CN104320881B - A kind of intelligent dimming controller in LED shadowless lamps illuminator - Google Patents

Abstract

The invention discloses the intelligent dimming controller in a kind of LED shadowless lamps illuminator,The image of surgical field of view is carried out by signature analysis and treatment using blurred vision image processing techniques,Extract brightness and the chrominance information of lighting environment,The current control parameter of LED light source is calculated by Intelligent treatment algorithm,The image of surgical field of view is converted into data image signal first,By video image acquisition module by the signal-obtaining to internal system,Parsed by computer assisted image processing module again,Extract illumination and Colour information,Corresponding illumination and such chroma feature vectors are extracted by illumination and chromaticity extraction module,The shadow region that shelter is produced is split through blurred picture segmentation module,The driving current value of corresponding LED light source is calculated by fuzzy neural network again,It is sent to the brightness that LED controller adjusts corresponding LED,Realize eliminating shade,Uniform-illumination and the purpose of permanent photocontrol,Realize control with an automatic light meter,It is not required to manual intervention,Shade is completely eliminated.

Description

A kind of intelligent dimming controller in LED shadowless lamps illuminator

Technical field

The present invention relates to the intelligent dimming control in lighting field, more particularly to a kind of LED shadowless lamps illuminator Device processed.

Background technology

The shadowless lamp that Hospitals at Present operating room is used is most of using common thermal light source (such as incandescent lamp, halogen tungsten lamp), Its light source caloric value is big, power is high, short life, and because operation shadowless lamp light efficiency is strong, the duration is long, so easily to doctor Eye damage.Additionally, also a kind of light source uses fluorescence radiation, belong to cold light source, such light source energy consumption is relatively low, brightness Greatly, but colour temperature selects single, life-span and thermal light source almost, and light decay is larger, and directive property is not enough, there is larger limitation Property.Light-Emitting Diode LED (Light Emitting Diode, be abbreviated as LED) belongs to cold light source, because with uniform illumination, sound The advantages of rapid, extra long life, environmental protection is answered, operation shadowless lamp is widely used in recent years, and will progressively substitute traditional light Source.

Operation shadowless lamp is one of indispensable critical medical devices of hospital operating room, the brightness of operation shadowless lamp and without shadow The quality of the performances such as degree is directly connected to surgical quality and patient health.Adjusted using manual type more than traditional shadowless lamp, this It is that operator artificially adjusts according to the comfort level of itself to plant regulation, and the degree of accuracy of brightness is difficult to ensure that there is very big office It is sex-limited, its light position and brightness can obtain accurately, timely adjust and will have influence on being normally carried out for operation.Operation in addition When, the body of doctor, head, hand and apparatus can cause to block to operative site, form shade, will shadow if eliminated not in time Ring surgical quality.Although existing shadowless lamp is furnished with brightness regulator, but brightness regulation is carried out, it can however not the moon is completely eliminated Shadow, can only weaken the influence of shade；Meanwhile, this regulation is manually completed by operator, real with certain ambiguity Shi Xing, accuracy are not high enough, and easily cause surgical environments pollution, and influence operation is normally carried out.At present, the reality of operation shadowless lamp When technology with an automatic light meter report it is still rare.With digitlization, informationization, the intelligentized development, hand of global medical equipment The automatic digital light regulating technology of art shadowless lamp just progressively turns into a study hotspot.

It can be seen that, at present existing operation shadowless lamp technology exist in-convenience in use, the problems such as regulating effect is undesirable.This hair Bright purpose is to provide a kind of LED shadowless lamp intelligent dimming controllers based on fuzzy logic, it is intended to solve existing technology Present in in-convenience in use, the problems such as regulating effect is undesirable.

The content of the invention

The technical problem to be solved in the present invention is to provide a kind of LED shadowless lamps intelligent dimming control based on fuzzy logic Device, solve present in existing technology in-convenience in use, a kind of LED shadowless lamps illumination system the problems such as regulating effect is undesirable Intelligent dimming controller in system.

In order to solve the above-mentioned technical problem, the invention provides the intelligent dimming control in a kind of LED shadowless lamps illuminator Device processed, the present invention is the intelligent processing method for realizing Medical shadowless lamp brightness adjustment control, to reach purpose with an automatic light meter, by information Change treatment technology, with blurred vision treatment technology and artificial neural network theories, devise a kind of intelligentized luminance detection And track algorithm, realize a kind of intelligentized LED shadowless lamps light adjusting controller.The controller can be adopted according to CMOS camera The video image information of collection, the illumination and Colour information of working region are obtained by graphical analysis and Processing Algorithm, are passed through The shadow region and position that blurred picture partitioning algorithm produces illumination region because blocking are separated, by fuzzy neural network The driving current value of the LED light source that need to be adjusted is mapped out, the luminosity of LED light source is controlled, realized to LED shadowless lamps workspace The accurate regulation and Self Adaptive Control of domain illumination and colour temperature, so as to reach the brightness of real-time regulation shadowless lamp so that working region is shone Degree purpose uniform, constant, without shadow.

The present invention realizes the closed-loop control of system using video sensor technology.The controller includes：Image capture module, figure As analysis and processing module, it is single that blurred picture splits module, illumination and chromaticity extraction module, Fuzzy Neural Network System etc. Unit.The system employs a kind of method based on blurred vision image procossing, images scene with digital image processing techniques The video image that head is obtained carries out signature analysis and treatment, extracts brightness and the chrominance information of lighting environment.Implement process It is as follows：The video image of illumination region is gathered by CMOS camera and data image signal is converted to, is adopted by video image Collect module by the signal-obtaining to internal system, then parsed video signal by computer assisted image processing module, carry Illumination and Colour information are taken, corresponding illumination and such chroma feature vectors are extracted by illumination and chromaticity extraction module, Feeding Fuzzy Neural Network System；Meanwhile, the shadow region that shelter is produced is split by blurred picture segmentation module, and Fuzzy neural network is sent the location parameter of cut zone as input value into, it is powerful finally by fuzzy neural network inside Mapping and computing capability calculate the driving current value of corresponding LED light source, and transmit these information to LED controller, adjust The brightness of corresponding LED reaches the purpose for eliminating shade and permanent photocontrol.

Image is the visual basis in the human perception world, but in the great amount of images information that the mankind are obtained by vision, Required for not all information content is all us, thus need to divide the image into that several are specific, with uniqueness The region of property.Image segmentation, exactly divides the image into several regions specific, with unique properties and extracts and feel emerging Interesting target.Seek to find the position and region of the shelter for causing shade for shadowless lamp, then effectively split The characteristic information of shelter is obtained, rational segmentation result can preferably find the useful information in image and conveniently it is carried out Treatment.

(1) the Shadow segmentation algorithm based on fuzzy logic

The useful information obtained by image is exactly the shadow region that shelter is produced on working face, and image is intrinsic Inherent ambiguity brings many difficulties to image segmentation, but but uses force it for the application of fuzzy set and Systems Theory is provided Ground, so we are understood, represented, processed using fuzzy set and Systems Theory and object image is blocked in segmentation in the present invention.

Piece image possesses different characteristic values, and the present invention is split by gradation of image to image.For a width M × N images, its gray level is 0~255.Pre-segmentation is carried out to image by original partitioning algorithm first, is carried on the back by pre-segmentation Scape (Background Region, BR) and target area (0bject Region, OR).Randomly select limited background and target Area pixel point, its gray average is calculated with reference to grey level histogram, is obtainedWithRespectively background and target area threshold value.Obtain Target area OR is obtained, fuzzy region (Fuzzy Region, FR) and the tonal range of background area BR areAnd

Shelter in image and background are split, thus we need by image be divided into the background area of determination with Target area.Reference background region and object reference region are considered as two fuzzy subsets of gray scale collection [0,1 ..., F-1].Retouch Stating the method for fuzziness has a lot, for example quantity area method, correlation coefficient process, minimax method, absolute exponent method, nonparametric method Deng.The system from Study on similar degree method apart from exchange premium degree, wherein setting domain U={ x₁, x₂..., x_nTo arbitrary fuzzy set A is closed, fuzziness is：

The fuzziness that can calculate target area OR and background area BR by the formula is L_BRAnd L_OR。

To in confusion regionIt is split into scopeWithIt is added separately to OR and BR In, g_fIt is g_FRThe partition value of set, obtains two new fuzzy subsets, is designated as OR ' and BR '：

New L is calculated by formula (1)_OR′And L_BR′.In the case where fuzzy subset adds new element, its fuzziness letter Numerical value can become (i.e. L greatly_OR′＞ L_ORL_BR′＞ L_BR).So, by its respectively with L_ORAnd L_BRNormalize, obtain two fuzziness shadows The factor is rung, is designated as：

By comparing η₁And η₂Size, judge g_FRAddition be bigger influence to background or target area.If η₁＞ η₂, then g_FRIt is bigger on target area fuzzy subset influence, i.e., it is higher with target area similarity, so should be by g_FRPut background area under The fuzzy set in domain；Conversely, then by g_FRPut the fuzzy set for blocking object area under.Gray scale to fuzzy region does same treatment, then can There is a certain gray value g_d, make η₁(g_d)=η₂(g_d), then g_dIt is segmentation threshold.

(2) workspace brightness and the determination of colourity

Coloured image typically represents that all colours are all considered as 3 the red R of basic colors, green G, indigo plants with RGB color The various combination of B, therefore, RGB color can be set up in Cartesian coordinate system.RGB color biggest advantage It is comparing directly perceived, for screen display easily, has the disadvantage height correlation between tri- components of R, G, B, some component occurs Change the change that can influence whole image color.Coloured image also can use HSI color spaces to represent, HSI color spaces are from people Vision system set out with tone H, saturation degree S and brightness I to describe color.HSI color spaces can be described with conical space,

Although description is complicated, the situation of change of tone, brightness and saturation degree can be showed will be apparent that.Two kinds of color skies Between between there is transformational relation.The image of a given width RGB color form, normalizes in the range of [0,1] to any group of The value of RGB all corresponding HSI component values can be obtained by corresponding conversion formula.The conversion formula of RGB to HSI color spaces For：

S=1-3min (r, g, b) S ∈ [0,1]

Tone H (Hue), the wavelength with light wave is relevant, and it represents impression of the sense organ of people to different colours, such as red, green Color, blueness etc., it can represent the chrominance information of illumination region image, such as warm colour, cool colour.Intensity I (Intensity), correspondence Brightness of image and gradation of image, are the light levels of color, and it can represent the illuminance information of illumination region image.By above-mentioned face Color model, it may be determined that the brightness of illumination workspace and colourity.

(3) Fuzzy Clustering Neural Network (FCNN)

The system proposes a kind of visual pattern target identification method based on fuzzy neural network.The method is with fuzzy system Based on model, the target occlusion thing and the scene of background composition that identification will be needed in every frame video image regard a fuzzy system as System, with the location and shape information of the moving target extracted in each frame as characteristic vector, using this feature vector as fuzzy The input of clustering neural network (FCNN) system, using fuzzy clustering identification algorithm, structure is a kind of can be to the light distribution of LED Fuzzy Clustering Neural Network (FCNN) model for being mapped, the output to system is predicted, and provides one group in current scene The optimization control parameter of LED light source light distribution and position distribution under situation, by adjust LED lamp panel light source exposure intensity and Angle, realizes the permanent light in irradiation working region, the purpose without shadow.

The structure of Fuzzy Clustering Neural Network FCNN is as shown in Figure 2.Whole system is made up of two parts：Part I is Fuzzy Classifier, it is made up of three layers of BP networks.Input layer is made up of P node, P component of correspondence input vector； Hidden layer is made up of C node, and its i-th node represents the deviation between input vector and ith cluster center, their transmission Function is：

Output layer is also made up of C node, and the output of each node represents degree of membership of the input vector to a certain classification.It is defeated Connection weight between ingress and hidden node represents the cluster centre v of a certain class_i, its needs is carried out excellent by learning algorithm Change；Used between hidden node and output node and connected without weighting, it collectively constitutes third layer node with the output of each sub-network Input.Part II is made up of C sub- network, and each sub-network is made up of a double-layer network, connection weight matrix w_i= (w_i1, w_i2, ..., w_iQ)^T, wherein w_ij=(w_j0 ⁽ⁱ⁾, w_j1 ⁽ⁱ⁾, w_j2 ⁽ⁱ⁾..., w_jP ⁽ⁱ⁾), input vector θ_k=(1, x_k1, x_k2..., x_kP)^T, i-th sub-network be output as：It completes the k-th consequent output of the i-th rule-like of input sample Calculate, system is total to be output as

System exports y_kThe distribution map mapping of LED light source array in present image scene, mapping reflection LED light will be given Source will could realize making the illumination in area to be illuminated domain to reach the value of regulation and keep constant with what kind of Luminance Distribution, while eliminate again The purpose of workspace shade, exports y_kLED light source driving current and LED lamp panel crevice projection angle will be controlled as regulated quantity, so that Realize irradiation area perseverance light, the effect without shadow.

After the system, its advantage is：

1st, by Intelligentized Information technology, the automatic constant light regulation of illumination region illumination is realized.This controller is used Fuzzy logic theory, with reference to the correlation technique of Computer Vision, obtain the shade distribution of illumination region positional information and Illuminance information, the regulated quantity of the driving current of LED light source is calculated by fuzzy neural network, controls the brightness of LED light source, is made The illumination for obtaining illumination region can keep constant, the uniform effect without shadow.

2nd, deficiency and defect that traditional shadowless lamp is present are improved.The problem that the present invention exists for current shadowless lamp, carries Go out a kind of vision based on fuzzy logic and track technology with an automatic light meter, system uses fuzzy video image processing techniques and nerve net Network is theoretical, to being tracked because of the shade produced by the shelters such as operator, operating theater instruments in video image, being split, it is determined that cloudy The distributed intelligence of illumination and colourity in the position and working region in shadow zone domain, calculates needs and is adjusted by neural network model The driving current value of whole LED light source, then the light distribution and brightness of LED light source array are controlled by LED constant current controller, make work Making face and operative region depth can obtain an intensity of illumination distribution that is constant, meeting specification, while eliminating because of operator, hand Shade produced by the shelters such as art apparatus.

3rd, digitizing technique and information-based intellectual technology are incorporated into the design of shadowless lamp.The present invention is by with fuzzy Logical theory, video image processing technology and neutral net intellectualized technology realize the with an automatic light meter of shadowless lamp, make tradition without shadow The design philosophy of lamp there occurs fundamental change, and the five big weak points that traditional shadowless lamp is present have been broken away from one stroke：

1. it is not high without shadow effect.Light source reflected or the angle irradiated the more, obtained after convergence without shadow effect better, And the hot spot being combined into 12 single light source bulb irradiations, its shadow effect that disappears is surely not too preferable, such as increases radiation source again, Clearly it is difficult to walk；

2. structure very complicated.12 lamp holders with 3 transformer-supplieds, the complexity of its structure, the huge of profile is to think And know；

3. security reliability is poor.For several more bulbs and transformer greatly improve the rate of breakdown of whole machine, once One breaks down, and whole shadowless lamp performance impairment is bad；

4. adjust frequent and dull laborious.Because spot diameter is small, thickness of thin, the change palpus with operation face and depth Constantly to focus, positioning could obtain optimal illumination, this just causes excessive infection chance and fatigue, influence operation to patient Quality；

5. to the thermal pollution of surgical environments.The electric elements such as more bulb and transformer, make caloric value increase, though there is wind Fan radiating, but the difficult temperature rise eliminated around patient eventually, make surgical environments degenerate；

4th, this project utilizes informationization technology means, the illumination and control of the thermal light source that broken traditions using LED cold light source technologies Molding formula, is built by modern information technologies such as visual pattern treatment technology, Fuzzy Neural Network Theory, Computer Control Technologies A set of new, intelligent shadowless lamp system, makes lighting for medical use technology march toward digitlization, information-based and intellectualization times.

5th, the design is monolithic design, and corn module is integrated in one piece of SOC (on-chip system) chip internal, with knot Structure is simple, and low cost is low in energy consumption, small volume, high reliability, can reach LED light source luminous efficacy and energy-saving effect Optimum state.

This controller has following features：

1) the intelligent dimming control system based on blurred vision treatment technology is established, is realized to the automatic of working region Brightness adjustment control, makes the illumination of illumination region constant in setting value；

2) realize to the Real-time segmentation of the moving target in working region and positioning, effective detection and shelter can be judged Position and region, so as to shade is completely eliminated, realize shadowless lamp truly；

3) with the specific of workspace illumination continuously adjustabe, user can arbitrarily set a brightness value, and system just can be certainly Motion tracking is simultaneously locked on the brightness value of setting；

4) the design is a kind of intelligentized automatic control system, and user need to only pre-set control parameter, just can be real Now whole control process, is not required to manual intervention during regulation；

5) the design also has adjustable range wide, degree of regulation advantage high, and energy continuously smooth enters in adjustable range Row brightness regulation, the step pitch of regulation is small, flicker free and jump, and the hot spot uniformity is high, with compared with top adjustment quality.

Brief description of the drawings

Fig. 1 is the intelligent dimming controller architecture block diagram in a kind of LED shadowless lamps illuminator of the invention.

Fig. 2 is the intelligent dimming controller Fuzzy Clustering Neural Network in a kind of LED shadowless lamps illuminator of the invention FCNN structure charts.

Specific embodiment

The present invention is described in detail with reference to the accompanying drawings and detailed description, it is impossible to is not understood as to of the invention Limitation；

According to Fig. 1 and Fig. 2, the intelligent dimming controller in a kind of LED shadowless lamps illuminator of the invention.Using taking the photograph The closed-loop control of illumination region illumination is realized as sensing technology, system includes：Image capture module, computer assisted image processing mould The units such as block, blurred picture segmentation module, illumination and chromaticity extraction module, Fuzzy Neural Network System；System is employed A kind of method based on blurred vision image procossing, the video image for obtaining live camera with digital image processing techniques Signature analysis and treatment are carried out, brightness and the chrominance information of lighting environment is extracted, then illumination region is gathered by CMOS camera Video image and be converted to data image signal, by video image acquisition module by the signal-obtaining to internal system, then Video signal is parsed by computer assisted image processing module, is extracted illumination and Colour information, by illumination and Chromaticity extraction module extracts corresponding illumination and such chroma feature vectors, sends into Fuzzy Neural Network System；Meanwhile, by obscuring Image segmentation module splits the shadow region that shelter is produced, and the location parameter of cut zone is sent as input value Enter fuzzy neural network, corresponding LED light source is calculated finally by the powerful mapping in fuzzy neural network inside and computing capability Driving current value, and transmit these information to LED controller, the illumination patterns and intensity of brightness of corresponding LED are adjusted, to arrive Up to the purpose for eliminating shade, proportional illumination and permanent photocontrol.

Image is the visual basis in the human perception world, but in the great amount of images information that the mankind are obtained by vision, Required for not all information content is all us, thus need to divide the image into that several are specific, with uniqueness The region of property.Image segmentation, exactly divides the image into several regions specific, with unique properties and extracts and feel emerging Interesting target.Seek to find the position and region of the shelter for causing shade for shadowless lamp, then effectively split The characteristic information of shelter is obtained, rational segmentation result can preferably find the useful information in image and conveniently it is carried out Treatment.

(1) the Shadow segmentation algorithm based on fuzzy logic

The useful information obtained by image is exactly the shadow region that shelter is produced on working face, and image is intrinsic Inherent ambiguity brings many difficulties to image segmentation, but but uses force it for the application of fuzzy set and Systems Theory is provided Ground, so we are understood, represented, processed using fuzzy set and Systems Theory and object image is blocked in segmentation in the present invention.

Piece image possesses different characteristic values, and the present invention is split by gradation of image to image.For a width M × N images, its gray level is 0~255.Pre-segmentation is carried out to image by original partitioning algorithm first, is carried on the back by pre-segmentation Scape (Background Region, BR) and target area (0bject Region, OR).Randomly select limited background and target Area pixel point, its gray average is calculated with reference to grey level histogram, is obtainedWithRespectively background and target area threshold value.Obtain Target area OR is obtained, fuzzy region (Fuzzy Region, FR) and the tonal range of background area BR areAnd

Our purpose is to be split the shelter in image and background, so we need for image to be divided into determination Background area and target area.Reference background region and object reference region are considered as two of gray scale collection [0,1 ..., F-1] Fuzzy subset.The method for describing fuzziness has a lot, such as quantity area method, correlation coefficient process, minimax method, absolute exponent Method, nonparametric method etc..The system from Study on similar degree method apart from exchange premium degree, wherein setting domain U={ x₁, x₂..., x_n, it is right Arbitrary fuzzy set A, fuzziness is：

The fuzziness that can calculate target area OR and background area BR by the formula is L_BRAnd L_OR。

To in confusion regionIt is split into scopeWithIt is added separately to OR and BR In, g_fIt is g_FRThe partition value of set, obtains two new fuzzy subsets, is designated as OR ' and BR '：

New L is calculated by formula (1)_OR′And L_BR′.In the case where fuzzy subset adds new element, its fuzziness letter Numerical value can become (i.e. L greatly_OR′＞ L_ORL_BR′＞ L_BR).So, by its respectively with L_ORAnd L_BRNormalize, obtain two fuzziness shadows The factor is rung, is designated as：

By comparing η₁And η₂Size, judge g_FRAddition be bigger influence to background or target area.If η₁＞ η₂, then g_FRIt is bigger on target area fuzzy subset influence, i.e., it is higher with target area similarity, so should be by g_FRPut background area under The fuzzy set in domain；Conversely, then by g_FRPut the fuzzy set for blocking object area under.Gray scale to fuzzy region does same treatment, then can There is a certain gray value g_d, make η₁(g_d)=η₂(g_d), then g_dIt is segmentation threshold.

(2) workspace brightness and the determination of colourity

Coloured image typically represents that all colours are all considered as 3 the red R of basic colors, green G, indigo plants with RGB color The various combination of B, therefore, RGB color can be set up in Cartesian coordinate system.RGB color biggest advantage Comparing directly perceived, for screen display easily, have the disadvantage R,

Height correlation between tri- components of G, B, some component there occurs that change can influence the change of whole image color. Coloured image also can use HSI color spaces to represent, HSI color spaces from the vision system of people tone H, saturation degree S and Brightness I describes color.HSI color spaces can be described with conical space,

Although description is complicated, the situation of change of tone, brightness and saturation degree can be showed will be apparent that.Two kinds of color skies Between between there is transformational relation.The image of a given width RGB color form, normalizes in the range of [0,1] to any group of The value of RGB all corresponding HSI component values can be obtained by corresponding conversion formula.The conversion formula of RGB to HSI color spaces For：

S=1-3min (r, g, b) S ∈ [0,1]

Tone H (Hue), the wavelength with light wave is relevant, and it represents impression of the sense organ of people to different colours, such as red, green Color, blueness etc., it can represent the chrominance information of illumination region image, such as warm colour, cool colour.Intensity I (Intensity), correspondence Brightness of image and gradation of image, are the light levels of color, and it can represent the illuminance information of illumination region image.By above-mentioned face Color model, it may be determined that the brightness of illumination workspace and colourity, are that the segmentation of shadow image and the determination of locus provide foundation.

(3) Fuzzy Clustering Neural Network (FCNN)

The system proposes a kind of visual pattern target identification method based on fuzzy neural network.The method is with fuzzy system Based on model, the target occlusion thing and the scene of background composition that identification will be needed in every frame video image regard a fuzzy system as System, with the location and shape information of the moving target extracted in each frame as characteristic vector, using this feature vector as fuzzy The input of clustering neural network (FCNN) system, using fuzzy clustering identification algorithm, structure is a kind of can be to the light distribution of LED Fuzzy Clustering Neural Network (FCNN) model for being mapped, the output to system is predicted, and provides one group in current scene The optimization control parameter of LED light source light distribution and position distribution under situation, by adjust LED lamp panel light source exposure intensity and Angle, realizes the permanent light in irradiation working region, the purpose without shadow.

The structure of Fuzzy Clustering Neural Network FCNN is as shown in Figure 2.Whole system is made up of two parts：Part I is Fuzzy Classifier, it is made up of three layers of BP networks.Input layer is made up of P node, P component of correspondence input vector； Hidden layer is made up of C node, and its i-th node represents the deviation between input vector and ith cluster center, their transmission Function is：

Output layer is also made up of C node, and the output of each node represents degree of membership of the input vector to a certain classification.It is defeated Connection weight between ingress and hidden node represents the cluster centre v of a certain class_i, its needs is carried out excellent by learning algorithm Change；Used between hidden node and output node and connected without weighting, it collectively constitutes third layer node with the output of each sub-network Input.Part II is made up of C sub- network, and each sub-network is made up of a double-layer network, connection weight matrix w_i= (w_i1, w_i2, ..., w_iQ)^T, wherein w_ij=(w_j0 ⁽ⁱ⁾, w_j1 ⁽ⁱ⁾, w_j2 ⁽ⁱ⁾..., w_jP ⁽ⁱ⁾), input vector θ_k=(1, x_k1, x_k2..., x_kP)^T, i-th sub-network be output as：It completes the k-th consequent output of the i-th rule-like of input sample Calculate, system is total to be output as

System exports y_kThe distribution map mapping of LED light source array in present image scene, mapping reflection LED light will be given Source will could realize making the illumination in area to be illuminated domain to reach the value of regulation and keep constant with what kind of Luminance Distribution, while eliminate again The purpose of workspace shade, exports y_kLED light source driving current and LED lamp panel crevice projection angle will be controlled as regulated quantity, so that Realize irradiation area perseverance light, the effect without shadow.

Claims (1)

1. the intelligent dimming controller in a kind of LED shadowless lamps illuminator, it is characterised in that：Realized using video sensor technology The closed-loop control of illumination region illuminance, the controller core unit includes：Image capture module, computer assisted image processing mould Block, blurred picture segmentation module, illumination and chromaticity extraction module, Fuzzy Neural Network System unit, controller are used A kind of method based on blurred vision image procossing, the video image for obtaining live camera with digital image processing techniques Signature analysis and treatment are carried out, brightness and the chrominance information of lighting environment is extracted, LED light source is calculated by Intelligent treatment algorithm Current control parameter, so as to adjust the brightness of LED light source, the video image that system gathers illumination region by CMOS camera is simultaneously Be converted to data image signal, by video image acquisition module by the signal-obtaining to internal system, then by graphical analysis with Processing module is parsed video signal, extracts illumination and Colour information, is extracted by illumination and chromaticity Module extracts corresponding illumination and such chroma feature vectors, sends into Fuzzy Neural Network System；Meanwhile, module is split by blurred picture The shadow region that shelter is produced is split, and sends the location parameter of cut zone as input value into fuzznet Network, finally by the powerful mapping in fuzzy neural network inside and computing capability, calculates the driving current of corresponding LED light source Value, and LED controller is transmitted these information to, the brightness of corresponding LED is adjusted, to realize eliminating shade, uniform-illumination and perseverance The purpose of photocontrol；Shelter in image and background are split, it is necessary to image to be divided into background area and the mesh of determination Mark region, two fuzzy sons reference background region and object reference region being considered as on gray value domain [0,1 ..., F-1] Collection, the system from Study on similar degree method apart from exchange premium degree, wherein setting domain U={ x₁, x₂..., x_n, to arbitrary fuzzy set A is closed, fuzziness is：

$L\left(A\right)=\frac{1}{mn}\underset{i=1}{\overset{m}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}\frac{|{\mathrm{\μ}}_A\left(x_{ij}\right)-{\mathrm{\μ}}_{0.5}\left(x_{ij}\right)|}{{\mathrm{\μ}}_A\left(x_{ij}\right)+{\mathrm{\μ}}_{0.5}\left(x_{ij}\right)}---\left(1\right)$

The fuzziness for calculating target area OR and background area BR by the formula is L_BRAnd L_OR,

To in confusion regionIt is split into scopeWithIt is added separately in OR and BR, g_f It is g_FRThe partition value of set, obtains two new fuzzy subsets, is designated as OR ' and BR '：

${\mathrm{OR}}^{\′}=OR\∪\left\{g_{FR}\right\}=OR\∪\[\stackrel{\‾}{g_o},g_f\]---\left(2\right)$

${\mathrm{BR}}^{\′}=BR\∪\left\{g_{FR}\right\}=BR\∪\[g_f,\stackrel{\‾}{g_b}\]$

New L is calculated by formula (1)_OR′And L_BR′, in the case where fuzzy subset adds new element, its ambiguity function value Big (i.e. L can be become_OR′>L_OR L_BR′>L_BR), by its respectively with L_ORAnd L_BRNormalize, obtain two fuzziness factors, remember For：

${\mathrm{\η}}_1\left(g_{FR}\right)=\frac{L_{{\mathrm{OR}}^{\′}}\left(g_{FR}\right)}{L_{OR}}---\left(3\right)$

${\mathrm{\η}}_2\left(g_{FR}\right)=\frac{L_{{\mathrm{BR}}^{\′}}\left(g_{FR}\right)}{L_{BR}}$

By comparing η₁And η₂Size, judge g_FRAddition be bigger influence to background or target area, if η₁>η₂, then g_FRIt is bigger on target area fuzzy subset influence, i.e., it is higher with target area similarity, so should be by g_FRPut background area under Fuzzy set；Conversely, then by g_FRPut the fuzzy set for blocking object area under, the gray scale to fuzzy region does same treatment, then have certain One gray value g_d, make η₁(g_d)=η₂(g_d), then g_dIt is segmentation threshold；The determination method of workspace brightness and colourity is：Cromogram As being represented with RGB color, all colours all regard 3 red R of basic colors, green G, the various combination of indigo plant B, the RGB face as The colour space is set up in Cartesian coordinate system；Coloured image represents with HSI color spaces, vision of the HSI color spaces from people System is set out, and color is described with tone H, saturation degree S and brightness I；HSI color spaces are described with conical space, two kinds of colors Transformational relation is there is between space, the image of a width RGB color form is given, [0,1] scope is normalized to any group of The value of interior RGB all obtains corresponding HSI component values by corresponding conversion formula；The conversion formula of RGB to HSI color spaces For：

$\begin{array}{ccc}r=\frac{R}{R+G+B}& g=\frac{G}{R+G+B}& b=\frac{B}{R+G+B}\end{array}$

$\begin{array}{cc}H=\left\{\begin{array}{cc}\mathrm{\&theta;}& g\&GreaterEqual;b\\ 2\mathrm{\&pi;}-\mathrm{\&theta;}& g<b\end{array}\right.& H\&Element;\&lsqb;0,2\mathrm{\&pi;}\&rsqb;\end{array}$

S=1-3min (r, g, b) S ∈ [0,1]

$\begin{array}{cc}I=\frac{R+G+B}{3\&times;255}& I\&Element;\&lsqb;0,1\&rsqb;\end{array}$

Tone H (Hue) is relevant with the wavelength of light wave, and it represents impression of the sense organ of people to different colours, red, green, blueness Deng, the chrominance information of its expression illumination region image, warm colour, cool colour etc., intensity I (Intensity), correspondence brightness of image and figure It is the light levels of color as gray scale, it represents the illuminance information of illumination region image；Using the vision figure of fuzzy neural network As target identification method, be will be needed in every frame video image based on fuzzy system model the target occlusion thing of identification with The scene of background composition regards a fuzzy system as, with the location and shape information of the moving target extracted in each frame as spy Levy vector, using this feature vector as Fuzzy Clustering Neural Network (FCNN) system input, using fuzzy clustering identification algorithm, Build a kind of Fuzzy Clustering Neural Network (FCNN) model that can be mapped the light distribution of LED, the output to system It is predicted, provides one group of optimization control parameter of LED light source light distribution and position distribution under current scene situation, passes through The light source exposure intensity and angle of LED lamp panel are adjusted, the permanent light in irradiation working region, the purpose without shadow is realized；Whole system is by two Individual part composition：Part I is Fuzzy Classifier, and it is made up of three layers of BP networks, and input layer is made up of P node, right Answer P component of input vector；Hidden layer is made up of C node, and its i-th node represents input vector and ith cluster center Between deviation, their transmission function is：

$d_{ik}=\left|\right|x_k-v_i|{|}^2=\underset{j=1}{\overset{P}{\&Sigma;}}{(x_{kj}-v_{ij})}^2$

Output layer is also made up of C node, and the output of each node represents degree of membership of the input vector to a certain classification, input section Connection weight between point and hidden node represents the cluster centre v of a certain class_i, it is optimized by learning algorithm；Hidden layer section Used between point and output node and connected without weighting, it collectively constitutes the input of third layer node with the output of each sub-network, the Two parts are made up of C sub- network, and each sub-network is made up of a double-layer network, connection weight matrix w_i=(w_i1, w_i2, ..., w_iQ)^T, wherein w_ij=(w_j0 ⁽ⁱ⁾, w_j1 ⁽ⁱ⁾, w_j2 ⁽ⁱ⁾..., w_jP ⁽ⁱ⁾), input vector θ_k=(1, x_k1, x_k2..., x_kP)^T, i-th son Network is output as：y_i ^k=w_iθ_k, it completes k-th calculating of the consequent output of the i-th rule-like of input sample, and system is total It is output as

$y_k=\underset{i=1}{\overset{c}{\&Sigma;}}{u}_{ik}{y}_i^k=\underset{i=1}{\overset{c}{\&Sigma;}}{u}_{ik}{w}_i{\mathrm{\&theta;}}_k$

System exports y_kThe distribution map mapping of LED light source array in present image scene is provided, mapping reflection LED light source will be with What kind of Luminance Distribution could realize making the illumination in area to be illuminated domain to reach the value of regulation and keep constant, while eliminating workspace again The purpose of shade, exports y_kLED light source driving current and LED lamp panel crevice projection angle are controlled as regulated quantity.