JP2865774B2 - Image processing device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus, and more particularly to an improvement in an image processing apparatus suitable for use in detecting the number of customers waiting for a group management elevator.

[Conventional technology]

2. Description of the Related Art Various methods have been proposed as devices for processing an image information to detect an object. For example, as a traffic flow measurement system for automobiles, the IEICE Transactions '85 / 3 Vol.J68-D No3 pp. 308-315 (Reference 1) contains images taken from above a highway. There is disclosed a technology for processing by an algorithm in which various constants (parameters) are determined.

As a device for measuring the flow of a person, various thresholds and weights used in an image processing algorithm are first determined by learning, as described in IE80-73 (1980.11) (Reference 2). An image processing apparatus that performs image processing by using is known.

Further, as disclosed in Japanese Patent Application Laid-Open No. 1-241667 (Document 3), when a neural network is used for a speech recognition technique, a speech pattern for which a correct answer is known is input and prepared in a learning mode. It is known to correct the load (coupling) coefficient in a direction to eliminate the difference from the correct answer.

[Problems to be solved by the invention]

In each of the above prior arts, a large number of input patterns for which the correct answer to the processing result is clear are prepared, and these are repeatedly input, and learning is performed so that the correct answer can be obtained. Therefore, even if a correct answer can be obtained for a prepared input pattern, it cannot follow an environment in which input information can be obtained when an unpredictable change occurs at first.

For example, there is an image processing apparatus that cannot follow the change of the background, which will be described later, and produces an erroneous output.

SUMMARY OF THE INVENTION It is an object of the present invention to provide an image processing apparatus that can obtain an accurate processing output while following an environmental change in an image information object in an actual operation state.

It is another object of the present invention to accurately grasp the number of customers in a group management elevator and perform group management control with high efficiency for calls.

[Means for solving the problem]

One aspect of the present invention includes means for generating teacher information for an image processing result for an image to be processed in the image processing means, and means for correcting the image processing algorithm or its parameters based on the teacher information.

According to another aspect of the present invention, a means for controlling a controlled object using an output of an image processing means, a means for generating teacher information for an output of the image processing means, an output of the image processing means and the control Means for inputting information from the means and generating a start command to the teacher information generating means in response to the fact that they have an abnormal relationship; and a processing algorithm of the image processing means or a processing algorithm based on the teacher information. Means are provided for modifying parameters used in the algorithm.

Here, the parameters collectively refer to constants, threshold values, weighting factors, and the like used in the algorithm.

[Action]

The image processing means processes the input image information and outputs a result of the processing to a control device that needs it.

As the teacher information generating means, the same image information as the image information to be processed by the image processing means is input and processed, and more accurate image processing means or information other than the image is input and the same kind of information is input by another completely different principle. Detection means for obtaining an output may be used. If necessary, compare the output of the image processing means and the output of the teacher information generating means, for example, based on the teacher information when the comparison result is inconsistent,
It operates to modify the algorithm of the image processing means or parameters such as various thresholds and coefficients used in the algorithm.

In general, in an image processing apparatus that requires simplicity and high-speed processing, changes in the environment of the image information object, such as changes in lighting, presence or absence of external light, changes in background, and seasons when used for detecting the number of people. An error is likely to occur due to a change in clothes and a degree of congestion due to the change. On the other hand, in the case of performing high-level image processing that does not require high-speed processing and performs high-level processing, errors are unlikely to occur even if the above-described environmental change occurs.

Therefore, if the former is used as the image processing means used online and the latter is used as the teacher information generating means,
The processing algorithm of the image processing means can be grown online, and more accurate image processing can be performed.

Other objects, configurations and operations of the present invention will be clarified by the following embodiments.

〔Example〕

First, the basic concept configuration of the present invention will be described with reference to FIG.

Image information captured by the plurality of image capturing units 111 to 11n is processed by the image processing devices 121 to 12n, respectively, and the results a1 to an are input to the control device 14 via the transmission path 13. This control device
Reference numeral 14 denotes, for example, an elevator group management control device described later.

Further, the information a1 to an and the information 14a indicating the operation status of the control device 14 are sent to the suspicious information detecting means 151 of the teacher information generating unit 15.
To enter.

Output 151a of suspicious information detecting means 151 and control device 14
The output information 14a of the teacher information generator 1 of the teacher information generator 15
52, and output information 152 of the teacher information generating means 152.
a is input to the image processing devices 121 to 12n.

On the other hand, the information a1 to an, the output 16a of the illuminance detection means 16 and the output 17a of the time output means 17 are input to the situation determination means 18, and the output 18a of the situation determination means 18 is input to the image processing apparatus 1.
Enter 21 to 12n.

 Hereinafter, the operation will be described using the image processing device 121 as an example.

The image information b1 captured by the imaging unit 111 is stored in the image processing device 1
The image data is input to 21 image processing means 1211 and processed. In this process, data stored in the coefficient table 1212 is used to convert the data into information on features such as the number and shape of detected objects or processing information subjected to a process such as binarization. Using the processing information a1, the control device 14 optimally controls the controlled object.

On the other hand, the control device 14 outputs information 14a indicating the current status of the control target. The processing information a1 of the image processing unit 1211 is input to the suspicious information detecting unit 151, and is compared with the information 14a indicating the status of the control target to monitor the information.

That is, the suspicious information detection unit 151 determines whether or not the processing result of the image processing unit 1211 matches the state of the control target. If the result is inconsistent, the suspicious information detecting means 151 includes processing information a1 of the image processing means 1211, a signal indicating that the information a1 is suspicious, and
The information 151a including the image information b1 that is the source of a1 is output to the teacher information generating means 152.

The teacher information generating means 152 analyzes the suspicious information a1 and the image information b1 included in the information 151a in detail, and outputs accurate processing information (hereinafter referred to as teacher information) corresponding to the image information b1 captured by the imaging means 111. Call) 152a. This teacher information 15
2a is output to the learning means 1213 of the image processing device 121.

The learning unit 1213 uses the teacher information 152a to determine an image processing algorithm used in the image processing unit 121 or correct values of various thresholds and weighting coefficients used in the algorithm, and a coefficient table storing these coefficients. 12
Modify the contents of 12.

In addition, the teacher information generating means 152 includes a suspicious information detecting means 151.
According to the output information 151a, the teacher information 152a can be obtained from the information 14a indicating the control status of the control device 14 in some cases (described later).

The situation determination unit 18 determines the environment of the imaging area to which the imaging unit 111 is aimed, based on the output information a1 of the image processing unit 1211, the output information 16a of the illuminance detection unit 16 or the output information 17a of the time output unit 17, and the coefficient table 1212 Coefficient selection information for
Outputs a. The operation of the situation determining means 18 will be described with reference to FIG.

FIG. 2 is a detailed configuration diagram of a main part centering on the image processing device 121 in FIG. The image processing apparatus 121 is provided with a plurality of coefficient tables 191, 192,.
Based on the output information 18a, the most suitable one is selected from the coefficient table.

Here, the situation determination unit 18 determines, from the output information a1 of the image processing unit 1211, rough information, for example, whether the ratio of the detected object occupying the screen is large or small, or the output information 16a of the illuminance detection unit 16. , The lighting (including extraneous light) is bright or dark, or the night or daytime, and the season is summer or winter, etc., are determined based on the output information 17a of the time output means 17, and the environmental information is matched. It outputs coefficient selection information 18a for selecting the obtained coefficient table.

The learning means 1213 corrects the stored contents of the currently used coefficient table or the stored contents of all the coefficient tables for the common coefficients based on the teacher information 152a.

Hereinafter, an embodiment in which the image processing apparatus according to the present invention is applied to a group management elevator will be described with reference to FIG.

In this embodiment, a description will be given of an elevator group management control system that integrally controls four elevators.

The group management control device 20, the number control devices 211 to 214 for controlling each elevator, the display control devices 221 to 224, and the image control device 23 are connected to the serial transmission line 24 in a multi-drop manner. Standard input / output devices such as call buttons 261 to 264 installed in the halls are connected to one of the serial transmission lines 271 to 274 connected to the control devices 211 to 214 of each elevator. The other serial transmission line 251-254 has an operation panel 29 in a car 281-284.
Connect 1 to 294 respectively.

Further, the transmission path 30 connected to the display control devices 221 to 224
LED displays 311 installed on the n floor and m floor of the building
Connect ~ 318.

Similarly, a serial transmission path connected to the image
321 to 324 are image processing devices 331 installed on the nth and m floors.
Connect ~ 338. Imaging means 341 to 348 are connected to the image processing devices 331 to 338, respectively. Further, the other serial transmission lines 325 are connected to image processing devices 339 to 3312 installed in each car. Imaging means 349 to 3412 are connected to the image processing devices 339 to 3312, respectively. Reference numerals 351 to 354 denote load detecting means provided in each car.

 The operation in the above configuration will be described below.

The group management control device 20 includes a call button 261-2 installed in the hall.
64 and the information of the operation panels 291 to 294 in the car
Receive via 1-214. Further, the group management control device 20 includes:
The number of waiting passengers detected by the image processing devices 331 to 338 provided on the nth and m floors and the number of passengers in the car detected by the image processing devices 339 to 3312 provided on the car are received via the image control device 23. Elevator operation control is performed based on the information.
The display control devices 221 to 224 are the display units 311 to 311 on the n floor and the m floor.
Output guidance information to 18 to improve service for waiting customers.

The image processing devices 331 to 3312 perform simple and rough processing for the sake of simplification of processing speed, transmission speed and software.

The unit control devices 211 to 214 control the operation of the elevator based on the command from the group management control device 20 and control the input / output devices installed on each floor and in the car.

FIG. 4 is a detailed configuration diagram of a main part centering on the image control device 23 in FIG.

The image processing device 331 includes an image input unit 3311, an image processing unit 3312,
It comprises a transmission processing unit 3313. The image input unit 3311 includes the imaging unit 34
The image information (analog amount) output from 1 is converted into a digital amount. The image processing unit 3312 detects the number of waiting customers staying in the hall based on the digital amount. That is, from the relationship between the background image fetched at the time when no person is present and the image information fetched at an arbitrary time thereafter, a process of calculating the number of people existing in the image information is performed. This process requires high speed because it is used for on-line control, and can be said to be a rough process.

The transmission processing unit 3313 transmits the waiting customer information obtained as described above to the image management device 23 according to a predetermined format. This image control device 23 corresponds to the teacher information generating unit 15 in FIG. This image control device receives the number of people information from each of the image processing devices 331 to 3312 via the transmission processing units 321 to 235. The image processing unit 236 arranges the received information and transmits the information to the group management control device 20 via the transmission processing unit 237. As a result, the group management control device 20
Appropriate group management control is performed based on the number of customers waiting on the floor and the m-th floor hall and whether the number of customers in the car is taken into account. On the other hand, if there is suspicious information, the image information used for the processing is obtained from a memory (not shown) in the corresponding image processing apparatus, and the image processing unit 236 performs detailed processing. The accurate information obtained as a result, that is, teacher information, is transmitted to the group management control device 20 and the image processing device that has output the suspicious information. For the detailed processing algorithm, for example, the one described in the above-mentioned document 2 is used, and the image information obtained from the memory in the image processing apparatus is subjected to time-axis noise removal and spatial-axis noise removal. By performing such detailed preprocessing, the number of target persons can be accurately detected.

The image processing apparatus that outputs the suspicious information corrects the parameters used in the processing, that is, various thresholds and load (coupling) coefficients, with the accurate information (teacher information) so as to reduce the deviation from the detection result, thereby improving detection accuracy. improves. Details will be described later.

FIG. 5 shows an installation state of the image processing apparatus in the present embodiment.

The imaging means 341 is installed on the ceiling of the elevator hall on the nth floor, the image processing device 331 is installed in the hoistway on which the car 281 moves, and both are connected by a video signal cable. Further, the image processing device 331 is connected to a transmission line 321 laid in a hoistway, and the image
To transmit information.

In such an installation situation, for example, a hall waiter
A method for detecting 361 to 363 will be described.

FIG. 6 is an image photographed by the imaging device 341 and is a screen 3
Reference numeral 7 denotes an arrangement of 256 unit pixels 371 in the horizontal direction and 240 pixels in the vertical direction.

Although the number of pixels differs depending on the imaging device, the description will be made using the resolution of the image input unit 3311, that is, 256 × 240 pixels. In the figure, the background is gray
Assume that the binary number represents 128, the head is black, for example, zero, and the clothing is white, for example, a binary number representing 255.

 FIG. 7 shows the processing of the image processing device 331.

The image processing device 331 uses the background image g (x, y) captured at the time when there is no person at the time of activation and the image information f (x, y) captured at an arbitrary time, and first calculates a difference represented by the following equation. Find the image h (x, y).

h (x, y) = | f (x, y) -g (x, y) | As a result, the background portion is black and the object is gray. Next, this image is compared with an appropriate threshold value B to obtain a binary image i (x, y) represented by the following equation. As a result, the background portion is black and the object is white.

 area of h (x, y) B i (x, y) = 255 white area h (x, y) <B area i (x, y) = 0 black where x is 0 to 255, y is 0-239.

Obtains the binary image from the occupied object area (number of pixels) S _k white number of pixels, which is divided by the number of pixels S of per the following equation, obtains the number N _m.

For example, as shown in FIG. 7, if the area corresponding to the waiting customer 361 is S ₁ , the waiting customer 362 is S ₂ , and the waiting customer 363 is S ₃ , n in the above equation is 3.

Note that the number of passengers is similarly obtained for the cars 281 to 284.

FIG. 8 is a flowchart showing the processing of the image control device 23 described above.

That is, each of the image processing devices 331 to 33 is performed by step 400.
The information on the waiting passengers and passengers is collected from 12, and organized and stored in the transmission memory (not shown) of the transmission processing unit 237 shown in FIG. In step 401, the above information is compared with the current elevator operation status to determine whether there is any suspicious information. As a result, if there is suspicious information, first, at step 402, a message is set to the effect that the information of the image processing apparatus corresponding to the transmission memory of the transmission processing unit 237 is suspicious. Subsequently, image information is obtained from the corresponding image processing apparatus in step 403, and this image information is processed in detail in step 404, and the number of waiting passengers or passengers is accurately obtained.

In step 405, the information of the corresponding image processing apparatus collected in step 400 is compared with the teacher information obtained as a result of step 404. If the results do not match, the result of step 404 is transmitted to the transmission processing units 237 and 2 by step 406.
The data is stored in transmission memories (not shown) corresponding to 31 to 235. Thus, accurate information is transmitted to the group management control device 20, and parameters such as thresholds of the corresponding image processing device are corrected.

Here, an example of a procedure for correcting a threshold value used for an image processing algorithm in the image processing device 331 will be described with reference to FIG.

Normally, the image processing unit 3312 performs a process of converting image information input from the imaging device 341 into a digital amount by the image input unit 3311 and storing the digital amount in the image information storage unit 41. Based on the image information f (x, y) and the background image information g (x, y) stored in the background image storage unit 42 in advance, the arithmetic unit 43 calculates the difference image h
Find (x, y). The comparison means 44 binarizes the result with an appropriate binarization level B to obtain a binary image i (x, y). The binary image i (x, y) by measuring the area S _k occupied by the object by the arithmetic unit 45, the arithmetic unit 46 this, divided by the per capita area S, seeking waiting passengers number N _m. Therefore, as the adjustment elements (parameters) in the present embodiment, the binarized bell B and the area S per person can be considered.

Therefore, performing wait arrivals detection result to, determining that the image integration unit 23 is erroneous, when come to transmit the instruction information N _s, the image processing section 3312 utilizes the idle time of waiting passengers detection learning . That is, a deviation of the waiting passengers number N _m and teacher information N _s by deviation detecting means 471 of the correction amount in the arithmetic unit 47. If the deviation Δn is positive (negative), the correction amount calculating means 472 calculates the correction amount in the direction of increasing (decreasing) the binarization level B or increasing (decreasing) the area S per person. . The deviation Δn is expressed by equation (1). Based on this deviation Δn, Δn = N _m −N _s ... (1) The adjustment ΔB or ΔS obtained by the setting means 54 or 55 is set by the setting means 54 or 55. It subtracts (or adds) from the initial value B _o or S _o that has been set. This value respectively stored in the storage means 48 and 49 and held until the next instruction information N _s is input. The method of correcting the two thresholds B and S from the deviation Δn described above complicates the distribution of the correction amounts of both. Therefore, another procedure for simply correcting the two thresholds will be described.

First, the area S per person is fixed, and the binarization level B is adjusted. In other words, the number of waiting customers N
S _k0 necessary for matching _m with the teacher information N _s is obtained by equation (2). Then, the white area S _k after binarization becomes S _k0
S _k0 = N _s · S ₀ (2) The binarization process is repeated while changing the adjustment ΔB until the value becomes close to.
As a result, when S _k closest to S _k0 is obtained, the binarization level B is fixed, and the area S per person is adjusted. That is, the arithmetic unit 472 and the area S per capita required to match the waiting passengers number N _m the teacher information N _s determined by equation (3), determining the adjustment amount ΔS by addition (4). The adjustment is performed in this manner.

In the present embodiment, the binarization level is adjusted first, but the area per person may be adjusted first, or only one of the binarization level or the area per person may be adjusted.

According to the present embodiment, the binarization level and the area per person can be adjusted to the optimum values.
It can be improved to detect accuracy close to N _s.

In the image processing apparatus for detecting the number of persons in this embodiment, the higher the accuracy of the image processing is, the more desirable the output is used for group management control. However, it takes time to perform advanced, that is, detailed image processing, and it is not possible to output a detailed image processing result in a required short time. Therefore,
For example, the image processing apparatuses 331 to 3 shown in FIG.
As described above, the image processing apparatus 312 must be an image processing apparatus that performs coarse and quick image processing. Therefore, for example, an image processing device for performing detailed image processing is prepared in the image processing device 236 (constituting the teacher information generating unit) in FIG. 4 where quick image processing is not required, and the idle time of waiting customer detection is used. Thus, learning based on the teacher information is performed so that the image processing result of the coarse image processing means approaches a correct value even a little, and the parameters of the image processing are corrected.

With reference to FIG. 10, another embodiment of the procedure for correcting the threshold value will be described.

In the embodiment shown in FIG.
Because the revision was not taken into account at the time and the distribution of multiple revisions, the revision could not be optimized.

In this embodiment, in addition to the above-described embodiment,
A means 473 for storing N _m and teacher information N _{s corresponding} thereto is provided, and after the input / output characteristics of the image processing device 331 are grasped, the correction is performed to optimize the correction.

First, the storage unit 473 stores any time and waiting passengers information N _m outputted from the image processing unit 3312, the instruction information N _s which is input when the waiting passengers information N _m is determined to be abnormal.

Next, the correction means 472 learns what difference the stored information has with respect to the correct answer (teacher information). As a result, it is possible to grasp the input / output characteristics of the image processing unit 3312, that is, the image processing unit 3312 itself.

FIG. 11 is a graph showing an example of the input / output characteristics thus obtained. For example, assuming that the angle of view of the imaging means is 120 ° and the installation position is 3 m above the floor, a maximum of 25 people (including the head)
To the extent detectable.

Therefore, ideally, as shown by the solid line 50 in the drawing, it is desirable that the waiter information for the teacher information be linear from 0 to 25.

However, even if the coefficients and the like are actually optimized, when the number of waiting customers increases, an overlap occurs on the screen, so that an error occurs as shown by the dashed line 51 in the figure.

Also, when a clear difference image is easily obtained due to the background color or the lighting condition at the site, as shown by a dotted line 52, the output is more than the dashed-dotted line 51, and when it becomes an unclear difference image, it is illustrated. As shown by the dotted line 53, the output is made smaller.

Therefore, after learning the input and output characteristics of the image processing apparatus by the instruction information N _s in the field, the inflection point from the result, namely obtaining the number of occurrence of overlap.

The teacher information N _s input after finishing the above processing, to correct. That is, first, when the teacher information N _s the following inflection point number, make corrections in the same manner as in Example to determine the optimum binarization level B and the area S ₁ of the per. As a result, the corrected input / output characteristics become characteristics close to the one-dot chain line 51.

Then, when the teacher information N _s is equal to or greater than the inflection point number, find the optimum value S ₂ of area per person. As a result, input / output characteristics close to the solid line 50 can be obtained.

As a result of the above processing, one binarization level B and two areas S ₁ and S ₂ per person are obtained. With this, first,
Performs processing using a binarization level B and the area S ₁ of the per, if this result is above the inflection point number, reprocessing replacing area per the S _2.

According to the present embodiment, since the optimum correction can be performed with a small number of times, the detection result does not become unstable (the correction is not repeated).

 FIG. 12 shows another embodiment of the correction procedure.

This embodiment is different from the embodiment shown in FIGS. 9 and 10 in that the correction amount calculating section 47 is constituted only by the storage means 473. The storage means 473 includes a removable external storage medium, for example, an IC card 474.

With such a configuration, for example, a person refers to the contents of the non-volatile memory (IC card) 474 and makes corrections at another location in the manner shown in FIG. The result is stored in another area of the non-volatile memory, and is mounted on the device again.

As a result, the corrected contents correspond to ΔB and ΔS shown.

According to the present embodiment, since a person makes a troublesome correction, more accurate correction can be performed and the apparatus can be simplified.

Another embodiment of the teacher information generating means according to the present invention will be described with reference to FIGS. In this embodiment, the algorithms and parameters of the image processing devices 331 to 3312 are corrected using the outputs of the in-car load detectors 351 to 354.

Below the floor of the elevator car, a differential transformer type load detector for detecting the weight of the ride is installed, and the elevator control device uses this output to perform control when the elevator is full. Incidentally, the output accuracy of the load detector in terms of the number of people is said to be ± 0.1, and can be sufficiently used as the teacher information generating means in the present invention.

Now, in order to adjust the image processing devices 331 to 338 installed in the hall based on the outputs of the car load detectors 351 to 354,
You only need to know how many people were in the hall, in other words how many people got into the basket.

For example, as shown in FIG. 5, it is assumed that there are two passengers in the car and three waiters in the hall on the nth floor, and one person gets off the nth floor and gets on three people. Load detector 3 provided in basket 281
As shown in FIG. 13, the output of 51 changes from two persons' state a → one person getting off b (or three persons getting on c) → three person getting off d (or one person getting off e) → four persons getting f. (Usually, priority for passengers).
By extracting the period d (or c) having a positive slope from this figure, the number of passengers, that is, the number of hall waiters until arrival at the car, can be known. As described above, when attention is paid to the period T from when the elevator arrives at the n-th floor hall to when the elevator departs, it is possible to reliably detect the n-th floor hall waiter before that by detecting the load in the car. However, since the timing of this detection is late, it cannot be used directly for group management control. It is useful as a means for generating teacher information.

A process flowchart of the group management control device 20 for using the number of passengers thus obtained as teacher information will be described with reference to FIG.

When it is confirmed that the elevator has arrived at the n-th floor hall by step 600, detection of the number of passengers is started by step 601 and step 601 is continued until the door is confirmed to be closed by step 602. Thereafter, when it is confirmed that the door is closed in step 602, the above-mentioned operation is performed in step 603 (FIG. 13).
The detection result of the image processing device 331 provided in the hall is compared with the number of passengers obtained as described above. If the results match, the process ends, and no teacher information is output. But,
If the results of the comparison do not match, it is confirmed in step 604 that no hall call has reoccurred, and that no people remain in the hall. Output to

According to the present embodiment, since the existing load detector is used in the elevator car as the teacher information output means, it is not necessary to newly provide the teacher information output means, and it has a proven track record in terms of accuracy and reliability. Teacher information can be obtained.

Note that the image processing devices 339 to 3312 provided in the car can detect passengers in the car with the door of the car closed and have high accuracy because the environment is always constant. Therefore, it can be used as teacher information generating means instead of the load detector.

Referring to FIG. 15, another embodiment relating to the detection of a waiting customer in an elevator hall will be described. This embodiment is an image processing unit.
This is a case where the 3312 is configured by a neural net.

A plurality of input data 61a output from the image input unit 3311
Image processing unit 3312 consisting of ~ 61 l
To detect the number of waiting customers. The details will be described below.

Input layer Niyurarunetsuto, composed of three layers of the intermediate layer and the output layer, further, 12 Quai of Niyuron the input layer n ₁₁ ~n _112, the intermediate layer 6 Ke of Niyuron of n ₂₁ ~n _26, the output layer n
₃₁ composed of Niyuron the two positions of ~n _32. The output of each neuron of the input layer is connected to all the neurons of the intermediate layer, and the output of each of the neurons of the intermediate layer is connected to the neurons of all the output layers. At this time, a value obtained by multiplying the output of each neuron of the input layer and the intermediate layer by the indicated load (coupling) coefficient K _{11 to} K ₈₆ (−∞ to + ∞) is input to each neuron of the intermediate layer and the output layer. Is done. For example, the middle class Niuron n
_The value _Un21 input to ₂₁ is obtained by equation (5).

Here, V _n1i is the output of Niyuron n ₁₁ ~n ₁₁₂ of the input layer (0-1).

This value is substituted into the expression (6), and 0 to
A value of one to the input to each Niyuron n ₃₁ ~n ₃₂ of the output layer as output V _n21 of the Niyuron n _21. In the output layer,
Niyuron n ₃₁ and n ₃₂ in the same manner as in Niyuron n ₂₁
Find the output value of.

These output V _n31 and V _n32, calculates the waiting passengers number N _m (7) below, and outputs.

_{N m = 100 · V n31 +10} · V n32 ...... (7) As a result, for example, V _n31 = 0.1, if the _{V n32 = 0.2, N m =} 1
0 + 2 = 12 a [person], i.e., the case of the present example, Niyuron n ₃₁ is position 10, Niyuron n ₃₂ were as indicating one's place.

This result is transmitted by the transmission processing unit 3313 to the image
And the teacher information generating means 15 therein (the image processing unit shown in FIG. 4).
236). Further, it is also transmitted to the group management control device 20 via the image generalization device 23.

Further, input data 61a to 6a input to the image processing unit 3312
1l may be various feature amounts extracted from the image, or data on pixels of the image may be directly used. For example, the 16th
As shown in the figure, one method is to divide the screen 37 into twelve blocks 701 to 712, and use a value obtained by normalizing the area of the object determined by each block by the equation (8).

Here, V _a : the value of the input data 61 a N: the number of horizontal pixels (256) M: the number of vertical pixels (240) S ₁ : the area (number of pixels) of the object in the block 701.

In this embodiment, the weight (coupling) coefficient is
When adjusting the K ₁₁ ~K _86, it may be the same as the method performed generally known Niyurarunetsuto learning. That is, teacher information is input in the opposite direction to the output layers niron n _{31 to} n ₃₂ ,
A backpropagation method of adjusting each coefficient while propagating this in the reverse direction from the intermediate layer to the input layer is used. This method is a method known from the above-mentioned document 3 and the like.
As described in, the pattern to be learned is presented to the input layer of the neural network, the teacher signal to be outputted is presented to the output layer, and the teacher signal in the output layer is compared with the actual output value. A learning method for determining the coupling coefficient so as to reduce the difference can be adopted.

The teacher information generating means 15 in the image control device 23 is periodically triggered by a timer 56. Periodic learning is performed by a timer interrupt.

According to the present embodiment, by configuring the image processing unit with the neural net, it is possible to expect a high speed of the waiting customer detection process, and it is possible to automatically obtain an appropriate load coefficient by the back propagation method. In particular, if the learning based on the teacher information is performed periodically, an image processing apparatus that is expected to grow in a pattern-recognition manner and can follow changes in the surrounding environment can be obtained.

In the above embodiment, the correction of the image processing parameters is executed when there is a suspicion in the output of the image processing apparatus used online by comparing the teacher information and the output of the image processing apparatus, or a timer interrupt is performed. The ones that learn periodically are described. As described above, the image processing parameters may be corrected as necessary, such as when the necessity is recognized or periodically.

INDUSTRIAL APPLICABILITY The present invention is widely applicable not only to the above-described system for detecting a waiting person in an elevator hall, but also to a system for recognizing characters on a license plate of an automobile or a system for detecting the degree of congestion in an exhibition hall or the like.

〔The invention's effect〕

According to the present invention, the pattern is not a preset pattern,
By incorporating the teacher information generating means for the image processing result of the actual processed image, the processing program necessary for the image processing or the parameters (for example, various thresholds or weighting factors) used in the program can be set to the image processing target. It can be modified according to the situation of the thing. Therefore, for example, the detection accuracy of the image processing system placed in the actual operation state can be improved online.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the basic concept of an image processing apparatus according to the present invention, FIG. 2 is a detailed block diagram of a part thereof, and FIG. 3 is an overall view showing the image processing apparatus according to the present invention applied to a group control elevator. Configuration block diagram, FIG. 4 is a detailed configuration block diagram of a main part thereof, FIG. 5 is a side cross-sectional view for overviewing the arrangement of a waiting detector in an elevator hall, and FIG. 6 is a diagram showing an example of an imaging screen. , FIG. 7 is an explanatory diagram of an image processing procedure for obtaining the number of waiting guests from the imaging screen, FIG. 8 is a processing flowchart of an embodiment in the image control device, and FIG. 9 is a procedure of correcting a parameter on the image processing algorithm. FIG. 10 is a view for explaining an embodiment, FIG. 10 is a view for explaining another embodiment of a procedure for correcting a parameter on an image processing algorithm, FIG. 11 is a view for explaining the operation thereof, and FIG. Diagram explaining the thirteenth Fig. 14 is an explanatory diagram of the operation of means for generating teacher information for the number of hall waiters in an elevator using a load detector, Fig. 14 is a processing flow diagram, and Fig. 15 is a group of image processing devices using a neural network. FIG. 16 is a block diagram of a main part configuration showing the application of the present invention when used as a waiting means detecting means for a managed elevator. FIG. 16 is an explanatory diagram of image processing by a neural network.

──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Masachika Yamazaki 4026 Kuji-cho, Hitachi City, Ibaraki Prefecture Inside Hitachi, Ltd.Hitachi Research Laboratory (72) Inventor Naoto Onuma 4026 Kuji-cho, Hitachi City, Ibaraki Prefecture Hitachi Research, Ltd. In-house (72) Inventor Kiyoshi Nakamura 4026 Kuji-cho, Hitachi City, Ibaraki Prefecture Inside Hitachi, Ltd.Hitachi Research Laboratory Co., Ltd. (72) Inventor Yoshio Sakai 1070 Ichimo, Katsuta City, Ibaraki Prefecture Mito Plant, Hitachi Ltd. Person Kenji Yoneda 1070 Ma, Katsuta-shi, Ibaraki Pref., Mito Plant, Hitachi, Ltd. (72) Inventor Naofumi Nakata 1070 Mo, Ichimo, Katsuta-shi, Ibaraki Pref., Mito Plant, Hitachi, Ltd. 1070 Mo, Katsuta-shi, Japan Inside the Mito Plant of Hitachi Ltd. (56) References JP-A-2-95091 (J JP-A-1-211100 (JP, A) JP-A-64-53287 (JP, A) JP-A-58-177873 (JP, A) JP-A-58-95079 (JP, A) 49-67341 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) B66B 3/00 H04N 7/18 G06F 15/62 380

Claims (8)

(57) [Claims] An image processing apparatus comprising means for inputting image information, executing image processing by a predetermined algorithm, and outputting a processing result, wherein the input of the image processing means is independent of the image processing means. An image processing apparatus comprising: means for generating teacher information for the output regarding the same object as image information; and means for correcting the parameters of the image processing based on the teacher information. 2. An image processing apparatus according to claim 1, wherein said teacher information generating means has means for inputting the same image information as said image processing means and executing more detailed image processing than said image processing means. . 3. An image processing apparatus comprising first image processing means for inputting image information, executing image processing according to a predetermined algorithm, and outputting a processing result, wherein an image to be processed in said first image processing means is provided. A second image processing means for inputting the same image information and generating teacher information for the output of the first image processing means based on more detailed image processing than the first image processing means; An image processing apparatus comprising: means for correcting image processing parameters of the first image processing means based on teacher information output from the processing means. 4. A method according to claim 1, wherein said means for judging a condition including the brightness of the object to be processed and means for correcting said image processing parameters in response to an output of said judging means. Image processing device provided. 5. An image processing apparatus according to claim 1, wherein said correction means comprises means for selecting from a plurality of parameters prepared in advance. 6. An image processing apparatus comprising means for executing image processing for inputting image information and confirming the number of objects,
An image processing apparatus comprising: means for detecting the load of the object; and means for correcting the parameter of the image processing using the detected load as teacher information for the number of the objects. 7. An image processing apparatus comprising a plurality of image processing means for respectively inputting different image information, executing image processing, and outputting a processing result, comprising the steps of: selecting an image selected from the plurality of image processing means; Means for inputting an image to be processed input to the processing means, generating teacher information for the output of the image processing means, and means for correcting image processing parameters of the image processing means based on the teacher information Image processing device. 8. A plurality of elevator cars to be serviced between multiple floors, destination call means provided in these cars, and a car control for stopping the cars at least on the floor corresponding to the destination call. Means, hall call means provided at each floor landing, group management control means for determining at least a service car for this hall call, a plurality of imaging means provided at an arbitrary landing, respectively corresponding to these imaging means First image processing means for recognizing the number of waiting passengers at the landing, means for transmitting the number of waiting passengers output from these image processing means to the group management control means, and input to the plurality of imaging means. A second image processing means for inputting image information and recognizing the number of waiting passengers at the landing by an algorithm more detailed than the first image processing means; and a number of waiting persons of the second image processing means. Group management elevator controller having means for modifying the parameters of the image processing of the first image processing means a force as teacher data.