JPH02173732A - Controller for flashing device - Google Patents

Abstract

PURPOSE:To automatize a flashing device by controlling the angle and the light quantity of the flashing part of the flashing device and the irradiation angle in accordance with output from an output layer of a neurone computer. CONSTITUTION:The output from plural photoelectric conversion elements 102 for image and a photodetector 106 for detecting a distance is inputted in an input layer consisting of plural units which constitute the neurone computer 109. The output from each unit of the input layer is inputted in an intermediate layer consisting of plural units which are bonded with specified bonding strength obtained by previous learning. The output from each unit of the intermediate layer is inputted in the output layer consisting of plural units bonded with the specified bonding strength obtained by previous learning. The output from the output layer is taken as the correction information outputted from the neurone computer 109 so as to correct the angle and the light quantity of the flashing part and the irradiation angle which are automatically set by the flashing device in accordance with the information from a camera such as ISO sensitivity, focal distance and stop, etc. Thus, the irradiation angle and the angle and the light quantity of the flashing part are accurately controlled to be optimum value at high speed.

Description

[Detailed description of the invention] [Industrial use! Ff] The present invention relates to a control device for a flash device that controls the angle, light amount, and irradiation angle of a flash unit to optimal values in a flash device that is attached to a camera to perform flash photography.

[Conventional technology] In the past, flash photography using flash devices (strobes) has been automated, and the focus has been on the "focal length of the lens" and "distance to the subject."
Obtain information on one or both of these and adjust the illumination angle so that the entire field of view is illuminated at all times.

However, it is not always necessary to irradiate the entire screen with strobe light, and it may be sufficient to irradiate only a portion of the screen. In such a case, it is a waste of energy to irradiate the entire screen with strobe light. There are also cases where partial irradiation is actively desired.

Changing the angle of the flash section is especially necessary when you want to obtain a bounce effect in addition to correcting vararax, but in this case, the current situation is that it relies on image information other than the angle of view and the user's intuition.

In addition, the light amount is controlled according to the signal supplied from the camera side without changing the light emission curve on the strobe side, but it may deviate from the control range at short distances etc. do. Therefore, the current situation is that users use a diffuser plate or the like to reduce the amount of light.

[Question MB that the invention seeks to solve] As described above, flash photography still relies on human intuition in many areas, and it is difficult to formulate these. Therefore, if we attempt to automate the above flash control by patterning using current Neumann-type microcomputers, the number of patterns will be enormous, and the program and processing time will also be enormous. The problem was that Norman-type microcomputers were not suitable for control, and were an obstacle to automation.

The present invention has been made to solve the above problems, and is a flash device that can accurately and quickly control the irradiation angle, the angle of the flash section, and the amount of light to optimal values without relying on human intuition. The purpose is to provide a control device.

[Means for Solving the Problems] A control device for a flash device of the present invention includes a number of stages of image photoelectric conversion elements, an optical device for forming an image of a subject and its background on the image photoelectric conversion elements, and a number of stages of image photoelectric conversion elements. a distance detection light emitting element, a number of stages of light receiving elements that receive the signal of this distance detection light projecting element, and a number of stages of units into which the outputs of the above number of stages of photoelectric conversion elements and the plurality of light receiving elements are inputted. an input layer consisting of an input layer, an intermediate layer consisting of a plurality of units that are mutually coupled with each unit of this input layer with a predetermined coupling strength, and a plurality of units that are mutually coupled with each unit of this intermediate layer with a predetermined coupling strength. a neuron computer comprising an output layer consisting of units with a number of stages, and an angle of the flash section of the flash device according to the output from the output layer of the neuron computer;
It is characterized by comprising a control means for controlling the amount of light and the angle of illumination.

[Operation] According to the present invention, the outputs of a plurality of image photoelectric conversion elements and a distance detection light receiving element are input to an input layer consisting of a plurality of units constituting a neuron computer, and the output of each unit of this input layer is as follows. The input is input to a middle layer consisting of a plurality of units connected with a predetermined connection strength obtained by prior learning, and the output of each unit of this middle layer is
input to an output layer consisting of 181 units connected with a predetermined connection strength obtained by prior learning,
The output of this output layer is used as correction information output by the neuron computer to correct the angle, light intensity, and irradiation angle of the flash unit that is automatically set by the flash device according to information from the camera such as ISO sensitivity, focal length, and aperture. This allows the flash device to be optimally controlled.

[Embodiment] FIG. 1 is a block diagram showing an example of a control device for a flash device (strobe) according to the present invention.

In the figure, a wide-angle lens 101 as an optical device is a color area sensor 1021 as an image photoelectric conversion element.
- forms an image in a wider range than the angle of view of the lens. The color area sensor 102 extracts brightness signals and color signals of the image by photoelectrically converting the image formed by the biped wide-angle lens 101. The output of this color area sensor 102 is sent to an A/D converter 107.
A/D converter 107
The converter converts the contrast signal and color signal as analog signals outputted from the color area sensor 102 into digital signals. The output of this A/D converter 107 is supplied to a neuron computer 109 (details will be described later). Further, the output of the neuron computer 109 is controlled by the main CPU as a control means.
11. This main CPU
I11 is in charge of overall control of the flash device.

An infrared light emitting diode (hereinafter referred to as IRED) 103 as a distance detection light emitting element is connected to the main CPU 1.
The infrared rays emitted from this IRED 103 are transmitted to the color area sensor 10 by a wide-angle lens 104.
It can be expanded to the range of 2 images.

The wide-angle lens 105 receives all the light beams emitted from the IRED 103 using a position detection element (hereinafter referred to as PSD) as a light receiving element.
(referred to as an array) 106. This PSD array 106 extracts a distance signal to the subject by photoelectrically converting the light beam guided by the wide-angle lens 105. The output of this PSD array 106 is supplied to an A/D converter S108. The A/D converter 108 is connected to the PSD array 106.
This converts a distance signal as an analog signal output from a digital signal into a digital signal. This A/D converter 108
The output of is also supplied to the neuron computer 109.

The main CPU 11 also includes a switch group (SW group) 112 provided in the strobe, a display element 113,
Motor 11 that serves both to change the irradiation angle and to change the angle of the flash section
4, and a charging capacitor capacity changeover switch (hereinafter referred to as capacitor SW) for changing the amount of light. Furthermore, the main CPU 11 is also connected to the camera body 150.

Also, "24 Self Neuron Computer 1091"
A learning storage unit 110 is connected.

The learning storage section 110 stores predetermined information by the WJl setting device of the learning storage section 110. Details will be described later.

Next, the neuron computer 109 mentioned above will be explained in detail. FIG. 2 shows a model of a neuron computer (hereinafter referred to as NC).

This model was developed by Ru*olhart and others as P1! This is a proposed model called the Back Propagation model (hereinafter referred to as
It is called the BP model (abbreviated as BP model). A neurocomputer consists of many units (neurons), and the units are classified into input layer, intermediate layer, and output layer. Each unit is connected in the direction of input layer - middle layer -> output layer, forming a network (coral net). The strength of the connections between each unit is determined by learning. However, there is no mutual connection between units within each layer. Figure 3 shows the model of each unit.

Next, the principle of this BP model learning algorithm will be explained. When a certain pattern P is given to the input layer, the actual output value appearing in the output layer is O, and 1) J is the desired output value at that time (hereinafter referred to as the teacher signal).
Then, the difference E between the two is expressed as follows, pJ
PJ is done.

E, -1/2 (t, -〇, )
... (1)1) J pJ
In order to learn pJ, it is sufficient to change the strength of connections between all units so as to reduce this error E, pJ.

When the pattern P is given, the strength of the bond W from the i-th unit of the (K-1) layer to the first unit of the layer
The amount of change in jl is defined as follows.

Here, l is set to 0 at the output layer and increases as the input layer approaches.

(anet, ni/aW, K) -(3)
pJ Jiru.

Also, if f is a sigmoid (Slgmold) function,
When expressed as K K O-f(net), the formula (3) is transformed as follows. The sigmoid function is shown in FIG.

K K K+19E, /
FaW, -1δ, -0-(4)1)J
J I 9 J u
l where δ is the amount of backward propagation of the error in the layer, and δ , −−9E , /anet
, KpJ pJ
It is pJ. Therefore, equation (2) is transformed as follows. Here, η is a constant.

K K K+1 Δ pW, -η pot δ,
-O... (5)Ji pJpi 0,) ,O,=f(net,), so pJ
I)J
I) The backward propagation amount δ of the J output layer is as follows.

j δ , O pJ δ , K = 1) J aEp/19net , K pJ (ao /anet, K pi f) J monthf ′ (net, K pJ ... (6) In the case of intermediate units, within each layer Since there is no unit coupling at , the backward propagation port for the error is:

... (7) Equation (7) is a recursive function of δ.

A general formulation of ΔpW,'ゝ is as follows.

ΔpW, K (n+1) l −ηδpjK·0+αΔpWjIK(n)K+1 I (8) where ΔpWjIK (0)−0, and n represents the number of times of learning. The second term on the right side of equation (8) is added to reduce error vibration and speed up convergence.

From equation (8), the bond strength is updated as follows.

W, K (n+1) to K −WJ I (n)+Δp Wji (n)(K−
0, 1, 2, ...) ... (9) Here, the sigmoid function f1 is -net I) ... (10) f 1-1
/ (1+e) Since f'-f (1-f,) is defined, the amount of backward propagation can be simplified as shown in the following equation.

In case of output crab unit: (t-0,.) ...(11) pJ 9J In case of intermediate unit: K K K δ , -0, (1-0,) pJ l)J
pJ (12) As shown in section 2, the calculation of ΔW starts from the output layer unit and moves to the intermediate layer unit.

In this way, learning proceeds in the opposite direction to the processing of input data.

Therefore, learning using the BP model is performed as follows. First, data for learning is input and the results are output. Next, the strength of the coupling is changed to reduce the resulting error (the difference between the actual output and the teacher signal). Then, input the learning data again. This operation is repeated until ΔW converges.

FIG. 5 shows the basic circuit configuration of the BP model.

A random access memory (hereinafter referred to as RAM) 1 stores the strength of connection Wjl, and consists of N pages -i to N for each layer. The RAM 2 stores the amount of change ΔWj1 in the bond strength W when the pattern P is given, and is composed of N pages k-1 to N. The RAM 3 stores the amount of backward propagation of errors, δ, and consists of J (N+1) pages of k-0 to N. The RAM 4 stores the output value 0 of each unit and consists of (N+1)1)J pages of k-0 to N. 5 is an arithmetic circuit of O, and 6 is δ.

1) J I) J arithmetic circuit; 7 is an arithmetic circuit for Δpwj; 9 is a sequence controller that controls the entire sequence.

The learning process using the BP model shown in FIG. 5 will be explained.

Here, the operation when the BP model is simulated by a Neumann type computer will be explained with reference to the flowcharts of FIGS. 6 to 9. Figure 6 is 0,
7 is a flowchart of the calculation, FIG. 8 is a flowchart of the calculation of JW, and FIG. 9 is a flowchart of the determination of the learning level.

In step 1 (Sl), the coupling strength Wji in the RAMI is initialized to a random value. Input value O, N in step 2
+1 is set in the RAM 4, and in steps 3J to 9, the arithmetic circuit 5 calculates unit output values O and K in order from the input layer to the output layer.

Next, in steps 11 to 20 of FIG. 7, the arithmetic circuit 6 calculates the backward propagation amount δpj0 of the error in the layer to obtain the output value 0pj0 and a desired output.

Next, in steps 21 to 24 of FIG. 8, the arithmetic circuit 7 calculates the cut-off value ΔpWj, K (0) of the coupling strength Δpwj,0, which is all 0, according to equation (8). In step 25, (9) (1), Wj1' by *,n circuit 8
(1) is found. Thereafter, these are stored in RAM1 to RAM4 in a form that updates the initial data.

Next, learn the middle layer. Returning to the flowchart in FIG. 7, O obtained above by the arithmetic circuit 6,
. The backward propagation amounts J δ and K of the error are determined using . Next, in the flowchart J of FIG.
The bond strength Wj, K(1) is determined according to equation (9). Similar to the output layer, the data obtained above is stored in RAM
1 to RAM4 in an updated form. The above flow is sequentially repeated toward the input layer (K - N +1), and the first learning is completed.

By performing the above learning a plurality of times (n), the strength of the coupling between each unit Wjl is determined, and when an input value O indicating a certain input pattern P is given, 1) J desired output value P The network for obtaining , is self-G)
J will be formed dynamically.

Figure 9 shows the actual output value O, the teacher signal t, and 9J.
l) Mean square error E of j
It is a flowchart for calculating p. The smaller this value is, the closer the actual output value will be to the desired output value. Now E
If p is smaller than a certain threshold ε, learning is terminated, and if p is larger than ε, learning is repeated.

Although learning for one input cover turn P has been described above, it is also possible to perform learning in which a plurality of input cover turns are used and a plurality of output cover turns corresponding to each pattern are obtained. It is also possible to learn to output one specific output pattern for a plurality of input patterns.

The BP model described above can be realized with the Neumann-type microcomputers that are currently widely used in consumer devices, but as it is, one of the major strengths of neurocomputers, which is the ability to speed up through parallel processing, will not be utilized. . Therefore, it is preferable that the processes shown in FIGS. 6 to 9 be performed in parallel by a plurality of computers.

FIG. 10 shows the configuration of a parallel processing system for this purpose.

A plurality of microprocessors P1 to P□ are connected to host processor 11. The neural network shown in FIG. 2 is divided into n partial networks, and each is assigned to a microprocessor P1.P. The host processor 1 controls the mutual timing of the microprocessors P1 to Po, integrates data distributed among the microprocessors P1 to P, and performs processing such as pattern recognition. Each of the microprocessors P1 to Pn performs arithmetic operations on a plurality of consecutive columns of output values O shown in FIG. 5 according to the arithmetic procedure described above. Therefore, the microprocessors P1 to P are t
δpj necessary to calculate the output value corresponding to u.

It is equipped with a RAM for storing ΔWjl''jl and an arithmetic circuit. When the calculation of the output values of all the units in charge is completed, the processors P1 to P are synchronized and are updated to update the data. Communication is performed.The host processor 1 determines the learning achievement level and controls the mutual timing of the microprocessors Pl=Pn.

When performing processing such as pattern recognition based on the learned results, the final required output value P is calculated by performing output calculations from the input layer shown in FIG. is required. In this case as well, speeding up can be achieved due to the parallelism of the neural network by executing distributed processing using microprocessors with the number of stages as shown in FIG.

Although the circuit shown in FIG. 5 is basically required in the learning process, the configuration can be greatly simplified if only the learning results are applied.

FIG. 11 shows the basic circuit configuration in this case. The input data is inputted to the input unit 12 (for example, an A/D converter, etc.) to obtain output data 0 by sequentially performing calculations. , or a rewritable ROM.

FIG. 12 is a schematic block diagram of a learning system during manufacturing for products to which learning results are applied. The product 16 has a built-in ROM 17l that stores the coupling strengths W, , K. 18 is a learning device, and the combination of ROM 17 and learning device 18 is basically the same as the device shown in FIG. Device 18 is separated. Note that it is not necessary to perform learning each time for each product of the same type, so it is also possible to copy and use the ROM 17.

Next, referring to the drawings, the neuron computer 10
The operation of a strobe control device using 9 will be explained.

First, a wide-angle lens 101 forms an image on a color area sensor 102 that covers a wider range than the lens field of view.
Extracts the brightness and color signals of the image. The bright/dark signal and color signal output by the color area sensor 102 are converted into digital signals by the A/D converter 107 and N Ci 0
9.

On the other hand, the I RED 1.03 is caused to emit light according to the light emission timing supplied from the main CPU 11, and the light beam is spread by the wide-angle lens 104 to the range of the image of the color area sensor 102. Then, the light beam reflected from the subject is focused by a wide-angle lens 105, and 1RED
PSD array 1 serves as a light receiving element for all 103 light beams.
061-. The distance signal photoelectrically converted by this PSD array 106 is inputted to an A/D converter 108, converted into digital data, and inputted to the NC 109. At this time, the main CPU 11 sends the data import signal to the NC
Output to l09.

The main CPU III also performs photometry from the camera body 150 to the camera body 150, an aperture value based on the Apl distance,
Input the shutter speed, lens f value, film sensitivity, and distance information, determine the illumination angle of the angle of view based on the lens f value,
NC109, aperture value of camera body 150, shutter speed 2, film sensitivity.

Outputs distance information and irradiation angle based on lens f value.

Further, as described above, the NC 109 converts the output of the color area sensor 102 into a digital value by A/D conversion '5107 and the output of the PSD array 106.
The signal converted into a digital value by the D converter 108 and the aperture value of the camera body 150, shutter speed, film sensitivity, distance information, and strobe illumination angle based on the lens f value are input from the main CPU 11.

NC109 has the same configuration as in FIG. 11, and the connection weight W
lj is stored in NC10 from the learning storage unit 110 when the power is turned on.
9 is loaded into the RAM. NC109 corrects the illumination angle using the connection weight Wlj.

The angle of the flash unit, the light amount value, and the exposure correction data of the camera body 150 are output to the main CPU 11. Main CPU
I11 is the output of NC109, that is, the correction of the irradiation angle,
The motor 114 and capacitor 5 are determined from the angle of the flash and the light intensity value.
W115 and further outputs exposure correction data to the camera body 150.

FIG. 13 shows a Wlj setting device of the learning storage section 110. A model pattern input device 122 inputs data when using a strobe to a learning neuron computer (hereinafter referred to as TNC) 121.

At the same time, the teacher signal 123 changes the originally desired value T
Input to NC121. The TNC 121 once stores the coupling coefficient of each layer in the built-in RAM through the above series of operations,
After the learning is completed, the RAM of the TNC 121 is transferred to the learning storage unit 110.
Transfer data. The learning storage unit 110 stores W, j of the learning storage unit
It can be separated from the setting device and used by attaching it to the strobe body.

Next, learning by model pattern input in the Wlj setting device of the record learning storage unit 110 will be explained. The digital value of the pattern is assumed to be O and J N+1 with the N+1 model as the state used for strobe photography. 01. is divided into each block and associated in an independent state, and outputs a series of values to the output layer. The neuron network is composed of three layers: an input layer, a middle layer, and an output layer, as shown in FIG. FIG. 15 shows two basic model patterns.

In FIG. 15(a), the angle of view information indicates that the main subject is at the center and the distance is 2 m, and the surroundings are at a distance beyond the reach of the strobe light. The distance information from the camera is 2 m, which is the same as the center subject. In the above case, the color rear sensor 102
The information includes the color of the sky and the color and brightness of the main subject, and the PSD array 106 for distance detection outputs 2 m for the central subject and 10 for the surroundings as distance information corresponding to the color area sensor 102.

Based on this information, NC109 reduces the beam angle by half (50
%, β-0,5) and lower the flash angle by 100 (
α--10). This gives strobe light only to the central subject.

In FIG. 15(b), the viewing angle information shows that the main subject is indoors, the ceiling is whitish and the main subject is close to the wall, and when the 1'i1 strobe light is emitted, a strange shadow appears in the background. The distance information from the camera is the same as the main subject.

The color area sensor 102 outputs color information of the ceiling, etc., and the PSD array 106 outputs distance information of the ceiling, distance information of the main subject, etc.

Based on this information, NC109 sets the flash angle to 90.
°Up (α-90@), double the light intensity (K-2), -0.5 correction as exposure compensation value to the camera (C--0.5)
Do this. This allows the light to circulate sufficiently and there are no strange shadows.

Next, the parameters corrected by the NC 109 will be explained with reference to FIG. 16.

β: Irradiation angle The range covering the lens focal length is β-1
shall be.

α; Flash angle 48≦α≦9 where horizontal direction is α−〇
0; Light amount 2 Normal is set to -1.

C: Exposure correction data on the camera side is usually C-0 or higher.Two examples have been shown, but in reality, several hundred patterns are used for model stamping. In Fig. 13, NC109 is T N
It has the same configuration as C121, and the strength of the connection is set by the learned learning storage data Wlj, and the color sensor 10
2 output, output of PSD array 106 and camera body 15
0 aperture value, shutter speed, film sensitivity, distance information,
Data on the irradiation angle of the strobe based on the lens f value is given as O, and 11J calculations are performed and the results are output from the output layer.

As explained above, the learning function of the neuron computer makes it possible to control even things such as strobe photography, which have a huge number of patterns and are difficult to formulate and program.

Additionally, the interpolation effect of the neuron computer makes it possible to output correct output even for patterns that were not input during learning.

Furthermore, the parallel nature of the neuron computer allows it to process a huge amount of oral input data at high speed, making it possible to provide a control device suitable for controlling strobes.

Although the neuron computer described above is realized in software, it can also be realized in hardware by realizing the coupling strength W s j with a resistance value or PLD, etc. be effective.

[Effects of the Invention] As explained below, the present invention provides a flash device that can accurately and quickly control the irradiation angle, the angle of the flash section, and the amount of light to optimal values without relying on human intuition. control device can be provided.

[Brief explanation of the drawing]

The figures show one embodiment of the present invention, in which Fig. 1 is a block diagram of a control device for a flash device, Fig. 2 is a diagram showing a model of a neurocomputer, and Fig. 3 is a model of each unit constituting a network. FIG. 4 is a diagram showing a sigmoid function, FIG. 5 is a block diagram of a neurocomputer, and FIGS. 6 to 9 are flowcharts when the neurocomputer shown in FIG. 5 is simulated with a Neumann type computer. , FIG. 6 is a flowchart for determining the output O of each unit, FIG. 7 is a flowchart for determining the amount of propagation in the J direction after the error, δ, FIG. 8 is a flowchart for determining the coupling strength coefficient Wj[, and FIG.
The figure is a flowchart for determining the learning level, No. 10.
The figures are: 1 is a block diagram of the column processing system, FIG. 11 is a block diagram of a device that applies learning results, and FIG. 12 is a block diagram of a system that trains a device that applies learning results.
Fig. FIG. 3 is a diagram for explaining parameters. 101... Optical device (wide-angle lens) 1o2... Image lunar photoelectric conversion element (color area sensor) 103...
Distance detection light emitting element (infrared light emitting element), 106...
Light-receiving element (position detection element), 109... Neuron computer, 111... Control means (main CPU). Fig. 1 Applicant's agent Patent attorney Junji Tsuboi A-ron Fig. net (sum of inputs) Fig. Fig. 9 Fig. 10 Microprocessor Fig. 10 Fig. 11 Fig. 12 Fig. 13 Fig. 14 figure

Claims (1)

[Scope of Claims] A plurality of image photoelectric conversion elements, an optical device for forming an image of a subject and its background on the image photoelectric conversion elements, a plurality of distance detection light projecting elements, and a distance detection light emitting element. a plurality of light-receiving elements that receive signals from the light-emitting element; an input layer consisting of a plurality of units into which outputs of the plurality of photoelectric conversion elements and the plurality of light-receiving elements are input; each unit of the input layer and a predetermined A neuron consisting of an intermediate layer consisting of multiple units interconnected with a connection strength of , and an output layer consisting of multiple units interconnected with each unit of this intermediate layer with a predetermined connection strength A control device for a flash device, comprising: a computer; and a control means for controlling the angle, amount of light, and irradiation angle of a flash section of the flash device in accordance with the output from the output layer of the neuron computer.