JP3490466B2 - Image monitoring device and elevator control device using the image monitoring device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image monitoring apparatus for monitoring the number of objects to be used waiting for an elevator in an elevator hall, and an elevator control apparatus for controlling an elevator using the image monitoring apparatus. is there.

[0002]

2. Description of the Related Art Conventionally, by processing an image picked up by a surveillance camera, the movement of a person in a visual field is obtained,
Various image monitoring devices such as measuring the number of people have been considered. For example, as a device for detecting a moving object from a difference between images at consecutive time points, [Sakuma, Ito, Masuda, intrusion object detection using difference between frames is used. Law, Technical Report of the Television Society, vol. 14, no. 49, pp. 1-6,199
0] and [Takeuchi, Nagata, Study on crowd flow analysis method using images, IEICE Image Engineering Workshop IE90
-60, 1990] has been reported. In addition, if you detect a moving object and track it, or if the background and human brightness and color are quite different, enter an image at a certain time,
It is also considered that the number of persons is measured by detecting the portion of the image of brightness and color corresponding to humans from there.

By the way, recently, it has been considered to efficiently control the elevator operation by utilizing the presence / absence of people and the result of counting the number of persons, which are obtained by the image monitoring apparatus, in the operation of the elevator.

However, in general, the elevator hall is often provided at a place where many people come and go, and therefore, all the people in the elevator hall are not always waiting for the elevator, and many mere passers-by are included. Has been. Therefore, like the above-mentioned image monitoring device, if all the people who are in the field of view of the camera are detected and it is determined that they are the users of the elevator, for example, there is no person waiting for the elevator, A person passing in front of the vehicle will be judged to be an elevator user and the elevator will be moved to that floor, which may significantly reduce the operation efficiency of the elevator.

[0005]

As described above, in the conventional image monitoring apparatus, it is impossible to distinguish a person who is really waiting for an elevator from a person who simply passes in front of the elevator.
The use of such a monitoring result of the image monitoring device for elevator control has a problem of impairing the efficient control of the elevator.

The present invention has been made in view of the above circumstances, and an image monitoring apparatus and an image monitoring apparatus capable of accurately obtaining the number of users waiting for an elevator and realizing efficient elevator operation control. An object is to provide an elevator control device using a monitoring device.

[0007]

Image monitoring apparatus according to the present invention
The device is an image pickup means for picking up an image of a field of view almost vertically below, an object detection means for detecting an object part to be monitored in the image picked up by the image pickup means, and a waiting object given in advance. Based on the position information on the one side, it is determined whether the object on the image is a waiting object or a moving object from the position on the image of the object to be monitored, and the number of objects determined to be the waiting object is set to the position. According to
And a means for outputting a number obtained by multiplying the weight value as a waiting number .

[0008]

Further, the elevator control device according to the present invention is
Based on the waiting number output from the image monitoring device,
Predict beta usage and control elevator operation
It is configured by means .

[0010]

According to the present invention, an object waiting to be used by an elevator can be counted separately from an object that simply passes through the spot, whereby efficient elevator operation control can be realized.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows the schematic construction of the embodiment. In this case, a television camera 1 is installed as an image input means at a high place such as a ceiling so that the direction near the vertical downward direction is seen. Then, the moving image input from the television camera 1 is digitized by the A / D converter 2, the human being is detected by the object detecting means 3 from the digital signal, and further, the human being detected by the object detecting means 3 is truly detected. The number of people waiting for the elevator is determined by the waiting object determination means 4, and the result is given to the elevator operation group control means 5 to control the operation of the elevator 6.

In this case, the object detecting means 3 and the waiting object determining means 4 are realized by a digital computer. The software can be considered according to the installation environment of the elevator and the fineness of the operation control, which will be described below. .

First, the object detecting means 3 uses only an image at a certain time point, uses an image at a certain time point and images at a reference time point, uses a plurality of images at consecutive time points, and further combines the above methods. You can think of things like

First, the former two cases will be explained. In the case of an environment in which the brightness of the floor surface has a very high contrast with an object such as a human, the flow chart method shown in FIG. 2 can be used. First, the image captured by the television camera 1 is input (step 201). Next, the image portion other than the density value corresponding to the brightness of the floor is binarized and extracted (step 202). In this case, smoothing for removing noise may be inserted before the binarization, or 2
It is also possible to perform reduction / enlargement after the binarization to remove (smooth) fine shape components of the binary image.

Next, the numbering (labeling) of the connected areas is performed on the binarized image (step 203).
From this, a connected region having a size considered to be an object (human being viewed from above) is extracted (step 204). The number of connected regions extracted here is the number of people. In this case, when a plurality of humans are close to each other and the binarized result shows that the regions for the respective humans are connected to each other to form a large region, it may be determined that the part is not a human and an error may occur. . Therefore, in step 204, no upper limit is set in the determination of the size of the area, and only small objects are excluded. Then, the area of the region selected here is divided by the average value of the areas on the image corresponding to one person to calculate the number of people (step 205).

Here, for example, when there is a person who carries luggage by a dolly, the size of the dolly is converted into the number of people, but in the case of elevator control, the number of objects waiting for the elevator. This kind of thinking is rather convenient because the question is what the total amount of passengers and objects will be. On the other hand, as another simple method, the number of pixels included in the density value determined to be the object by binarization may be obtained, and the number of pixels may be estimated by dividing the number by the average value of the human size ( Then use the same method for all elevators,
If the relationship between the installation position of the camera and the floor surface is the same, the size average value on the human image is the same for all, so there is no need to divide it).

In the above description, it is assumed that there is a difference in density between the floor surface and the object, but if the floor surface has other characteristics, it can be used. For example, if the floor has a certain color, the color can separate the object and the floor. For this purpose, an image may be input with a color camera and a portion having a relationship of RGB density values different from that of the floor may be taken out. Alternatively,
Even if the density value is used in a black and white camera, if the floor has a color, a filter for transmitting or blocking the color of the floor is attached to the lens of the camera to make the processing more reliable. This applies not only to this method, but also to all other methods described below. Further, when there is a pattern (texture) on the floor, the texture may be detected, and a part where the texture cannot be detected may be the object. Alternatively, as shown in FIGS. 3 and 4, it may be possible to devise a floor in front of the elevators 301 and 401. In this case, some patterns (marks), beads that reflect light, infrared rays, or other light-emitting elements are evenly distributed on the floor as shown in FIG. 3, or in a specific pattern as shown in FIG. 4, or randomly. Marks 302 and 402 are installed. In this way, when there are no objects, it is known in advance where and how they will appear in the image, so it is checked whether such objects can be detected from the image at the time of actual operation. This can be examined by binarizing the periphery of a fixed part of the image to find a pattern or a small point, or using the pattern or the point as a template by template matching. Then, with such processing, the number of people can be estimated from the number of patterns and points that cannot be detected.

In the above description, the area having a certain property is taken out, but the boundary (edge) of the different area can be obtained by differentiating the image. For example, such a method is preferable when there are environmental changes such as changes in lighting conditions. Actually, the differential processing is shown in FIG.
It is put before the binarization (step 202), and in the binarization, the part having a large differential value is taken out. In this case, since the edge is easily cut, the adjacent edge is connected by the enlargement processing or the like, and the process proceeds to step 203. In this case, since the size of the connected region does not represent the area of the object, the area of the circumscribed rectangle of the connected region is used instead, or the hole filling process inside the contour line is performed, and then the process proceeds to step 203, which is correct. Area is required. As another method, instead of step 203, the contour line may be traced and the object may be obtained separately.

When a pattern such as a floor, a border between tiles, or other objects in the field of view is always detected as an object, its position is known in advance.
It suffices that they are not the target object.

Next, a method of using the image at a certain time point and the image at the reference time point as the object detecting means 3 will be described. The processing flow in this case is shown in FIG. First, an image is input from the TV camera 1 at a certain point (step 50).
1). Then, the difference between the input image and the reference image 502 that has been captured and stored in advance when there is no object (subtraction is performed for each pixel) is detected (step 5).
03). In this case, the input image may be subjected to preprocessing such as smoothing before the difference, or the reference image may be subjected to similar processing.

Then, by binarizing the difference result,
A part having a large difference from the reference image is extracted (step S50).
4). The subsequent processing of steps 505 to 507 is the same as the processing of steps 203 to 205 shown in FIG. 2 described above. Similarly, there is a convenient way to estimate the number of people from the total area without numbering the connected areas. Furthermore, in order to make it resistant to environmental changes, both images to be compared may be differentiated (the differentiated image may be stored as the reference image) and the difference may be taken. In particular, this method is effective when there is no pattern on the floor surface. After the binarization processing of the difference result, it is the same as the case of using the differential image in the method described above.

On the other hand, if the environmental change is small, a constant reference image may be used in the above processing. If this is not the case, the reference image is sometimes replaced with one that matches the situation at the time. For example, if no object is detected in the process, the image at that time is used as the reference image for the process from the next time. This may be performed every time when the object is not detected, or may be performed only when a certain time or more has elapsed.

By the way, the following three objects are used as the object detecting means 3 using only the image at a certain time point and the object detecting means 4 for the object detecting means 3 using the image at a certain time point and the image at the reference time point. Things can be considered.

First, the first waiting object determination means 4 uses knowledge of the target environment and the target object. First, the simplest example will be described with reference to FIG. FIG. 6 shows an example of an environment in which the simplest method for determining the number of waiting objects can be used.
03 and a white circle 604 are persons detected by the object detecting means 3. In this case, in the illustrated example, the elevator entrance 601
Due to the presence of columns and other stationary objects 602 on both sides of the vehicle, people who simply pass in front of the elevator will typically pass through a section that is continuous with the aisle. Therefore, white circle 6
It is determined that the object detected in the portion indicated by 04 is not the person waiting for the elevator.

That is, when the detection result from the object detecting means 3 is received as shown in the processing flow of FIG. 7A, it is checked from the position information of the object whether the position of the object is a position determined to be a waiting object (step 701). ), The total number of objects determined to be waiting objects is output (step 70).
2). In this case, the number of waiting objects is determined by excluding objects outside the predetermined range, but in reality, such unrelated parts of the field of view of the image are the object rather than removing the object after detecting it. It is more efficient not to process by the detection means 3. This can be realized by extracting and processing only a necessary portion of the input image with the mask image showing the region of interest, or skipping the processing of the pixel of the portion outside the range in each image processing. Further, in this case, if the camera can be installed so that such a irrelevant region does not enter the field of view, the installation position will replace such means.

By the way, in the case of elevator control, the relative relationship between the number of people waiting on each floor or the number of people waiting on each floor is more important than the exact absolute value of the number of people waiting. In such a case, as described above, it is not simply divided into the inside and the outside, but as shown in FIG. You may make it determine a waiting number by giving a weight. That is, normally, a person waiting for an elevator stands near the entrance 801 or at a place a little away near the button 802 for instructing stop. Therefore, a large weight is given to this part.
Also, there may be few people standing in front of the elevator, but since it is not necessary to lower the weight in particular, the same weight or an appropriate weight is given here. After that, a smaller weight is given as the distance from the elevator increases.

As the processing in this case, as shown in the processing flow of FIG. 7 (b), weighting by the object position is performed based on the weight data 703 stored in advance (step 704), and the weighted value is calculated. The sum is calculated (step 705) to determine the size of the needs of the elevator on the floor. In this case, it is effective to arrange beads or light emitting elements on the floor according to the weight, as described in FIG.

On the other hand, such an idea can be applied to other than the environment of the structure shown in FIG. For example, when there are few people who pass in front of the elevator and the people who come near the elevator are mainly people who get on the elevator, or other than the causes as shown in FIG. In this case, it is a very limited gathering. This method may be used by changing the color of the floor around the elevator or drawing a line so as to surround the elevator to promote such a phenomenon.

If this method is used, it is necessary to obtain knowledge of how to wait for an elevator and install it in the device in advance. In the description so far, it is assumed that a person thinks beforehand and incorporates the program into the program. However, as shown in FIG. 9, for example, an image captured by the television camera 1 is directly monitored, or an image once recorded in the recording device 904 is monitored. A keyboard 90 is used to select areas in which a human observes through 901 and there is a high probability that there is a person waiting, and the weighting of the probability.
5, the mouse 902 or the light pen 903 may be designated and given to the microcomputer 906, and the weight data calculated here may be loaded into the memory 907. Alternatively, the input image may be directly or recorded on a recording medium such as a video tape or a disc to determine whether or not it is a waiting object by a more advanced method as described below, and use the data to determine the waiting object area or Weighting may be performed.

Next, the second waiting object number determination means 4 is a method suitable for the case where there is a person passing in front of the elevator but the frequency is not so large. Here, knowledge about the speed of motion of the target object on the image is used.
FIG. 10 is a front view of the environment in which the camera 1 is installed. Here, the minimum value of the height of the target human is hs and the maximum value thereof is ht. Also, the field of view 1
If the side length is wmin, the camera must be installed and selected so that the installation height hc of the camera and the angle of view θ of the camera satisfy the following formula (this is defined by other items described later). This is also the case for the method, however, if the whole body image is not needed to extend the observation range, then ht is not the height maximum, but the minimum object height that must be captured as an image). wt = 2 (hc-ht) tan (θ / 2) ≧ wmin (1)

Now, assuming that the camera 1 is installed as shown in FIG. 10, the size of the visual field of the camera 1 is w at the heights of ht and hs from the floor surface, respectively.
t and ws. Here, a person passing in front of the elevator moves at a speed sp or higher, and a person waiting for the elevator moves at a speed sw.
I think it only works below. In other words, it can be assumed that those who pass in front of the elevator continue to walk at a normal walking speed and then pass by, while those who wait for the elevator move little or move at most around their speed. .

Considering the motion on the image in such a situation, it is when a person of height hs moves at the speed sp that the smallest speed is observed on the image for a person passing by. On the contrary, for a person waiting for an elevator, the person who has the height ht is the one who can see the fastest movement on the image.
This is the case when moving with w.

In this case, considering only the speed of a person passing in front of the elevator, the processing flow is as shown in FIG. Now, assume that the object detection information is received from the object detection means 3 (step 1101). After a little time ts has passed from this state (step 1102), the object detection means 3 is instructed to detect an object (step 1103),
Perform object detection. If some people are detected even at this point, it is output as the waiting number (step 1
104). That is, in such an environment, it is assumed that people who pass by do not appear one after another, so after waiting for a while, if people are still detected, it is determined that they are waiting for an elevator. How long to wait here
That is, the value of the time ts is ts = ws / sp, considering that the object moves the slowest on the image, and further takes a further value when walking diagonally in the visual field of the image. That is, if sp is 1 m / s and ws is 10 m, then 10 seconds must be waited. In practice, this is the worst case, and the latency may be a little shorter. However, in any case, this method requires waiting for a certain amount of time, which is effective when the field of view of one camera is limited to a small range. That is, it is effective when the field of view is covered by a plurality of cameras, or when the place waiting for the elevator is limited to a small size.

Next, the third waiting object number judging means 4 judges the passing person and the waiting person at the speed on the image. In this case, assuming that the image is square and one side has n pixels, the minimum motion mp of the passing person on the image during the unit time and the maximum movement mw of the waiting person are mp = n · sp / ws ... ( 2) mw = n · sw / wt (3)

In general, ws is about twice or less than wt, and sp is considerably larger than sw. Therefore, the position change between two points on the image can sufficiently separate the passing person and the waiting person. In this case, assuming that the size of a human being viewed from above is almost equal, the size (the amount of the dimension of length, not the area) on the image of the human of height ht is k / wt, and the height h
The person of s is represented by k / ws (k is a certain constant). Consider the following two times. tp = (k / ws) / mp = k / (n · sp) (4) tw = (k / wt) / mw = k / (n · sw) (5)

These times represent the times when there is no overlap between the objects on the image between the two time points. FIG. 12 is a diagram explaining this. In this case, the circle A in each of the drawings (a) to (d) is the position of the object at the present time, and the circle B is the position of the object after the time tp or tw. Where tw is tp
On the other hand, it is quite large. Therefore, tp <t <tw (6)

If t is selected so as to obtain the object and the object is detected again, in the case of a person waiting for the elevator, circles A and B are shown in the image as shown in FIGS. The overlap is maintained, but in the case of a person passing by, the overlap disappears as shown in (c) to (d) in the figure. In this case, since the size of a human being is not constant in practice, tp and tw do not necessarily represent both extremes. For example, the shortest time for waiting people to disappear
That is, although it was considered that the object moves fastest on the image, it is considered that the size of the body becomes smaller as the height becomes shorter, and this is a factor that shortens the time until the overlap disappears.
There is a possibility that w will not necessarily be the upper limit. But,
Since the values of both tw and tp obtained here are quite different,
It is possible to choose t as considered here.

FIG. 13 shows a processing flow based on the above idea. In this case, when the object detection information is received from the object detection means 3 (step 1301), the object detection means 3 detects the object after the time t determined by the equation 6 (step 1302). In this case, it is determined whether or not the object is a waiting object from the overlap of the object areas in the detected images for two times (step 1303). In this case, in order to perform step 1303 by inter-image calculation, first the logical product of the images after the object detection processing at both times is taken (step 130).
4). The connected regions are numbered for the result (step 1305), and those having a small area are excluded (step 1306). By looking at the area number of the first object detection result at the position corresponding to the position of the image of the area obtained as a result and the detection result after time t, the area corresponding to the waiting object is obtained (step 1307).

With the above processing, the areas correspond to each other one-to-one, and the object detecting means 3 performs step 205 shown in FIG.
In step 507 indicated by, the area of the connected area is large and it may be determined that the area corresponds to a plurality of people. If the number of people in the determination is the same, the number of people is set as the waiting number. 1
If the number of persons is different, the smaller one is the number of waiting objects. In the case where one area at the previous time point corresponds to a plurality of areas at the later time point, the smaller one of the sum of the number of decision people in the plurality of areas and the number of decision people at the previous time point is set as the waiting number. As the processing, the first two cases are included in the last case (corresponding area is 1), so this may be performed (step 1308).

In addition to the above method, if the detection result is represented by symbol data such as a circumscribed rectangle of the object area,
It is also possible to make an approximate determination based on whether or not the rectangles overlap. This method is suitable for implementation on a computer that does not have image processing hardware.

As described above, instead of using the same object detecting means 3 after the time t, the method of directly comparing the image after the time t and the image at the original time point based on the idea of the third method described above. There is also.

It is assumed that an image is input after a certain time t. Here, if the object is a passerby, the object should be moving more than t · mp, and if it is a waiting person, the motion is t
・ It should be less than mw.

Therefore, as shown in FIG. 14, the template 140 is surrounded by the surroundings of the object region detected at the original time.
1, the template 1401 is moved to a range 1402 slightly larger than the distance t · mw around the center (appropriate point such as the center of gravity) of the region of the image after the time t, and the correlation coefficient at each position is calculated. Ask. In the figure, range 140
Although 2 is shown as a circle, a square may be used instead for simplicity. Alternatively, instead of the correlation coefficient, the sum of the absolute values of the pixel value differences may be obtained. If there is a portion showing a good correlation in this range 1402, it can be determined that the object corresponding to the area is a waiting object because the movement is small. In this case, t need not satisfy Expression 6. In order to limit the area for calculating the correlation as much as possible, tm
It is sufficient to use a small t as long as p is equal to or larger than the number of pixels clearly observed on the image.

Further, after the time t, the same object detecting means 3 is again provided.
Alternatively, the above-described idea may be used. That is, within a distance slightly larger than the distance t · mw, the time t
If an object that is considered to be compatible is detected later, it is determined to be a waiting object. In this case, the determination of the corresponding region may be regarded as corresponding only by the existence of the object, but the similarity of attributes such as area and shape, or the correlation between the regions of both images by the method described above. May be obtained, and the similarity of a certain value or more may be requested. There is another method of directly using the image after the time t, but for convenience of description, this will be described later in the description of the method of using the continuous image.

Next, the above-mentioned two types of object detecting means and 3
How to combine the types of waiting object determination means will be described. Basically, any combination is possible. However, among the first method of waiting object determination, the method of changing the weight depending on the position and the third method require position information of the target object, and therefore, only the area of the entire object can be obtained from the object detection means. No convenient method can be used. Note that the method of weighting the waiting object determining means by position can be used in combination with the second and third methods. That is, the object detected by these methods is weighted according to its position and used for elevator control. The above-described counting process of waiting persons starting from the object detection is performed at appropriate time intervals.

There is also a method of using images at a plurality of consecutive time points for object detection and waiting object determination. First, a method of using images at two consecutive time points will be described with reference to the processing flow shown in FIG. In this case, first, the difference between the image 1501 at time t0 and the image 1502 at time t1 after the time t is calculated (step 1503). Here, if the object 1601 having a larger density value than the background moves from time t0 to t1 as shown in FIG. 16A, the image of the difference result is
As shown in FIG. 16B, positive and negative parts occur. When an object whose density value is smaller than the background moves, the positive and negative positions are reversed, but this process does not need to consider the moving direction of the object, so this is not a problem.

Hereinafter, the processing is performed separately for the positive portion and the negative portion. First, the binarization takes out a portion of the positive portion which is equal to or larger than the threshold value (step 1504). Similarly, for the negative portion, the portion whose absolute value is equal to or larger than the threshold value is extracted (step 1508). Then, the process for the binarization result (steps 1505-1507 and 1509-)
1511) is the same in both cases, so the processing for the positive part will be described here. In this case, since the movement of the person waiting for the elevator is small, there is a possibility that the difference image has a narrow area as shown in FIG. Since such areas may be divided into small areas, they are connected by expansion (expansion) processing (step 150).
5). Then, the connected areas are numbered (step 1
506), small objects are removed (step 1507). For the sake of convenience, the processing up to this point is performed by the object detection means 3.

The waiting object determining means 4 can be realized by processing the information on the positive and negative regions obtained as described above. Here again, the concept of the third waiting object determination method described above is used. Especially, it is the same idea as the method using the correlation described at the end of the explanation. In other words, for a certain negative (or positive) area, it is checked whether or not there is a positive (or negative) area near the same object that seems to correspond to the same object. If so, this object is a waiting object. The value of the distance for this determination of the closeness is determined in the same manner as that considered in the third method. As described above, during the time t, the waiting object moves less than t · mw and the passing object moves more than t · mp.

Therefore, the distance threshold is t · m
An appropriate value smaller than p may be selected. However, if the object moves largely between two time points as shown in FIG. 18, the movement of the center of the positive and negative regions (center of gravity, center of the circumscribed rectangle, etc.) becomes the movement of the object, but the movement is small and as shown in FIG. In this case, the position movement of the positive and negative areas does not match the movement of the object. If the center of the circumscribed rectangle is used as the representative point, the deviation will be a little smaller, but the distance between the center points will still be larger than the actual movement. In FIG. 17, the actual movement is indicated by an arrow, but the center of the circumscribed rectangle (indicated by a cross in the figure) moves by the amount indicated by the thick arrow in the figure. This shift is related to the size of the object, and may be several tens of percent of that size. Therefore, if the value of t is selected so that the value of t · mp is close to or larger than the size of the object, the threshold value for distance determination can be selected in the order of the size of the object. The problem of error is avoided. If the value of t is too large, the amount of movement of the object becomes large and the correspondence becomes difficult to understand. On the other hand, if it is too small, the change between images in the waiting object may be too small to be detected. Considering these two points and the detection error of the movement amount, t
Need to decide. As a standard, the value of t is selected such that the value of t · mp is about 1 to 5 times the average size of the object on the image.

This method has an advantage that it can be used even in a place where the environment changes greatly such as outside light because it uses images at close times. In this method, the difference processing is performed on the input image as it is, but the difference processing may be performed after performing preprocessing such as noise removal. Alternatively, the input image may be differentiated and the difference between the differentiated images may be used. In this case, since the corresponding region of the contour of the object is obtained, the method described in the first method of detecting the object using the differential image can be used for detecting the positive or negative region.

In the method using continuous images, if the movement of the person waiting for the elevator is very small, it may be missed. Actually, since human beings are moving to some extent, such a problem is not so large, but when the field of view of the camera is considerably large, a problem may occur because the motion on the image becomes small. In such a case, the following improved method can be used. One is a method used in combination with a method of obtaining the difference between the reference image and the current image. There are also several ways to do this.

In the first method, when a small positive or negative area is detected, it is not removed because it is small, and the image at that point (either t0 or t1) and the reference image are not removed. By taking the difference, it is checked whether or not an object is detected around the area in the same manner as the second method of the object detecting means 3 described above. If detected, it is a waiting object, and if not detected, it is noise. The second method is t
The difference processing of continuous images is performed by selecting t so that the value of mw becomes small enough not to be detected on the image (for example, 1 to 2 pixels or less). That is, if small objects are removed in this process, the objects are detected only by the passing person. Along with this processing, a difference processing from the reference image (either of t0 and t1 may be performed) is also performed. This is performed by the second method of the object detecting means 3 described above. Of the areas detected here, one that does not overlap with the area obtained by the continuous difference is selected and determined as a waiting object.

Further, the second method may be considered focusing on the difference processing with the reference image (or the first method of the object detecting means). Since the number of waits is calculated, it is useless to always perform continuous difference in which no waiting object is detected, so this is the better method. First, object detection is performed at a certain time point, and when an object is detected, an image after time t is input,
Then, the difference processing of the image at the previous time point is performed. As described above, whether or not the object is a waiting object is determined by the amount of movement in the positive and negative areas.

In addition, if a small area is detected from the continuous difference image, then the image is input once or more, the difference values between the continuous images are added around the small area, and the object is really a waiting object. , Or it can be judged whether it is noise.
If the object is a waiting object, the motion is accumulated, and if a region having a large difference value is added, the motion is widened, so that it can be determined.

As a method of using continuous images, the logical product between the absolute value of the difference between the former two and the absolute value of the difference between the latter two is applied to the images at three consecutive time points (or differential images thereof). You may use the method of taking. By using this as an object detecting means, the waiting object can be determined by using the constraint of the amount of motion in the same manner as in the case of the two images.

In the above embodiments, the image processing part is realized by a digital computer such as a microcomputer. Of course, it is also possible to carry out using dedicated hardware for image processing capable of performing inter-image calculation and numbering of connected areas at high speed. The following image input system may be devised and used for the above image processing means.

First, if the environment is likely to be affected by outside light, infrared illumination is performed, and if an infrared transmission filter is attached to the camera, it becomes stronger against the influence of outside light. In addition, when performing the difference between images, the brightness variation due to the AC lighting of the illumination may be a problem. To avoid this, if there is a significant difference between the images, a method of retaking the images can be used. Further, in order to prevent such a situation from occurring easily, a power source in which the environment lighting is set to a high frequency by an inverter may be turned on. Further, since it is not necessary to look at a fine portion in the case of counting the number of people, if necessary, the focus of the lens of the camera may be slightly deviated to cause blurring, and noise may be removed. Further, if an infrared camera is used for input instead of a normal television camera, by extracting pixels having a value corresponding to human temperature by binarization, that is, by using the first-mentioned method of the object detecting means. Humans can be reliably detected. Next, a case where the image monitoring device as described above is applied to control of an elevator will be described.

FIG. 19 shows a schematic structure thereof. In this example, two TV cameras 1901a
f is connected to the camera switching device 1902, and the signal of the selected television camera is sent to the image monitoring device 1903. Then, the waiting number measured by the image monitoring device 1903 is converted into appropriate information by the waiting number correcting means 1908 and sent to the elevator group control device 1904.
The group control device 1904 predicts the use of the elevator 1905 based on the information on the number of waits and the destination floor instruction made by the person in each elevator, and controls the operation as appropriate.

In this example, two cameras are provided on each floor and the number of waits is counted by one, but the number of cameras and the number of image monitoring devices can be freely set as required. The relationship between the number of cameras and the number of image monitoring devices is determined in consideration of the capability of the counting device so that it can be processed at the time interval necessary to catch the time change of the number of people. When the signals of a plurality of cameras are used as counterparts in one image monitoring device as in this example, the processing is performed in a time division manner. In this case, basically, the processing of the signal of a certain camera is made to come around at regular time intervals, or a large amount of memory for recording the image is required, but input the image of each camera appropriately. The subsequent processing is performed by using the computer for time division. Based on either of these, the priority of camera processing may be changed according to the operation of the elevator.

This is performed by the counting time point control means 1906. Here, as shown in FIG. 20, normally, a plurality of cameras are switched at regular time intervals (step 200).
1). When the elevator stops at a floor, it receives the signal (step 2002) and prevents the camera signal of that floor from being processed during the stop (step 200).
3). This is because it is difficult for the number of people to measure when there is an elevator getting on and off, and the measurement itself does not have much meaning when it is stopped at a certain floor. However, as will be described later, when the image monitoring device is used for other purposes such as safety confirmation when getting on and off the elevator, on the contrary, the camera on the stop floor has priority. Although this will be described in detail later, if a gathering of people to a specific floor is expected and the situation of the gathering is important for the operation prediction, etc., prioritize the processing of cameras on such floors or related floors. To do so. Such information is received from the building use management means 1907, and the priority order of camera processing is determined (step 2004). Or, if an elevator stops at a floor, and then waits for that floor after a certain amount of time,
As shown in steps 2101, 2102 of FIG. 21, it is possible to receive the data on the operation status of the elevator from the group control device 1904 and determine the camera to be used.

By the way, the efficiency of elevator group control is
It depends on how accurately it is possible to predict how many people will move from which floor to which floor. The above-mentioned image monitoring device is used for that purpose, but it basically counts people waiting, and which floor the waiting person wants to go, first of all, whether it wants to go up or not. I don't know if I want to go to. Therefore,
In addition to counting the number of people waiting, it is possible to use a knowledge / information about the flow of people in the building to improve the accuracy of prediction.

Elevator waiting number correction means 1 shown in FIG.
908 does this. This is realized by a microcomputer or the like, but as shown in FIG. 22, for example, from the knowledge 2201 about allocation based on the use characteristics of each floor fixed in advance and the knowledge 2202 about allocation based on the usage status of the building on the day, the number of waits counted Allocate the number of people going up and the number of people going down. The knowledge about the usage characteristics of each floor stores the knowledge about allocation, which is determined structurally and empirically of the building. For example, most people waiting on the ground floor face up (all without underground),
The top floors are all facing down. For the other floors as well, the ratio of the average number of people waiting upward and downward is checked and stored from the past usage status. This may be a function of time (the proportion of people going up and going down depends on the time).

For example, a phenomenon such as an increase in the number of people heading to the cafeteria floor during meals can generally occur. In such a case, it is stored as a table for each time, and at the time of use, necessary data is extracted and used according to the time by the clock 2204 inside the computer. In the allocation 2203 of the waiting number,
The number of waits is assigned to the stored ratio in the upward and downward directions and output to the group controller 1904. The data of this allocation determined empirically may be corrected by the actual operation data. In addition, this allocation is
If it changes due to factors such as month, day of the week, holidays, weather, and holidays such as nearby shops, data for each of these factors may be stored and the data applicable at the time of operation may be used.

Further, as a thing depending on the floor, it may be considered that the allocation is influenced not by each floor alone but by a plurality of floors. For example, when there are many people waiting on the upper floor, there are few people going up, among the people on the middle floor,
Or vice versa may be observed in some buildings. In a building where such a phenomenon occurs, a table for determining allocation based on the number of people waiting on multiple floors or a formula for determination is prepared and stored and used.

The same allocation prediction can be made from the building usage information on the day. For example, if there is a large conference or event on a floor, many people will move to that floor near the start of the event. Therefore, it is expected that the number of people who go to each floor will increase. further,
If many of the participants have information on which floor the office is located, they give a large percentage of the number of elevator waits on that floor towards the venue floor. A value determined empirically may be used as the allocation value. Such usage information is input from the keyboard 2205 or the mouse 2206 connected to the computer. Alternatively, if the computer manages the schedule of each person and the reservation of the conference room, etc., the information is stored in the network 220.
You may make it input through 7.

In the above two cases, the allocation value is empirically determined. This is obtained by manually examining the usage status of the elevator. Alternatively, it is possible to determine the image of the camera of the waiting number counting means directly or as a video or the like by observing the human.

As an alternative, it is possible to correct the predicted value by learning using an image monitoring device. FIG. 23 shows an example thereof. This part is also realized in the microcomputer circuit of the waiting number correcting means 1908. The image monitoring device 1903 counts the number of waits before and after boarding an elevator on a certain floor (2301,
2302), the difference, that is, among the number of waiters, find the number of waiters who seem to have boarded the elevator that arrived (if it is a ratio, divide this by the number of people before arrival) (2303),
The learning means compares the calculated allocation value (2304) with the stored allocation value (2304) to learn a future predicted value (23).
05).

In this case, it is also possible to simply divide the number of people after arrival by the number of people before arrival and calculate the reverse direction ratio without using the difference between before and after arrival. The simplest method of learning is to take the average of a certain number of past times as the new predictor. If the time, other factors, and the correlation with the number of waits on other floors are considered, model those relationships appropriately and find the parameters in them from the observed values. Alternatively, a possible factor may be used as an input layer and a ratio of a certain direction may be graded with an appropriate step size as an output layer to find a relationship with a multilayer neural network. Instead of performing learning automatically, the correctness value may be recorded and a person may look at it to determine the prediction method. Further, in order to know the correct answer to the allocation, the number of passengers may be estimated from the weight change of the elevator. That is, the weight increase at the time when the passenger has finished getting on the elevator as compared with the minimum value after the door is opened (the minimum value after the door is opened), and this is calculated as the average human weight. Break. In this case, it is sufficient that a person who completely descends gets on the bicycle after getting off, but if both directions come and go at the same time, the error becomes large.

Although the above is used to predict the use based on various kinds of knowledge, the use can be predicted by guiding a place waiting for an elevator. This is not related to the upward / downward prediction, but when the destination floors of the elevators are separated as shown in FIG. 24, for example, people who are near the respective elevators are waiting for those elevators. As group control data. In this case, the area division and weighting methods described in the first method of the waiting number determination means are used. People in a certain area may simply wait for an elevator, but, as in the case of covering the scene with multiple cameras described later, weighting the intermediate area and determining the number of people You may do it. In FIG. 24, the determination boundary is shown as a semicircle as an example of the former.

As a method of positively using the same method, as shown in FIG. 25 (a), as shown in FIG.
Alternatively, a display of waiting places 2501 to 2503 for a specific floor may be provided to count the number of people there. Alternatively, instead of the floor, as shown in FIG. 25B, the display device 2504 may be provided on a wall or the like and the number of people gathering around the display device 2504 may be counted.

Another information available for allocation is shown in FIG.
A button 2601 for requesting stop as shown in FIG.
Alternatively, when the destination floor designating device 2602 is installed, it is the information of the desired floor of the user indicated by it.
With these, if only the upward direction is specified, the waiting person can basically determine that he or she wants to go in that direction. Of course, the above-mentioned information for predicting the usage may be used alone or in an appropriate combination.

When the waiting number counting device is used for an elevator, it can be used not only for the group control as described above but also for the following purposes, and the performance as an elevator system can be improved.

One is to turn on the lamp of the stop request button 2701 as shown in FIG. 27 when the waiting person is detected. If the elevator is traveling in one direction, such as the top floor or the bottom floor, all you have to do is turn it on when it is detected. The user does not need to press a button. If the detection fails, the user will have to press the button, but if the user does not have a lamp, the user will only have to press the button, and the operation of the device will force the user to perform an unnatural operation. There is no. On floors that may go up and down, this method is simply not available. If the trend of the waiter is known by the method described in the usage prediction, the lamp may be turned on accordingly. However, if only a small number of people are waiting, there is a risk of making an incorrect decision, so it is generally better to let the user press the button.

When the number of people is counted and the operation plan of the elevator is completed, the elevator display board 2702 may display which direction or destination is the earliest elevator. Other information may be added to the display, such as expected time to arrival. It is considered that this display allows the user to change the waiting position of the elevator depending on the purpose of use. After the display, the waiting number was counted again after a certain time, and the control of the elevator could be more finely controlled from the waiting number in front of each elevator. However,
It will be controlled under the limited condition that the displayed contents and arrival order are not different. Further, the following additional functions can be realized by using the object detecting means 3 of the image monitoring device. First, if an object is always detected in the same place for a certain period of time, there is a suspicion of a suspicious object. In that case, as shown in FIG. 19, a buzzer 1909 arranged in a guard room or the like is sounded to call attention, and the television image is displayed on the monitor 1910. This can improve security. Going forward, we will install cameras in places other than elevators that are necessary for safety,
If you do object detection, match the elevator,
A comprehensive building image monitoring system can be configured.

The object detecting means 3 can also be used to safely close the elevator door. For example, after the door is opened, object detection is performed at short time intervals. This object detection may be limited to just in front of the open door. Now that the object is no longer detected, close the door after a while. Alternatively, without performing such continuous processing, the object detection means 3 may confirm that there is no person in front of the elevator and close the door when a certain time approaches to close the door of the elevator. . This allows the door to be closed safely. In the case of the former method, once the user gets in and out of the elevator, the user can start immediately without pressing the opening / closing button in the elevator, which is convenient for the user.

When the third method of the waiting object determining means is used, the door is closed when it is detected that an object at a speed not determined to be the waiting object is approaching the door during the time when the door is open. You may wait for a while.
This is because it is determined that some people are rushing to the elevator when they see the elevator open. In order to perform the above processing, when a plurality of camera processings are shared by one image monitoring device, it is necessary to give priority to the processing of the elevator stop floor.

When the present invention is applied to an elevator, the camera is mainly mounted on the ceiling. In this case, it is preferable if the ceiling is high due to a blow-through or the like, but if not, it is necessary to devise a method to cover the area where the waiting number is to be measured. Therefore, this problem is solved here by using a plurality of cameras.

When using a plurality of cameras, the sharing of the field of view becomes a problem. The overlap of fields of view is not that big,
If a rough count is sufficient, simply sum the results from each camera,
It may be the value of the whole count. To improve the accuracy a little more, use the following method.

If the mounting position of the camera is not so high, the situation as shown in FIG. 28 will result. There are two ways to handle the processing results of multiple cameras: when the images of adjacent cameras are taken at almost the same time, and when there is some time before the images are taken.
There is a street. In the former case, there may be an image monitoring device for each camera, or images may be taken at short time intervals and then processed. The latter is a case where one image monitoring device is used by switching to a plurality of cameras. If images are taken almost at the same time, there is a big problem that the fields of view of the cameras are counted at the boundary portion.

In the example of FIG. 28, it is necessary that camera a2801 and camera b2802 do not count the same person twice. For that purpose, it is necessary to specify the part corresponding to the same person in the two images.

The shaded area in FIG. 28 is the field of view commonly observed by the two cameras. The tallest target (height h
To ensure that t) is all within the field of view of the camera,
As shown in FIG. 28, the fields of view of the cameras must overlap (although it will be described later, it is not always necessary to install the cameras so that they can fit in).

Here, a problem when the correspondence of the same part is obtained by the two cameras is that the position on the image of the person standing on the specific floor position does not become a fixed place. That is, the position of the image of the head or body part changes depending on the height. FIG. 29 shows an example of images of the cameras a and b. In this case, the positions in the images of the tall person t2803 and the short person s2804 in FIG.
The positions of the head and the body are different as in 9 (a) and 9 (b). However, the position of the feet on the floor is the same. Further, it is possible to obtain in advance where in the images the same position on the floor moves in both images.

Therefore, as shown in FIG. 30, it is checked whether or not each connected area (object) in one of the count result images of both images is in the same floor position in the other connected area. For a certain connected area, the position on the center side of the image is obtained (step 3001). For example, it is a portion such as f in FIG. Then, assuming that it is the floor position, the corresponding floor position f'of the other image is read from the data stored in advance (step 3002). This data can be created for each point in the image by finding the coordinates of the corresponding point in the other image. Then, from the connected regions of the other image, the one in which the image center side is closer to the corresponding position is checked. If there is such a region, it is determined that both are derived from the same person, and double counting is not performed (step 3003). Note that in the case of an image input condition in which the positional deviation error is not so large, as a simple method, the correspondence may be checked at the center position of the image region instead of at the foot position.

If there is a certain time difference between the image pickup times of the adjacent cameras, the same method may be used, but since there is a possibility that the target has moved between the time points, the following procedure is performed without such treatment. It should be processed as follows. This is the same method as that in which the weight is considered in the first method of the waiting object determination means. As shown in FIG. 31, in the area where the visual fields of the cameras overlap, the object detected at that position is multiplied by an appropriately determined weight. In FIG. 31, the shaded area is the camera a28.
The area that may be observed not only by 01 but also by the camera b2802. Therefore, consider weights as shown in FIG. In FIG. 31, the position on the floor indicates the position on the image. The camera b2802 is similarly weighted. The sum of those multiplied by the weight (round up or 4 if necessary)
The number of waits is made (rounding down to the nearest 5). Also in this case, the foot position, that is, the position closer to the center of the image in the image area is used as the image position. However, if the error is not so large, the position on the image may be substituted by a value such as the center of the area. This method may be used as a simplified method when images are input from a plurality of cameras simultaneously or almost simultaneously.

If it is desired to have a wide field of view with a small number of cameras, it is not necessary to use the camera arrangement so that all persons of high height can be within the field of view as described above. As shown in FIG. 32, when the position from the floor is low, the adjacent cameras 3201,
The visual field boundaries 3202 may intersect. In this case, in the periphery of the visual field, the human 3203 head portion cannot be seen, and only the torso and legs can be seen. In this case, when the object moves around, the part that enters the field of view also changes, so changes due to movement and changes due to different parts of the object entering (or leaving) the image are visible on the image. Will be observed.

Therefore, the apparent movement on the image becomes large. Since there is a large difference in speed between the waiting object and the passing object, the speed-related judgment of the third method of the waiting object judging means is made so that the waiting object is not regarded as the passing object in consideration of such peripheral effects. It is possible to choose a threshold.
When the camera is used in this way, this must be taken into consideration when selecting the judgment threshold value. Alternatively, different determination thresholds may be used for the image peripheral portion. This is the same as the coping method for the wide-angle lens and the fisheye lens described below.

Another method of expanding the field of view is to use a wide-angle lens or a fisheye lens. Of course, this and the method using a plurality of cameras described so far can be used together. Generally, when using a lens with a wide angle of view, there are problems of insufficient peripheral light quantity and peripheral distortion. The former is solved by changing the threshold value for binarization and the like depending on the position of the image or the density value of each part of the image. The latter is not a problem in this respect because it is not necessary to consider the fine shape in the present invention (however, it may be necessary to consider the point that affects the size of the object, which will be described later).

In the present invention, the effect of the wide-angle lens, that is, the wide field of view, must be taken into consideration rather than these general problems. One is the change in the size of the object on the image due to perspective, and the other is that when you look at a person standing on the floor from above, those near the center of the optical axis of the camera are mainly heads and shoulders. However, it is possible to see the whole body in the surroundings, but there is a large change in the appearance depending on the position.

FIG. 33 is a diagram for explaining this. In this case, the person a near the center of the optical axis of the camera 1 looks like c in the figure in the image, while the person b around the person d
The whole body looks like. In both of the above cases, the size changes on the image, and attention must be paid to the determination of the determination thresholds for the size and the motion speed in the processing. In the latter case, for example, the head and the leg are likely to be separated as two objects and detected as an object, and an error may occur in the measured value if care is not taken. In addition to that, as shown in FIG.
Even if 02 are standing apart from each other, the problem that the images appear to overlap with each other as shown in a in FIG.

The change in the size on the image is detected by the object detecting means 3
Step 205 of FIG. 2 or Step 507 of FIG.
It becomes a problem when using the average value of one average person.
Further, in the third method of the waiting object determination means 4, in the method of using the overlap between two time points, the size of the object is involved in determining the time interval between the two time points. However, since the size range of an object image is almost determined by the position on the image, the average value of the appropriate human size is calculated in advance for each image position and stored. However, this may be used at the time of determination or the like.

FIG. 35 is a diagram for explaining the change in the size of a certain point of the image in the case of the ordinary plane mapping. In this case, the imaging surface 3501 is behind the lens, but is also on the front surface of the lens for ease of explanation. As mentioned before, humans are standing on the floor, so it is convenient to use the position of their feet to locate the image. That is, the position of the portion closer to the image center of the object is set as the position on the image of the object. However, it may be possible to substitute the center position of the object depending on the conditions because only the approximate range is considered. Further, as shown in FIG. 35, the size of the person 3502, 3503, 3504 standing on the same position also changes depending on the height. For each point in the image, consider the highest and the lowest target objects and determine the approximate range where the area may change. In addition, considering an object of average size, the average area is calculated.

In general, the difference between the maximum target (the tallest person) that may appear and the minimum size is large in the surroundings, and therefore, it is handled like step 205 in FIG. 2 or step 507 in FIG. Can't be exact. In order to accurately measure the number of people, the image in the area may be examined and, for example, a process such as counting the number of parts considered to be the head may be inserted. However, if a slight error is allowed, the method of step 205 or step 507 described above is used in consideration of the height of a human being who appears most frequently and the size thereof.

Regarding the time interval between the two time points, it is advisable to use a different value depending on the location in consideration of the change depending on the position of the image in such an area range. However, in this case, an image is input at some time intervals and an image at an appropriate time is used depending on the position of the image. Since it is necessary to input some images, if it is not appropriate, the waiting object determination means may use another method among those described above.
However, in the case where the motion speed is used to determine whether or not the object is a waiting object, the maximum speed of the waiting object and the minimum speed of the passing object on the image were considered with the minimum height and the maximum height in the previous explanation. However, when there is a large change in the size of the image, it may not be possible to make an appropriate determination. In such a case, an extremely high or low value that does not appear very often is not considered, and the lowest and highest values are set within a range that often appears to determine the determination value.

In any case, the value used for the determination depends on the position of the image. Therefore, as shown in FIG. 36, a table 3601 in which a value required for the determination can be taken out at the position of the image is prepared, and step 30 in FIG.
5 or step 507 of FIG. 5 and the third method of the waiting object determination means.

So far, the solution to the problem of the change in the size of the objects and the overlap has been described. Another problem when using a wide-angle lens, that is, a region on an image for an object is likely to be divided into a plurality of regions, can be similarly solved by the determination value table 3601 based on the image position.
As an example of this, step 30 of FIG.
5 or 507 in FIG. 5 is added with a portion as shown in FIG.

If the minimum object area written in the judgment value table according to the image position in the image or the lower limit of the range of the size of a human that often appears (except for very small ones considered to be noise) ) If there is a region (step 3701, region A in FIG. 38), it is checked whether or not there is another region around it. The existence may be checked only for the target small area in the same distance from the center of the image or in the direction farther from the center. When such other areas meet (step 3702, area B in FIG. 38), the size of the case where they and the target area are combined is checked, and if they are within the area range of one person, Collectively judge them as one person. In some cases, the total area may correspond to the area of multiple people.
For example, in FIG. 38, if A has a size of about 0.5 people and B has a size of about 1.5 people, the total is 2 people.

In such a case, the combined area is determined as the number of people corresponding to the size. Therefore, in the case of the example, it is determined that there are two people. It should be noted that whether to use the minimum object area for determining the size or the minimum area among the sizes that often appear is determined depending on the environment / lens system. Which one is selected may be changed depending on the position of the image. Any position dependence may be written in advance in the image position determination table 3601 and used.

Even when a fish-eye lens is used, it can be handled in the same manner as a normal wide-angle lens by using the image position determination table 3601. In this case, although the projection method is different, as with a normal lens, for an object on the surface of the floor, the size change range on the image depends on the position of the floor, that is, the position on the image. Determined. Therefore, although the size of the image is determined depending on the projection method of the fish-eye lens used, it can be written in the determination table 3601 based on the image position and used. For example, there are various types of fish-eye lenses, but in a fish-eye lens of equal solid angle projection, the point of incidence at the zenith angle Φ (incident angle to the lens optical axis) is sin (Φ /
Projected to a position proportional to 2). Therefore, FIG.
As shown in (a) and (b), an image 3903 captured by the television camera 3902 is obtained via the fisheye lens 3901. Since the target area range is obtained from such a relationship, the same treatment as in the case of the wide-angle lens will be performed thereafter.

However, the movement speed of the target was considered by the second and third methods of the waiting object determination means.
Be careful with fisheye lenses. With a plane projection lens, the velocity on the image of a constant height object walking on the floor does not depend on the position of the image. However, since the part that can be seen changes depending on the position, it is necessary to be careful in handling the speed accordingly. In the case of fisheye lenses, this change in appearance is even greater, and in addition, the floor (image) is essentially
Depending on the position of, the velocity on the image of the object is different even for the same velocity motion. As shown by concentric circles in FIG. 40, in spherical projection, if an object moves on a surface that is equidistant from the center of the projection sphere, the motion on the image is the same wherever the object is on that surface. . However, the floor surface is not such a spherical surface. In the means for determining the waiting object, the maximum speed of the waiter is the movement of the person with the maximum height, and the minimum speed of the passerby is the movement of the person with the minimum height. With the fisheye lens, in order to consider the minimum and maximum velocities, it is necessary to consider the position on the image. When the field of view shown by a fan in FIG. 40 is viewed, the maximum speed of the waiter is that of the person A having the maximum height standing on the center of the optical axis, without considering the influence of the change in the appearance of the object. Further, the minimum speed of the passerby is that of the person B having the minimum height around the visual field. Therefore, the judgment value is determined with this as the extremes. However, when there is no significant difference between the two, that is, when a clear determination cannot be made, this speed range is determined for each position of the image as in the case of the above-described change in appearance. Therefore, the determination method for each position may be written in the determination table based on the image position. In the case of a fish-eye lens, the change in appearance is also large, and therefore both the effect and the effect described here are observed. Therefore, this method is particularly effective.

In fact, even in the case of a wide-angle lens, the appearance like a fish-eye lens becomes more noticeable especially in the peripheral portion as the angle of view increases. Therefore, even if an object of the same height moves on the floor surface, the speed on the image changes depending on the position of the image. In such a case, the same treatment as that of a fisheye lens may be performed if necessary.

Even if the optical axis of the lens is not perpendicular to the floor surface, the size of the object changes depending on the position of the floor. If the degree cannot be ignored, it depends on the image position as in the case of the fisheye lens. A method using the determination table 3601 may be used.

The change in the size of the appearance of the object on the image as described above may be obtained from the projection formula, or it may be obtained by collecting actual image data and calibrating the camera system. May be. In particular, since the influence of the distortion of the wide-angle lens cannot be obtained from the projection formula, the method using actual data is suitable. The calibration may be performed by an actual person, or as shown in FIG. 41, the calibration objects 4101 and 4102 such as the minimum height and the maximum height or the cylinders having the lower and upper limits of the normal appearing range are prepared and used. You may.

Finally, the mounting of the camera will be described. A camera 4202 is attached to the ceiling 4201 as shown in FIGS. The method (a) is effective in preventing illumination light from entering the lens. This effect can be obtained even with a lens hood. In addition, FIG. 43 (a) (b)
A camera 4301 may be installed in the lighting equipment as shown in FIG. In this case, the light from the illumination 4302 is directly reflected by the camera 43.
A hood 4303 is provided around the camera 4301 so as not to enter 01.

[0105]

As described above, according to the present invention, the number of objects waiting to be used can be determined separately from the objects that simply pass through the spot. And when this is used for elevator control, it becomes possible to predict the usage situation of users based on the number of waiting people, and it is possible to appropriately perform group control to reduce the waiting time on each floor of the elevator. Elevator control can be realized.

[Brief description of drawings]

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

FIG. 2 is a flowchart for explaining an object detecting means.

FIG. 3 is a diagram showing a mark arrangement example for object detection.

FIG. 4 is a diagram showing an example of mark arrangement for object detection.

FIG. 5 is another flowchart for explaining the object detecting means.

FIG. 6 is a diagram showing an example in which a waiting object determination method using knowledge about the environment can be used.

FIG. 7 is a flowchart for explaining a waiting object determination method using knowledge about the environment.

FIG. 8 is a diagram showing an example of weighting by position.

FIG. 9 is a diagram for explaining how to set weight data by an interactive method.

FIG. 10 is a diagram for explaining a field of view by a television camera.

FIG. 11 is a flowchart for explaining a waiting object determination method by a method of waiting for a passing object to pass.

FIG. 12 is a diagram for explaining movements on an image of a waiting object and a passing object at two times.

FIG. 13 is a flowchart for explaining a waiting object determination method based on a speed difference.

FIG. 14 is a diagram for explaining a waiting object determination method using correlation.

FIG. 15 is a flowchart for explaining a waiting number counting method using images at continuous time points.

FIG. 16 is a diagram for explaining appearance of positive and negative regions in continuous images.

FIG. 17 is a diagram for explaining a difference result when the motion of the object is small.

FIG. 18 is a diagram for explaining a difference result when the motion of an object is large.

FIG. 19 is a diagram showing an application example of the image monitoring apparatus of the present invention to an elevator.

FIG. 20 is a diagram for explaining the processing by the counting time point control means.

FIG. 21 is a diagram for explaining an addition process to the counting time point control means.

FIG. 22 is a diagram for explaining a process in a waiting number correction unit.

FIG. 23 is a diagram for explaining learning of an allocation prediction method.

FIG. 24 is a diagram showing a situation where the destination floor of the elevator is divided. .

FIG. 25 is a diagram showing a case where an upward and downward meeting place is indicated by display.

FIG. 26 is a diagram showing a case where a stop request button is used.

FIG. 27 is a diagram showing an example of the relationship between the number of wait counts and display.

FIG. 28 is a diagram for explaining overlapping of fields of view of a plurality of cameras.

FIG. 29 is a diagram for explaining an image observed in an overlapping portion of the visual fields of a plurality of cameras.

FIG. 30 is a flowchart illustrating a double counting prevention process when the fields of view overlap.

FIG. 31 is a diagram for explaining a method of coping with overlapping of fields of view by weighting.

FIG. 32 is a diagram for explaining a case where the fields of view of adjacent cameras are small.

FIG. 33 is a diagram showing a change in appearance of an object in an image of a wide-angle lens.

FIG. 34 is a diagram for explaining a case where images of people in the vicinity overlap.

FIG. 35 is a diagram for explaining a difference in size of images of people of different heights.

FIG. 36 is a diagram for explaining the use of a judgment value table based on image positions.

FIG. 37 is a flowchart for explaining a process of integrating separated objects.

FIG. 38 is a diagram showing an example of an image when integrated.

FIG. 39 is a diagram showing a relationship between the object of the fisheye lens of the equisolid angle mapping and the size of the image.

FIG. 40 is a diagram illustrating conditions under which the maximum and minimum velocities on an image are observed in the case of a fisheye lens.

FIG. 41 is a diagram showing an example of a calibration object.

FIG. 42 is a diagram showing an example of a method of mounting the television camera.

FIG. 43 is a diagram for explaining a method of incorporating a camera into a lighting device.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Television camera, 2 ... A / D converter, 3 ... Object detection means, 4 ... Waiting object determination means, 5 ... Group control means, 6 ... Elevator, 901 ... Monitor, 902 ... Mouse, 903 ... Light pen, 904 … Recording device, 905… Keyboard, 9
06 ... Microcomputer, 907 ... Weight data memory, 1902 ... Camera switcher, 1903 ... Image monitoring device, 1904 ... Elevator group control device, 1905 ... Elevator, 1906 ... Counting time point control means, 1907 ... Building utilization management means, 1908 ... Waiting number correcting means, 1909 ...
Buzzer, 3601 ... Judgment value table by image position.

Continuation of the front page (56) Reference JP-A-52-112949 (JP, A) JP-A-63-258380 (JP, A) JP-A 1-220688 (JP, A) JP-A 64-69480 (JP , A) JP 50-90045 (JP, A) JP 4-191269 (JP, A) JP 5-39176 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB) (Name) B66B 3/00-3/02 H04N 7/18-7/18

Claims (2)

(57) [Claims] 1. An image pickup means for picking up an image of a target field of view substantially vertically downward, an object detecting means for detecting an object portion to be monitored from an image picked up by the image pickup means, and a waiting object given in advance. While determining whether the object on the image is a waiting object or a moving object from the position on the image of the object to be monitored based on position information regarding the waiting method , to the number of objects determined to be waiting objects, Rank
An image monitoring apparatus comprising: a unit that outputs a number obtained by multiplying a weight value according to the position as a waiting number . 2. Output from the image monitoring device according to claim 1.
Elevator utilization status is predicted based on the number of waiting
An elevator control device comprising means for controlling the operation of the elevator.