination to the result.

(2.2) morphology is handled

For the differentiated video image of background, we change into binary map to them, white portion in this binary map comprises the cap of motion and the noise that is much brought by illumination, so we adopt morphological operation to remove some fragmentary noises and disconnect the connected region that some should not connect together originally.We detect the number of connected region in the picture on this basis, write down the area of each connected region in real time, the position of central point and the length scale on two coordinate axis are so that we utilize whether area and some simple shapes are that our interested target is made a strategic decision to connected region.

Conclude that whether to be the decision process of cap see that exactly whether its area is greater than some thresholdings, described thresholding is by the decision of the size of cap in the image, also promptly choose a definite value slightly littler than cap area, this thresholding can be got rid of those smaller connected regions of being brought by illumination, and the length of connected region and wide size will satisfy certain thresholding, this thresholding is by the length of cap and wide decision, if oversize or too narrow just think the worker with adjunct, its objective is get rid of the worker with the influence of spin off, can obtain our interested moving target thus, note the center position of this target, list in to be equipped with and follow the trail of row.

In the described step (3), employing Kalman filtering method is followed the trail of and is equipped with the each point of following the trail of in the row, does the operation that adds or subtract when it satisfies the condition of our setting.The advantage of utilizing Kalman is position and the velocity information that this method is utilized tracking point, can overcome the influence of pedestrian's motion speed, improved the precision that detects, and complexity is lower, realizes than being easier to.

Kalman filtering is the sixties in 20th century, and Hungary mathematician Kalman proposes, and he is the method for describing a linear dynamic system with a state equation and measurement equation at first.It is its mathematical tool with state equation that this method adopts time domain method, with multivariable Control, and optimum control and estimation and the self-adaptive controlled main contents that are made as.

The problem that Kalman filtering will solve is to seek x under least mean-square error _kEstimated value

Its feature is to calculate x with recurrence method _kConcretely, establish the state equation of known dynamic system and measure equation, they are respectively

x _k＝A _kx _k-1+ω _k-1 (1)

y _k＝C _k+v _k (2)

In the formula: A _k--n * n ties up matrix;

C _k--m * n ties up matrix, is called the measurement matrix;

x _k---n ties up state vector;

y _k--m ties up observation vector;

ω _k--n dimension average is zero white noise vector, process noise;

υ _k--m dimension average is zero white noise vector, measurement noise.

Suppose:

(1) ω _kWith υ _kAll be that average is zero normal white noise, and ω _kWith υ _kUncorrelated mutually.

(2) original state x ₀Be random vector, it and ω _k, υ _kIndependent, its statistical property is given

From state equation and measurement equation A as can be known _kWith C _kBe known, y _kBeing the data that measure, also is known certainly.Problem is how from y _kAnd x _K-1Try to achieve

At first, temporarily do not consider ω _kWith υ _k, the x that obtain by formula (5) and (6) this moment _kWith y _kUse respectively

With

Expression, then a step recursion result is

${\hat{x}}_k^{\′}=A_k{\hat{x}}_{k-1}---\left(3\right)$

${\hat{y}}_k^{\′}=C_k{\hat{x}}_k^{\′}=C_k{A}_k{\hat{x}}_{k-1}---\left(4\right)$

Here,

Be x _K-1Estimated value.Again will

With y _kThe actual observation value make comparisons, their difference is used

Expression has ${\tilde{y}}_k=y_k-{\hat{y}}_k^{\′},$ Obviously

Implied ω _K-1With υ _kInformation, in other words

Implied current (up-to-date) observed value y _kInformation, so claim Be new breath.

Then, we will

Multiply by a certain K _kRevise original

Value can be estimated better

${\hat{x}}_k=A_k{\hat{x}}_{k-1}+K_K(y_k-{\hat{y}}_k^{\′})=A_k{\hat{x}}_{k-1}+K_K(y_k-C_k{A}_k{\hat{x}}_{k-1})---\left(5\right)$

Therefore,

With true value x _kSquare error be an error square formation.If we can try to achieve the K under this error battle array minimal condition _κ, then with this K _κSubstitution formula (9), then resulting Be exactly to x _kLinear optimal estimate.

Let us is according to (5) now, and (6) and (9) ask the K under the square error battle array minimal condition _κUse P _kExpression square error battle array then has:

Because hypothesis ω _kWith υ _kAll be that average is zero normal white noise, and ω _kWith υ _kAll be that average is zero normal white noise, and ω _kWith υ _kUncorrelated mutually, promptly

$E\left[{\mathrm{\ω}}_k\right]=0;\mathrm{cov}[{\mathrm{\ω}}_k,{\mathrm{\ω}}_j]=E\left[{\mathrm{\ω}}_k{\mathrm{\ω}}_j^{\mathrm{\τ}}\right]=Q_k{\mathrm{\δ}}_{\mathrm{kj}}---\left(7\right)$

$E\left[{\mathrm{\υ}}_k\right]=0;\mathrm{cov}[{\mathrm{\υ}}_k,{\mathrm{\υ}}_j]=E\left[{\mathrm{\υ}}_k{\mathrm{\υ}}_j^{\mathrm{\τ}}\right]=R_k{\mathrm{\δ}}_{\mathrm{kj}}---\left(8\right)$

$\mathrm{cov}[{\mathrm{\ω}}_k,{\mathrm{\υ}}_j]=E\left[{\mathrm{\ω}}_k{\mathrm{\υ}}_j^{\mathrm{\τ}}\right]=0;k,j=\mathrm{0,1,2},\mathrm{\Λ}$

Here

Find the solution the least mean-square error battle array, we obtain following one group of Kalman, one step recursion formula:

$P_k^{\′}=A_k{P}_{k-1}{A}_k^{\mathrm{\τ}}+Q_{k-1}---\left(9\right)$

$K_k=P_k^{\′}{C}_k^{\mathrm{\τ}}{(C_k{P}_k^{\′}{C}_k^{\mathrm{\τ}}+R_k)}^{-1}---\left(10\right)$

P _k＝(I-K _kC _k)P _k′ (11)

${\hat{x}}_k=A_k{\hat{x}}_{k-1}+K_k(y_k-C_k{A}_k{\hat{x}}_{k-1})---\left(12\right)$

If original state x ₀System performance E[x ₀] and var[x ₀] known, and order ${\hat{x}}_0=E\left[x_0\right]={\mathrm{\μ}}_0,$ Again

According to following algorithm, can obtain all estimated values and relevant information.

In the present invention, we utilize the simplest Kalman method for tracing, have mainly utilized the coordinate that is equipped with tracking point and this some speed on two coordinate axis (described coordinate axis promptly is two level and vertical limit at image place) to realize; Setting speed (i.e. speed on two coordinate axis) is invariable, simultaneously sets many lines at graphic interface: set successively between the upper and lower boundary line at graphic interface and enter end lines (a), the initial that comes out (b), enter initial (c), the end lines of coming out (d).Enter and be surveyed area between the following boundary line of initial and graphic interface, entering initial with entering between the end lines is trace regions.Described tracing process is specific as follows: (1) is when personnel enter: at first, in a single day personnel are detected in surveyed area, then be put into and follow the trail of in the sequence fully, served as to enter and begun behind the initial to follow the trail of, active value when tracking this target in next frame (value) adds 2, follow the trail of and then do not subtract 1, (this threshold value is that of drawing by training makes counting precision than higher value when active value surpasses certain threshold value, by artificial setting), when target is passed through a line, and entering below the initial when initial detected reference position, we just add 1 to the number that enters to speed for upwards the time, otherwise do not add.Why we will establish, and to enter initial be because if the error of number plus-minus is very big near entering end lines when having the people to rock back and forth, utilizes to enter this condition of initial and just can address this problem.When (2) personnel came out: the process when this process and personnel enter was similar, two lines are arranged equally, and (initial comes out, the end lines of coming out) retrains, detect near entering end lines that we just charged to out following the trail of fully in the sequence of personnel to it when speed is arranged was downward, and write down detected initial position, begin tracing process simultaneously, it is same if active value adds 2 when tracking in next frame, otherwise subtract 1, when active value reaches certain thresholding, and when target is passed through the d line, initial position is on b, and we just added 1 to personnel's number of coming out when speed was downward, otherwise did not add.These four lines that are provided with are frameworks of turnover personnel counting, and the personnel that solved stop or rock problem.

The present invention compared with prior art, have following advantage and beneficial effect: (1) the inventive method can be big at the colliery illumination effect, background is complicated, the miner is mobile big and carry in the particular surroundings that power tool differs, successfully detect the miner, recognition capability is strong, and the personnel that solved problem such as stop or rock, and anti-interference is good; (2) can carry out monitoring intuitively in real time and writing down the number that mine passes in and out the miner, and then borehole operation miner total number of persons is controlled, in time find accident potential.Can provide accurate data for carrying out rescue work timely and effectively after the accident generation simultaneously, help taking suitably decision-making and measure to organize accident to rescue, improve the emergency management and rescue reaction velocity, and for the ex-post analysis accident provides relevant image document, real-time and accuracy are all comparatively desirable; (3) the inventive method is except helping to have solved the mine safety problem, can also provide data accurately and reliably for the safety in production decision-making, for determining that interim trouble free service target and organization security quality dynamic chek provide foundation, can adapt to actual needs preferably, in safety in production, can bring into play tangible effect, have remarkable social benefit.

Description of drawings

Fig. 1 is the operational flowchart of the inventive method.

Fig. 2 is that moving object detects the image that utilizes the background subtraction point-score to form in the step in the method shown in Figure 1.

Fig. 3 is the image after imagery exploitation morphology shown in Figure 2 is handled.

Fig. 4 is the employed graphic interface of Kalman method for tracing; The implication of each line title is among the figure: enter end lines (a), the initial that comes out (b), enter initial (c), the end lines of coming out (d).

Embodiment

The present invention is described in further detail below in conjunction with embodiment and accompanying drawing, but embodiments of the present invention are not limited thereto.

Embodiment

Fig. 1 shows operating process of the present invention, and as seen from Figure 1, this miner automatic monitoring method of coming in and going out comprises the steps---

(1) original video input: utilize the camera that is installed in top, mine gateway, stationkeeping to obtain continuous sequence of video images;

(2) moving object detects: in the time of in the miner appears at the video boundary, it is detected processing, isolate the miner from complicated background, orient initial position, direction of motion that the miner occurs, list in to the corresponding record of its work and with it and follow the trail of sequence fully;

In the described step (2), the fundamental purpose that detects step is exactly to judge whether two field picture the inside has the existence of the number of people, follows the trail of sequence fully if exist then it is listed in, follows the trail of the back through some frames and participates in the decision-making of coming in and going out.Before detecting, need video pictures is carried out a series of pre-service, mainly comprise following step:

(2.1) background subtraction is handled

The purpose of this step is to extract foreground image with the background subtraction separating method, and also can get rid of the noise of some illumination to a certain extent.Mainly contain for the moving object detection method under the fixed scene: frame-to-frame differences, optical flow method and background subtraction point-score.What we adopted in this project is that the background subtraction point-score extracts moving target, why adopt this method to be because frame-to-frame differences method instability, because the speed of personnel's walking is different in the colliery, and also exist in some cases actionless the time, if so do many frame-to-frame differencess method with m width of cloth image, the m here is not easy to determine, adopts fixing m result very undesirable.And the illumination effect that optical flow method is subjected to is too big, and a lot of illumination noises is exactly arranged under the colliery.And at us in this case, the brightness of prospect generally is the brightness greater than background, so we adopt the background subtraction point-score, when nobody passes through below the colliery, utilizes manual extraction one width of cloth background as the reference image.

The background difference is chosen a background frames exactly as the reference image, does difference with present frame and reference picture, if reference picture is chosen suitably, then can be partitioned into moving object more accurately.When background image is static, promptly do not change in time, and the image with moving target changes except the pixel value of motion target area, all the other belong to background parts, remain unchanged, such situation is the ideal situation of background difference, but this situation is non-existent in actual conditions, so generally will take a real-time context update.Overall process in this system's operation all has a real-time context update process.When detecting present frame and do not have moving object, we just background is once upgraded with present frame, and better to realize, differential effect more accurately.

The background difference formula that the present invention adopts is:

D _ij＝I _ij-B _ij

D wherein _IjRepresent background difference recoil mark i, the pixel value at j place, I _Ij, B _IjRepresent original image and background image i respectively, the pixel value at j place.Can well under the situation of accurate extraction prospect, contain the noise that some are unnecessary again like this.Remaining image comprised the head part of motion and illumination effect (see figure 2) at random after the background image that utilization extracts was done the background difference, can utilize following morphological operation to remove the influence of illumination to the result.

(2.2) morphology is handled

For the differentiated video image of background, we change into binary map to them, white portion in this binary map comprises the cap of motion and the noise that is much brought by illumination, so we adopt morphological operation to remove some fragmentary noises and disconnect the connected region that some should not connect together originally.We detect the number of connected region in the picture on this basis, write down the area of each connected region in real time, the position of central point and the length scale on two coordinate axis are so that we utilize whether area and some simple shapes are that our interested target is made a strategic decision to connected region.

Conclude that whether to be the decision process of cap see that exactly whether its area is greater than some thresholdings, described thresholding is by the decision of the size of cap in the image, also promptly choose a definite value slightly littler than cap area, this thresholding can be got rid of those smaller connected regions of being brought by illumination, and the length of connected region and wide size will satisfy certain thresholding, this thresholding is by the length of cap and wide decision, if oversize or too narrow just think the worker with adjunct, its purpose venting remove the worker with the influence of spin off, we can obtain our interested moving target (see figure 3) thus, note the center position of this target, list in to be equipped with and follow the trail of row.

(3) follow the trail of statistics: for miner's number of people of following the trail of fully in the sequence, follow the trail of one by one, if in image sequence frame thereafter, can track continuously and according to information such as its direction of motion, and meeting some requirements, (as: number of people is crossed counting line, direction forward, the initial position that occurs at certain bar line with inferior), at this moment system advances/goes out on the number sum to add up at it accordingly, and then can be advanced/go out miner's number of mine real-time and accurately.

In the described step (3), employing Kalman method for tracing is followed the trail of and is equipped with the each point of following the trail of in the row, does the operation that adds or subtract when it satisfies the condition of our setting.The advantage of utilizing Kalman is position and the velocity information that this method is utilized tracking point, can overcome the influence of pedestrian's motion speed, improved the precision that detects, and complexity is lower, realizes than being easier to.

In the present invention, we utilize the simplest Kalman method for tracing, have mainly utilized the coordinate and the speed on two coordinate axis that are equipped with tracking point to realize; Because we got 30 two field pictures in 1 second, so can setting speed be invariable.Also can in operating process, add some more stable factors, such as colouring information etc.As shown in Figure 4, set many lines at graphic interface: set successively between the upper and lower boundary line at graphic interface and enter end lines (a), the initial that comes out (b), enter initial (c), the end lines of coming out (d).Enter and be surveyed area between the following boundary line of initial and graphic interface, entering initial with entering between the end lines is trace regions.Described tracing process is specific as follows: (1) is when personnel enter: at first, in a single day personnel are detected in surveyed area, then be put into and follow the trail of in the sequence fully, served as to enter and begun behind the initial to follow the trail of, active value when tracking this image in next frame (value) adds 2, follow the trail of and then do not subtract 1, (this threshold value makes counting precision than higher value for of drawing by training when active value surpasses certain threshold value, by artificial setting), when target is passed through a line, and initial detected reference position is entering below the initial, and we just add 1 to the number that enters to speed for upwards the time, otherwise do not add.Why we will establish, and to enter initial be because the error of number plus-minus is very big when having the people to rock back and forth near entering end lines, utilizes to enter this condition of initial and just can address this problem.When (2) personnel came out: the process when this process and personnel enter was similar, two lines are arranged equally, and (initial comes out, the end lines of coming out) retrains, detect near entering end lines that we just were designated as out following the trail of fully in the sequence of personnel to it when speed is arranged was downward, and write down detected initial position, the beginning tracing process, it is same if active value adds 2 when tracking in next frame, otherwise subtract 1, when active value reaches certain thresholding, and when target is passed through the d line, initial position is on b, and we just added 1 to personnel's number of coming out when speed was downward, otherwise did not add.These four lines that are provided with are frameworks of turnover personnel counting, and the personnel that solved stop or rock problem.

The foregoing description is a preferred implementation of the present invention; but embodiments of the present invention are not restricted to the described embodiments; other any do not deviate from change, the modification done under spirit of the present invention and the principle, substitutes, combination, simplify; all should be the substitute mode of equivalence, be included within protection scope of the present invention.

Claims (7)

1, a kind of miner who is used for the colliery automatically method of monitoring of coming in and going out is characterized in that:

(1) original video input: utilize the camera that is installed in top, mine gateway, stationkeeping to obtain continuous sequence of video images;

(2) moving object detects: in the time of in the miner appears at the video boundary, it is detected processing, make corresponding record then and it is listed in to follow the trail of sequence fully;

(3) follow the trail of statistics: for miner's number of people of following the trail of fully in the sequence, follow the trail of one by one, and add up, and then advanced/go out miner's number of mine to advancing/go out the number sum accordingly.

2, the miner who the is used for the colliery according to claim 1 automatically method of monitoring of coming in and going out is characterized in that: in the described step (2), needed video pictures is carried out pre-service before detecting, comprise that background subtraction handles and the morphology processing procedure.

3, the miner who the is used for the colliery according to claim 2 automatically method of monitoring of coming in and going out, it is characterized in that: described background subtraction is handled the background difference formula that adopts and is

D _ij＝I _ij-B _ij

D wherein _IjRepresent background difference recoil mark i, the pixel value at j place, I _Ij, B _IjRepresent original image and background image i respectively, the pixel value at j place.

4, the miner who the is used for the colliery according to claim 2 automatically method of monitoring of coming in and going out, it is characterized in that: described morphology processing procedure comprises binaryzation step and morphological operation step; Described binaryzation step is that the differentiated video image of background is changed into binary map; Described morphological operation step is to utilize basic open and close computing that image is done further processing.

5, the miner who the is used for the colliery according to claim 1 automatically method of monitoring of coming in and going out is characterized in that: in the described step (3), adopt the Kalman method for tracing to follow the trail of and be equipped with the each point of following the trail of in the row, do the operation that adds or subtract when it satisfies the condition of setting.

6, the miner who the is used for the colliery according to claim 1 automatically method of monitoring of coming in and going out, it is characterized in that: in the described step (3), set four lines at graphic interface: set successively between the upper and lower boundary line at graphic interface and enter end lines, the initial that comes out, enter initial, the end lines of coming out; Enter and be surveyed area between the following boundary line of initial and graphic interface, entering initial with entering between the end lines is trace regions.

7, the miner who the is used for the colliery according to claim 1 automatically method of monitoring of coming in and going out, it is characterized in that: the described tracing process of step (3) is specific as follows

When (1) personnel enter: at first, in a single day personnel are detected in surveyed area, then be put into and follow the trail of in the sequence fully, served as to enter and begin behind the initial to follow the trail of, active value adds 2 when tracking this image in next frame, follows the trail of then not subtract 1, when active value surpasses certain threshold value and crosses a line when target, initial detected reference position is entering below the initial, and we just add 1 to the number that enters to speed for upwards the time, otherwise do not add;

When (2) personnel came out: the process when this process and personnel enter was similar, detect near entering end lines that we just were designated as out following the trail of fully in the sequence of personnel to it when speed is arranged was downward, and write down detected initial position, and the beginning tracing process, same if active value adds 2 when tracking in next frame, otherwise subtract 1, when active value reaches certain thresholding, and cross the d line, initial position is on b, we just added 1 to personnel's number of coming out when speed was downward, otherwise did not add.

Images