JPH11259625A - Number of persons detecting device and number of users detecting device for elevator - Google Patents

Abstract

PROBLEM TO BE SOLVED: To constitute the number of persons detecting device which can detect the correct number of persons even if the state of its monitoring area changes. SOLUTION: An input image from an image input device 1 is stored in a current image memory 2. Further, an input image in a no-person state is stored in a background image memory 6. A person part extracting means 7 compares the current image and the background image with each other to detect person's parts from their differences, a person's part area extracting means 8 finds the area of the person's part, and a person's part group number extracting means 9 finds the number of person's part groups. A person number counting means 10 determines the mean weight between the number of persons found from the person's part area and the number of persons found from the number of person's part groups and the number 11 of person's is calculated and outputted.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a device for detecting the number of people in a monitoring area from input image data obtained by using an image input device such as a CCD camera and a device for detecting the number of passengers of an elevator.

[0002]

2. Description of the Related Art FIG. 12 shows, for example, Japanese Patent Application No. 9-123983.
FIG. 1 is an overall configuration diagram showing a conventional number detection device shown in FIG. In the figure, 1 is a CCD camera, an artificial retinal chip,
An image input device such as an infrared camera captures an image of a monitoring area such as an elevator hall, and stores the input image in the current image memory 2 as a current image. The input image is stored in the original background image memory 4 as the original background image via the original background image updating means 3. The input image is stored in the background image memory 6 as a background image via the background image updating means 5, and is updated with the current image and the original background image.

[0006] The current image and the background image are compared by a person extracting means 7, and the number of people 11 is counted by a number counting means 10 and output. Then, when binarized by the human part extracting means 7, the threshold value is adjusted by the threshold value adjusting means 12 and the difference between the current image and the background image is extracted while automatically updating the background. Binarized, the size of the area of the differential binarized image,
Alternatively, the number of persons 11 is detected from the number of blocks or the average of both.

[0004]

In the above-described conventional number-of-people detecting device, the number of people 11 is detected from the size of the area of the difference binary image between the current image and the background image. In an indoor environment such as an elevator hall where indoor lighting is the center, when subtracting a background image from a current image, a shadow of a person may be detected in addition to a human part. Normally, in an indoor environment, the contrast of a shadow is lower than that of a person, so that the shadow can be separated by setting a higher threshold value for binarization.

However, in an environment where a large number of illuminating lamps are installed and thus shadows are generated in multiple directions, when the number of people increases and the shadows of other people overlap, the shadows of those portions become dark and the human part becomes dark. Falsely detected. Therefore, when the number of people increases, the difference area increases due to the overlapping of shadows,
The accuracy of detecting the number of people decreases. Also, when the number of people increases, the people are connected by a shadow, and the number of blocks in the difference image decreases. Also, when an average is taken from both, the reliability is inferior because each uncertainty element increases. There are problems such as.

The present invention has been made in order to solve the above-mentioned problem, and is intended to detect the correct number of persons even if the number of persons increases and the shadows overlap, thereby affecting the difference area and the number of blocks. It is an object of the present invention to provide a detection device and a device for detecting the number of passengers of an elevator.

[0007]

According to a first aspect of the present invention, there is provided an apparatus for detecting the number of persons, wherein the number of persons obtained from the area and the number of clumps of a human part image extracted from a difference between a current image and a background image is determined by: The weight is changed at the time of averaging the number of persons, and the number of persons is output.

Further, according to a second aspect of the present invention, in the first aspect of the present invention, in the first aspect, a weight for averaging is set according to the extracted area and the number of blocks, and the number is changed by changing the weight. Is output.

[0009] Further, the number detection device according to a third aspect of the present invention is the device according to the first aspect, wherein the number of people is extracted from a table indicating the number of people corresponding to the extracted area and the number of blocks, and is output. .

[0010] In the fourth aspect of the present invention, in the first aspect, the number of persons is obtained by fuzzy inference using a rule and a membership function obtained from the number of persons output from each person. It is like that.

Further, the number detection device according to a fifth aspect of the present invention is the device according to the first aspect, wherein the number of persons is obtained by using a neural network obtained by learning from the number of persons output by each person. It is.

A person detection apparatus according to a sixth aspect of the present invention estimates a difference detection area other than a human part resulting from overlapping of shadows of a person on an image plane, and corrects the shadow to correct the number of persons. Is to ask for.

[0013] Further, the number detection apparatus according to a seventh aspect of the present invention is the apparatus for detecting the number of persons according to the sixth aspect of the present invention, wherein the correction is performed using the following correction formula. X / a = (A / a + kB) / (1 + k) where A: difference area from background image a: unit area per person L: number of links k: area increase rate with respect to a per link X: True area

An elevator passenger number detecting apparatus according to an eighth aspect of the present invention is obtained by applying the number of passenger detecting apparatus according to any one of claims 1 to 7 to an apparatus for detecting the number of passengers using an elevator hall or a car as a monitoring area. It is.

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiment 1 1 to 5 are diagrams showing one embodiment of the first and second inventions of the present invention. FIG. 1 is an overall configuration diagram, FIG. 2 is an explanatory diagram of the influence of illumination, FIG. (A) is an actually measured difference image, (B) is an original difference image, FIG. 4 is a flowchart of the number of people detecting operation, and FIG. 5 is a weight setting diagram when the number of people is averaged.

In FIG. 1, reference numeral 1 denotes an image input device such as a CCD camera, an artificial retinal chip, an infrared camera, etc., 2 denotes a current image memory for storing signals from the image input device 1 as a current image, and 3 denotes a current image. Original background image updating means for updating an original background image, 4 is an original background image memory for storing an original background used when updating a background image, 5 is a background image updating means for updating a background image, and 6 is a background image. This is a background image memory to be executed.

Reference numeral 7 denotes a human part extracting means for extracting a human part using the current image and the background image; 8, a human part area extracting means for obtaining the area of the human part image extracted by the human part extracting means 7; The human part mass number extracting means 10 for obtaining the number of masses of the human part image extracted by the part extracting means 7, the area extracted by the human part area extracting means 8 and the number of masses extracted by the human body mass number extracting means 9 The number counting means 11 counts the number of persons from, and 11 is the detected number of persons.

In FIG. 2, 21 is an illumination lamp, 22 is a person,
23 is a shadow of the person 22, and 24 is an overlapping portion of the shadow 23, which is a dark shadow. In FIG. 3, reference numeral 31 denotes 2 of the shadow 23 of the human part.
The area at the time of binarization, 32, is the area of the connection portion sucked up at the time of binarization due to the overlap of the shadow 23. The measured difference in FIG. 3A is smaller than the original difference in FIG. The area has increased due to the overlap.

Next, the operation of this embodiment will be described with reference to FIGS. First, a switch (not shown)
To start the apparatus of FIG. 1 (“STA” in FIG. 4).
RT "). In step S1, the image input device 1 captures an image in the monitoring area. The captured image is stored in the current image memory 2. In step S2, the original background image updating means 3 does nothing when fetching a signal immediately after the apparatus is started,
The image in the current image memory 2 is passed to the original background image memory 4 as it is. After that, the image data of the original background image memory 4 is
The image data in the current image memory 2 is updated so as to approach the current image with a predetermined time delay.

In step S3, the background image updating means 5 does nothing at the time of signal capture immediately after the apparatus is started, and passes the image of the current image memory 2 to the background image memory 6 as it is. Thereafter, the image data in the background image memory 6 is updated using the image data in the current image memory 2 and the original background image memory 4.
Normally, the background image memory 6 is updated so as to approach the current image with a predetermined time delay.

Next, in step S4, the human part extracting means 7
The image data in the background image memory 6 is subtracted from the image data in the current image memory 2 to separate the human part from the background, and only the human part is extracted. Thereafter, binarization is performed using a predetermined threshold value. In the binarization, one or both of binarization using an absolute value for extracting the entire human part and black binarization for extracting a human head are executed.

In step S5, the human part area extracting means 8 obtains the value of the binarized area of the absolute value extracted by the human part extracting means 7. This can be realized by counting the number of pixels that have become black due to binarization. The area of binarization of black may be obtained. In step S6, the human body mass number extracting means 9
The number of binarized blocks of black extracted by the human part extracting means 7 is obtained. This can be realized by squeezing the blocks in order from the surroundings to obtain one point, and counting the number. Note that the number of bins for binarizing the absolute value may be obtained.

In step S7, the number counting means 10 may consider one of the two methods as the number 11 detected as it is in step S8. Seeking average. However, if the number of people is small, this is acceptable, but if the number of people is large, the following problems occur.

That is, in an indoor environment such as a normal elevator hall, a large number of illuminating lights 21 are installed on the ceiling, and the elevator waiters are also illuminated by this, and shadows 23 are generated in multiple directions. However, since the shadow 23 is extremely thin and is sufficiently smaller than the contrast of the human part when binarized by the human part extracting means 7, it is usually excluded by the binarization, and the area and the area of the human part are reduced. It does not affect the number of chunks. However, as the number of people increases, the individual shadows 23 in multiple directions overlap with each other, the density increases, and when binarized, the shadows are sucked up as a human part, which may cause a decrease in accuracy.

Although the brightness of the illuminating lamps 21 that cause the shadows 23 also affects the brightness, the illuminating lamps 21 on the ceiling are usually installed in large numbers of the same type in most cases. The effect is greater as the distance is closer to 22, and smaller as the distance is greater. That is, it is inversely proportional to the double of the distance from the characteristics of light. Therefore, the size of the shadow 23 is most affected by the illumination lamp 21 above the head and is small because the shadow 23 is illuminated from above. The shadow 23 generated by oblique light is thin.

Thus, it can be seen that the influence of the shadow 23 illuminated from above is large. That is, when the person 22 stands adjacent and the shadow 23 illuminated from above overlaps, the shadow 23 becomes darker and is taken up as a human part in binarization. Further, when the number of people is large, many shadows 23 overlap, and even though each shadow 23 is thin, the shadows 23 may be darkened as a total and may be sucked up as a human part in binarization.

Since the shadow 23 extracted in this manner is in contact with the foot of the person 22, the portion of the shadow 23 is also extracted by binarization, and as shown in FIG. I will. That is, when the person 22 is adjacent, the person part is sucked up as a person part, so that the human part area 31 often becomes large in binarization. Also, at that time, the number of lumps is reduced because the number of connected persons 22 is increased.

The number of lumps may be divided by one person depending on the color of the clothes of the person 22. In the case of a person who cuts clothes of the same color as the floor, the head and feet are observed to be separated due to the difference from the background image. Depending on how it is done,
The number of chunks may be 2 for one person.

The following can be understood from the above tendency. When the number of people increases or when the person 22 stands adjacent to each other, an overlapping portion 24 of the shadow 23 occurs, and the area is displayed relatively large. When the number of people increases or the person 22 stands adjacent to each other, an overlapping portion 24 of the shadow 23 occurs, and the number of blocks is displayed as a small number. The number of chunks may be divided and increased depending on the clothing color. When the person's part is divided by the clothing color, the area to be displayed is smaller because there is no clothing part to be displayed at the time of binarization.

From the above tendency, simply averaging the number of people from the area and the number of people from the number of lump, especially when the number of people increases and the overlapping portion 24 of the shadow 23 occurs,
It can be seen that highly accurate detection of the number of people is difficult to expect. Therefore, taking these features into account, the average weight of both is changed according to the size of the area and the size of the number of lumps to improve the accuracy of detecting the number of people. FIG. 5 shows the setting of the size of the area, the size of the number of blocks, and the average weight at that time.

Here, the area and the number of lumps are measured by setting a camera having a viewing angle of about 45 degrees on a ceiling having a height of 5 m. Larger than people. This number is appropriately changed according to the angle of view and the installation height of the camera lens. (1) When the area is small and the number of lumps is small, the numbers of both are close to each other, and both are counted normally. Therefore, the weights are averaged as equal. (2) When the area is small and the number of lumps is medium, the person 22 is in a state of being slightly separated, and the number of lumps is larger than in reality.
Therefore, the average is set with a little weight on the area side. (3) When the area is small and the number of lumps is large, the person 22 is in a separated state, and the number of lumps is larger than the actual one and is inaccurate. Therefore, weighting is performed on the area side and averaging is performed. (4) If the area is small and the number of lumps is small, the people 22 are in a dense state, and the number of lumps is smaller than the actual number. Also, shadow 23
, The area is slightly larger than the actual area of the human part. Therefore, the average is set with a little weight on the area side. (5) When the area is medium and the number of blocks is medium, although the area is slightly larger, the weights are equal and averaged because the numbers are close to each other. (6) If the area is large and the number of lumps is large, the person 22 is in a state of being slightly separated, and the number of lumps is larger than in reality. Therefore, the average is set with a little weight on the area side. (7) If the area is large and the number of lumps is small, the people 22 are in a dense state, and the number of lumps is considerably smaller than the actual number. Also,
The area is slightly larger than the actual area due to the influence of the shadow 23. Therefore, weighting is performed on the area side and averaging is performed.

(8) If the area is large and the number of blocks is medium,
The number of clumps is slightly smaller due to the density. Also, shadow 2
Due to the influence of 3, the area is slightly larger than the actual area of the human part. Therefore, the weights are averaged as equal. (9) When the area is large and the number of lumps is large, the number of lumps is slightly smaller and the area is slightly larger due to the influence of the shadow 23, but it is considered that the detection is almost normal. Therefore, the weights are averaged as equal.

Here, the method of weighting is, for example, "equal".
What is expressed is 5: 5, "We put a little weight"
What is expressed is 6: 4, and what is expressed as "weighting" is 8: 2. In this way, it is possible to appropriately weight the average of the area and the number of blocks, and it is possible to detect the number of persons according to each state.
This is a microcomputer "microcomputer"
Can be executed by simple program processing. In addition,
The weight ratio is merely an example, and may be changed to one suitable for the use environment or use conditions. It is more desirable to check the weight ratio by on-site measurement or the like, and to change the weight ratio to a more appropriate one, because more accurate measurement can be performed.

Embodiment 2 FIG. 6 is a diagram showing one embodiment of the first and third inventions of the present invention, and is a diagram showing the number of output persons with respect to the relationship between the area and the number of blocks. 1 to 4 are also used in the second embodiment. In the first embodiment, appropriate average weighting is performed based on the relationship between the area and the number of chunks. However, in the second embodiment, the number of output persons is one-to-one corresponding to the number of persons from the area and the number of persons from the number of chunks. Prepare the corresponding table,
The number of people is obtained by a table lookup method.

For example, depending on the installation location and environment of the camera,
If you want to slightly change the weighting ratio or perform an actual measurement experiment on site and want to detect the number of people using the value obtained there,
The table lookup method is also convenient. This is because the value may be directly rewritten in the corresponding column of the table. The table can be easily changed by using, for example, an EPROM microcomputer. The row direction in FIG. 6 shows the number of people from the area, and the column direction shows the number of people from the number of blocks.

Embodiment 3 FIGS. 7A and 7B are diagrams showing one embodiment of the first and fourth inventions of the present invention. FIG. 7A is a diagram showing rules of fuzzy inference, FIGS. 7B to 7D are membership function diagrams, and FIG. ) Is the area membership function,
(C) is a membership function of the number of chunks, and (D) is a membership function of the weight. 1 to 4 are also used in the second embodiment.

In the first embodiment, the average weight is determined from the relationship between the area and the number of blocks. As described above, for an event in which the tendency of the input and the output can be obtained well, fuzzy inference can be used by obtaining a rule and a membership function, and a more precise value can be obtained. Using FIG. 7, fuzzy inference of two inputs and one output is performed, and a weight corresponding to the area and the number of blocks is obtained. Of course, the output may be the number of people.

The fuzzy inference can be executed by setting a program on a microcomputer. Alternatively, the result of the fuzzy inference may be written in a table shown in FIG. 6 in advance and extracted by a table lookup method.

Embodiment 4 FIG. FIG. 8 is an explanatory diagram of a neural network showing one embodiment of the first and fifth aspects of the present invention. 1 to 4 are also used in the second embodiment. In the first embodiment, the weight at the time of averaging was determined from the relationship between the area and the number of blocks. As described above, in a relationship in which an input value and an output value are obtained, the output value is learned as a teacher signal using a neural network, a synapse coupling coefficient of the neural network is obtained, and a neural network is formed. A method of outputting the number of people using a network is also effective. of course,
The output may be a weight instead of the number of people.

FIG. 8 shows an example of a neural network. In this example, the input layer is composed of two neurons, one from the area and the number from the mass, the middle layer is composed of five neurons, and the output layer is composed of one neuron. Are connected by the synaptic connection coefficient obtained by The configuration of this neuron is stored in the microcomputer,
The number of people can be obtained by inputting the number of people based on the area and the number of people based on the number of blocks, and as an output.

In the case where the number is stored on the microcomputer, the relationship between the number of people from the area and the number of people from the number of blocks according to the installation environment of the area is learned online and stored in the RAM of the microcomputer. Change the synaptic coupling coefficient,
It is also possible to perform more accurate measurement. Further, when the RAM capacity of the microcomputer is small, it is possible to store the synapse coupling coefficient learned offline in advance in the ROM, or to use the table lookup method as described in the second embodiment. .

Embodiment 5 FIG. This is an embodiment of the sixth and seventh inventions of the present invention, and shares FIGS. In the case of a number detection device that calculates the number of people from the difference between the current image input from the image input device 1 and the automatically updated background image,
When the number of persons to be detected increases, the problem described in the first embodiment may occur.

That is, when the person 22 is adjacent, the shadow 23 overlaps and is sucked up as a person, so that the area of the person becomes large in binarization. Further, since the adjacent persons 22 are connected to each other, the number of blocks is reduced. Therefore, when the number of blocks is smaller than the number of people, a value larger than the actual area is determined. In the fifth embodiment, the above-described increase is inferred, and the area is corrected by the area correcting means to increase the accuracy of detecting the number of people.

FIG. 3A shows a state where three persons 22 are integrated. An inference correction formula for removing the area 32 increased by the overlap of the shadows 23 is obtained, and the influence of the increase is excluded. Assuming that the difference area from the background image is A, the unit area per person is a, the number of links (the number of areas 32) is L, and the area increase rate per unit area a per link is k, the true area X X = A−k · a · L (1)

X / a = A / a−k · L (2) when both sides of the equation (1) are divided by a. Here, X / a: true number of people (from area) A / a: number of people from area before correction, [true number of people] = [number of people from area before correction]-[increase per link] Rate] / [Number of links] (3)

When two persons are adjacent and integrated into one body, the number of connecting parts, that is, the number L of links is one. When three persons are adjacent and united, the number L of links becomes two. However, in the case of clustering, the number L of links is three. Here, for the purpose of simple modeling, we do not consider the case of a dumpling. In fact, in places such as elevator halls, people 22 unconsciously take distance from others,
It is rare for the person 22 to clump in a dumpling, and it is not necessary to consider much.

When four people are adjacent and united, the number of links L
Becomes 3. Considering this, the number of links when n people are adjacent can be n-1. 4 + 2 +
The number of links L when two people form two blocks is 2, 3 when the number of links is 2 and 5 people are 2 + 3, and the number of links L is 3. Considering this way, it can be seen that the number of links L is the difference between the number of people and the number of blocks. If the number of people is X / a and the number of blocks is B, the number of links L is: L = X / a-B (4)

From the equations (2) and (4), X / a = A / a−k · (X / a−B) (5) When both sides are arranged, X / a = (A / a + k) B) / (1 + k) (6), and if the area increase rate k is known, the true number of people (X / a)
Is found.

Next, it was confirmed through experiments that the value k of the number of pixels per link increases. Height 170c
m, a plurality of adults with medium flesh were photographed many times from 5 m above with an artificial retinal chip camera having a 45 ° angle of view lens, and the value of the pixel number increase rate k was determined. From the result, k =
0.7 was obtained. This value may vary depending on the angle of view of the camera, the installation height, and the number of pixels, but may be obtained through experiments or calculations. In a microcomputer program, it is only necessary to rewrite the coefficients, and it can be easily realized.

The link has been considered to be rare when the person 22 is in the form of a dumpling. However, in order to further improve the accuracy, it is inferred from the shape of the lump whether or not the link is a dumpling. Is effective even if the number of links L is added accordingly.

Embodiment 6 FIG. 9 and 10 are diagrams showing an embodiment of the eighth invention of the present invention. FIG. 9 is an explanatory diagram for detecting the number of passengers in an elevator hall, and FIG. 10 is an explanatory diagram for detecting the number of passengers in an elevator car. . 1 to 4
Are also used in the sixth embodiment.

In FIG. 9, reference numeral 41 denotes a person detection device installed on the ceiling of an elevator hall, which has been described in the first to fifth embodiments. 42 is a waiting customer, 43 is an integrated unit for each floor that collects signals from the plurality of people detecting devices 41 on each floor, 44 is an integrated unit for all floors that collects signals for all floors, and 45 is a group management device for elevators.

In FIG. 10, reference numeral 51 denotes a person detection device installed on the ceiling of a car 52 of an elevator, which has been described in the first to fifth embodiments. 53 is a passenger in the car 52, 54 is an all-car integration unit for collecting signals from all the cars 52, and 55 is an elevator group management device.

The waiting hall 42 and the car 52 of the elevator hall
The number of passengers 53 or both of them is measured by any of the methods described above, and the group management of the elevators is performed. For example, when there are many waiters 42 on a certain floor, two cars 52 are dispatched there, or when the inside of the car 52 is full, control is performed such as distributing another car, and the elevator user is controlled. The waiting time is shortened, and the unpleasant feeling due to unloading is reduced.

[0057]

As described above, according to the first aspect of the present invention, the number of persons determined from the area of the human part image extracted from the difference between the current image and the background image and the number of blocks are used.
According to the second invention, the weight at the time of averaging is changed according to the extracted area and the number of lumps, and the weight is changed and the number is output. Therefore, it is possible to weight the average according to the situation, and to obtain a highly accurate measurement result according to the situation.

Further, in the third invention, the number of persons is extracted from the table indicating the number of persons corresponding to the extracted area and the number of blocks, and is output, so that a highly accurate measurement result according to the situation can be obtained. Yes, and by rewriting the table, measurement suitable for the installation environment can be performed.

Further, in the fourth invention, rules and membership functions are created from the number of people extracted and the number of lump numbers to perform fuzzy inference. In the fifth invention, neural networks are learned and formed. Since the number of persons is obtained, accurate weighting can be performed, and highly accurate measurement can be performed.

Further, in the sixth invention, a difference detection area amount other than the human part resulting from the overlapping of the shadow of the person on the image plane is estimated, and the shadow part is corrected. Since the correction formula for correcting the portion is used, more accurate detection of the number of people can be performed.

In the eighth invention, the number-of-people detecting device is applied to the device for detecting the number of passengers by using the elevator hall or the inside of the car as a monitoring area, so that high-precision information on the number of passengers required for controlling the elevator is output. A detection device can be configured.

[Brief description of the drawings]

FIG. 1 is an overall configuration diagram showing a first embodiment of the present invention.

FIG. 2 is an explanatory diagram of an influence of illumination according to the first embodiment of the present invention.

3A and 3B are explanatory diagrams illustrating a difference image according to the first embodiment of the present invention, wherein FIG. 3A is an actual difference image, and FIG. 3B is an original difference image.

FIG. 4 is a flowchart of a number detection operation according to the first embodiment of the present invention.

FIG. 5 is a diagram showing a weight setting when the number of people is averaged according to the first embodiment of the present invention;

FIG. 6 is a diagram showing the number of output persons with respect to the relationship between the area and the number of blocks according to the second embodiment of the present invention.

FIG. 7 is a diagram showing a third embodiment of the present invention, wherein (A)
Is a diagram showing rules of fuzzy inference, (B) to (D) are membership function diagrams of fuzzy inference, (B) is a membership function of area, (C) is a membership function of lump count, and (D) Is the weight membership function.

FIG. 8 is a diagram illustrating a neural network according to a fourth embodiment of the present invention.

FIG. 9 is a diagram for explaining the detection of the number of waiting customers in an elevator hall according to the sixth embodiment of the present invention.

FIG. 10 is an explanatory view of detecting the number of passengers in a car of an elevator according to a sixth embodiment of the present invention.

FIG. 11 is an overall configuration diagram showing a conventional person detection device.

[Explanation of symbols]

Reference Signs List 1 image input device, 2 current image memory, 3 background image memory, 7 person section extracting means, 8 person section area extracting means, 9
Means for extracting the number of human masses, means for counting 10 people, 11 people, 2
1 Illumination lamp, 22 persons, 23 shadows, 24 shadow overlapping part, 31 shadow area, 32 shadow connection area, 41 person detection device, 42 waiting customers, 51 person detection device, 52
Basket, 53 passengers

Claims (8)

[Claims] The present invention compares a current image of a monitoring area input from an image input device with a background image of the monitoring area created in advance, and determines a size of a human part image and the number of clumps of the human part image from the difference. In a device for extracting and calculating the number of persons from the extracted area and the number of blocks, averaging the numbers, and outputting the number of persons, the weight at the time of the averaging is set, and the weight is calculated by the number of persons. A number detection device comprising a number counting means for changing and outputting the number of people. 2. The number-of-people counting means is configured to set a weight at the time of averaging in accordance with the extracted area and the number of chunks, and to change the weight to output the number of people. The number detection device according to claim 1. 3. The number detecting apparatus according to claim 1, wherein said number counting means is configured to extract and output the number of persons from a table indicating the number of persons corresponding to the extracted area and number of blocks. 4. The number-of-people counting means is configured to determine the number of persons by fuzzy inference using a rule and a membership function obtained from the number of persons output according to each number of persons. The number detection device described. 5. The number detecting apparatus according to claim 1, wherein said number counting means is configured to calculate the number of persons by using a neural network obtained from the number of persons output according to each number of persons. 6. A current image of the monitoring area input from the image input device is compared with a background image of the monitoring area created in advance, and the area of the human part image and the number of clumps of the human part image are determined from the difference. In a device that extracts and obtains the number of persons from each of the extracted area and the number of clumps and outputs the number of persons by averaging the numbers, a part other than the part of the person resulting from the overlapping of the shadow of the person on the image plane is used. Characterized in that it comprises an area correcting means for estimating the difference detection area amount and correcting the shadow portion to obtain the number of people. 7. The area correction means is defined as X / a = (A / a + k
B) It is configured to perform correction using a correction formula represented by (1 + k), where A is the difference area from the background image, a is the unit area per person, L is the number of links, k is 1
7. The apparatus according to claim 6, wherein the area increase rate relative to a per link and X is a true area. 8. The number of passengers of an elevator, wherein the number of passengers detecting device according to any one of claims 1 to 7 is applied to a device for detecting the number of passengers using an elevator hall or a car as a monitoring area. Detection device.