JP7311299B2 - Human recognition system and human recognition program - Google Patents

Description

The present invention relates to a human recognition system and a human recognition program, and in particular, for example, based on a distance image from a distance image sensor provided in areas where people stay or pass through, such as passages, gates, shops, event venues, and residences. The present invention relates to a human recognition system and a human recognition program for recognizing human attributes.

This type of technology can be used, for example, to count the number of people who pass through the entrance or gate of a room or building, to investigate the flow of people in an area, and to conduct various types of personal information-free video targeting of the general public who do not have a device. It can be used for measurement and the like.

For example, Patent Document 1 discloses an example of a people counting device that serves as the background of the present invention. In Patent Document 1, rectangle information is output as a person presence area, and the person presence state is estimated according to the aspect ratio or the length and width of the rectangle information. It is possible to count the number of people even if people overlap by determining that it is occurring and detecting the upper half, lower half, right half, and left half of the body.

JP 2012-108785 A [G06T 1/00, 7/00, G06M 11/00]

The people counting device of Patent Document 1 detects only people based on the rectangular ratio, so it is difficult to accurately count people when a plurality of people are in close contact with each other.

SUMMARY OF THE INVENTION Therefore, a primary object of the present invention is to provide a novel human recognition system and human recognition program.

Another object of the present invention is to provide a human recognition system and a human recognition program capable of accurately detecting the positions and counting the number of people by processing distance images.

In order to solve the above problems, the present invention employs the following configuration. It should be noted that reference numerals in parentheses, supplementary explanations, etc. indicate the correspondence with the embodiments described later in order to facilitate understanding of the present invention, and do not limit the present invention in any way.

A first invention includes a distance image sensor that photographs a measurement area in which a person stays or passes through from above and outputs a distance image, and a lump similar to the shape of a person is selected from the distance image and the lump is included. a chunk distance image extracting unit for extracting a chunk distance image; a cross-sectional image generating unit for generating a plurality of cross-sectional images obtained by cutting the range image from the chunk distance image at a plurality of cutting planes with different heights; Based on the cross-sectional image, the position and number of people included in the clump distance image are recognized, and the number of clump distance images and the positions and numbers of people included in the clump distance image are recognized in the measurement area. A human recognition system comprising a human recognition unit that recognizes the positions and number of people who

In the first invention, a distance image sensor (12) outputs a distance image obtained by photographing a measurement area in which a person stays or passes through from above. A clump distance image extraction unit (14, S9) extracts a clump distance image resembling a human shape from the distance image. A cross-sectional image generation unit (14, S11, S21, S23, S41) generates a plurality of cross-sectional images obtained by cutting the distance image from the mass distance image at a plurality of cutting planes having a height passing through at least a part of the human body. to generate A person recognition unit (14, S25-S35, S43) recognizes the positions and number of people included in the lump distance images based on the plurality of cross-sectional images, and recognizes the number of lump distance images and the number of lump distance images. The positions and number of people present in the measurement area are recognized according to the positions and number of people included in the measurement area.

According to the first invention, it is possible to accurately detect the positions of people or count the number of people in the measurement area by processing the distance image.

A second invention is according to the first invention, wherein the human recognition unit includes a head detection unit that detects a head by detecting an object that can be estimated as a human head from a plurality of cross-sectional images, The human recognition system recognizes the positions and the number of the heads detected by the head detection unit as the positions and the number of people present in the measurement area.

In the second invention, the human recognition unit (14, S25-S35, S43) includes a head detection unit (14 , S25), and further recognizes the positions and numbers of the detected heads as the positions and numbers of people present in the measurement area.

According to the second invention, since the head of a person is detected and the position and number of the head are recognized as the position and number of the person, the exact position and number of the person existing in the measurement area can be measured. .

A third invention is according to the second invention, wherein the head detection unit detects the head based on the size and shape of contour lines of an object included in the plurality of cross-sectional images. , is a human recognition system.

According to the third invention, the head can be detected more accurately.
A fourth invention is according to the second or third invention , wherein the head detection unit excludes an object that cannot be estimated as an independent head from the size, shape and range within a certain height range. A person recognition system that includes an exclusion unit.
In the fourth aspect, the exclusion unit (14, S31-S33), for example, in a certain height range, objects that cannot be separated from other objects (objects integrated with other objects) or Objects that cannot be estimated as the head are excluded as excluded objects.
In the fourth invention, for example, an object with a non-rounded shape such as a rectangle (a shape presumed to be luggage or the like), an object that is too large or too small for a head, etc. are excluded as excluded objects. , the person recognition unit can accurately recognize the position and number of people.

A fifth invention is according to any one of the second to fourth inventions, and the human recognition unit detects a plurality of heads in two or more cross-sectional images when the head detection unit detects a plurality of heads. It is a human recognition system that detects the number of people present in a measurement area according to the mutual distance of heads.

According to the fifth invention, even when a plurality of people are close to each other, it is possible to accurately count the number of people.

A sixth aspect of the present invention is a human recognition system computer equipped with a distance image sensor that photographs a measurement area in which a person stays or passes through from above and outputs a distance image , and detects a mass resembling a human shape from the distance image. Lump distance image extraction unit that selects and extracts a clump distance image containing the clump, and cross section that generates a plurality of cross-sectional images by cutting the distance image from the clump distance image at a plurality of cutting planes with different heights. Recognizing the position and number of people included in the clump distance images based on the surface image generation unit and the plurality of cross-sectional images, and determining the number of clump distance images and the positions and numbers of people included in the clump distance images It is a human recognition program that functions as a human recognition unit that recognizes the positions and number of people existing in the measurement area in response to .

Also in the sixth invention, the same effect as in the first invention can be expected.

According to this invention, it is possible to accurately detect the positions of people and count the number of people by processing the distance image.

The above object, other objects, features and advantages of the present invention will become more apparent from the following detailed description of the embodiments with reference to the drawings.

FIG. 1 is an illustrative view showing a human recognition system according to one embodiment of the present invention. FIG. 2 is a block diagram showing the FIG. 1 embodiment. 3 is an illustrative view showing one example of a memory map of the memory shown in FIG. 2; FIG. FIG. 4 is a flowchart showing an example of human recognition operation executed by the computer shown in FIG. FIG. 5 is an illustrative view showing one example of a distance image acquired in the FIG. 4 embodiment. FIG. 6 is an illustrative view showing one example of a distance image preprocessed in the embodiment of FIG. 4 . FIG. 7 is a detailed flow diagram of the person recognition step in the FIG. 4 embodiment. FIG. 8 is an illustrative view showing one example of a recognition result in the person recognition step. FIG. 9 is an illustrative view showing another example of recognition results in the person recognition step. FIG. 10 is an illustrative view showing another example of recognition results in the person recognition step. FIG. 11 is a flowchart showing an example of human recognition operation in another embodiment, which is executed by the computer shown in FIG. FIG. 12 is a flow diagram showing in detail the person recognition step in the FIG. 11 embodiment. FIG. 13 is an illustrative view showing one example of a recognition result in the person recognition step.

<First embodiment>
Referring to FIG. 1, a human recognition system 10 of this embodiment includes a range image sensor 12 provided above a gate GT that defines a measurement area in which a person PSN stays or passes. Also, the number of distance image sensors 12 included in the human recognition system 10 may be singular (one) or plural (two or more). In this embodiment, the human recognition system 10 includes one range image sensor 12 . As an example, the distance image sensor 12 has a viewing angle or angle of view of 40 degrees in the front, rear, left, and right directions.

This embodiment is an embodiment for recognizing a person existing in the measurement area defined by the gate GT. good. That is, the present invention can also be used as a system for recognizing people existing in a certain area of an open space, and the human recognition system of the present invention can also be used as a system for recognizing people in an open space. Let me point out in advance.

The distance image sensor 12 is a sensor for acquiring a distance image of a predetermined area (capturing area). However, the imaging area of the distance image sensor 12 includes at least the measurement area defined by the gate GT. That is, the distance image sensor 12 is a sensor for acquiring a distance image of the measurement area. The distance image sensor 12 emits light such as infrared light or laser, and captures the light reflected from the object (reflected light) by an optical sensor such as a CCD (Charge Coupled Device) sensor. The distance image sensor 12 measures the time taken for light to return for each pixel or calculates the phase difference, thereby measuring the actual distance to the object. A product called Xtion (registered trademark) manufactured by ASUS (registered trademark) is adopted as the distance image sensor 12 of the embodiment. Note that in other embodiments, the distance image sensor 12 uses a Microsoft (registered trademark) Kinect (registered trademark) sensor, a Panasonic (registered trademark) three-dimensional distance image sensor D-IMager (registered trademark), or the like. It is also possible to This type of sensor is sometimes called a three-dimensional distance measuring sensor, a 3D scanner, or the like.

As the distance image sensor, an omnidirectional laser rangefinder such as LiDAR (for example, an imaging unit LiDAR (HDL-32e) (trade name) manufactured by Velodyne), a stereo camera, and the like can be used.

As shown in FIG. 2, the distance image signal from the distance image sensor 12 is input to a computer 14, which stores the distance image signal as digital data in a memory 16 such as a RAM. Although the details will be described later, the computer 14 recognizes the number of people present in the measurement area from the result of processing the distance image data stored in the memory 16, and displays the result on the display 18 and/or a printer (Fig. not shown).

Memory 16 includes program storage area 20 and data storage area 22, as shown in FIG.

In the program storage area 20, a data acquisition program 20a, a preprocessing program 20b, a mass recognition program 20c and a person recognition program 20d are preset.

The data acquisition program 20a acquires the distance image signal from the distance image sensor 12 as distance image data (three-dimensional distance image data) of the entire distance image (overall distance image) of the shooting area, and stores it in the data storage area 22. This is a program stored in the obtained distance image data area 22a.

The preprocessing program 20b is a program for performing preprocessing such as noise removal on the range image data stored in the obtained range image data area 22a.

The lump recognition program 20c extracts the distance image data (lump distance image data) for the lump distance image having a predetermined shape from the distance image data for which preprocessing has been completed, and stores the lump distance image data in the data storage area 22. This is a program stored in area 22c.

The human recognition program 20d extracts the shape of a person's head from the lump distance image, detects the number of heads included in the lump distance image, and counts the number of heads in each lump distance image. , and stores it in the human recognition data area 22 d of the data storage area 22 .

The data storage area 22 includes a background distance image data area 22b in addition to the acquired distance image data area 22a, the cluster distance image data area 22c, and the human recognition data area 22d. This background distance image data area 22b is an area for storing a distance image (background distance image) from the distance image sensor 12, which is only the background without people at the gate G shown in FIG. The image data is used for preprocessing by the preprocessing program 20b as described later.

FIG. 4 is a flowchart showing the main processing of the human recognition operation of the computer 14 in the human recognition system 10 of this embodiment.

When the computer 14 starts the main processing of the human recognition operation, in step S1, according to the data acquisition program 20a, the distance image signal from the distance image sensor 12 is acquired as distance image data and stored in the acquired distance image data area 22a. do. An example of a distance image (three-dimensional distance image) 24 indicated by this distance image data is shown in FIG. This drawing, like the similar drawings that follow, shows in gray scale the distance image actually obtained and processed by the inventors in their experiments.

However, in FIG. 5, a dark gray portion and a light gray portion can be seen, and the dark portion is closer to the distance image sensor 12 than the light portion. Also, FIG. 5 shows a distance image when a plurality of people are present in the photographing area.

Next, in steps S3, S5 and S7, the computer 14 performs noise removal (step S3), distortion correction (step S5), and background information removal (step S7) on the range image as shown in FIG. 5 according to the preprocessing program 20b. preprocessing.

Since various noises are mixed in the measurement values of the range image sensor 12, such noises are removed by smoothing, median filtering, or other filters in step S3, if necessary. Specifically, outliers in the measured values are removed, smoothed (coarse quantization), objects that are too small and too thin are removed, and the result is stored in the obtained range image data area 22a (FIG. 3) or another Store in an area (not shown).

In the next step S5, distortion correction is performed as part of the preprocessing. In the case of a distance image sensor such as an infrared laser that measures a wide range by bending the optical axis with a lens or the like, distortion aberration or the like occurs. Stored in data area 22a (FIG. 3) or another area (not shown).

In step S7, background information removal processing is executed. To detect only people and moving objects, it learns the reflections of static objects and removes them from measurements. Specifically, background distance image data stored in advance is read from the background distance image data area 22b of the data storage area 22 (FIG. 3), and this background distance image data is subtracted from the distance image data being processed. At the same time, the data outside the preset measurement area is erased, and the result (distance image data after preprocessing) is stored in the obtained distance image data area 22a (FIG. 3).

An example of a distance image 24 after such preprocessing is shown in FIG. In FIG. 6, a dark gray portion 26 is the distance image remaining as a result of the preprocessing, and a light gray portion 28 is removed for reasons such as being out of range.

In step S9, lump separation processing is executed. Here, from the distance image data (FIG. 6) that has undergone preprocessing, lumps similar to the shape of a person are selected from the dark gray portion 26, and an image of a predetermined shape (clump distance image) 30 (FIG. 8A), and separated into distance image data (lump distance image data) for each lump distance image 30 . As a method for sorting out the clump distance image 30, for example, in the dark gray portion 26 included in the preprocessed distance image, when considering the two-dimensional size, the X axis and the Y axis are each longer than a predetermined length. An image of a size that has a length and a predetermined length or less, the ratio of the area of the rectangle circumscribing the image to the area of the image is a certain amount or more, and the aspect ratio is within the range of the predetermined ratio. An image whose outline resembles the shape of a person is extracted as a block distance image. Therefore, an image that does not resemble the shape of a person, such as an image with a shape that is not rounded such as a rectangle, or an image that is too large or too small, is not extracted as a clustered distance image. In other words, an image that does not resemble the shape of a person is excluded from the object of human recognition processing. Then, the lumped distance image data separated from the distance image data for which preprocessing has been completed is stored in the lumped distance image data area 22c.

As another selection method of the cluster distance image 30, instead of the distance image, a cross-sectional image of a predetermined height (for example, 1200 mm) is generated, and out of the images included in the cross-sectional image, the contour is An image similar to the shape of a person may be extracted as a block distance image.

In subsequent step S11, human recognition processing is executed for each cluster distance image 30 according to the subroutine shown in FIG. Briefly, the human recognition process generates a plurality of cross-sectional images 32 from the clump distance image data of the clump distance image 30, and based on the plurality of cross-sectional images 32, the clump distance image 30. This is a process for recognizing the number of people included. Each of the plurality of cross-sectional images 32 is an image obtained by cutting the lump distance image data at an arbitrary height (in the Z-axis direction) within a range (cutting range) of, for example, 600-1800 mm (see FIG. 8 ( B), FIGS. 9(B) and 10(B)).

When starting the human recognition process, the computer 14 sets the cutting height to the first height in the first step S21. However, the first height is either the upper limit or the lower limit of the cutting range. In this embodiment, the first height is the upper limit height of the cutting range, for example 1800 mm.

Subsequently, in step S23, as shown in FIGS. 8(B), 9(B) and 10(B), a cross-sectional image 32 is generated by cutting the lump distance image at the cutting height. In step S25, the object 34 is detected from the generated cross-sectional image 32. FIG. However, the object 34 is an object that resembles the shape of a human head (an object that can be estimated as a human head) included in the cross-sectional image 32 . Here, based on the cutting height of the cross-sectional image 32, the size of the contour line of the object included in the cross-sectional image 32, the shape of the contour line (for example, a substantially round shape), and the like, the human head in the cross-sectional image 32 Detects whether an object 34 that can be estimated as exists. In other words, it can be estimated that a substantially round closed area within a certain height range is a person's head. Further, in step S25, if a plurality of objects 34 that can be estimated as human heads are separated from each other, they are detected as separate objects 34. FIG.

Subsequently, in step S27, it is determined whether or not a new object exists. A new object is an object 34 that is different from the object 34 already detected in the cross-sectional image 32 of another cutting height, which is detected in the previous step S25.

Whether or not the object 34 detected in the previous step S25 is different from the already detected object 34 depends on the height of the cross-sectional image 32, the distance between the center positions of each object 34, and the overlap between the objects 34. It is judged according to the ratio of the area. In other words, it is determined whether the object 34 detected in the previous step S25 is the same object 34 as the already detected object 34 or a different object 34 according to these parameters for identity determination. . A case where a plurality of objects 34 exist in one lump distance image is a case where a plurality of people are clinging to each other, embracing each other, holding a child, or when a plurality of people are in close contact. are close to each other (FIG. 8(A) or FIG. 10(A)). Assuming such a case, it is determined whether the object 34 detected in step S25 is different from the object 34 already detected.

For example, in the third cross-sectional image 32C from the top in FIG. 8B or the second cross-sectional image 32B from the top in FIG. A different second object 34b is detected. In these examples, the first object 34a and the second object 34b are recognized as different objects from the height of the cross-sectional image 32, the distance between the center positions, the ratio of overlapping areas, and the like. That is, the second object 34b is recognized as a new object. Therefore, in the examples of FIGS. 8B and 10B, it is determined that a new object exists when the second object 34b is detected.

Returning to FIG. 7, if "YES" in step S27, that is, if it is determined that a new object exists, the new object is registered in the tracking list in step S29, and the process proceeds to step S35. The tracking list is a list for tracking an object 34 detected from a certain cross-sectional image 32 and determining the change tendency of the object 34 on the cross-sectional image 32 of another cutting height. This tracking list is stored in the person recognition data area 22d.

On the other hand, if "NO" in step S27, that is, if it is determined that the new object does not exist, it is determined in step S31 whether an excluded object exists. An excluded object is an object that cannot be estimated as an independent head from the size, shape, range, etc. within a certain height range. For example, excluded objects include an object 34 that cannot be separated from other objects 34 (an object that is integrated with other objects) within a certain height range, and an object 34 that cannot be estimated as a head based on the changing tendency of the object 34. etc.

Note that the object 34 that cannot be estimated as the head includes an object 34 that has a shape that is not rounded such as a rectangle (a shape that is estimated to be luggage or the like), or an object 34 that is too large or too small as a head. Object 34 etc. correspond.

Also, even if an object is integrated with another object at a relatively low height (for example, 1000 mm or less), if it is separated within a certain height range, it will not be recognized as an excluded object. There is

For example, in the first cross-sectional image 32A from the top of FIG. 9B, a first object 34a and a second object 34b are detected. Since the first object 34a and the second object 34b are separated from each other, they are recognized as separate objects at this stage. Accordingly, each of the first object 34a and the second object 34b are individually registered on the tracking list. However, in the cross-sectional images 32B-32D below the second from the top of FIG. 9B, only one object is detected, including both the first object 34a and the second object 34b. In this case, it is determined that one object is originally detected in the first cross-sectional image 32A, but two objects are detected due to slight unevenness or the like. Therefore, in cross-sectional images 32B-32D, only one object is tracked and the other object is recognized as an excluded object.

In the example shown in FIG. 9, only the first object 34a having a large area in the cross-sectional image 32A is tracked, and the second object 34b is recognized as an excluded object. It should be noted that slight unevenness as shown in FIG. 9A may occur due to the posture of the person included in the clump distance image 30, the headgear such as a hat, and the like.

Returning to FIG. 7, if "YES" in step S31, that is, if it is determined that an excluded object exists, in step S33 the excluded object is excluded from the tracking list (deleted from the tracking list), and in step S35 proceed to In this way, objects that cannot be estimated as independent heads to a certain height range are removed (pruned) from the tracking list. On the other hand, if "NO" in step S31, that is, if it is determined that the exclusion object does not exist, the process proceeds to step S35 without going through step S33.

Then, in step S35, it is determined whether or not the cutting height is the second height. However, the second height is the height of the upper limit or the lower limit of the cutting range, whichever is different from the first height. In this embodiment, the first height is the upper limit height of the cutting range, so the second height is the lower limit height of the cutting range, eg, 600 mm.

If "YES" in step S35, that is, if it is determined that the cutting height is the second height, the process proceeds to step S41, which will be described later. On the other hand, if "NO" in step S35, that is, if it is determined that the cutting height is not the second height, it is determined in step S37 whether the area (cross-sectional area) of the object 34 is equal to or larger than a predetermined area. do. Here, the total area of all the objects 34 included in the clump distance image 30 is a predetermined percentage (75 to 90%) of the total area of the clump distance image 30 (the total area of the cross-sectional image 32). determine whether to exceed

If "NO" in step S37, that is, if it is determined that the area of the object is not equal to or greater than the predetermined area, then in step S39 the cutting height is changed and the process returns to step S23. Here, the cutting height is changed from the first height to approach the second height. That is, the cutting height is changed from either the upper limit or the lower limit of the cutting range so as to approach the other height. In this example, the cutting height is changed from the upper cutting range height (1800 mm) to approach the lower cutting range height (600 mm).

The width (change width) by which the cutting height is changed is, for example, 5 to 40 mm. However, the amount of change in the cutting height is appropriately adjusted depending on the detection accuracy of the distance image sensor 12 and the condition of the object. In this embodiment, the change width of the cutting height is initially set to 10 to 40 mm in consideration of the detection accuracy of the distance image sensor 12 . However, if the total area of the objects included in the cross-sectional image 32 exceeds a predetermined percentage (10 to 20%) of the total area of the clump distance image 30 (the total area of the cross-sectional image 32), or If a plurality of objects are included in the cross-sectional image 32, the change width of the cutting height is set small (for example, 5 to 20 mm). Further, when the immediately preceding cross-sectional image 32 does not include an object, the change width of the cutting height may be set large.

On the other hand, if "YES" in step S37, that is, if it is determined that the area of the object is equal to or larger than the predetermined area, then in step S41, the number of heads of the objects 34 registered in the tracking list is The number of people included in the distance image 30 is counted, and in step S43, the position of the head of the object 34 registered in the tracking list is detected as the position of the person included in the clump distance image 30, and the main processing is performed. return.

As described above, according to the first embodiment, by executing the human recognition process for each lumped distance image 30, the positions and numbers of persons for each lumped distance image 30 are recognized. The number of people present in the measurement area (whole distance image) can be recognized by totaling the number of people for each cluster distance image 30 included in the whole distance image. In addition, it is possible to detect the position of a person existing in the measurement area (entire distance image). That is, by processing a one-frame distance image captured at a certain timing, it is possible to accurately count the number of people and detect the positions of people in the one-frame distance image.

Furthermore, according to the first embodiment, it is detected whether or not there is an object 34 that can be estimated as a human head from the size and shape of the outline of the object included in the cross-sectional image 32. , the head can be detected accurately.

Furthermore, according to the first embodiment, when a plurality of objects 34 are detected in a certain cross-sectional image 32, another Since it detects whether it is an object 34 or not, the number of people can be accurately counted even when a plurality of people are close to each other. For example, even when a parent is holding a child, the parent and child can be detected separately.

<Second embodiment>
The human recognition system 10 of the second embodiment omits the lump separation process, generates a plurality of cross-sectional images 32 from the overall distance image 24, and recognizes the number of people present in the measurement area (entire distance image). It is the same as the first embodiment except that it is made to The human recognition system 10 of the second embodiment will be described below, but the content overlapping the description of the first embodiment will be omitted.

As shown in FIG. 11, in the human recognition system 10 of the second embodiment, the cluster separation process of step S9 described in the first embodiment is omitted in the main process of the human recognition operation. Therefore, in the second embodiment, human recognition processing is executed in step S11 based on the overall distance image 24 for which the preprocessing up to step S7 has been completed.

When the computer 14 starts the human recognition process, in step S21, the cutting height is set to the first height. to generate

Then, the processing from step S25 to step S35 is executed in the same manner as in the first embodiment, and if "YES" in step S35, that is, if it is determined that the cutting height is the second height, In step S53, the number of objects 34 registered in the tracking list is counted as the number of persons included in the overall distance image 24 (measurement area). is detected as the position of the person included in the overall distance image 24 (measurement area), and the process returns to the main processing. Also, in the human recognition process of the second embodiment, step S37 described in the first embodiment is omitted. Therefore, if "NO" in step S35, that is, if it is determined that the cutting height is not the second height, then in step S39 the cutting height is changed and the process returns to step S23.

As described above, according to the second embodiment, as shown in FIG. 13, the number of objects 34 included in the overall distance image 24 is calculated by executing the human recognition process based on the overall distance image 24. By counting the number of people included in the image 24, the number of people present in the measurement area can be recognized. In addition, it is possible to detect the position of a person existing in the measurement area (entire distance image). That is, it is possible to accurately count the number of people by processing a one-frame distance image captured at a certain timing.

In the above embodiment, all the steps in FIG. 4 (FIGS. 7, 11, and 12) were described as being executed by one computer 14, but multiple computers may be used, or specific processes may be executed. It may be processed by a dedicated processing circuit such as a DSP instead of a computer.

It should be noted that the specific numerical values such as the angles, distances, lengths of time, ratios, and widths of change given above are merely examples, and can be changed as needed.

DESCRIPTION OF SYMBOLS 10... Person recognition system 12... Range image sensor 14... Computer 16... Memory 18... Display

Claims (6)

a distance image sensor that captures a measurement area in which people stay or pass through from above and outputs a distance image;
a lump distance image extraction unit that selects a lump similar to a human shape from the distance image and extracts a lump distance image containing the lump;
a cross-sectional image generating unit for generating a plurality of cross-sectional images obtained by cutting the distance image from the clustered distance image at a plurality of cutting planes having different heights; and generating the clustered distance image based on the plurality of cross-sectional images. recognizing the positions and number of people included in the measurement area, and recognizing the positions and the number of people present in the measurement area according to the number of the clump distance images and the positions and the number of people included in the clump distance images. A human recognition system comprising a human recognition unit that The human recognition unit includes a head detection unit that detects a head by detecting an object that can be presumed to be a human head from the plurality of cross-sectional images, and detects the head detected by the head detection unit. 2. The person recognition system according to claim 1, wherein the positions and numbers are recognized as the positions and numbers of persons existing in said measurement area. 3. The human recognition system according to claim 2, wherein said head detection unit detects the head based on the size and shape of outlines of objects included in said plurality of cross-sectional images. 4. The human recognition system according to claim 2 , wherein said head detection unit includes an exclusion unit that excludes an object that cannot be estimated as an independent head from the size, shape and range within a certain height range. When the head detecting unit detects a plurality of heads, the human recognizing unit detects a person existing in the measurement region according to the mutual distances of the plurality of heads in the two or more cross-sectional images. 5. The human recognition system according to any one of claims 2 to 4, which detects the number of . A human recognition system computer equipped with a distance image sensor that captures from above a measurement area where a person stays or passes through and outputs a distance image,
a lump distance image extraction unit that selects a lump similar to a human shape from the distance image and extracts a lump distance image containing the lump;
a cross-sectional image generation unit for generating a plurality of cross-sectional images obtained by cutting the distance image from the clustered distance image at a plurality of cutting planes having different heights; and generating the clustered distance image based on the plurality of cross-sectional images. and recognizes the positions and number of people present in the measurement area according to the number of clump distance images and the positions and numbers of people included in the clump distance images. A human recognition program that functions as a human recognition unit.

Images