CN104408724A - Depth information method and system for monitoring liquid level and recognizing working condition of foam flotation - Google Patents

Abstract

The invention discloses a depth information method and system for monitoring liquid level and recognizing working condition of foam flotation. The depth information method includes 1, establishing a Kinect-based foam flotation working condition monitoring system in the aspects of soft hardware; 2, collecting flotation foam color and depth data through a Kinect sensor, subjecting depth data to depth information extracting and filtering, and aligning the color data and the depth data in the aspects of time sequence and position and storing the same correspondingly; 3, according to the color data and the depth data, extracting three-dimensional characteristics (including the depth information) such as color, area, volume, speed and damage rate of the foam; 4, performing liquid level monitoring by analyzing relation between the current foam surface liquid level characteristics and the overflow well edge height; 5, subjecting the foam image characteristics to clustering analysis by an improved k-means algorithm, allowing on-line recognition of the flotation working condition achieved. The depth information method and system is applicable to working monitoring and real-time working condition recognition of the foam flotation field, automatic control and optimized operation of the flotation production are realized, and resource utilization rate is improved.

Description

Based on froth flotation level monitoring and the operating mode's switch method and system of depth information

Technical field

The present invention relates to mineral manufacture field, especially a kind of froth flotation level monitoring based on depth information and operating mode's switch method and system.

Background technology

Working condition in flotation operating mode and Floating Production Process, identifies that operating mode is most important to instructing flotation to produce timely and accurately.Froth flotation is most widely used a kind of beneficiation method in mineral processing, the physicochemical change related to is extremely complicated, be all artificially judge current working according to experienced workman by foam surface characteristics in visual inspection flotation cell (color, size, speed etc.) change all the time, and control the floating operations such as chemical feeding quantity with this.But the judgement of different workmans to foam surface characteristics does not have unified quantitative criteria, cause the subjectivity of operating mode's switch and even operation and randomness comparatively large, make flotation be difficult to be in optimum state and cause production run unstable, affecting the quality of production.Along with the develop rapidly of the technology such as machine vision and image procossing, the work carrying out Intelligent Recognition operating mode in conjunction with flotation site foam characteristics has made great progress.By identifying the operating mode classification of flotation site fast and accurately, flotation production control system can adjust manufacturing parameter in time, makes Floating Production Process remain at optimum state.

Mineral pulp level is an important indicator of flotation site operating mode's switch, but be difficult to owing to being covered by froth bed measure, simultaneously, the floatation foam image gathered based on the froth flotation monitoring of working condition of traditional industry video camera and recognition system and method (following summary is legacy system and method) only has plane picture information, not there is depth information, the Real-Time Monitoring to the change of flotation pulp liquid level cannot be obtained, be difficult to the automatic identification and the control automatically that realize liquid level.In addition, do not have depth information also can bring larger difference to the extraction of determination (from work human stereo vision) and the monitor value of characteristics of image empirical value (plane visual from legacy system and method), cause the instability of operating mode's switch result and inaccurate, the application of legacy system and method is extremely restricted.

Summary of the invention

The object of this invention is to provide a kind of froth flotation level monitoring based on depth information and operating mode's switch method and system, by in visual field foam top layer a little apart from bottom land distance, obtain the foam top layer liquid level feature relevant to mineral pulp level, have effectively achieved the online level monitoring based on machine vision; The flotation froth surface three-dimensional feature extracted by kinect sensor, is effectively breached the limitation of planar foam feature, can identify current working more exactly.

For achieving the above object, the invention provides a kind of froth flotation level monitoring based on depth information and operating mode's switch method, comprise the following steps:

Step one: build the froth flotation working-condition monitoring system based on kinect from software and hardware aspect;

Step 2: gather, process and preserve the flotation froth color data and depth data that are obtained by kinect;

Step 3: the color obtain step 2 and depth data carry out feature extraction, color combining and depth data extract three-dimensional (with depth information) feature such as color, area, volume, speed, percentage of damage of foam;

Step 4: adopt distribution fitting method to carry out statistical study to the depth data that step 2 obtains; Method for parameter estimation is adopted to obtain foam top layer liquid level feature; And then the relation obtained between foam top layer liquid level feature and overflow groove edge height, for on-line monitoring and the operating mode's switch of froth flotation liquid level;

Step 5: foam characteristics constitutive characteristic vector step 3 obtained, adopts the k-means algorithm improved to carry out the cluster analysis of off-line, obtain several cluster centres; The foam characteristics of extract real-time flotation site, and carry out real-time matching with cluster centre, the online operating mode obtaining current flotation.

The present invention also provides a kind of froth flotation level monitoring based on depth information and working condition recognition system, it is characterized in that, comprise kinect sensor, high frequency light source, computing machine, kinect sensor is fixed on apart within the scope of flotation cell liquid level 1.2m-3.5m, and its camera place plane is parallel with flotation cell bottom surface; High frequency light source provides illumination for kinect sensor gathers color data, the color collected and depth data is flowed through USB data line and is sent to computing machine, and computing machine realizes data processing, real-time working condition monitoring and result display thereof; Software systems, under cross-platform framework Openni, adopt interactive programming interface, to realize the Real-time Obtaining of color and depth data stream, process, preservation and feature extraction and operating mode's switch.

Beneficial effect: the flotation froth surface three-dimensional feature that the present invention is extracted by kinect sensor, effectively breaches the limitation of planar foam feature, and combines with extraction foam top layer liquid level feature, more accurately can identify current working.

Accompanying drawing explanation

Fig. 1 is the froth flotation level monitoring of the embodiment of the present invention based on depth information and the hardware structure diagram of operating mode's switch method and system.

Fig. 2 is based on the froth flotation level monitoring of depth information and the software flow pattern of operating mode's switch method and system shown in Fig. 1.

Fig. 3 is the color image that in the embodiment of the present invention, kinect gathers.

Fig. 4 is the depth image that in the embodiment of the present invention, kinect gathers.

Embodiment

Kinect is a novel sensor (camera) of Microsoft's development, can the color of Real-time Obtaining target area and the degree of depth two kinds of data, breach the limitation of the two dimensional image feature of traditional camera, the three-dimensional feature information more close with real world can be obtained, be applied to field of play at present, but be not also widely used in industry spot.

For this reason, we are incorporated into froth flotation industry spot Kinect sensor first, the depth data gathered by Kinect sensor, obtain the foam top layer liquid level feature relevant to mineral pulp level, carry out level monitoring and operating mode; In conjunction with the color data that Kinect sensor gathers, extract three-dimensional (with depth information) feature such as color, area, volume, speed, percentage of damage of foam, realize the real-time working condition identification to froth flotation production run.

Below with reference to the drawings and specific embodiments, the present invention is described in further details.

As shown in Figure 2, a kind of froth flotation level monitoring based on depth information of the present embodiment and operating mode's switch method comprise the following steps:

Step one: build the froth flotation working-condition monitoring system based on kinect from software and hardware aspect;

According to the froth flotation monitoring of working condition built shown in Fig. 1 based on depth information and recognition system, hardware is primarily of compositions such as kinect sensor 1, high frequency light source 4, computing machines 5.High frequency light source 4 provides illumination for kinect sensor 1 gathers color data.Kinect sensor 1 is fixed on apart from flotation cell 6 liquid level 2m place, its camera 2 place plane must be parallel with flotation cell 6 bottom surface, the color collected in the scope of camera visual field 3 and depth data are flowed through USB data line and is sent to computing machine 5, computing machine 5 realizes data processing, real-time working condition monitoring and result display thereof.In addition, add dust cover (not shown in FIG.) at high frequency light source 4 and kinect sensor 1 outside, be used for weakening the impact of extraneous factor (as corrosion, polluting camera lens etc.), strengthen the reliability of system.Flotation cell 6 one end is provided with overflow groove 7, and flotation cell 6 inside is equiped with stirring apparatus 8 and scraper plate 9.In figure, Hb represents the distance of camera place plane to flotation cell bottom land, Hd represents that in visual field, foam top layer point is to the distance of camera place plane, Hh represents that in visual field, foam top layer point is to the distance of bottom land, Ht represents certain some place froth bed thickness in visual field (certain point is to the distance of mineral syrup liquid), Hk represents mineral pulp level, and Hy represents overflow groove edge height.

The froth flotation working-condition monitoring system based on kinect is built from software aspect, under cross-platform framework Openni (OpenNatural Interaction), adopt interactive programming interface, to realize the Real-time Obtaining of color and depth data stream, process, preservation and feature extraction and operating mode's switch.

Step 2: gather, process and preserve the flotation froth color data and depth data that are obtained by kinect;

Comprise following steps:

Step 1: image data;

Color data stream is set as RGB888 (RGB), resolution is 640*480, and frame per second is 30FPS; Depth data stream format is set as PIXEL_FORMAT_DEPTH_1_MM form (data precision is millimeter), resolution is 640 × 480; Frame per second is 30FPS; Adopt " Alternative View " instrument under cross-platform framework Openni, the visual angle of correction of color camera and depth camera, to realize color data stream and depth data stream aiming on field of regard.

Start kinect equipment and carry out image acquisition; Adopt the data layout of above-mentioned steps setting to create color data stream and depth data stream simultaneously, obtain two kinds of data that sequential is as shown in Figure 3 and Figure 4 aimed at: 640 × 480 × 3 rank color matrix C and 640 × 480 rank matrix of depths D.Such as in color image, some P ₁(243,133) are labeled as " RGB:242,241,239 ", represent that the value of this point in the color matrix that kinect gathers is C (243,133,1)=242, C (243,133,2)=241, C (243,133,3)=239.In depth image, some P ₁(243,133) are labeled as " Index:661 ", represent that the value of this point in the color matrix that kinect gathers is D (243,133)=661; Mark " RGB:0.918,0.918,0.918 " obtains for showing by depth value normalization, and in depth image, color is darker just represents that depth value is less, and namely this foam top layer distance camera distance is nearer, by comparing foam head point P ₁index value (661) and the foam edge point P at place ₂the Index value (693) at place can be easy to verify above-mentioned saying.

Step 2: process data;

(1) depth information in depth data is obtained;

The matrix of depths D obtained by kinect sensor is 16 integer data, its high 13 store be depth information, and low 3 store be index information, obtain depth information by shifting function:

D ₁=D/2 ³formula 1

D in formula ₁for only storing the matrix of depths of depth information (and not having index information).

(2) color combining data carry out filtering completion process to depth data;

Adopt associating bilateral filtering method to matrix of depths D ₁carry out filtering process, by the depth image completion owing to blocking disappearance, obtain filtered matrix of depths D ₂, it is at the depth value D at point (x, y) place ₂(x, y) is:

$D_2(x,y)=\frac{1}{w_p}\underset{(i_1,j_1)\∈{\mathrm{\Ω}}_1}{\mathrm{\Σ}}w(i_1,j_1)\×D_1(i_1,j_1)$ Formula 2

Ω in formula ₁for the filtered reference neighborhood of required point (x, y), x ∈ [1, a], y ∈ [1, b], (i ₁, j ₁) ∈ Ω ₁, i ₁∈ [x-r ₁, x+r ₁], j ₁∈ [y-r ₁, y+r ₁], r ₁for the radius of neighbourhood, D ₁(i ₁, j ₁) be matrix D ₁at point (i ₁, j ₁) depth value at place, w _pfor normalized parameter:

$w_p=\underset{i_1,j_1\∈{\mathrm{\Ω}}_1}{\mathrm{\Σ}}w(i_1,j_1)$ Formula 3

W (i in formula ₁, j ₁) be the neighborhood Ω of required point (x, y) ₁interior each point (i ₁, j ₁) for the weights of required point, by depth image spatial domain weight w _swith coloured image gray scale territory weight w _rcomposition, that is:

W (i ₁, j ₁)=w _s(i ₁, j ₁) × w _r(i ₁, j ₁) formula 4

Wherein:

$w_s(i_1,j_1)=\exp (-\frac{{(i_1-x)}^2+{(j_1-y)}^2}{{{2\mathrm{\σ}}_s}^2})$ Formula 5

$w_r(i_1,j_1)=\exp (-\frac{{(\mathrm{Gray}(i_1,j_1)-\mathrm{Gray}(x,y))}^2}{{{2\mathrm{\σ}}_r}^2})$ Formula 6

Wherein, σ _rand σ _sbe respectively weight w _sand w _rcorresponding Gaussian function standard deviation, Gray (x, y), Gray (i ₁, j ₁) be respectively pixel (x, y) and neighborhood Ω thereof ₁interior each point (i ₁, j ₁) gray-scale value at place, computing formula is:

Gray (x, y)=0.02989 × C (x, y, 1)+0.5870 × C (x, y, 2)+0.1149 × C (x, y, 3) formula 7

Wherein C (x, y, 1), C (x, y, 2), C (x, y, 3) be respectively the color value on a × b rank matrix mid point (x, y) of R, G, B color component in the color matrix C of rank, a × b × 3, traversal obtains a little gray matrix Gray.

Choose suitable Gaussian function standard deviation sigma _rand σ _s, at suitable filtered reference neighborhood Ω ₁in, effectively can eliminate depth information and the noise spot of the region disappearance that is blocked.

(3) reflection foam top layer is obtained apart from bottom land distance H _hmatrix of depths D ';

As shown in Figure 1, H _hrepresent that in visual field, foam top layer point (x, y), apart from the distance of bottom land, can be obtained by following formula:

H _h=H _b-H _dformula 8

Wherein H _bfor kinect sensor camera place plane is to the distance (being definite value after installing) of flotation cell bottom land, H _dfor foam top layer point in visual field is to the distance of kinect sensor camera place plane, when kinect sensor camera place plane is parallel with flotation cell bottom surface, H _dnamely the matrix of depths D matrix of depths D after displacement and filtering completion for being obtained by kinect sensor ₂the value of middle corresponding point, i.e. H _d=D ₂(x, y).

Obtain reflection H _hcertain point (x, y) depth value D ' (x, y) be:

D ' (x, y)=H _b-D ₂(x, y) formula 9

In the present embodiment, obtain depth information D in depth data by formula 1 shifting function ₁.Get required point (x, y) filtered reference neighborhood Ω ₁radius of neighbourhood r ₁=3, x ∈ [1,640], y ∈ [Isosorbide-5-Nitrae 80], i ₁∈ [x-r ₁, x+r ₁], j ₁∈ [y-r ₁, y+r ₁], by formula 2 ~ 6 couples of matrix of depths D ₁carry out filtering process, obtain filtered matrix of depths D ₂.Carry out gray processing by formula 7 pairs of color data and obtain gray matrix Gray.Reflection foam top layer is obtained apart from bottom land distance H by formula 8 ~ 9 _hmatrix of depths D ', wherein H _b=4m.

Step 3: preserve data;

Color matrix C and matrix of depths D ' is stored.With Motion JPEG AVI coded system access color data, with Motion JPEG2000 (16) coded system access depth data.

Step 3: the color obtain step 2 and depth data carry out feature extraction, color combining and depth data extract three-dimensional (with depth information) feature such as color, area, volume, speed, percentage of damage of foam; Comprise following sub-step:

Step 1: extract color characteristic;

The color obtain step 2 and depth data carry out feature extraction, color combining and depth data, are extracted the color characteristic of foam: R average, G average, B average and gray average Gray by following formula 12 ~ 15 _m.

Suppose that the depth value scope of object in visual field is for [d _min, d _max], definition datum distance is:

Db=(d _max-d _min)/2 formula 10

Suppose that the color being positioned at reference range place point is standard value, then when intensity of illumination is identical, from distance of camera lens more away from (being greater than reference range) some color decay larger, need to strengthen its value, and from distance of camera lens more close to the point of (being less than reference range), need to weaken its color value, for this reason, by setting weight function, carry out standardization in conjunction with depth data to color data, formula is:

Q (x, y)=(D'(x, y)-db) ³/ 10 ⁶+ 1 formula 11

Color characteristic after color value standardization is defined as:

R average: $R_m=\frac{\underset{x=1}{\overset{a}{\mathrm{\Σ}}}\underset{y=1}{\overset{b}{\mathrm{\Σ}}}C(x,y,1)\·q(x,y)}{a\×b}$ Formula 12

G average: $G_m=\frac{\underset{x=1}{\overset{a}{\mathrm{\Σ}}}\underset{y=1}{\overset{b}{\mathrm{\Σ}}}C(x,y,2)\·q(x,y)}{a\×b}$ Formula 13

B average: $B_m=\frac{\underset{x=1}{\overset{a}{\mathrm{\Σ}}}\underset{y=1}{\overset{b}{\mathrm{\Σ}}}C(x,y,3)\·q(x,y)}{a\×b}$ Formula 14

The gray matrix Gray obtained by step 2, calculates gray average:

${\mathrm{Gray}}_m=\frac{\underset{x=1}{\overset{a}{\mathrm{\Σ}}}\underset{y=1}{\overset{b}{\mathrm{\Σ}}}\mathrm{Gray}(x,y)\·q(x,y)}{a\×b}$ Formula 15

Step 2: extract area features;

The intensity profile Gs that the gray matrix Gray obtained step 2 calculates a two field picture is obtained the gray matrix Gray ' of contrast strengthen by formula 16 ~ 17; Adopt watershed algorithm to split matrix Gray ', obtain foam regions two values matrix A by formula 18; Travel through the pixel that a two field picture is all, foam regions is numbered, obtain the label information Ps of marked region _n; Add up the area of all N number of foam regions respectively, obtain area distributions Area; N number of area value in area distributions Area is obtained Area ' according to order rearrangement from small to large, is extracted the area features Area of foam by formula 22 _m.

The gray matrix Gray of step 2 acquisition is calculated to the intensity profile Gs of a two field picture, represent the number of the pixel meeting Gray (x, y)=g with Gs (g), wherein g ∈ [0,255].

From e=0, get e=0 respectively, e=1 ..., e=e ', e ' ∈ (0,255), when e=e ' time, meet:

$\frac{\underset{g\∈[o,e^{\′}]\∪[e^{\′},255]}{\mathrm{\Σ}}\mathrm{Gs}\left(g\right)}{a\×b}>0.01$ Formula 16

Obtain the gray matrix Gray ' of contrast strengthen, its gray-scale value at point (x, y) place is:

${\mathrm{Gray}}^{\′}(x,y)=\left\{\begin{array}{cc}0& \mathrm{Gray}(x,y)\∈(0,e^{\′})\\ \frac{255}{255-2e}(\mathrm{Gray}(x,y)-e^{\′})& \mathrm{Gray}(x,y)\∈(e,255-e^{\′})\\ 255& \mathrm{Gray}(x,y)\∈(255-e^{\′},255)\end{array}\right.$ Formula 17

Adopt watershed algorithm to split matrix Gray ', obtain foam regions two values matrix A, it is designated as A (x, y) at the gray-scale value at point (x, y) place, adopts formula below to carry out the differentiation of foam regions point and frontier point:

Travel through the pixel that a two field picture is all, foam regions be numbered, obtain the label information of marked region:

Ps _n: { (x ₁, y ₁), (x ₂, y ₂), L, (x _{area (n)}, y _{area (n)}) formula 19

Ps in formula _nrepresent the set of all pixels of the n-th foam regions, n=1,2, L, N, N are foam regions sum.Area (n) represents the number of the pixel in the n-th foam regions, be region area, add up the area of all N number of foam regions respectively, obtain area distributions: Area:{Area (1), Area (2) ... Area (n) ... Area (N) }, calculate the overall area area of the N number of foam regions of a two field picture:

$\mathrm{SArea}=\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\mathrm{Area}\left(n\right)$ Formula 20

N number of area value in area distributions Area is obtained Area ' according to order rearrangement from small to large: { Area ' (1), Area ' (2) L Area ' (n) L Area ' (N) }, wherein Area ' (1) is foam area minimum value, and Area ' (N) is foam Maximum Area.

From f ₁=1 starts, and gets f respectively ₁=1, f ₁=2 ..., f ₁=f ₁', f ₁' ∈ (1, N), works as f ₁=f ₁in ' time, meet:

$\frac{\underset{n\∈[1,{f_1}^{\′}]}{\mathrm{\Σ}}{\mathrm{Area}}^{\′}\left(n\right)}{\mathrm{SArea}}>0.2$ Formula 21

Then defining foam area features is:

${\mathrm{Area}}_m=\frac{\underset{n={f_1}^{\′}+1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{Area}}^{\′}\left(n\right)}{N-{f_1}^{\′}-1}$ Formula 22

Step 3: extract volume characteristic;

The region barycenter of the n-th foam regions is calculated by formula 22 ~ 28 with the long and short semiaxis CB of region sub-elliptical _n, DB _n; The pixel coordinate of transverse end points is calculated: c by formula 29 ~ 30 ₁(u _n1, v _n1), c ₂(u _n2, v _n2); The matrix of depths D ' that integrating step two obtains, calculates transverse end points and depth value l corresponding to region barycenter by formula 31 _n1, l _n2, l _nc, suppose that foam is the spheroid of standard, foam real space height h can be obtained by formula 32 _n; Major axis physical length CR is obtained by formula 33 ~ 36 _n, in like manner can obtain minor axis physical length DR _n; Volume V (n) of the n-th foam is obtained by formula 37.

The bulking value of N number of foam regions distribution Vs is obtained Vs ' according to order rearrangement from small to large; The volume characteristic V of foam is extracted by formula 38 _m.

For region two values matrix A and the zone marker Ps of a two field picture _n, (p+q) rank rule square of its n-th foam regions is defined as:

$M_n(p,q)=\underset{(x,y)\∈\mathrm{Ps}\left(n\right)}{\mathrm{\Σ}}{x}^p{y}^{q}A(x,y)$ Formula 23

The region barycenter of the n-th foam regions can be tried to achieve by its 1 rank rule distance:

${\mathrm{AC}}_n:({\stackrel{\‾}{x}}_n,{\stackrel{\‾}{y}}_n)=(M_n\left(\mathrm{1,0}\right),M_n\left(\mathrm{0,1}\right))$ Formula 24

Then (p+q) rank centre distance of a two field picture is:

$U_n(p,q)=\underset{(x,y)\∈\mathrm{Ps}\left(n\right)}{\mathrm{\Σ}}{(x-{\stackrel{\‾}{x}}_n)}^p{(y-{\stackrel{\‾}{y}}_n)}^{q}A(x,y)$ Formula 25

If only consider the second-order moment around mean collection of a two field picture, then foam regions can be approximately an ellipse centered by the barycenter of region, then have:

${\mathrm{CB}}_n={\left(\frac{U_n\left(\mathrm{2,0}\right)+U_n\left(\mathrm{0,2}\right)+{[{(U_n\left(\mathrm{2,0}\right)-U_n\left(\mathrm{0,2}\right))}^2+{4U}_n{\left(\mathrm{1,1}\right)}^2]}^{1/2}}{U_n\left(\mathrm{0,0}\right)/2}\right)}^{1/2}$ Formula 26

${\mathrm{DB}}_n={\left(\frac{U_n\left(\mathrm{2,0}\right)+U_n\left(\mathrm{0,2}\right)-{[{(U_n\left(\mathrm{2,0}\right)-U_n\left(\mathrm{0,2}\right))}^2+{4U}_n{\left(\mathrm{1,1}\right)}^2]}^{1/2}}{U_n\left(\mathrm{0,0}\right)/2}\right)}^{1/2}$ Formula 27

${\mathrm{\θ}}_n=\frac{1}{2}\arctan \left(\frac{{2U}_n\left(\mathrm{1,1}\right)}{U_n\left(\mathrm{2,0}\right)-U_n\left(\mathrm{0,2}\right)}\right)$ Formula 28

CB _nrepresent the major semi-axis of sub-elliptical, DB _nfor minor semi-axis, θ _nfor inclination angle.

And then calculate the pixel coordinate of oval long (short) axle head point: c ₁(u _n1, v _n1), c ₂(u _n2, v _n2), formula is:

$u_{n1}={\stackrel{\‾}{x}}_n-{\mathrm{CB}}_n\×\cos \left({\mathrm{\θ}}_n\right),v_{n1}={\stackrel{\‾}{y}}_n-{\mathrm{CB}}_n\×\sin \left({\mathrm{\θ}}_n\right)$ Formula 29

$u_{n2}={\stackrel{\‾}{x}}_n+{\mathrm{CB}}_n\×\cos \left({\mathrm{\θ}}_n\right),v_{n2}={\stackrel{\‾}{y}}_n+{\mathrm{CB}}_n\×\sin \left({\mathrm{\θ}}_n\right)$ Formula 30

The matrix of depths D ' that integrating step two obtains, calculates transverse end points and depth value l corresponding to region barycenter _n1, l _n2, l _nc:

l _n1＝D′(u _n1,v _n1)

L _n2=D ' (u _n2, v _n2) formula 31

$l_{\mathrm{nc}}=D^{\′}({\stackrel{\‾}{x}}_n,{\stackrel{\‾}{y}}_n)$

Suppose that foam can be approximately the spheroid of standard, definition foam real space height h _nfor:

H _n=l _nc-min (l _n1, l _n2) formula 32

View parameter in conjunction with kinect sensor: horizontal view angle δ ₁with vertical angle of view δ ₂, calculate the three dimensional space coordinate C that major axis end points is corresponding _n1and C _n2:

C _n1: (δ ₁* [2u _n1l+ (a-1) (L-l _n1)], δ ₂[2 (b-v _n1-1) L+ (b-1) (L-l _n1)], L-l _n1) formula 33

C _n2: (δ ₁* [2u _n2l+ (a-1) (L-l _n2)], δ ₂[2 (b-v _n2-1) L+ (b-1) (L-l _n2)], L-l _n2) formula 34

Wherein L represents the distance of depth camera to reference image plane (volume coordinate z=0).

And then obtain major axis physical length:

${\mathrm{CR}}_n=C_{1n}{C}_{2n}\sqrt{{{\mathrm{\δ}}_1}^2{[2L(u_{1n}-u_{2n})-(a-1)(l_{1n}-l_{2n})]}^2+{{\mathrm{\δ}}_2}^2{[2L(v_{1n}-v_{2n})+(b-1)(l_{1n}-l_{2n})]}^2+{(l_{1n}-l_{2n})}^2}$

Formula 35

Suppose that the depth value that major axis two end points are corresponding is equal, major axis actual (tube) length angle value can be reduced to:

${\mathrm{CR}}_n=2L\sqrt{{{\mathrm{\δ}}_1}^2{(u_{1n}-u_{2n})}^2+{{\mathrm{\δ}}_2}^2{(v_{1n}-v_{2n})}^2}$ Formula 36

In like manner can obtain minor axis physical length DR _n.

Finally can obtain the volume (supposing that foam is the spheroid of standard) of the n-th foam:

$V\left(n\right)=\frac{4}{3}\mathrm{\π}(\frac{{\mathrm{CR}}_n}{2}\×\frac{{\mathrm{DR}}_n}{2}\×h_n)$ Formula 37

The thinking of a two field picture area features is obtained according to sub-step in step 32, by N number of Domain Volume Distribution value: Vs:{Vs (1), Vs (2) ... Vs (n) ... Vs (N) }, Vs ' is obtained: { Vs ' (1), Vs ' (2) ... Vs ' (n) ... Vs ' (N) } according to order rearrangement from small to large;

From f ₂=1 starts, and gets f respectively ₂=1, f ₂=2 ..., f ₂=f ₂', f ₂' ∈ (1, N), works as f ₂=f ₂in ' time, meet:

$\frac{\underset{n\∈[1,f_2^{\′}]}{\mathrm{\Σ}}{V}^{\′}s\left(n\right)}{\underset{n\∈[1,N]}{\mathrm{\Σ}}{V}^{\′}s\left(n\right)}>0.2$ Formula 38

Then define foam volume to be characterized as:

$V_m=\frac{\underset{n=f_2^{\′}+1}{\overset{N}{\mathrm{\Σ}}}{V}^{\′}s\left(n\right)}{N-f_2^{\′}-1}$ Formula 39

Step 4: extraction rate feature;

(1) Harris Corner Detection;

First, supposing to strengthen certain point (x, y) in gray matrix Gray ' is angle point, gets its neighborhood Ω ₂radius of neighbourhood r ₂=5, i ₂∈ [x-r ₂, x+r ₂], j ₂∈ [y-r ₂, y+r ₂], by formula 40 ~ 42, calculate this angle point neighborhood Ω ₂interior fractional value G a little, gets the h of fractional value higher than mark 70 up to standard ₁individual, sort from high to low by fractional value and obtain point set ${\mathrm{Hs}}_1:\{(x_1,y_1),(x_2,y_2)...(x_{h_1},y_{h_1})\}.$

Suppose that certain point (x, y) strengthened in gray matrix Gray ' is angle point, its neighborhood Ω ₂interior to be a little changed to along the mean intensity (gray-scale value) in certain side-play amount (Vx, Vy) direction:

$I=\underset{(i_2,j_2)\∈{\mathrm{\Ω}}_2}{\mathrm{\Σ}}{({\mathrm{Gray}}^{\′}(i_2+\mathrm{Vx},j_2+\mathrm{Vy})-{\mathrm{Gray}}^{\′}(i_2,j_2))}^2$ Formula 40

I in formula ₂∈ [x-r ₂, x+r ₂], j ₂∈ [y-r ₂, y+r ₂], r ₂for Ω ₂the radius of neighbourhood.

The calculating of Strength Changes that all side-play amount directions of angle point (x, y) are averaged, Taylor series expansion is carried out to formula 40, omits higher order term, be described as with matrix form:

$I\≈\left[\begin{array}{cc}\mathrm{Vx}& \mathrm{Vy}\end{array}\right]\left[\begin{array}{cc}\mathrm{\Σ}{\left(\frac{{\mathrm{\δGray}}^{\′}}{{\mathrm{\δh}}_x}\right)}^2& \mathrm{\Σ}\frac{{\mathrm{\δGray}}^{\′}}{{\mathrm{\δh}}_x}\frac{{\mathrm{\δGray}}^{\′}}{{\mathrm{\δh}}_y}\\ \mathrm{\Σ}\frac{{\mathrm{\δGray}}^{\′}}{{\mathrm{\δh}}_x}\frac{{\mathrm{\δGray}}^{\′}}{{\mathrm{\δh}}_y}& \mathrm{\Σ}{\left(\frac{{\mathrm{\δGray}}^{\′}}{{\mathrm{\δh}}_y}\right)}^2\end{array}\right]\left[\begin{array}{c}\mathrm{Vx}\\ \mathrm{Vy}\end{array}\right]=\left[\begin{array}{cc}\mathrm{Vx}& \mathrm{Vy}\end{array}\right]Q\left[\begin{array}{c}\mathrm{Vx}\\ \mathrm{Vy}\end{array}\right]$ Formula 41

In formula, Q is covariance matrix, and two eigenwert represents that certain puts the mean intensity change of maximum mean intensity change and vertical direction thereof in all side-play amount directions, and definition scoring functions is:

G=Det (Q)-kTrace ²(Q) formula 42

Wherein Det (Q) is eigenwert sum, i.e. determinant of a matrix value, and Trace (Q) amasss for eigenwert is, i.e. matrix trace.When two eigenwerts are all larger, fractional value is higher.

Definition G _rfor mark up to standard, traversal angle point (x, y) neighborhood Ω ₂interior a little, obtain the h of fractional value higher than mark up to standard ₁individual, sort from high to low by fractional value and obtain point set

Secondly, angle point neighborhood Ω is got ₃radius of neighbourhood r ₃=7, i ₃∈ [x-r ₃, x+r ₃], j ₃∈ [y-r ₃, y+r ₃], by formula 43, the most value process in gray-scale value local is carried out to enhancing gray matrix Gray'; Determine that gray-scale value local is worth a little most by formula 44 ~ 45; From point set Hs ₁in filter out the point meeting local gray level and be worth condition most, obtain by h ₂the point set of individual composition ${\mathrm{Hs}}_2:\{(x_1,y_1),(x_2,y_2)...(x_{h_2},y_{h_2})\}.$

Carry out the most value process in gray-scale value local to enhancing gray matrix Gray', formula is:

Graymax(x,y)＝max(Gray′(i ₃,j ₃)),(i ₃,j ₃)∈Ω ₃

Formula 43

Graymin(x,y)＝min(Gray′(i ₃,j ₃)),(i ₃,j ₃)∈Ω ₃

Wherein Graymax (x, y) is the value at gray matrix Graymax mid point (x, the y) place after gray-scale value suboptimize, Graymin (x, y) be the value at gray matrix Graymin mid point (x, y) place after gray-scale value local minimum, Ω ₃for the neighborhood of point (x, y), its radius of neighbourhood is set to r ₃, and i ₃∈ [x-r ₃, x+r ₃], j ₃∈ [y-r ₃, y+r ₃],

Following formula determination gray-scale value local is adopted to be worth most a little:

From point set Hs ₁in filter out the point meeting local gray level and be worth condition most, obtain by h ₂the point set of individual composition ${\mathrm{Hs}}_2:\{(x_1,y_1),(x_2,y_2)...(x_{h_2},y_{h_2})\}.$

Finally, to matrix of depths D' carry out with ask gray matrix Gray' gray-scale value local value order the same operation, obtain depth value and be locally worth most a little, from point set Hs ₂in filter out the point meeting partial-depth and be worth condition most, obtain by h ₃the angle point point set of individual composition ${\mathrm{Hs}}_3:\{(x_1,y_1),(x_2,y_2)...(x_{h_3},y_{h_3})\}.$

(2) screen to through three angle points the angle point point set Hs determined ₃in point follow the trail of, extraction rate feature.

Suppose the same angle point (x in the two field picture of front and back two _h, y _h) intensity constant (namely gray-scale value is constant), that is:

Gray' _t(x _h, y _h)=Gray' _t+Vt(x _h+ Vu, y _h+ Vv) formula 46

Wherein Gray' _t(x _h, y _h) be that t (previous frame image) strengthens gray matrix Gray' mid point (x _h, y _h) gray-scale value, Gray' _t+Vt(x _h+ Vu, y _h+ Vv) for being the t+Vt moment (a rear two field picture) strengthen gray matrix Gray' mid point (x _h+ Vu, y _h+ Vv) gray-scale value, (Vu, Vv) is angle point (x _h, y _h) side-play amount that occurs within the Vt time, Vt is adjacent two frame time intervals, h=1,2, L, h ₃.

Then there is basic light stream constraint equation:

$\frac{{\∂\mathrm{Gray}}^{\′}}{{\∂x}_h}\mathrm{Vu}+\frac{{\∂\mathrm{Gray}}^{\′}}{{\∂y}_h}\mathrm{Vv}=-\frac{{\∂\mathrm{Gray}}^{\′}}{\∂\mathrm{Vt}}$ Formula 47

Suppose angle point (x _h, y _h) and neighborhood Ω ₄in side-play amount a little consistent, neighborhood Ω ₄radius is set to r ₄, by angle point (x _h, y _h) and neighborhood Ω ₄a little ((2r is total in institute ₄+ 1) ²individual) bring optical flow constraint equation into, the equation number exceeding unknown number number (Vu, Vv two) can be obtained, by iterative, certain angle point (x can be obtained _h, y _h) side-play amount (Vu, Vv), then displacement size is:

$\mathrm{Vw}=\sqrt{{\mathrm{Vu}}^2+{\mathrm{Vv}}^2}$ Formula 48

Carry out iterative to all angle points detected, asking for average displacement is:

$w=\frac{\underset{i=1}{\overset{h_3}{\mathrm{\Σ}}}{\mathrm{Vw}}_i}{h_3}$ Formula 49

Adjacent two two field picture interval times are Vt, then define foam velocity characteristic to be:

$\mathrm{Rate}=\frac{w}{\mathrm{Vt}}$ Formula 50

In the present embodiment, screening through three angle points the angle point point set Hs determined ₃in get angle point (x _h, y _h), h=1,2 ..., h ₃, its neighborhood Ω ₄radius of neighbourhood r ₄=1, i ₄∈ [x-r ₄, x+r ₄], j ₄∈ [y-r ₄, y+r ₄], obtain point set Hs by formula 46 ~ 49 ₃the displacement Vw of interior each point and average displacement w a little; Adjacent two two field picture interval times are Vt=0,33s, obtain foam velocity characteristic Rate by formula 50.

In tracing process, needing continuous deletion not need the angle point followed the trail of, the angle point with shifting out sight line as little especially in displacement Vw, when deleting a part of angle point, and when residue angle point number is less than 50, need again detects angle point by abovementioned steps and following the trail of.

Step 5: extract percentage of damage feature;

Step 3 is obtained matrix of depths D ', with a two field picture N number of foam regions label information point set Ps _none_to_one corresponding, obtains the set PD of depth value corresponding to each pixel of the n-th foam regions _n: (d ₁, d ₂, L d _{area (n)}), obtain the minimum value d of depth value _min, obtain the height PH of each pixel of foam regions _n(d ₁-d _min, d ₂-d _min, L d _{area (n)}-d _min), ask for height and PH by formula 51 _nsum.

PH _nsum=Σ PH _nformula 51

For a rear two field picture, read and previous frame data point set Ps _ncorresponding depth data PD _n' (d ₁', d ₂', L d ' _{area (n)}), obtain the height PH of each pixel of foam regions _n' (d ₁'-d _min, d ₂'-d _min, L d ' _{area (n)}-d _min), ask for height and PH ' by formula 52 _nsum.

PH ' _nsum=Σ PH _n' formula 52

Work as PH _nsum'/PH _nsumduring < 0.75, be considered as the bubble collapse that label is n, in like manner all N number of regions processed, count on a total BS bubble collapse, extracted the percentage of damage feature Break of foam by formula 53.

$\mathrm{Break}=\frac{\mathrm{BS}}{N}$ Formula 53

Step 4: adopt the depth data that distribution fitting method obtains step 2, carry out statistical study, method for parameter estimation is adopted to obtain the distribution of foam top layer height, and then the relation obtained between foam top layer height and overflow groove edge height, for on-line monitoring and the operating mode's switch of froth level; Comprise following sub-step:

Step 1: the matrix of depths D ' obtained step 2, extracts foam top layer liquid level feature by formula 55 ~ 60: average level L _a, liquid level smoothness L _u, minimum liquid level L _l, the highest liquid level L _h.

As shown in Figure 1, for foam top layer point (x, y) in kinect visual field, meet:

H _h=H _t+ H _kformula 54

Wherein H _kfor mineral pulp level (supposing that mineral syrup liquid is parallel with flotation cell bottom surface), H _trepresent certain point (x, y) place froth bed thickness (in visual field, certain point is to the distance of mineral syrup liquid).H _kbe an important indicator of flotation site operating mode's switch, but be difficult to because mineral syrup liquid is covered by froth bed measure, for this reason, by the foam top layer distance bottom land distance H a little of institute in visual field _h=D ' (x, y), obtains the foam top layer liquid level feature relevant to mineral pulp level, carries out operating mode's switch.

Use a large amount of off-line data to carry out fitting of distribution to the data in matrix of depths D ', obtain the result of its Normal Distribution.The point estimation method is adopted to estimate the parameter of normal distribution:

The average of normal distribution: $\mathrm{\&mu;}=\frac{\underset{(x,y)=\left(\mathrm{1,1}\right)}{\overset{(a,b)}{\mathrm{\&Sigma;}}}{D}^{\&prime;}(x,y)}{a\&times;b}$ Formula 55

The standard deviation of normal distribution: $\mathrm{\&sigma;}=\sqrt{\frac{\underset{(x,y)=\left(\mathrm{1,1}\right)}{\overset{(a,b)}{\mathrm{\&Sigma;}}}{(D^{\&prime;}(x,y)-\mathrm{\&mu;})}^2}{a\&times;b}}$ Formula 56

Define 4 foam top layer liquid level features:

Average level: L _a=μ formula 57

Liquid level smoothness: L _u=σ formula 58

Minimum liquid level: L _l=u-3 σ formula 59

The highest liquid level: L _h=u+3 σ formula 60

Step 2: by statistical study, obtains foam top layer liquid level feature and overflow groove edge height H _ybetween relation, H _y=3m; Judge current level operating mode by formula 61, when this classification operating mode continues more than 10 minutes, judge that recognition result comes into force.

Step 5: foam characteristics constitutive characteristic vector step 3 extracted, adopts the k-means algorithm improved to carry out the cluster analysis of off-line, obtain several cluster centres; The foam characteristics of extract real-time flotation site, and carry out real-time matching with cluster centre, the online operating mode obtaining current flotation.

Step 5 comprises following sub-step:

Step 1: 8 foam characteristics constitutive characteristic vector d that step 3 is extracted _s:

Ds=[Graym Rm Gm Bm Aream Vm Rate Break] formula 62

Step 2: off-line obtains cluster centre;

By the monitoring record of a period of time, sampling g group obtains characteristic data set:

$\mathrm{DS}=\left[\begin{array}{c}{\mathrm{ds}}_1\\ {\mathrm{ds}}_2\\ .\\ .\\ .\\ {\mathrm{ds}}_g\end{array}\right]=\left[\begin{array}{cccccccc}{\mathrm{Graym}}_1& {\mathrm{Rm}}_1& {\mathrm{Gm}}_1& {\mathrm{Bm}}_1& {\mathrm{Aream}}_1& {\mathrm{Vm}}_1& {\mathrm{Rate}}_1& {\mathrm{Break}}_1\\ {\mathrm{Graym}}_2& {\mathrm{Rm}}_2& {\mathrm{Gm}}_2& {\mathrm{Bm}}_2& {\mathrm{Aream}}_2& {\mathrm{Vm}}_2& {\mathrm{Rate}}_2& {\mathrm{Break}}_2\\ & & & & .& & & \\ & & & & .& & & \\ & & & & .& & & \\ {\mathrm{Graym}}_g& {\mathrm{Rm}}_g& {\mathrm{Gm}}_g& {\mathrm{Bm}}_g& {\mathrm{Aream}}_g& {\mathrm{Vm}}_g& {\mathrm{Rate}}_g& {\mathrm{Break}}_g\end{array}\right]$ Formula 63

Traversal cluster numbers k value, and k ∈ (1,2 ..., g), for each k value, carry out K-means cluster, definition evaluation function, determines preferable clustering number kb, obtains kb cluster centre: [cc ₁, cc ₂... cc _kb], form cluster centre set KB:

$\mathrm{KB}=\left[\begin{array}{c}{\mathrm{cc}}_1\\ {\mathrm{cc}}_2\\ .\\ .\\ .\\ {\mathrm{cc}}_{\mathrm{kb}}\end{array}\right]=\left[\begin{array}{cccccccc}{\mathrm{Graym}}_{c_1}& {\mathrm{Rm}}_{c_1}& {\mathrm{Gm}}_{c_1}& {\mathrm{Bm}}_{c_1}& {\mathrm{Aream}}_{c_1}& {\mathrm{Vm}}_{c_1}& {\mathrm{Rate}}_{c_1}& {\mathrm{Break}}_{c_1}\\ {\mathrm{Graym}}_{c_2}& {\mathrm{Rm}}_{c_2}& {\mathrm{Gm}}_{c_2}& {\mathrm{Bm}}_{c_2}& {\mathrm{Aream}}_{c_2}& {\mathrm{Vm}}_{c_2}& {\mathrm{Rate}}_{c_2}& {\mathrm{Break}}_{c_2}\\ & & & & .& & & \\ & & & & .& & & \\ & & & & .& & & \\ {\mathrm{Graym}}_{c_{\mathrm{kb}}}& {\mathrm{Rm}}_{c_{\mathrm{kb}}}& {\mathrm{Gm}}_{c_{\mathrm{kb}}}& {\mathrm{Bm}}_{c_{\mathrm{kb}}}& {\mathrm{Aream}}_{c_{\mathrm{kb}}}& {\mathrm{Vm}}_{c_{\mathrm{kb}}}& {\mathrm{Rate}}_{c_{\mathrm{kb}}}& {\mathrm{Break}}_{c_{\mathrm{kb}}}\end{array}\right]$

Formula 64

Step 3: online operating mode's switch.

According to the froth flotation working-condition monitoring system based on depth information that step one builds, Real-time Obtaining froth flotation color and depth data, the real-time froth images structural feature proper vector by step 3 is extracted:

Cc _current=[Graym _currentrm _currentgm _currentbm _currentaream _currentvm _currentrate _currentbreak _current] formula 65

Euclidean distance between definition real-time characteristic and cluster centre feature is characteristic similarity:

$\mathrm{KN}=\left[\begin{array}{c}|{\mathrm{cc}}_1-{\mathrm{cc}}_{\mathrm{current}}|\\ |{\mathrm{cc}}_2-{\mathrm{cc}}_{\mathrm{current}}|\\ .\\ .\\ .\\ |{\mathrm{cc}}_{\mathrm{kb}}-{\mathrm{cc}}_{\mathrm{current}}|\end{array}\right]$ Formula 66

Wherein | cc _kb-cc _current| represent vector (cc _kb-cc _current) mould, that is:

$|{\mathrm{cc}}_{\mathrm{kb}}-{\mathrm{cc}}_{\mathrm{current}}|=\sqrt{{({\mathrm{Graym}}_{c_{\mathrm{kb}}}-{\mathrm{Graym}}_{\mathrm{current}})}^2+{({\mathrm{Rm}}_{c_{\mathrm{kb}}}-{\mathrm{Rm}}_{\mathrm{current}})}^2+...+{({\mathrm{Break}}_{c_{\mathrm{kb}}}-{\mathrm{Break}}_{\mathrm{current}})}^2}$ Formula 67

A large amount of off-line data is obtained to the cluster centre of optimum number by formula 62 ~ 64; According to the froth flotation working-condition monitoring system based on kinect that step one builds, Real-time Obtaining froth flotation color and depth data, extract real-time froth images structural feature proper vector cc by step 3 _current, by formula 66 using with the highest cluster centre of real-time characteristic similarity as current working, complete operating mode.

Claims (8)

1., based on froth flotation level monitoring and the operating mode's switch method of depth information, it is characterized in that, comprise the following steps:

Step one: build the froth flotation working-condition monitoring system based on kinect from software and hardware aspect;

Step 2: gather, process and preserve the flotation froth color data and depth data that are obtained by kinect;

Step 3: the color obtain step 2 and depth data carry out feature extraction, color combining and depth data extract the stereoscopic features such as color, area, volume, speed, percentage of damage of foam;

Step 4: adopt distribution fitting method to carry out statistical study to the depth data that step 2 obtains; Method for parameter estimation is adopted to obtain foam top layer liquid level feature; And then the relation obtained between foam top layer liquid level feature and overflow groove edge height, for on-line monitoring and the operating mode's switch of froth flotation liquid level;

Step 5: foam characteristics constitutive characteristic vector step 3 obtained, adopts the k-means algorithm improved to carry out the cluster analysis of off-line, obtain several cluster centres; The foam characteristics of extract real-time flotation site, and carry out real-time matching with cluster centre, the online operating mode obtaining current flotation.

2. the froth flotation level monitoring based on depth information according to claim 1 and operating mode's switch method, it is characterized in that, described step one comprises following sub-step:

Step 1: build the froth flotation working-condition monitoring system based on kinect from hardware aspect, hardware comprises kinect sensor, high frequency light source, computing machine, be fixed on by kinect sensor apart within the scope of flotation cell liquid level 1.2m-3.5m, its camera place plane is parallel with flotation cell bottom surface; High frequency light source provides illumination for kinect sensor gathers color data, the color collected and depth data is flowed through USB data line and is sent to computing machine, and computing machine realizes data processing, real-time working condition monitoring and result display thereof;

Step 2: build the froth flotation working-condition monitoring system based on kinect from software aspect, under cross-platform framework Openni, adopt interactive programming interface, to realize the Real-time Obtaining of color and depth data stream, process, preservation and feature extraction and operating mode's switch.

3. the froth flotation level monitoring based on depth information according to claim 1 and operating mode's switch method, it is characterized in that, described step 2 comprises following sub-step:

Step 1: image data;

(1) setting data form;

Color data stream is set as RGB888, and resolution is the pixel quantity that pixel quantity that a × b, a × b represents in horizontal direction is multiplied by vertical direction, and frame per second is 30FPS; Depth data stream format is set as PIXEL_FORMAT_DEPTH_1_MM form, and data precision is millimeter, and resolution is a × b; Frame per second is 30FPS;

(2) hardware aims at color data and depth data;

Adopt " Alternative View " instrument under cross-platform framework Openni, the visual angle of correction of color camera and depth camera, to realize color data stream and depth data stream aiming on field of regard;

(3) software aims at color data and depth data;

Start kinect equipment and carry out image acquisition; Adopt the data layout of above-mentioned steps setting to create color data stream and depth data stream simultaneously, obtain two kinds of data that sequential is aimed at: rank, a × b × 3 and a × b rank matrix of depths D, color matrix C are by representing that 3 a × b rank matrixes of image R, G, B color component form respectively;

Step 2: process data;

(1) depth information in depth data is obtained;

The matrix of depths D obtained by kinect sensor is 16 integer data, its high 13 store be depth information, and low 3 store be index information, obtain depth information by shifting function:

D ₁=D/2 ³formula 1

D in formula ₁for only storing the matrix of depths of depth information;

(2) color combining data carry out filtering completion process to depth data;

Adopt associating bilateral filtering method to matrix of depths D ₁carry out filtering process, by the depth image completion owing to blocking disappearance, obtain filtered matrix of depths D ₂, it is at the depth value D at point (x, y) place ₂(x, y) is:

$D_2(x,y)=\frac{1}{w_p}\underset{(i_1,j_1)\&Element;{\mathrm{\&Omega;}}_1}{\mathrm{\&Sigma;}}w(i_1,j_1)\&times;D_1(i_1,j_1)$ Formula 2

Ω in formula ₁for the filtered reference neighborhood of required point (x, y), x ∈ [1, a], y ∈ [1, b], (i ₁, j ₁) ∈ Ω ₁, i ₁∈ [x-r ₁, x+r ₁], j ₁∈ [y-r ₁, y+r ₁], r ₁for the radius of neighbourhood, D ₁(i ₁, j ₁) be matrix D ₁at point (i ₁, j ₁) depth value at place, w _pfor normalized parameter:

$w_p=\underset{i_1,j_1\&Element;{\mathrm{\&Omega;}}_1}{\mathrm{\&Sigma;}}w(i_1,j_1)$ Formula 3

W (i in formula ₁, j ₁) be the neighborhood Ω of required point (x, y) ₁interior each point (i ₁, j ₁) for the weights of required point, by depth image spatial domain weight w _swith coloured image gray scale territory weight w _rcomposition, that is:

W (i ₁, j ₁)=w _s(i ₁, j ₁) × w _r(i ₁, j ₁) formula 4

Wherein:

$w_s(i_1,j_1)=\exp (-\frac{{(i_1-x)}^2+{(j_1-y)}^2}{2{{\mathrm{\&sigma;}}_s}^2})$ Formula 5

$w_r(i_1,j_1)=\exp (-\frac{{(\mathrm{Gray}(i_1,j_1)-\mathrm{Gray}(x,y))}^2}{2{{\mathrm{\&sigma;}}_r}^2})$ Formula 6

Wherein, σ _rand σ _sbe respectively weight w _sand w _rcorresponding Gaussian function standard deviation, Gray (x, y), Gray (i ₁, j ₁) be respectively pixel (x, y) and neighborhood Ω thereof ₁interior each point (i ₁, j ₁) gray-scale value at place, computing formula is:

Gray (x, y)=0.02989 × C (x, y, 1)+0.5870 × C (x, y, 2)+0.1149 × C (x, y, 3) formula 7

Wherein C (x, y, 1), C (x, y, 2), C (x, y, 3) be respectively the color value on a × b rank matrix mid point (x, y) of R, G, B color component in the color matrix C of rank, a × b × 3, traversal obtains a little gray matrix Gray;

Choose suitable Gaussian function standard deviation sigma _rand σ _s, at suitable filtered reference neighborhood Ω ₁in, effectively can eliminate depth information and the noise spot of the region disappearance that is blocked;

(3) reflection foam top layer is obtained apart from bottom land distance H _hmatrix of depths D ';

H _hrepresent that in visual field, foam top layer point (x, y), apart from the distance of bottom land, can be obtained by following formula:

H _h=H _b-H _dformula 8

Wherein H _bfor kinect sensor camera place plane is to the distance of flotation cell bottom land, H _dfor foam top layer point in visual field is to the distance of kinect sensor camera place plane, when kinect sensor camera place plane is parallel with flotation cell bottom surface, H _dnamely the matrix of depths D matrix of depths D after displacement and filtering completion for being obtained by kinect sensor ₂the value of middle corresponding point, i.e. H _d=D ₂(x, y),

Obtain reflection H _hcertain point (x, y) depth value D ' (x, y) be:

D ' (x, y)=H _b-D ₂(x, y) formula 9

Step 3: preserve data;

Color matrix C and matrix of depths D ' is stored, with Motion JPEGAVI coded system access color data, with Motion JPEG2000 (16) coded system access depth data.

4. the froth flotation level monitoring based on depth information according to claim 3 and operating mode's switch method, it is characterized in that, described step 3 comprises following sub-step:

Step 1: extract color characteristic;

The color obtain step 2 and depth data carry out feature extraction, and color combining and depth data extract the color characteristic of foam,

Suppose that the depth value scope of object in visual field is for [d _min, d _max], definition datum distance is:

Db=(d _max-d _min)/2 formula 10

Suppose that the color being positioned at reference range place point is standard value, then when intensity of illumination is identical, be greater than reference range from distance of camera lens more away from some color decay larger, need to strengthen its value, and be less than reference range from distance of camera lens more close to point, need to weaken its color value, for this reason, by setting weight function, carry out standardization in conjunction with depth data to color data, formula is:

Q (x, y)=(D'(x, y)-db) ³/ 10 ⁶+ 1 formula 11

Color characteristic after color value standardization is defined as:

R average: $R_m=\frac{\underset{x=1}{\overset{a}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{b}{\mathrm{\&Sigma;}}}C(x,y,1)\&CenterDot;q(x,y)}{a\&times;b}$ Formula 12

G average: $G_m=\frac{\underset{x=1}{\overset{a}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{b}{\mathrm{\&Sigma;}}}C(x,y,2)\&CenterDot;q(x,y)}{a\&times;b}$ Formula 13

B average: $B_m=\frac{\underset{x=1}{\overset{a}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{b}{\mathrm{\&Sigma;}}}C(x,y,3)\&CenterDot;q(x,y)}{a\&times;b}$ Formula 14

The gray matrix Gray obtained by step 2, calculates gray average:

${\mathrm{Gray}}_m=\frac{\underset{x=1}{\overset{a}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{b}{\mathrm{\&Sigma;}}}\mathrm{Gray}(x,y)\&CenterDot;q(x,y)}{a\&times;b}$ Formula 15

Step 2: extract area features;

The gray matrix Gray of step 2 acquisition is calculated to the intensity profile Gs of a two field picture, represents the number of the pixel meeting Gray (x, y)=g with Gs (g), wherein g ∈ [0,255],

From e=0, get e=0 respectively, e=1 ..., e=e ', e ' ∈ (0,255), when e=e ' time, meet:

$\frac{\underset{g\&Element;[o,e^{\&prime;}]\&cup;[e^{\&prime;},255]}{\mathrm{\&Sigma;}}\mathrm{Gs}\left(g\right)}{a\&times;b}>0.01$ Formula 16

Obtain the gray matrix Gray ' of contrast strengthen, its gray-scale value at point (x, y) place is:

${\mathrm{Gray}}^{\&prime;}(x,y)=\left\{\begin{array}{cc}0& \mathrm{Gray}(x,y)\&Element;(0,e^{\&prime;})\\ \frac{255}{255-2e}(\mathrm{Gray}(x,y)-e^{\&prime;})& \mathrm{Gray}(x,y)\&Element;(e,255-e^{\&prime;})\\ 255& \mathrm{Gray}(x,y)\&Element;(255-e^{\&prime;},255)\end{array}\right.$ Formula 17

Adopt watershed algorithm to split matrix Gray ', obtain foam regions two values matrix A, it is designated as A (x, y) at the gray-scale value at point (x, y) place, adopts formula below to carry out the differentiation of foam regions point and frontier point:

formula 18

Travel through the pixel that a two field picture is all, foam regions be numbered, obtain the label information of marked region:

Ps _n: { (x ₁, y ₁), (x ₂, y ₂), L, (x _{area (n)}, y _{area (n)}) formula 19

Ps in formula _nrepresent the set of all pixels of the n-th foam regions, n=1,2, L, N, N are foam regions sum; Area (n) represents the number of the pixel in the n-th foam regions, be region area, add up the area of all N number of foam regions respectively, obtain area distributions: Area:{Area (1), Area (2) ... Area (n) ... Area (N) }, calculate the overall area area of the N number of foam regions of a two field picture:

$\mathrm{SArea}=\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{Area}\left(n\right)$ Formula 20

N number of area value in area distributions Area is obtained Area ' according to order rearrangement from small to large: { Area ' (1), Area ' (2) L Area ' (n) L Area ' (N) }, wherein Area ' (1) is foam area minimum value, and Area ' (N) is foam Maximum Area;

From f ₁=1 starts, and gets f respectively ₁=1, f ₁=2 ..., f ₁=f ₁', f ₁' ∈ (1, N), works as f ₁=f ₁in ' time, meet:

$\frac{\underset{n\&Element;[1,{f_1}^{\&prime;}]}{\mathrm{\&Sigma;}}A{\mathrm{rea}}^{\&prime;}\left(n\right)}{\mathrm{SArea}}>0.2$ Formula 21

Then defining foam area features is:

${\mathrm{Area}}_m=\frac{\underset{n={f_1}^{\&prime;}+1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{Area}}^{\&prime;}\left(n\right)}{N-{f_1}^{\&prime;}-1}$ Formula 22

Step 3: extract volume characteristic;

For region two values matrix A and the zone marker Ps of a two field picture _n, (p+q) rank rule square of its n-th foam regions is defined as:

$M_n(p,q)=\underset{(x,y)\&Element;\mathrm{Ps}\left(n\right)}{\mathrm{\&Sigma;}}{x}^p{y}^{q}A(x,y)$ Formula 23

The region barycenter of the n-th foam regions can be tried to achieve by its 1 rank rule distance:

$A{C}_n:({\stackrel{\&OverBar;}{x}}_n,{\stackrel{\&OverBar;}{y}}_n)=(M_n\left(\mathrm{1,0}\right),M_n\left(\mathrm{0,1}\right))$ Formula 24

Then (p+q) rank centre distance of a two field picture is:

$U_n(p,q)=\underset{(x,y)\&Element;\mathrm{Ps}\left(n\right)}{\mathrm{\&Sigma;}}{(x-{\stackrel{\&OverBar;}{x}}_n)}^p{(y-{\stackrel{\&OverBar;}{y}}_n)}^{q}A(x,y)$ Formula 25

If only consider the second-order moment around mean collection of a two field picture, then foam regions can be approximately an ellipse centered by the barycenter of region, then have:

${\mathrm{CB}}_n={\left(\frac{U_n\left(\mathrm{2,0}\right)+U_n\left(\mathrm{0,2}\right)+{[{(U_n\left(\mathrm{2,0}\right)-U_n\left(\mathrm{0,2}\right))}^2+{4U}_n{\left(\mathrm{1,1}\right)}^2]}^{1/2}}{U_n\left(\mathrm{0,0}\right)/2}\right)}^{1/2}$ Formula 26

${\mathrm{DB}}_n={\left(\frac{U_n\left(\mathrm{2,0}\right)+U_n\left(\mathrm{0,2}\right)-{[{(U_n\left(\mathrm{2,0}\right)-U_n\left(\mathrm{0,2}\right))}^2+{4U}_n{\left(\mathrm{1,1}\right)}^2]}^{1/2}}{U_n\left(\mathrm{0,0}\right)/2}\right)}^{1/2}$ Formula 27

${\mathrm{\&theta;}}_n=\frac{1}{2}\arctan \left(\frac{{2U}_n\left(\mathrm{1,1}\right)}{U_n\left(\mathrm{2,0}\right)-U_n\left(\mathrm{0,2}\right)}\right)$ Formula 28

CB _nrepresent the major semi-axis of sub-elliptical, DB _nfor minor semi-axis, θ _nfor inclination angle;

And then calculate the pixel coordinate of oval long (short) axle head point: c ₁(u _n1, v _n1), c ₂(u _n2, v _n2), formula is:

$u_{n1}={\stackrel{\&OverBar;}{x}}_n-{\mathrm{CB}}_n\&times;\cos \left({\mathrm{\&theta;}}_n\right),v_{n1}={\stackrel{\&OverBar;}{y}}_n-{\mathrm{CB}}_n\&times;\sin \left({\mathrm{\&theta;}}_n\right)$ Formula 29

$u_{n2}={\stackrel{\&OverBar;}{x}}_n+{\mathrm{CB}}_n\&times;\cos \left({\mathrm{\&theta;}}_n\right),v_{n2}={\stackrel{\&OverBar;}{y}}_n+{\mathrm{CB}}_n\&times;\sin \left({\mathrm{\&theta;}}_n\right)$ Formula 30

The matrix of depths D ' that integrating step two obtains, calculates transverse end points and depth value l corresponding to region barycenter _n1, l _n2, l _nc:

l _n1＝D′(u _n1,v _n1)

L _n2=D ' (u _n2, v _n2) formula 31

$l_{\mathrm{nc}}=D^{\&prime;}({\stackrel{\&OverBar;}{x}}_n,{\stackrel{\&OverBar;}{y}}_n)$

Suppose that foam can be approximately the spheroid of standard, definition foam real space height h _nfor:

H _n=l _nc-min (l _n1, l _n2) formula 32

View parameter in conjunction with kinect sensor: horizontal view angle δ ₁with vertical angle of view δ ₂, calculate the three dimensional space coordinate C that major axis end points is corresponding _n1and C _n2:

C _n1: (δ ₁* [2u _n1l+ (a-1) (L-l _n1)], δ ₂[2 (b-v _n1-1) L+ (b-1) (L-l _n1)], L-l _n1) formula 33

C _n2: (δ ₁* [2u _n2l+ (a-1) (L-l _n2)], δ ₂[2 (b-v _n2-1) L+ (b-1) (L-l _n2)], L-l _n2) formula 34

Wherein L represents the distance of depth camera to reference image plane space coordinate z=0;

And then obtain major axis physical length:

${\mathrm{CR}}_n=C_{1n}{C}_{2n}=\sqrt{{{\mathrm{\&delta;}}_1}^2{[2L(u_{1n}-u_{2n})-(a-1)(l_{1n}-l_{2n})]}^2+{{\mathrm{\&delta;}}_2}^2{[2L(v_{1n}-v_{2n})+(b-1)(l_{1n}-l_{2n})]}^2+{(l_{1n}-l_{2n})}^2}$

Formula 35

Suppose that the depth value that major axis two end points are corresponding is equal, major axis actual (tube) length angle value can be reduced to:

${\mathrm{CR}}_n=2L\sqrt{{{\mathrm{\&delta;}}_1}^2{(u_{1n}-u_{2n})}^2+{{\mathrm{\&delta;}}_2}^2{(v_{1n}-v_{2n})}^2}$ Formula 36

In like manner can obtain minor axis physical length DR _n;

Finally can obtain the volume (supposing that foam is the spheroid of standard) of the n-th foam:

$V\left(n\right)=\frac{4}{3}\mathrm{\&pi;}(\frac{{\mathrm{CR}}_n}{2}\&times;\frac{{\mathrm{DR}}_n}{2}\&times;h_n)$ Formula 37

The thinking of a two field picture area features is obtained according to sub-step in step 32, by N number of Domain Volume Distribution value: Vs:{Vs (1), Vs (2) ... Vs (n) ... Vs (N) }, Vs ' is obtained: { Vs ' (1), Vs ' (2) according to order rearrangement from small to large ... Vs ' (n) ... Vs ' (N) };

From f ₂=1 starts, and gets f respectively ₂=1, f ₂=2 ..., f ₂=f ' ₂, f ' ₂∈ (1, N), works as f ₂=f ' ₂time, meet:

$\frac{\underset{n\&Element;[1,f_2^{\&prime;}]}{\mathrm{\&Sigma;}}{V}^{\&prime;}s\left(n\right)}{\underset{n\&Element;[1,N]}{\mathrm{\&Sigma;}}{V}^{\&prime;}s\left(n\right)}>0.2$ Formula 38

Then define foam volume to be characterized as:

$V_m=\frac{\underset{n=f_2^{\&prime;}+1}{\overset{N}{\mathrm{\&Sigma;}}}{V}^{\&prime;}s\left(n\right)}{N-f_2^{\&prime;}-1}$ Formula 39

Step 4: extraction rate feature;

(1) Harris Corner Detection;

Suppose that certain point (x, y) strengthened in gray matrix Gray ' is angle point, its neighborhood Ω ₂interior along the mean intensity gray-value variation in certain side-play amount (Vx, Vy) direction be a little:

$I=\underset{(i_2,j_2)\&Element;{\mathrm{\&Omega;}}_2}{\mathrm{\&Sigma;}}{({\mathrm{Gray}}^{\&prime;}(i_2+\mathrm{Vx},j_2+\mathrm{Vy})-{\mathrm{Gray}}^{\&prime;}(i_2,j_2))}^2$ Formula 40

I in formula ₂∈ [x-r ₂, x+r ₂], j ₂∈ [y-r ₂, y+r ₂], r ₂for Ω ₂the radius of neighbourhood;

The calculating of Strength Changes that all side-play amount directions of angle point (x, y) are averaged, Taylor series expansion is carried out to formula 40, omits higher order term, be described as with matrix form:

$I\&ap;\left[\begin{array}{cc}\mathrm{Vx}& \mathrm{Vy}\end{array}\right]\left[\begin{array}{cc}\mathrm{\&Sigma;}{\left(\frac{\mathrm{\&delta;}{\mathrm{Gray}}^{\&prime;}}{{\mathrm{\&delta;h}}_x}\right)}^2& \mathrm{\&Sigma;}\frac{{\mathrm{\&delta;Gray}}^{\&prime;}}{{\mathrm{\&delta;h}}_x}\frac{{\mathrm{\&delta;Gray}}^{\&prime;}}{{\mathrm{\&delta;h}}_y}\\ \mathrm{\&Sigma;}\frac{{\mathrm{\&delta;Gray}}^{\&prime;}}{{\mathrm{\&delta;h}}_x}\frac{{\mathrm{\&delta;Gray}}^{\&prime;}}{{\mathrm{\&delta;h}}_y}& \mathrm{\&Sigma;}{\left(\frac{{\mathrm{\&delta;Gray}}^{\&prime;}}{{\mathrm{\&delta;h}}_y}\right)}^2\end{array}\right]\left[\begin{array}{c}\mathrm{Vx}\\ \mathrm{Vy}\end{array}\right]=\left[\begin{array}{cc}\mathrm{Vx}& \mathrm{Vy}\end{array}\right]Q\left[\begin{array}{c}\mathrm{Vx}\\ \mathrm{Vy}\end{array}\right]$ Formula 41

In formula, Q is covariance matrix, and two eigenwert represents that certain puts the mean intensity change of maximum mean intensity change and vertical direction thereof in all side-play amount directions, and definition scoring functions is:

G=Det (Q)-kTrace ²(Q) formula 42

Wherein Det (Q) is eigenwert sum, i.e. determinant of a matrix value, and Trace (Q) amasss for eigenwert is, i.e. matrix trace; When two eigenwerts are all larger, fractional value is higher;

Definition G _rfor mark up to standard, traversal angle point (x, y) neighborhood Ω ₂interior a little, obtain the h of fractional value higher than mark up to standard ₁individual, sort from high to low by fractional value and obtain point set

Carry out the most value process in gray-scale value local to enhancing gray matrix Gray', formula is:

Graymax (x, y)=max (Gray ' (i ₃, j ₃)), (i ₃, j ₃) ∈ Ω ₃formula 43

Graymin(x,y)＝min(Gray′(i ₃,j ₃)),(i ₃,j ₃)∈Ω ₃

Wherein Graymax (x, y) is the value at gray matrix Graymax mid point (x, the y) place after gray-scale value suboptimize, Graymin (x, y) be the value at gray matrix Graymin mid point (x, y) place after gray-scale value local minimum, Ω ₃for the neighborhood of point (x, y), its radius of neighbourhood is set to r ₃, and i ₃∈ [x-r ₃, x+r ₃], j ₃∈ [y-r ₃, y+r ₃],

Following formula determination gray-scale value local is adopted to be worth most a little:

formula 44

formula 45

From point set Hs ₁in filter out the point meeting local gray level and be worth condition most, obtain by h ₂the point set of individual composition ${\mathrm{Hs}}_2:\{(x_1,y_1),(x_2,y_2)...(x_{h_2},y_{h_2})\};$

Finally, to matrix of depths D' carry out with ask gray matrix Gray' gray-scale value local value order the same operation, obtain depth value and be locally worth most a little, from point set Hs ₂in filter out the point meeting partial-depth and be worth condition most, obtain by h ₃individual composition angle point point set

(2) screen to through three angle points the angle point point set Hs determined ₃in point follow the trail of, extraction rate feature;

Suppose the same angle point (x in the two field picture of front and back two _h, y _h) intensity constant, that is:

Gray' _t(x _h, y _h)=Gray' _t+Vt(x _h+ Vu, y _h+ Vv) formula 46

Wherein Gray' _t(x _h, y _h) be that t (previous frame image) strengthens gray matrix Gray' mid point (x _h, y _h) gray-scale value, Gray' _t+Vt(x _h+ Vu, y _h+ Vv) for being the t+Vt moment (a rear two field picture) strengthen gray matrix Gray' mid point (x _h+ Vu, y _h+ Vv) gray-scale value, (Vu, Vv) is angle point (x _h, y _h) side-play amount that occurs within the Vt time, Vt is adjacent two frame time intervals, h=1,2, L, h ₃;

Then there is basic light stream constraint equation:

$\frac{{\&PartialD;\mathrm{Gray}}^{\&prime;}}{{\&PartialD;x}_h}\mathrm{Vu}+\frac{{\&PartialD;\mathrm{Gray}}^{\&prime;}}{{\&PartialD;y}_h}\mathrm{Vv}=-\frac{{\&PartialD;\mathrm{Gray}}^{\&prime;}}{\&PartialD;\mathrm{Vt}}$ Formula 47

Suppose angle point (x _h, y _h) and neighborhood Ω ₄in side-play amount a little consistent, neighborhood Ω ₄radius is set to r ₄, by angle point (x _h, y _h) and neighborhood Ω ₄a little (2r is total in institute ₄+ 1) ²individually bring optical flow constraint equation into, obtain the equation number exceeding unknown number number (Vu, Vv two), by iterative, certain angle point (x can be obtained _h, y _h) side-play amount (Vu, Vv), then displacement size is:

$\mathrm{Vw}=\sqrt{{\mathrm{Vu}}^2+{\mathrm{Vv}}^2}$ Formula 48

Carry out iterative to all angle points detected, asking for average displacement is:

$w=\frac{\underset{i=1}{\overset{h_3}{\mathrm{\&Sigma;}}}V{w}_i}{h_3}$ Formula 49

Adjacent two two field picture interval times are V t, then define foam velocity characteristic to be:

$\mathrm{Rate}=\frac{w}{\mathrm{Vt}}$ Formula 50

In tracing process, need continuous deletion not need the angle point followed the trail of, the angle point with shifting out sight line as little especially in displacement Vw, when deleting a part of angle point, and when residue angle point number is less than certain predefined threshold value, need again detects angle point by abovementioned steps and follow the trail of;

Step 5: extract percentage of damage feature;

Step 3 is obtained matrix of depths D ', with a two field picture zone marker information point set Ps _none_to_one corresponding, obtains the set PD of depth value corresponding to each pixel of the n-th foam regions _n: (d ₁, d ₂, L d _{area (n)}), obtain the minimum value d of depth value _min, draw the height PH of each pixel of foam regions _n(d ₁-d _min, d ₂-d _min, L d _{area (n)}-d _min), ask for height and:

PH _nsum=Σ PH _nformula 51

For a rear two field picture, read and previous frame data point set Ps _ncorresponding depth data PD ' _n(d ' ₁, d ' ₂, L d ' _{area (n)}), obtain the height PH ' of each pixel of foam regions _n(d ' ₁-d _min, d ' ₂-d _min, L d ' _{area (n)}-d _min), ask for height and:

PH ' _nsum=Σ PH ' _nformula 52

Work as PH _nsum'/PH _nsumduring < 0.75, be considered as the bubble collapse that label is n, in like manner process all N number of regions, count on a total BS bubble collapse, then bubble collapse rate is:

$\mathrm{Break}=\frac{\mathrm{BS}}{N}$ Formula 53.

5. the froth flotation level monitoring based on depth information according to claim 4 and operating mode's switch method, it is characterized in that, described step 4 comprises following sub-step:

Step 1: the matrix of depths D ' obtained step 2, extracts foam top layer liquid level feature: average level L _a, liquid level smoothness L _u, minimum liquid level L _l, the highest liquid level L _h;

For foam top layer point (x, y) in kinect visual field, meet:

H _h=H _t+ H _kformula 54

Wherein H _kfor mineral pulp level, suppose that mineral syrup liquid is parallel with flotation cell bottom surface, H _trepresent certain point (x, y) place froth bed thickness, namely in visual field, certain point arrives the distance of mineral syrup liquid, H _kbe an important indicator of flotation site operating mode's switch, but be difficult to because mineral syrup liquid is covered by froth bed measure, for this reason, by the foam top layer distance bottom land distance H a little of institute in visual field _h=D ' (x, y), obtains the foam top layer liquid level feature relevant to mineral pulp level, carries out operating mode's switch;

Use a large amount of off-line data to carry out fitting of distribution to the data in matrix of depths D ', obtain the result of its Normal Distribution, adopt the point estimation method to estimate the parameter of normal distribution:

The average of normal distribution: $\mathrm{\&mu;}=\frac{\underset{(x,y)=\left(\mathrm{1,1}\right)}{\overset{(a,b)}{\mathrm{\&Sigma;}}}{D}^{\&prime;}(x,y)}{a\&times;b}$ Formula 55

The standard deviation of normal distribution: $\mathrm{\&sigma;}=\sqrt{\frac{\underset{(x,y)=\left(\mathrm{1,1}\right)}{\overset{(a,b)}{\mathrm{\&Sigma;}}}{(D^{\&prime;}(x,y)-\mathrm{\&mu;})}^2}{a\&times;b}}$ Formula 56

Define 4 foam top layer liquid level features:

Average level: L _a=μ formula 57

Liquid level smoothness: L _u=σ formula 58

Minimum liquid level: L _l=u-3 σ formula 59

The highest liquid level: L _h=u+3 σ formula 60

Step 2: by statistical study, obtains foam top layer liquid level feature and overflow groove edge height H _ybetween relation:

formula 61

Step 3: extract real-time current foam top layer liquid level feature, application of formula 61 judges, obtains the operating mode classification that current foam top layer liquid level feature reflects; When this classification operating mode continues more than 10 minutes, judge that recognition result comes into force.

6. the froth flotation level monitoring based on kinect according to claim 5 and operating mode's switch method, it is characterized in that, described step 5 comprises following sub-step:

Step 1: 8 foam characteristics constitutive characteristic vector d that step 3 is extracted _s:

Ds=[Graym Rm Gm Bm Aream Vm Rate Break] formula 62

Step 2: off-line obtains cluster centre;

By the monitoring record of a period of time, sampling g group obtains characteristic data set:

$\mathrm{DS}=\left[\begin{array}{c}{\mathrm{ds}}_1\\ {\mathrm{ds}}_2\\ .\\ .\\ .\\ {\mathrm{ds}}_g\end{array}\right]=\left[\begin{array}{cccccccc}{\mathrm{Graym}}_1& {\mathrm{Rm}}_1& {\mathrm{Gm}}_1& {\mathrm{Bm}}_1& {\mathrm{Aream}}_1& {\mathrm{Vm}}_1& {\mathrm{Rate}}_1& {\mathrm{Break}}_1\\ {\mathrm{Graym}}_2& {\mathrm{Rm}}_2& {\mathrm{Gm}}_2& {\mathrm{Bm}}_2& {\mathrm{Aream}}_2& {\mathrm{Vm}}_2& {\mathrm{Rate}}_2& {\mathrm{Break}}_2\\ & & & & .& & & \\ & & & & .& & & \\ & & & & .& & & \\ {\mathrm{Graym}}_g& {\mathrm{Rm}}_g& {\mathrm{Gm}}_g& {\mathrm{Bm}}_g& {\mathrm{Aream}}_g& {\mathrm{Vm}}_g& {\mathrm{Rate}}_g& {\mathrm{Break}}_g\end{array}\right]$ Formula 63

Traversal cluster numbers k value, k ∈ (1,2 ..., g), for each k value, carry out K-means cluster, definition evaluation function, determines preferable clustering number kb, obtains kb cluster centre: [cc ₁, cc ₂... cc _kb], form cluster centre set KB:

$\mathrm{KB}=\left[\begin{array}{c}{\mathrm{cc}}_1\\ {\mathrm{cc}}_2\\ .\\ .\\ .\\ {\mathrm{cc}}_{\mathrm{kb}}\end{array}\right]=\left[\begin{array}{cccccccc}{\mathrm{Graym}}_{c_1}& {\mathrm{Rm}}_{c_1}& {\mathrm{Gm}}_{c_1}& {\mathrm{Bm}}_{c_1}& {\mathrm{Aream}}_{c_1}& {\mathrm{Vm}}_{c_1}& {\mathrm{Rate}}_{c_1}& {\mathrm{Break}}_{c_1}\\ {\mathrm{Graym}}_{c_2}& {\mathrm{Rm}}_{c_2}& {\mathrm{Gm}}_{c_2}& {\mathrm{Bm}}_{c_2}& {\mathrm{Aream}}_{c_2}& {\mathrm{Vm}}_{c_2}& {\mathrm{Rate}}_{c_2}& {\mathrm{Break}}_{c_2}\\ & & & & .& & & \\ & & & & .& & & \\ & & & & .& & & \\ {\mathrm{Graym}}_{c_{\mathrm{kb}}}& {\mathrm{Rm}}_{c_{\mathrm{kb}}}& {\mathrm{Gm}}_{c_{\mathrm{kb}}}& {\mathrm{Bm}}_{c_{\mathrm{kb}}}& {\mathrm{Aream}}_{c_{\mathrm{kb}}}& {\mathrm{Vm}}_{c_{\mathrm{kb}}}& {\mathrm{Rate}}_{c_{\mathrm{kb}}}& {\mathrm{Break}}_{c_{\mathrm{kb}}}\end{array}\right]$

Formula 64

Step 3: online operating mode's switch;

According to the froth flotation working-condition monitoring system based on kinect that step one builds, Real-time Obtaining froth flotation color and depth data, the real-time froth images structural feature proper vector by step 3 is extracted:

Cc _current=[Graym _currentrm _currentgm _currentbm _currentaream _currentvm _currentrate _currentbreak _current] formula 65

Euclidean distance between definition real-time characteristic and cluster centre feature is characteristic similarity:

$\mathrm{KN}=\left[\begin{array}{c}|{\mathrm{cc}}_1-{\mathrm{cc}}_{\mathrm{current}}|\\ |{\mathrm{cc}}_2-{\mathrm{cc}}_{\mathrm{current}}|\\ .\\ .\\ .\\ |{\mathrm{cc}}_{\mathrm{kb}}-{\mathrm{cc}}_{\mathrm{current}}|\end{array}\right]$ Formula 66

Wherein | cc _kb-cc _current| represent vector (cc _kb-cc _current) mould, that is:

$|{\mathrm{cc}}_{\mathrm{kb}}-{\mathrm{cc}}_{\mathrm{current}}|=\sqrt{{({\mathrm{Graym}}_{c_{\mathrm{kb}}}-{\mathrm{Graym}}_{\mathrm{current}})}^2+{({\mathrm{Rm}}_{c_{\mathrm{kb}}}-{\mathrm{Rm}}_{\mathrm{current}})}^2+...+{({\mathrm{Break}}_{c_{\mathrm{kb}}}-{\mathrm{Break}}_{\mathrm{current}})}^2}$ Formula 67

Choose operating mode corresponding to the minimum value of KN matrix as current working, complete operating mode's switch.

7. the froth flotation level monitoring based on depth information and working condition recognition system, it is characterized in that, comprise kinect sensor, high frequency light source, computing machine, kinect sensor is fixed on apart within the scope of flotation cell liquid level 1.2m-3.5m, and its camera place plane is parallel with flotation cell bottom surface; High frequency light source provides illumination for kinect sensor gathers color data, the color collected and depth data is flowed through USB data line and is sent to computing machine, and computing machine realizes data processing, real-time working condition monitoring and result display thereof; Software systems, under cross-platform framework Openni, adopt interactive programming interface, to realize the Real-time Obtaining of color and depth data stream, process, preservation and feature extraction and operating mode's switch.

8. the froth flotation level monitoring based on depth information according to claim 7 and working condition recognition system, it is characterized in that, the mode of the Real-time Obtaining of color and depth data stream, process, preservation and feature extraction and operating mode's switch is: gather, process and preserve the flotation froth color data and depth data that are obtained by kinect; Carry out feature extraction to the color obtained and depth data, color combining and depth data extract the stereoscopic features such as color, area, volume, speed, percentage of damage of foam; Distribution fitting method is adopted to carry out statistical study to the depth data obtained; Method for parameter estimation is adopted to obtain foam top layer liquid level feature; And then the relation obtained between foam top layer liquid level feature and overflow groove edge height, for on-line monitoring and the operating mode's switch of froth flotation liquid level; By the foam characteristics constitutive characteristic vector obtained, adopt the k-means algorithm improved to carry out the cluster analysis of off-line, obtain several cluster centres; The foam characteristics of extract real-time flotation site, and carry out real-time matching with cluster centre, the online operating mode obtaining current flotation.