CN106250820A - A kind of staircase mouth passenger flow based on image procossing is blocked up detection method - Google Patents

Abstract

The invention discloses a kind of staircase mouth passenger flow based on image procossing to block up detection method, including to gathering the pretreatment of video image, reduce the time-consuming of detection process；In conjunction with project background, set monitoring region, reduce the staircase mouth periphery passenger interference to detection；Passenger's target based on particular model anisotropic filter positions, and according to passenger crown arc model, carries out Model Matching；And set passenger flow and block up threshold value, carry out judgement of blocking up.The present invention utilizes the image procossing knowledge such as particular model anisotropic filter and simple Model Matching, realizes staircase mouth passenger flow statistics and intelligent decision of blocking up, it is possible to achieve the extraction one by one to passenger crown target, so that passenger counts accurately.Method disclosed by the invention, can apply in staircase monitoring system, coordinates staircase controller communication, can replace the nurse work of people, well ensures the safe operation of the place such as market or subway station staircase, the generation of minimizing accident.

Description

A kind of staircase mouth passenger flow based on image procossing is blocked up detection method

Technical field

The present invention relates to the technical field of staircase security monitoring and image procossing, refer in particular to a kind of based on image procossing Staircase mouth passenger flow block up detection method.

Background technology

Along with the fast development of China's economy, and the continuous pursuit that people are to quick life style so that answering of staircase With more and more universal, the especially public place such as megastore, subway station.But, bring easily to our life at staircase Meanwhile, some safety problems are also worth we note that, as at staircase mouth, passenger flow is blocked up, easily occur that passenger falls down, the thing such as tramples Therefore, therefore block up detection to carrying out real-time passenger flow at staircase mouth, just seem and be highly desirable to.

For the problems referred to above, current common countermeasure: arrange personnel, on-the-spot nurse.But there is manpower in this method Cost nurse high, artificial is it is difficult to ensure that drawbacks such as real-times.Therefore, utilize image procossing knowledge, realize intelligent monitoring and reality Time detection, a kind of more efficient way of can yet be regarded as.

By image procossing knowledge, realize above-mentioned Detection task, relate to the detection and location of passenger's target, passenger's target Tracking and judge whether occur that passenger flow is blocked up according to specific judgment condition.Wherein, the detection of (1) passenger's target Location, it is contemplated that the colors such as passenger's clothing, floor plates are easy to the complicated factor of people crown color mixture, and the present invention proposes one Plant method based on particular model anisotropic filter, carry out passenger flow target location.(2) judgment condition that passenger flow is blocked up, In view of the requirement of real-time of detection, the present invention mainly number in the monitoring region set this because usually considering.

Summary of the invention

It is an object of the invention to the shortcoming overcoming prior art with not enough, it is provided that a kind of staircase mouth based on image procossing Passenger flow is blocked up detection method, it is achieved the Intelligent Measurement to staircase mouth passenger flow congestion status, can apply in staircase monitoring system, Coordinate staircase controller communication, the nurse work of people can be replaced, well ensure the peace of the place such as market or subway station staircase Row for the national games, the generation of minimizing accident.

For achieving the above object, technical scheme provided by the present invention is: a kind of staircase mouth passenger flow based on image procossing Block up detection method, comprise the following steps:

1) video image acquisition, 360 ° shot downwards particular by fixing vertical floor plates on the ceiling rotatable Hemisphere photographic head obtains；

2) video image collected is carried out gray processing process, to reduce the time-consuming of algorithm, as follows:

I (x, y)=Gray (Image (x, y))

Wherein, (x, is y) image that arrives of acquired original to Image, and (x y) is the image after gray processing to I；

3) set monitoring region, be the appointment region for carrying out passenger's target location, only take advantage of in this appointment region Visitor's target detection, and passenger flow statistics；

4) passenger's target based on particular model anisotropic filter location

It is difficult to be blocked according to the people crown, and it has the feature of arc model, by arc model anisotropic filter Be identified and position passenger's target, including particular model anisotropic filter kernel and detective operators, plots peak search and Matching treatment process, specific as follows:

Described particular model anisotropic filter kernel, in cartesian coordinate system, by original definition be:

$F_{\mathrm{\θ}}(x,y)=e^{(-d^2/2{\mathrm{\σ}}_1^2-s^2/2{\mathrm{\σ}}_2^2)}$

Wherein,

$d\&ap;\frac{r^2-{(x-rcos\mathrm{\&theta;})}^2-{(y-rsin\mathrm{\&theta;})}^2}{2r},(d<<r)$

$s=\sqrt{x^2+y^2}$

In formula, (x y) is the coordinate of any point in image；D is current point and the particular model anisotropic filter of definition Distance between circular arc；S is the distance between current point and the particular model anisotropic filter center of definition；σ₁、σ₂It it is gaussian kernel Standard deviation；θ is the direction of particular model anisotropic filter core；

Described detective operators, for a given direction θ, detective operators DF_θ(x, y) is defined as:

${\mathrm{DF}}_{\mathrm{\&theta;}}(x,y)=\frac{\&part;F_{\mathrm{\&theta;}}(x,y)}{\&part;x}cos\mathrm{\&theta;}+\frac{\&part;F_{\mathrm{\&theta;}}(x,y)}{\&part;y}sin\mathrm{\&theta;}$

Then by the input picture I after gray proces, (x, y) with detective operators DF_θ(x, y) does convolution, has further and turns Change image I_θ' (x, y):

I_θ' (x, y)=I (x, y) * DF_θ(x,y)

At transition diagram as I_θ' (x, y) in, the curvature marginal position in curvature is in set point, find maximum peak Value response, searches for for plots peak；

Described plots peak is searched for, and uses a kind of newton searching method realized based on quadratic function models, for each Individual pixel (x_i,y_i), each Newton step is a length of:

$\left[\begin{array}{c}{\mathrm{\&Delta;x}}_i\\ {\mathrm{\&Delta;y}}_i\end{array}\right]=-{\left[\begin{array}{cc}\frac{{\&part;}^2{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;x^2}& \frac{{\&part;}^2{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;x\&part;y}\\ \frac{{\&part;}^2{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;y\&part;x}& \frac{{\&part;}^2{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;y^2}\end{array}\right]}^{-1}\left[\begin{array}{c}\frac{\&part;{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;x}\\ \frac{\&part;{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;y}\end{array}\right]$

If meeting | (Δ x_i+Δy_i) |≤0.5 (pixel), then it is assumed that position (x_i,y_i) it is a plots peak, this district The acutance of territory peak value can pass through plots peak point (x_i,y_i) Grad investigate, if gradient in the x, y direction meet such as Lower condition:

${\left|\frac{\&part;{I_{\mathrm{\&theta;}}}^{\&prime;}(x,y_i)}{\&part;x}\right|}_{(x=x_i-1,x_i+1)}\&GreaterEqual;T_x---\left(1\right)$

${\left|\frac{\&part;{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y)}{\&part;y}\right|}_{(y=y_i-1,y_i+1)}\&GreaterEqual;T_y---\left(2\right)$

Wherein, T_x、T_yBe threshold value constant, then this position is defined as an initial search point, by the time all in transition diagram picture Initial search point found after, start on each initial search point perform matching treatment process, extract passenger's head with this Top feature；

Described matching treatment process, it is assumed that during previously, has n initial search point S_i(x_i,y_i)|_{I=0 ... n-1}Determined Justice, based on each initial search point S_i(x_i,y_i), perform matching treatment process, for determining whether the crown profile of passenger is positioned at This position, this process starts from S_i(x_i,y_i) position, execution is with path based on arc model, the half of arc model Footpath is R, and its value is that the mean radius according to passenger crown profile is chosen；

The central point C of this arc model_i(x_o,y_o) it is defined as:

$\left\{\begin{array}{c}{x}_o=x_i-R{\mathrm{cos\&theta;}}_o\\ y_o=y_i+R{\mathrm{sin\&theta;}}_o\end{array}\right.$

If the direction θ of detective operators_oEqual to 0, initial point is (x_i,y_i), then have:

C_i(x_o,y_o)=(x_i-R,y_i)

This matching treatment process is to realize according to below scheme: with central point C_i, scope 0～2 π, chooses different directions θ_j, along radial direction θ_jOn, in set point, performing marginal point search, this detection range is formed as follows:

$\left\{\begin{array}{c}x=d_{j-1}*{\mathrm{cos\&theta;}}_j+x_o\\ y=d_{j-1}*{\mathrm{sin\&theta;}}_j+y_o\end{array}\right.{|}_{d-\mathrm{\&xi;}<d_{j-1}<d+\mathrm{\&xi;}}$

Wherein, d_j-1It is central point C_iAnd θ_j-1Distance between step-size in search marginal point on direction；ξ is detection range, logical It it is often several pixel size；

This edge point position is to have the position of greatest gradient value in the ξ of region of search 2, and gradient is quilt in the direction of search It is defined as:

$G(x,y)=\frac{\&part;{I^{\&prime;}}_{\mathrm{\&theta;}}(x,y)}{\&part;x}{\mathrm{cos\&theta;}}_j+\frac{\&part;{I^{\&prime;}}_{\mathrm{\&theta;}}(x,y)}{\&part;y}{\mathrm{sin\&theta;}}_j$

After marginal point is found, by calculating marginal point at θ_j-1And θ_jConnectedness on direction, checks new edge Whether point belongs to the identical profile in previous steps；For central point C_iAnd θ_jBetween the marginal point searched on direction away from From d_j, its limited region, this region is to come based on the minimum and maximum radius that the passenger crown profile estimated is possible Definition；

When by these initial search points S_i(x_i,y_i)|_{I=0 ... n-1}Guide after all of region is scanned for, through ash In input picture after degreeization, position, all of passenger top area just can be found；

5) passenger flow is blocked up threshold decision

The patronage detected in the monitoring region that mainly statistics sets, is filtered based on particular model direction by above-mentioned , if there is prison in passenger's target location of ripple device, it is possible to the passenger crown number detecting and navigating in the monitoring region set Control range statistics number exceedes setting threshold value, then be determined as passenger flow and block up.

The present invention compared with prior art, has the advantage that and beneficial effect:

1, use image procossing and machine vision knowledge, carry out detection that the intelligence of staircase passenger flow is blocked up, this solution party Case is the most novel.

2, the method that the anisotropic filter based on particular model proposed carries out head of passenger detection and location, can well Reduce the interference of the factor such as large piece article and ambient light change entrained by passenger's clothing, floor plates color, passenger, and to taking advantage of Visitor crown target carries out extraction one by one, so that passenger's counting is more accurate.

3, it is applied in staircase monitoring system, coordinates staircase controller communication, well instead of the artificial of staircase operation Nurse task, reduces cost of labor, and the real-time of detection of accomplishing to block up staircase mouth passenger flow, can well ensure market or The smooth and easy operation of safety of the place staircases such as subway station.

Accompanying drawing explanation

Fig. 1 is the detection method overall flow block diagram of the present invention.

Fig. 2 be the present invention detection method in photographic head scheme of installation.

Fig. 3 is the schematic diagram of photographic head in terms of staircase side.

Fig. 4 is the monitoring region set, and A, B are rectangle summit.

Fig. 5 is in the detection method of the present invention, the geometric model of particular model anisotropic filter core in cartesian coordinate system.

Fig. 6 is the passenger crown based on detection method outline process, a be initial search point, b be the field of search Territory point, C are central points.

Detailed description of the invention

Below in conjunction with specific embodiment, the invention will be further described.

Staircase mouth passenger flow described in the present embodiment is blocked up detection method, mainly by the knowledge of image procossing, realizes Intelligent Measurement to staircase mouth passenger flow congestion status, as it is shown in figure 1, comprise the following steps:

1) video image acquisition

As shown in Figures 2 and 3,360 ° shot downwards particular by fixing vertical floor plates on the ceiling rotatable Hemisphere photographic head 1 obtains, and ceiling 2 is used for installing fixing camera 1, and staircase mouth floor plates 3 is that the shooting of photographic head 1 is right As, for staircase mouth passenger flow 4 is monitored.Photographic head 1 is apart from highly preferred 4 meters of staircase mouth floor plates 3.

2) video image collected is carried out gray processing process, to reduce the time-consuming of algorithm, as follows:

I (x, y)=Gray (Image (x, y))

Wherein, Image (x, y) is the image that arrives of acquired original, I (x y) is the image after gray processing.

3) set monitoring region, be the appointment region for carrying out passenger's target location, only take advantage of in this appointment region Visitor's target detection, and passenger flow statistics；As shown in Figure 4, monitoring region is the boxed area of icon, wherein A (x_a,y_a)、B (x_b,y_b).I.e. monitoring region is:

$\left\{\begin{array}{c}{x}_a\&le;x\&le;x_b\\ y_a\&le;y\&le;y_b\end{array}\right.$

4) passenger's target based on particular model anisotropic filter location

It is difficult to be blocked according to the people crown, and it has the feature of a certain degree of arc model, passes through arc model Anisotropic filter is identified and positions passenger's target, as shown in Figure 5 and Figure 6, including particular model anisotropic filter kernel And detective operators, plots peak search and matching treatment process, specific as follows:

Described particular model anisotropic filter kernel, in cartesian coordinate system, by original definition be:

$F_{\mathrm{\&theta;}}(x,y)=e^{(-d^2/2{\mathrm{\&sigma;}}_1^2-s^2/2{\mathrm{\&sigma;}}_2^2)}$

Wherein,

$d\&ap;\frac{r^2-{(x-rcos\mathrm{\&theta;})}^2-{(y-rsin\mathrm{\&theta;})}^2}{2r},(d<<r)$

$s=\sqrt{x^2+y^2}$

In formula, (x y) is the coordinate of any point in image；D is current point and the particular model anisotropic filter of definition Distance between circular arc；S is the distance between current point and the particular model anisotropic filter center of definition；σ₁、σ₂It it is gaussian kernel Standard deviation, preferably σ² ₁=1/6, σ² ₂=1/36；θ is the direction of particular model anisotropic filter core；

Described detective operators, for a given direction θ, detective operators DF_θ(x, y) is defined as:

${\mathrm{DF}}_{\mathrm{\&theta;}}(x,y)=\frac{\&part;F_{\mathrm{\&theta;}}(x,y)}{\&part;x}cos\mathrm{\&theta;}+\frac{\&part;F_{\mathrm{\&theta;}}(x,y)}{\&part;y}sin\mathrm{\&theta;}$

Then by the input picture I after gray proces, (x, y) with detective operators DF_θ(x, y) does convolution, has further and turns Change image I_θ' (x, y):

I_θ' (x, y)=I (x, y) * DF_θ(x,y)

At transition diagram as I_θ' (x, y) in, be in a range of curvature marginal position in curvature, maximum peak value rings Should be found, search for for plots peak；

Described plots peak is searched for, and uses a kind of newton searching method realized based on quadratic function models, in order at I_θ' (x, y) in find out all of plots peak point, for those brightness values beyond certain threshold value pixel for, pixel-by-pixel search be Necessary；For each pixel (x_i,y_i), each Newton step is a length of:

$\left[\begin{array}{c}{\mathrm{\&Delta;x}}_i\\ {\mathrm{\&Delta;y}}_i\end{array}\right]=-{\left[\begin{array}{cc}\frac{{\&part;}^2{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;x^2}& \frac{{\&part;}^2{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;x\&part;y}\\ \frac{{\&part;}^2{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;y\&part;x}& \frac{{\&part;}^2{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;y^2}\end{array}\right]}^{-1}\left[\begin{array}{c}\frac{\&part;{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;x}\\ \frac{\&part;{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;y}\end{array}\right]$

If step distance value is in a pixel, i.e. | (Δ x_i+Δy_i) |≤0.5 (pixel), then position (x_i,y_i) it is one Individual plots peak, the acutance of this plots peak can pass through plots peak point (x_i,y_i) Grad investigate, if gradient Meet following condition in the x, y direction:

${\left|\frac{\&part;{I_{\mathrm{\&theta;}}}^{\&prime;}(x,y_i)}{\&part;x}\right|}_{(x=x_i-1,x_i+1)}\&GreaterEqual;T_x---\left(1\right)$

${\left|\frac{\&part;{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y)}{\&part;y}\right|}_{(y=y_i-1,y_i+1)}\&GreaterEqual;T_y---\left(2\right)$

Wherein, T_x、T_yIt it is threshold value constant；

Meet above-mentioned (1) and (2), then this position is defined as an initial search point, by the time all of in transition diagram picture After initial search point is found, starts to perform matching treatment process on each initial search point, extract the passenger crown with this Feature；

Described matching treatment process, it is assumed that during previously, has n initial search point S_i(x_i,y_i)|_{I=0 ... n-1}Quilt Definition.Based on each initial search point S_i(x_i,y_i), perform matching treatment process, for determining the crown profile whether position of passenger In this position, this process starts from S_i(x_i,y_i) position, execution is with path based on arc model, this circular arc mould The radius of type is R, and R value is that the mean radius according to passenger crown profile is chosen, and R preferred value is 15 (pixel size)；

The central point C of this arc model_i(x_o,y_o) it is defined as:

$\left\{\begin{array}{c}{x}_o=x_i-R{\mathrm{cos\&theta;}}_o\\ y_o=y_i+R{\mathrm{sin\&theta;}}_o\end{array}\right.$

If the direction θ of detective operators_oEqual to 0, initial point is (x_i,y_i), then have:

C_i(x_o,y_o)=(x_i-R,y_i)

This matching treatment process is to realize according to below scheme: with central point C_i, scope 0～2 π, chooses different directions θ_j, along radial direction θ_jOn, in a little scope, performing marginal point search, this detection range is formed as follows:

$\left\{\begin{array}{c}x=d_{j-1}*{\mathrm{cos\&theta;}}_j+x_o\\ y=d_{j-1}*{\mathrm{sin\&theta;}}_j+y_o\end{array}\right.{|}_{d-\mathrm{\&xi;}<d_{j-1}<d+\mathrm{\&xi;}}$

Wherein, d_j-1It is central point C_iAnd θ_j-1Distance between step-size in search marginal point on direction；ξ is detection range, logical Being often several pixel size, ξ is preferably 3 (pixel size)；

This edge point position is to have the position of greatest gradient value in the ξ of region of search 2, and gradient is quilt in the direction of search It is defined as:

$G(x,y)=\frac{\&part;{I^{\&prime;}}_{\mathrm{\&theta;}}(x,y)}{\&part;x}{\mathrm{cos\&theta;}}_j+\frac{\&part;{I^{\&prime;}}_{\mathrm{\&theta;}}(x,y)}{\&part;y}{\mathrm{sin\&theta;}}_j$

After marginal point is found, by calculating marginal point at θ_j-1And θ_jConnectedness on direction, checks new edge Whether point belongs to the identical profile in previous steps.For central point C_iAnd θ_jBetween the marginal point searched on direction away from From d_j, its limited region, as shown in Figure 6 d_maxAnd d_min.This region is can based on the passenger crown profile estimated The minimum and maximum radius of energy defines.If edge exceeds this range areas, then this matching treatment process is considered It is invalid.d_maxAnd d_minIt is preferably 30 (pixel size) and 7 (pixel size) respectively.

When by these initial search points S_i(x_i,y_i)|_{I=0 ... n-1}Guide after all of region is scanned for, through ash In input picture after degreeization, position, all of passenger top area just can be found；

5) passenger flow is blocked up threshold decision

The patronage detected in the monitoring region that mainly statistics sets, if occurring, monitoring region statistical number of person exceedes Set threshold value, be then determined as passenger flow and block up.Wherein, positioned by above-mentioned passenger's target based on particular model anisotropic filter, Can detect and navigate to the passenger crown number N in the monitoring region set_f, set staircase mouth passenger flow and block up threshold value N₀, N₀ Preferably 5, then adjudicating judgment condition of blocking up is:

In the monitoring region set as shown in Figure 4

$\left\{\begin{array}{c}{x}_a\&le;x\&le;x_b\\ y_a\&le;y\&le;y_b\end{array}\right.$

In,

$result=\left\{\begin{array}{c}true,N_f\&GreaterEqual;N_0\\ false,otherwise\end{array}\right.$

The Intelligent Measurement that staircase mouth passenger flow is blocked up can be realized.

Embodiment described above is only the preferred embodiments of the invention, not limits the practical range of the present invention with this, therefore The change that all shapes according to the present invention, principle are made, all should contain within the scope of the present invention.

Claims (1)

1. a staircase mouth passenger flow based on image procossing is blocked up detection method, it is characterised in that comprise the following steps:

1) video image acquisition, particular by fixing 360 ° of rotatable hemisphere that vertical floor plates shoots downwards on the ceiling Photographic head obtains；

2) video image collected is carried out gray processing process, to reduce the time-consuming of algorithm, as follows:

I (x, y)=Gray (Image (x, y))

Wherein, (x, is y) image that arrives of acquired original to Image, and (x y) is the image after gray processing to I；

3) set monitoring region, be the appointment region for carrying out passenger's target location, only carry out passenger's mesh in this appointment region Mark detection, and passenger flow statistics；

4) passenger's target based on particular model anisotropic filter location

It is difficult to be blocked according to the people crown, and it has the feature of arc model, is entered by arc model anisotropic filter Row identifies and location passenger's target, including particular model anisotropic filter kernel and detective operators, plots peak search and coupling Processing procedure, specific as follows:

Described particular model anisotropic filter kernel, in cartesian coordinate system, by original definition be:

$F_{\mathrm{\&theta;}}(x,y)=e^{(-d^2/2{\mathrm{\&sigma;}}_1^2-s^2/2{\mathrm{\&sigma;}}_2^2)}$

Wherein,

$d\&ap;\frac{r^2-{(x-rcos\mathrm{\&theta;})}^2-{(y-rsin\mathrm{\&theta;})}^2}{2r},(d<<r)$

$s=\sqrt{x^2+y^2}$

In formula, (x y) is the coordinate of any point in image；D is current point and the particular model anisotropic filter circular arc of definition Between distance；S is the distance between current point and the particular model anisotropic filter center of definition；σ₁、σ₂It it is the mark of gaussian kernel Quasi-deviation；θ is the direction of particular model anisotropic filter core；

Described detective operators, for a given direction θ, detective operators DF_θ(x, y) is defined as:

${\mathrm{DF}}_{\mathrm{\&theta;}}(x,y)=\frac{\&part;F_{\mathrm{\&theta;}}(x,y)}{\&part;x}cos\mathrm{\&theta;}+\frac{\&part;F_{\mathrm{\&theta;}}(x,y)}{\&part;y}sin\mathrm{\&theta;}$

Then by the input picture I after gray proces, (x, y) with detective operators DF_θ(x, y) does convolution, has transition diagram further As I_θ' (x, y):

I_θ' (x, y)=I (x, y) * DF_θ(x,y)

At transition diagram as I_θ' (x, y) in, the curvature marginal position in curvature is in set point, find maximum peak value and ring Should, search for for plots peak；

Described plots peak is searched for, and uses a kind of newton searching method realized based on quadratic function models, for each picture Vegetarian refreshments (x_i, y_i), each Newton step is a length of:

$\left[\begin{array}{c}{\mathrm{\&Delta;x}}_i\\ {\mathrm{\&Delta;y}}_i\end{array}\right]=-{\left[\begin{array}{cc}\frac{{\&part;}^2{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;x^2}& \frac{{\&part;}^2{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;x\&part;y}\\ \frac{{\&part;}^2{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;y\&part;x}& \frac{{\&part;}^2{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;y^2}\end{array}\right]}^{-1}\left[\begin{array}{c}\frac{\&part;{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;x}\\ \frac{\&part;{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y_i)}{\&part;y}\end{array}\right]$

If meeting | (Δ x_i+Δy_i) |≤0.5 (pixel), then it is assumed that position (x_i,y_i) it is a plots peak, this peak, region The acutance of value can pass through plots peak point (x_i, y_i) Grad investigate, if gradient meets following bar in the x, y direction Part:

$|\frac{\&part;{I_{\mathrm{\&theta;}}}^{\&prime;}(x,y_i)}{\&part;x}{|}_{(x=x_i-1,x_i+1)}\&GreaterEqual;T_x---\left(1\right)$

$|\frac{\&part;{I_{\mathrm{\&theta;}}}^{\&prime;}(x_i,y)}{\&part;y}{|}_{(y=y_i-1,y_i+1)}\&GreaterEqual;T_y---\left(2\right)$

Wherein, T_x、T_yBe threshold value constant, then this position is defined as an initial search point, by the time all of in transition diagram picture at the beginning of After beginning Searching point is found, start to perform matching treatment process on each initial search point, extract the passenger crown with this special Levy；

Described matching treatment process, it is assumed that during previously, has n initial search point S_i(x_i,y_i)|_{I=0 ... n-1}It is defined, Based on each initial search point S_i(x_i,y_i), perform matching treatment process, for determining whether the crown profile of passenger is positioned at this Point position, this process starts from S_i(x_i,y_i) position, execution is with path based on arc model, the radius of arc model Being R, its value is that the mean radius according to passenger crown profile is chosen；

The central point C of this arc model_i(x_o,y_o) it is defined as:

$\left\{\begin{array}{c}{x}_o=x_i-R{\mathrm{cos\&theta;}}_o\\ y_o=y_i+R{\mathrm{sin\&theta;}}_o\end{array}\right.$

If the direction θ of detective operators_oEqual to 0, initial point is (x_i,y_i), then have:

C_i(x_o,y_o)=(x_i-R,y_i)

This matching treatment process is to realize according to below scheme: with central point C_i, scope 0～2 π, chooses different directions θ_j, edge Radial direction θ_jOn, in set point, performing marginal point search, this detection range is formed as follows:

$\left\{\begin{array}{c}x=d_{j-1}*{\mathrm{cos\&theta;}}_j+x_o\\ y=d_{j-1}*{\mathrm{sin\&theta;}}_j+y_o\end{array}\right.{|}_{d-\mathrm{\&xi;}<d_{j-1}<d+\mathrm{\&xi;}}$

Wherein, d_j-1It is central point C_iAnd θ_j-1Distance between step-size in search marginal point on direction；ξ is detection range, it is common that several Individual pixel size；

This edge point position is to have the position of greatest gradient value in the ξ of region of search 2, and gradient is defined in the direction of search For:

$G(x,y)=\frac{\&part;{I^{\&prime;}}_{\mathrm{\&theta;}}(x,y)}{\&part;x}{\mathrm{cos\&theta;}}_j+\frac{\&part;{I^{\&prime;}}_{\mathrm{\&theta;}}(x,y)}{\&part;y}{\mathrm{sin\&theta;}}_j$

After marginal point is found, by calculating marginal point at θ_j-1And θ_jConnectedness on direction, checks that new marginal point is The no identical profile belonged in previous steps；For central point C_iAnd θ_jDistance d between the marginal point searched on direction_j, Its limited region, this region is to define based on the minimum and maximum radius that the passenger crown profile estimated is possible 's；

When by these initial search points S_i(x_i,y_i)|_{I=0 ... n-1}Guide after all of region is scanned for, through gray processing After input picture in, position, all of passenger top area just can be found；

5) passenger flow is blocked up threshold decision

The patronage detected in the monitoring region that mainly statistics sets, by above-mentioned based on particular model anisotropic filter Passenger's target location, it is possible to detect and navigate to set monitoring region in passenger crown number, if monitored space occurs Territory statistical number of person exceedes setting threshold value, then be determined as passenger flow and block up.