JPS62177689A - Device for counting the number of persons - Google Patents

Abstract

PURPOSE:To minimize the influence of an ambient condition and to count the number of persons with a high accuracy by detecting the central position of the black circle in the concentration changing direction, detecting the persons and counting the number of persons after the head part of the persons is identified by the concentration characteristic. CONSTITUTION:When a view illuminated by an infrared ray dropping illuminating means 31 is photographed by a high angle photographing means 32, the image signal is stored into an image memory circuit 34 corresponding to the position. At such a time, the head part image of the person comes to be a black circle. The direction value of respective picture elements of the image is calculated by a concentration changing direction calculating circuit 35 and stored into a concentration changing direction memory circuit 37. A comparing circuit 39, by using the direction value and the address shifting vector of a direction extracting template 38, executes the weighting to respective attention point picture elements, obtains the accumulative value, compares with the set value and obtains the center of the black circle. A concentration comparing circuit 40 reads the concentration value of plural images at the near circumference from the circuit 34, decides the concentration characteristic and at the time of valid, outputs a counting-up signal. A counter 41 outputs the number of the persons in the picture after an address generating circuit 36 scans all picture elements.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a people counting device that counts the number of people present within a set area by image processing.

[Conventional technology]

FIG. 4 is a flowchart showing a conventional method of counting people. In the figure, 1 is the process of capturing an image when no one is present, 2 is the process of resetting the counter J and CNT,
3 is the process of capturing the image when it should be counted, 4 is the process of resetting the counter I, and 5 is the pixel (I, J) of the image obtained by process 1 and the pixel (I, J) of the image obtained by process 3.

6 is a threshold process to check whether there is a change greater than a certain "threshold value" as a result of process 5, and 7, 8.10 is a process to calculate the difference in the density value between CNT, I, and J). In the process of counting up, at 9°11, the count values of counters I and J are the set value I□. .

12 is a process for determining whether or not J ENO has been reached, and a process for obtaining the value of the counter CNT.

Next, a conventional method of counting people will be explained with reference to FIG. When counting the number of people, first, an image when no one is present is captured and stored in the memory 1 in order to compare it with an image for counting the number of people (processing 1).

Next, capture the image when counting the number of people and use it in memory 2.
(Process 3). The values stored in both memory 1 and memory 2 are subtracted for each pixel, the result is compared with a certain threshold value TH, and when the value is greater than the threshold value TH, the count value CNT is incremented by one (processing 2).゜4〜11
). This count value CNT is the number equivalent to the number of people (
Processing 12).

[Problem that the invention seeks to solve]

Conventional people counting methods are as described above, so for example, it is possible to count even small lumps (noise components such as dust on the screen), or when two people overlap, the image is treated as a smaller lump than the actual two people. There were problems such as undercounting the number of people counted.

In response, the applicant has developed a method for counting people that does not count small noise components and can also count the number of people who overlap over 2 Å (patent application). 60-186074)
.

FIG. 5 is a flowchart showing the method of counting the number of people described in the above-mentioned Japanese Patent Application No. 186074/1983. In the figure, 15 is the process of capturing the screen when no one is present, 16 is the process of capturing the image when counting is to be done, and 17 is the process 15
and process 16, 18 is a threshold process to check whether there is a change of more than a certain threshold value as a result of process 17, and 19 is process 18
20 is a process of labeling connected regions with a value of 1 from the result of 20 is a process of counting the area of each labeled region, 2
1 is a process in which the area of each label region obtained in process 20 is divided by a certain set area value M and quantized, and 22 is a process in which the sum of the quantization numbers of all labels is output as a number of people count output.

Next, this method of counting the number of people will be explained.

When counting the number of people using this method, first, an image when no one is present is captured and stored in the memory 1 in order to compare it with an image for counting the number of people (process 15). Next, an image for counting the number of people is captured and stored in the memory 2 (process 16). The values stored in memory 1 and memory 2 are subtracted for each pixel to obtain the absolute value, and the result is compared with a certain threshold value TH.
If larger, set 1”, otherwise set IIQll.
is stored in memory 2 (processing 17.18).

Next, in a labeling process 19, the areas connected vertically and horizontally at the point 61'' in the memory 2 are regarded as one area, and each group is labeled.

Process 20 is a process for finding the area of each label by counting the pixels of the same label. For example, if the area of each label found is M, and the area on the human image is M, then if Xi < M (Xi is the area of the connected region of label i), this is determined to be noise and deleted. . Further, when Xi=M, this corresponds to one person, so the quantization number is set to 1. When X i >M, when several people overlap, Xi/M is expressed as a quantized number obtained by rounding off to an integer. This quantization process is performed for each label (
Processing 21). Finally, find the sum of the quantization numbers of all labels,
This is output as the detected number of people in process 22.

However, in such a people counting method, it is assumed that the image of a person in an image is expressed by the area of a pixel region that changes by a threshold value TH or more, and the human image is also uniformly set. Since it is quantized for each area with an area M, it is easily affected by the light distribution to the human subject and the floor background, the density of the people, and the accompanying changes in the shape of shadows, making it difficult to count with high precision. Met.

This invention was made to solve the above-mentioned problems, and an object thereof is to obtain a people counting device that can count the number of people with high accuracy.

[Means for solving problems]

The people counting device according to the present invention includes an infrared light transmitting illumination means, a prone imaging means that has a filter that transmits light in the vicinity of the infrared wavelength, and that images an illuminated scene from above. an image storage circuit that stores the density gradation of the image; a density change direction calculation circuit that converts the direction of density change from white to black near each pixel of the image into a quantized direction value; and a density change direction calculation circuit that stores the converted value. a scanning address generation circuit that scans the pixel of interest of the image storage circuit and the density change direction memory circuit; and an equal dividing point on a circumference with a predetermined radius that adds to the address centering on the pixel of interest. When the direction determined by the direction extraction template storing the pixel address shift vector and the direction value in the density change direction storage circuit is opposite to the direction determined by the address shift vector of the direction extraction template, the direction is maximized. A weight that decreases as the deviation becomes smaller is applied to the equally divided point pixels of each pixel of interest, and this is accumulated for each pixel of interest, and the accumulated value and the set value are compared for each pixel of interest. a comparison circuit that outputs a black detection signal, and a density comparison circuit that receives the black detection output signal and compares the density characteristics of a set area including the pixel of interest in a density storage circuit with a set value and increments a counter. It has been established that

[Effect]

In this invention, when a scene including a human being is illuminated by the infrared light transmissive illumination means and imaged by the photographic imaging means, the human head is covered with a black disc due to the infrared light absorption property of hair. When the obtained image is captured and stored in the density storage circuit, the direction value of each pixel of the image is calculated by the density change direction calculation circuit, and the comparison circuit compares this direction value with the address shift.
The equally divided point pixels of each target pixel are weighted using the vector and the cumulative value is determined, and the target pixel that is the center for black is determined by comparing the cumulative value with the set value. The density comparison circuit compares the density characteristics of the set area including the pixel of interest with the set value, and the counter counts up. In this way, the human head is emphasized and input, and the spatial information based on the density value is converted into density. The information is replaced with spatial information in the direction of change and detected, and then the concentration/spatial information is returned and confirmed before counting is performed.

〔Example〕

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

1 to 3 show a people counting device according to an embodiment of the present invention. In FIG. 1, reference numeral 31 denotes infrared light projection illumination means for illuminating a scene including a person, 32 a horizontal image capturing means for capturing an overhead image of the illuminated scene including a person, and 33 an optical axis of the downward view image capturing means 32. 34 is an image storage circuit that stores the density gradation of the image; 35 is a filter for each pixel of the image in the image storage circuit 34; A density change direction calculation circuit 37 calculates the direction of density change from white to black from the density gradation and the density gradation of neighboring pixels, and converts this into a quantized direction value. 36 is a scanning address generation circuit that sequentially scans the pixel of interest in the image storage circuit 34 and the density change direction storage circuit 37 over the entire image; 38 is a scanning address generation circuit that sequentially scans the pixel of interest in the image storage circuit 34 and the density change direction storage circuit 37; This is a direction extraction template that stores address shift vectors of equally divided point pixels on a circumference with a predetermined radius to be added.

Further, reference numeral 39 indicates a direction determined by the direction value extracted from the density change direction storage circuit 37 and a direction extraction template 38.
When the direction determined by the address shift vector extracted from 40 is a comparator circuit which accumulates this for each target pixel and outputs a black detection signal when the accumulated value is larger than a set value for each target pixel; 40 receives the black detection signal from the comparator circuit 39; 41 is a counter that counts up the number of detections; 42 is a point of interest that stores the address of the already counted pixel of interest; This is an address accumulation table.

Next, the operation will be explained. When the area where the number of people is to be counted is illuminated by the infrared light projection illumination means 31 located above, and the scene is imaged from above by the overhead imaging means 32 via the filter 33, the obtained image signal is The image is stored at an address in the image storage circuit 34 with a value corresponding to the density gradation value. At this time, the image of the human head shows a low gradation with respect to the surrounding background due to the infrared absorption characteristics of the hair, and the distribution width of the human height is a low ratio compared to the imaging distance, so it is It becomes a black circle with a constant diameter.

The density change direction calculation circuit 35 compares the increase/decrease in density value between each pixel in the image storage circuit 34 and its neighboring pixels by, for example, spatial differentiation processing, and determines the direction in which the largest density difference occurs for each pixel. directional resolution n set accordingly
It is replaced with a density change direction value (0, 1, . . . , n-1) corresponding to the density change direction.

The results of performing this operation on all pixels in the image storage circuit 34 are stored in the density change direction storage circuit 37 at addresses corresponding to the respective positions.

On the other hand, the direction extraction template 8 includes the radius r on the image, which is almost uniquely determined by the image conditions, and the previous direction resolution n.
Address shift vectors (Xo, . . . , , −1) to equally divided points on the circumference centered on the point of interest determined by are stored.

FIG. 2(C) shows the direction value in the density change direction memory circuit 37, and FIG. 2(81, Cb) shows the method for determining the direction/value.
Please refer to the figure, which shows the concept of the direction extraction template 38. Here, the directional resolution is n=8. The density change direction is statistically displayed with an arrow in the direction from high to low gradation, and the direction value (
0, ..., 7) is shown.

When the calculated values of all pixels are stored in the density change direction storage circuit 37, the scan address generation circuit 36 sequentially generates scan addresses, and the point of interest is from the upper left corner to the lower right end on the density change direction storage circuit 37. The entire screen is scanned. At this time, for each pixel point of interest, address shift vector corresponding point n pixels given by the direction extraction template 38, that is, n pixels on the circumference of a radius r centered around the point of interest.
All direction values stored at the equal division points are read out. The read direction value is sent to the comparator circuit 3 for each pixel of interest.
9, a comparison is made. That is, here, address shift vector direction (direction value XK is 'Lc' when XK
= mod (K+n/2.n) (however, mod (d+
, dz) is the non-negative integer remainder of d+/dz)) and the direction in which
(When Xm + Vb, n) = 4) The maximum weight value is
A weight value that decreases as it moves away from the exact opposite direction (the deviation in direction becomes smaller) is given to each circumferential division point, and the cumulative value obtained by adding this value over the entire circumference (K from 1 to n) is used to calculate the circularity. Therefore, the cumulative value obtained for each pixel of interest is compared with the threshold set in advance, and if it is larger than the threshold, this point of interest is expressed in the human head image described above. It is determined that it is a certain center point for black, and a black detection signal is output to the density comparison circuit 40.

The density comparison circuit 40 stores the density values of a plurality of pixels in an appropriate neighborhood range including the current address in the density storage circuit 34.
, and determine appropriate quantities between the density values, such as density average value, density variance value, and combinations thereof, by comparison with appropriately set threshold values. When the determination is valid, the concentration comparison circuit 40 outputs the counter circuit 4.
1 outputs an output signal. Counter circuit 41
After the address generation circuit 36 scans all pixels, it outputs the number of people on the screen, and the series of processing ends.

The apparatus of this embodiment as described above takes advantage of the fact that the human head is imaged as a black color of uniform size due to the infrared light absorption property of hair, and uses the direction of density change to capture a black image. By detecting the center position of the person and then counting the number of people after identifying the person's head based on the density characteristics, the number of people can be counted. It is possible to reduce the influence and improve the accuracy of people counting.

In the above embodiment, the black circle, that is, the hair on the head, is used as the number of human heads, but when the cumulative value expressing the circularity is less than the set threshold in the comparison circuit 39, before moving the point of interest, When the read direction value and address shift vector direction value XK are in the same direction (i.e. when mod (XK+V,,n)=Q)
A comparison circuit is configured such that the maximum weight value is assigned, and the weight value is decreased according to the direction shift, and the weight value is set as the cumulative value, and an output signal is output when the weight value is larger than a separately provided set value. At the same time, the density comparison circuit 40 identifies the output signal on this path as the black detection signal, and in the density feature comparison, when comparing the density average value with the corresponding threshold, white density above a certain level is possible. It is also possible to configure a concentration comparison circuit that performs a threshold value judgment, thereby making it possible to obtain a highly accurate people counting device that includes not only black hair but also bald heads based on their roundness in the human head count. becomes.

In addition, the detected and counted addresses are stored in a separately configured counted point-of-interest address accumulation table 42, and when the scanning address generation circuit 36 generates a new point-of-interest address, it is combined with each stored address. If this value is smaller than the set value determined by the imaging conditions such as the physical size of the person, the imaging distance, and the imaging magnification, no further comparisons are made and the next focus is calculated. By checking the point address, it is possible to detect noise with higher accuracy, such as a shadow image of a person.

〔Effect of the invention〕

As described above, according to the present invention, the human head is input with emphasis, spatial information based on density values is replaced with spatial information in the direction of density change, and then the density is detected.

Back to spatial information V1! Since the system is configured to count the number of people after recognition, it is possible to improve the accuracy of counting the number of people by reducing the effects of the density of five people in the light distribution, changes in shadow shape, and the like.

[Brief explanation of drawings]

FIG. 1 is a configuration diagram showing a people counting device according to an embodiment of the present invention, FIG. ) is a conceptual diagram of the direction value in the density change direction memory circuit, Figure 3 is a conceptual diagram of the direction extraction template, Figure 4 is a flowchart of the conventional people counting method, and Figure 5 is a diagram developed by the applicant. It is a diagram showing a flowchart of the method for counting the number of people. 31 is an infrared light projection illumination means, 32 is an overhead imaging means, 33
34 is a filter, 34 is an image storage circuit, 35 is a density change direction calculation circuit, 36 is a scanning address generation circuit, 37 is a density change direction storage circuit, 38 is a direction extraction template, 39 is a comparison circuit, 40 is a density comparison circuit, 41 is a counter, 42
is a detected hot spot address accumulation table.

Claims (3)

[Claims] (1) An infrared light projection illumination device that projects infrared light from above; an overhead imaging device that includes a filter that transmits light around the wavelength of the infrared light, and that captures an overhead image of the illuminated scene; an image storage circuit that stores the density gradation of the image in an address corresponding to the position; a density change direction calculation circuit that determines the direction of density change leading to black and converts it into a quantized direction value; a density change direction storage circuit that stores the converted value for each pixel; and the density change direction storage circuit. and a scanning address generation circuit that sequentially scans the pixel of interest in the image storage circuit over the entire image, and an address shift/shift of the equally divided point pixels on the circumference of a predetermined radius that adds to the address centering on the pixel of interest. A direction in which a pair of a direction extraction template storing a vector, a direction determined by the direction value extracted from the density change direction storage circuit, and a direction determined by the address shift vector extracted from the direction extraction template face each other. A weight that is maximum when the combination of A comparison circuit that outputs a black circle detection signal when the accumulated value is larger than a set value, and a comparison circuit that outputs a black circle detection signal when the black circle detection signal from the comparison circuit is received, sets the density characteristics in the set area including the corresponding address in the density storage circuit to the set value. A people counting device comprising: a concentration comparison circuit that compares the number of people with the number of people and outputs a count-up signal; and a counter that counts up based on the count-up signal and outputs the count value as a count value of the number of people. (2) After the comparison and judgment of the extracted direction values, the comparison circuit sets the weight to the maximum value for a combination in the same direction when the accumulated weight value is smaller than the set value, and sets the weight to decrease as the direction shifts. This is accumulated for each equally divided point pixel, and when the accumulated value is larger than a set value, an output signal is output, and the density comparison circuit identifies this output signal as a black circle detection signal, Claim 1, characterized in that the determination of density characteristics is performed by comparison with a separately provided setting value.
People counting device as described in section. (3) When comparing and determining each pixel of interest, the comparison circuit compares the address of the pixel of interest from the address generation circuit with each value in the already counted address address accumulation table, and Claim 1 or 2, characterized in that when the distance on the image is smaller than a set value, the pixel of interest of the next address from the scanning address generation circuit is compared and judged. People counting device.