JPH07244714A - Object detection device - Google Patents

Abstract

PURPOSE:To provide the object detection device which has high detection precision and superior practicality. CONSTITUTION:Plural distance measuring sensors Sij are arranged uniformly over the entire area above a monitor area, and plural three-dimensional data in the monitor area are obtained from fitting position data and measurement data of the distance measuring sensors Sij. Those three-dimensional data are classified into plural groups by a classifying process part 14. A feature quantity calculation part 15 calculates the feature quantities of the respective groups. Then when an attribute decision part 16 decides the attributes of the groups from the feature quantities of the respective groups, a path specification part 18 extracts only a group having an attribute pertinent to a human being. An extracted group storage part 17 is stored with the latest extracted data and extracted data which are one sample before, and the path specification part 18 compares both of them to specify respective path. When a specified path passes a specific position in the monitor area, an entry/exit counting part 20 detects that and counts how many persons enter and exit.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an object detecting device for detecting a predetermined object within a specific surveillance area, and more particularly, the present invention relates to a route along which a moving object such as a human passes through the surveillance area. The present invention relates to an object detection device that has a function of tracking an object and a function of detecting passage of a specific position and counting the number of passages.

[0002]

2. Description of the Related Art Conventionally, in order to detect a moving path of a predetermined object in a specific surveillance area, a television camera is installed above the area and an image obtained from the camera is processed in real time to achieve a purpose. A method of detecting a moving object is used.

Further, in order to detect a person or an object passing through a gate or a door and to count entrances and exits, a plurality of photoelectric switches or touch switches are provided along the traveling direction, and a detection signal of each switch is provided. The method of detecting the movement route according to the transition of the above and counting the number of times that the movement route has passed the position of the gate or the door is implemented.

[0004]

In the first method, the cost is high because the camera and the image processing device are expensive, and accurate data cannot be obtained unless the camera is installed in an optically stable environment. Therefore, there is a problem that it is not practical.

Further, in the second method, since the switches are arranged along the traveling direction, sufficient accuracy can be obtained for detection and counting of, for example, luggage carried in at a constant interval. However, since it is impossible to recognize the three-dimensional shape of an object with this method, when the detection target is a human being or an animal, the motion or baggage is erroneously detected, and the accuracy of the data is remarkably high. descend.

The present invention has been made in view of the above problem, and measures a plurality of three-dimensional data in a monitoring area, divides these three-dimensional data into groups, and extracts a group having a specific attribute. Therefore, the detection accuracy is high,
An object is to provide an object detection device having excellent practicability.

[0007]

SUMMARY OF THE INVENTION The present invention is an object detecting device for detecting the presence or absence of a predetermined object in a specific monitoring area, which is data for measuring a plurality of three-dimensional data in the monitoring area. Measuring means, grouping means for classifying each three-dimensional data measured by the data measuring means into a plurality of groups, and characteristic amount calculating means for calculating a characteristic amount for each group classified by the grouping means And an attribute determination unit that determines the attribute of each group based on the characteristic amount calculated by the characteristic amount calculation unit, and an attribute that corresponds to the predetermined object in the monitoring area based on the determination result of the attribute determination unit. And a detection means for detecting the presence / absence of a group that the user has.

According to a second aspect of the present invention, the data measuring means includes a plurality of distance measuring sensors that are evenly arranged over the entire upper surface of the monitoring area and measure the distance in the vertical direction.

In the invention of claim 3, the detecting means detects the presence or absence of a group having an attribute corresponding to a human.

In the invention of claim 4, the attribute determining means is means for determining the attribute of each group by comparing the feature amount of each group with a reference vector learned by a neural network program. According to the fifth aspect of the invention, the feature amount of each group is substituted into a predetermined discriminant function formula to determine whether each group has an attribute corresponding to the predetermined object. Claim 6
In the invention, the attribute determining means is means for determining the attribute of each group by executing a fuzzy inference operation using the feature amount of each group.

According to a seventh aspect of the present invention, there is provided an object detecting device for detecting a movement path of a predetermined object within a specific monitoring area, wherein the data measuring means measures a plurality of three-dimensional data within the monitoring area. Grouping means for classifying each three-dimensional data measured by the data measuring means into a plurality of groups, and feature amount calculating means for calculating a feature amount for each group classified by the grouping means, An attribute determining unit that determines the attribute of each group based on the characteristic amount calculated by the characteristic amount calculating unit, and a group that has an attribute corresponding to the predetermined object in the monitoring area based on the determination result of the attribute determining unit Position detecting means for detecting the position of the object, and the position of a group having an attribute corresponding to the predetermined object by repeatedly operating the respective means at predetermined time intervals. Detecting a temporal change, but this change in the position of the configuration of a route specifying means for specifying a movement path of said predetermined object.

An eighth aspect of the present invention is an object detecting device for detecting that a predetermined object passes through a specific position within the specific monitoring area. According to the present invention, in addition to the configuration of the invention of claim 7, there is provided a specific position passage detecting means for detecting that the movement path specified by the path specifying means has passed a specific position in the monitoring area. .

[0013]

Operation: A plurality of three data in the monitoring area are provided by the data measuring means.
After measuring the dimensional data, the grouping means classifies these data into a plurality of groups. The feature amount calculating means calculates the feature amount for each group, and the attribute determining means determines the attribute of the group from the obtained feature amount. The detection means detects a group having the attribute of the predetermined object as a detection target from each group.

Similarly, in the invention of claim 7, after a plurality of three-dimensional data in the monitoring area are grouped to calculate the feature amount for each group, the attribute of the group is determined from the obtained feature amount. To be done. The position detecting means detects the position of a group having a predetermined object attribute from each of these groups. The route specifying unit repeatedly executes the above-described operation at predetermined time intervals, detects a change in the position of a group having a predetermined object attribute, and specifies this as a moving path to be detected.

Further, in the invention of claim 8, the specific position detecting means detects that the movement route detected by the above procedure has passed a specific position in the monitoring area.

According to the second aspect of the present invention, since the distance measuring sensors are evenly arranged over the entire upper surface of the monitoring area, the distance measuring sensor outputs the distance data output from the mounting positions of the respective distance measuring sensors.
Three-dimensional data for the entire monitoring area is required.

According to the third aspect of the present invention, a group having an attribute that particularly corresponds to a human being is detected separately from an object such as baggage or a motion of a limb.

In the inventions of claims 4, 5, and 6, the accuracy of attribute determination of each group is improved by using the neural network, the discriminant function, and the fuzzy inference, respectively.

[0019]

1 (1) to 1 (3) show an entrance / exit gate 1 to which an entrance / exit detector according to an embodiment of the present invention is applied. The front view is shown in FIG.
A plan view is shown in (2), and a vertical sectional view of the gate is shown in FIG. 1 (3).

The entrance / exit detector according to this embodiment detects a person who enters or exits through the gate 1 and counts the number of entrance / exit. The ceiling surface 2 of the gate 1 is shown in FIG. As shown in the drawing, I distance measuring sensors S _ij are arranged in the horizontal direction and J pieces in the vertical direction (total of K pieces), which are evenly arranged at intervals d ₀ .

The distance measuring sensor S _ij has a structure in which a pair of photosensors and arithmetic circuits are provided on a microchip, and a reflection pattern to be detected by these photosensors is provided in the horizontal central portion of the ceiling surface 2. The light projection pattern generation unit 4 for generating the light is arranged. As shown in FIG. 3A, the light projection pattern generation unit 4 includes an auxiliary light source 5 such as an LED or a miniature bulb 5 and a pair of cylindrical lenses 6a and 6a.
b, a visible light cut filter 7, and a striped band film 8. The plurality of auxiliary light sources 5 are arranged at equal intervals in the lateral direction of the central portion of the ceiling surface 2 (here, the distance d ₀ is the same as that of the distance measuring sensor).
For each of the cylindrical lenses 6a, 6
By passing through b, the visible hole cut filter 7 and the band film 8, a striped light projection pattern as shown in FIG. 3B is generated.

As shown in FIG. 1C, this projection pattern is applied to the entire floor surface 9 directly below the gate 2, and the floor surface 9,
Alternatively, it is reflected by an object or a person on the floor surface 9 to generate a striped reflection pattern.

Each distance measuring sensor S _ij is installed so that the direction in which the above-mentioned pair of photosensors are arranged is orthogonal to the striped pattern of the strip film 8, whereby each photosensor detects a vertical striped reflection pattern. . There is a phase difference in the detection pattern of each photo sensor due to the difference in the position of the photo sensor, and the calculation unit of the distance measuring sensor executes a predetermined calculation using this phase difference and the distance between the photo sensors. ,
The distance from the mounting position of the distance measuring sensor to the reflecting surface is calculated.

In FIG. 1, numeral 3 indicates an entrance / exit passage, arrow a indicates the direction of entry, and arrow b indicates the direction of exit.

With the above structure, the space directly below the gate 1 functions as a monitoring area for detecting a person passing through the gate. FIG. 4 shows the monitoring area 40 in a plan view. Within this monitoring area, the x-axis is set in the horizontal direction, the y-axis is set in the vertical direction, and the z-axis is set in the height direction, and K based on the mounting position of the distance measuring sensor S _ij and the measured value of each sensor is set.
The three-dimensional data TR are specified. The entrance / exit detection device detects a person passing through the gate 1 and counts the number thereof by performing a process described later using these three-dimensional data TR. In addition, LN in the drawing is a check line set in the monitoring area 40 for use in the counting process described later.

FIG. 5 shows a circuit configuration of the entrance / exit detection device. In addition to the distance measuring sensor S _ij described above, a distance calculation unit 10, an environment setting unit 11, an environment storage unit 12, a comparison processing unit 1 are provided.
3, classification processing unit 14, feature amount calculation unit 15, attribute determination unit 1
6, it includes an extraction group storage unit 17, a route identification unit 18, a movement information storage unit 19, an entrance / exit counting unit 20, a display unit 21, a control processing unit 22, and the like as its configuration.

The distance data obtained by each of the distance measuring sensors S _ij is supplied to the distance calculation unit 10 and converted into height data from the floor surface to the object. FIG. 6 shows the relationship between the height data and the measurement data of the distance measuring sensor S _ij . LE in the figure
N _ij is the mounting height of the distance measuring sensor S _ij , that is, the distance from the floor surface to the distance measuring sensor S _ij , and SEN _ij is the distance measuring sensor S _ij.
The measurement data measured by _ij , that is, the distance from the distance measuring sensor to the object (human in the illustrated example) is shown.

Assuming that the height of the object is H _ij , this height H _ij can be calculated by the following equation (1).

[0029]

[Equation 1]

Returning to FIG. 5, the distance calculating section 10 determines the distance measuring sensors S _ij at predetermined time intervals (for example, 10 ms).
The measurement data SEN _ij is read from and the above equation (1) is executed to calculate the height H _ij of the object. This height H _ij is
It is output to the environment setting unit 11 at the initial setting and to the comparison processing unit 13 at the operating state.

The environment setting section 11 grasps beforehand the fluctuation range of the output value due to optical noise or machine difference for each distance measuring sensor S _ij , and obtains the height H obtained by the above equation (1). _This is for removing the error of _ij . The height H _ij obtained for each distance measuring sensor S _ij is set as an initial value IN _ij of height, and this is set as the initial value IN _ij of the environment storage unit. Output to 12. The environment storage unit 12 stores the initial value IN _ij of the height obtained from each distance measuring sensor S _ij, and also stores the mounting height LEN _ij of each distance measuring sensor and the mounting position of each sensor described above. .

The comparison processing unit 13 compares the height H _ij given by the distance calculation unit 10 with the initial value IN _ij of the height stored in the environment storage unit 12 when the apparatus is in operation, and the comparison result. The height H _ij is corrected to the optimum value by.

The classification processing unit 14 uses the three-dimensional data consisting of the x and y coordinates of the mounting position and the finally obtained height H _ij for each distance measuring sensor S _ij , and the grouping processing described later. Then, each three-dimensional data is classified into a plurality of groups.

The characteristic amount calculation unit 15 executes a predetermined calculation using the data of each group classified by the classification processing unit 14 to calculate various characteristic amounts described later.

The attribute determination unit 16 operates by a learning vector quantization (LVQ) program, which is a type of neural network, and is given in advance for objects having different attributes such as humans, luggage, animals, and floor surfaces. The feature amount is calculated from a plurality of models, and the feature amount pattern for each attribute is stored in the form of a multidimensional reference vector. Further, the attribute determining unit 16 calculates a distance between each feature vector of each group calculated by the feature calculating unit 15 and each reference vector when the apparatus is in operation, and as a result, the reference vector at the shortest distance is obtained. Is determined to be the closest to the group, and the attribute of this reference vector is confirmed as the attribute of the group.

The route specifying unit 18 checks the attributes given to each group, extracts a group having an attribute corresponding to a human, and extracts the respective feature quantities from the extracted group storage unit 17
To memorize. The extracted group storage unit 17 stores the latest data and the data one sample before regarding the group extracted as a person. The route identifying unit 18 corresponds to the latest data 1
The data before the sample is extracted, and the movement route of the group is specified from the change of both data.

The movement information storage unit 19 is for storing the route specified by the route specifying unit 18, and stores the latest position data of the center of the group and the position data of one sample before.

The entrance / exit counting unit 19 counts the number of people entering and leaving by detecting that the group determined to be a human moves through the check line LN to the entrance side of arrow a or the exit side of b. The moving route of each group stored in the moving information storage unit 19 is sequentially compared with the position of the check line LN to count the number of visitors.

The display unit 21 is an LED or liquid crystal display device, a CRT display device, or the like.
The number of visitors and the number of exits counted by the entrance / exit counting unit 19 are displayed in real time.

The control processing section 22 includes a CPU 23 and a ROM 2
4, RAM 25, CPU23, R
Based on the control program stored in the OM 24, the operation of each unit is controlled in series. Further, the RAM 25 includes a plurality of areas as work areas for operating the respective processing units, in addition to an area for storing the measurement data SEN _ij and the height H _ij for each distance measuring sensor S _ij .

In FIG. 7, the data structure of each memory of the RAM 25, the extraction group storage unit 17, and the movement information storage unit 19 is shown as an environment storage unit 12, a comparison processing unit 13, and a classification processing unit 1.
4, feature quantity calculation unit 15, attribute determination unit 16, route identification unit 1
8 shows the relationship between each part of the entrance / exit counting unit 20.

In the RAM 25, as the above-mentioned work area, a three-dimensional data storage area 26 for storing three-dimensional data corresponding to each distance measuring sensor S _ij and data for each group obtained in the course of the classification process. A group data storage area 27 for storing (hereinafter referred to as “group data”), a characteristic amount storage area 28 for storing the characteristic amount calculated for each group, and the like are provided.

The three-dimensional data storage area 26 is provided with _K areas r _{1 to} r K corresponding to the number of distance measuring sensors S _ij , and each area is stored in the environment storage unit 12. and mounting positions of each distance measuring sensor S _ij, the height H _ij and the three-dimensional data consisting of each distance measuring sensor S _ij given from the comparison processing unit _{13 (x k, y k,} z k) are stored. The classification processing unit 14 reads each three-dimensional data, performs a predetermined classification process, and stores the result in the group data storage area 27.

The group data storage area 27 stores s _{1 to} s for the N groups obtained by the classification process.
_In each area of _N , the central coordinates G _n of each group, and the three-dimensional data E _n extracted as elements belonging to the group
(Hereinafter referred to as “element data E _n ”), an average value AD _n of the distance between each element and the center of the group (hereinafter referred to as “average distance AD _n between element / center”), each of an invalidation flag F _n described later, and the like. Data is stored.

The stored contents of the group data storage area 27 are rewritten many times according to the processing of the classification processing section 14. The feature amount calculation unit 15 reads the final group data, and as the feature amount for each group, the number of belonging elements S _n , the group center x coordinate CX _n , the y coordinate CY _n , and the z coordinate of each element. Mean value AZ _n and variance value VZ _n , and the distance to the group at the shortest position among other groups (hereinafter referred to as the “distance to the shortest group”) DM
Each _n is calculated and sequentially stored in the areas f _{1 to} f _N of the feature amount storage area 28.

Of the above feature quantities, the number of elements S _n corresponds to the area of the object, and the x and y coordinates CX of the center of the group.
_n and CY _n represent the position of the object. Further, the average value AZ _n and the variance value VZ _n of the z-coordinates of each element are data on the three-dimensional shape of the object, and the distance DM _{n from the} shortest group represents the distance to an object existing nearby. It is a thing.

The feature quantity of each group is stored in the feature quantity storage area 28.
Stored in a predetermined area f _n of the attribute determination unit 1
In step 6, the group attribute determination process using these feature quantities is performed. As a result of this determination, the attribute data Q _{n of the} group is obtained and stored in the area f _n as the seventh feature amount.

The extraction group storage section 17 stores the latest data in the latest data storage area NM (hereinafter simply referred to as "area N").
M ”) and a comparison data storage area OM (hereinafter simply referred to as“ area OM ”) that stores data of one sample before, that is, 10 ms before. The area NM is
N stored in the feature amount storage area 28 in the RAM 25
This is for extracting M groups having attribute data corresponding to a human from the individual groups and storing the characteristic amounts thereof. When the route specifying unit 18 performs this extraction work, Each feature amount is sequentially stored in the area MN.

In the figure, in each area of g _{1 to} g _M , for each group extracted from the feature amount storage area 28, the number of elements NS _{m of} the group and the x-coordinate N of the group center N
X _m , y coordinate NY _m , z coordinate average value NA _m of each element,
Variance value NV _{m of} z coordinate, distance ND to the shortest group
_m is stored in each.

Each data in the area NM is moved to the area OM before the next sampling by the distance measuring sensor. In the illustrated example, the number of elements OS is set in each region of h _{1 to} h _L for each of _L groups extracted one sample before.
_l, a group center x coordinate OX _l, y coordinates OY _l, mean OA _l and variance values OV _l of z coordinates of each element, each feature quantity distance OD _l of the shortest groups are stored.

The movement information storage section 19 stores a change in the position of the center of the group as movement information for each group stored in the area NM. The route identifying unit 18 extracts the data one sample before for each group stored in the region NM from the region OM, and in the regions q _{1 to} q _M of the movement information storage unit 19, the latest position of the center of each group. (NX _m , NY _m ) and the position (PX _m ,
PY _m ) and

The group data storage area 27, the feature quantity storage area 28, and the N of the extraction group storage unit 17 are stored.
The M and OM areas and the movement information storage unit 19 have sufficient areas to store the data that varies with each sampling, and unused areas 29 to 33 remain in each memory even after the data is stored. Has been done.

The detailed control procedure in each of the above-mentioned parts will be described below in order using a flow chart. First of all,
A procedure for _obtaining an initial height value IN _ij (hereinafter simply referred to as “initial value IN _ij ”) for each distance measuring sensor S _ij will be described. After each part of the device is reset in step 1, step 2,
At 3, the counter j in the y-axis direction and the counter i in the x-axis direction are set to 1, respectively. Next, the distance calculation unit 10
_{Takes in} the measurement values of each distance measuring sensor S _{ij a} predetermined number of times, and determines the maximum value as SEN _ij (step 4).

In the next step 5, the distance calculation unit 10 executes the above-mentioned equation (1) to calculate the height H _ij, and outputs this value to the environment setting unit 11. The environment setting unit 11 determines whether or not the given height H _ij is larger than 0. When it is larger than 0, a value obtained by adding a predetermined offset value α to this height H _ij is set as an initial value IN _ij . (Step 7). When the height H _ij is 0 or less, the substantial value of the height H _ij at this time is determined to be "0", and the offset value α is set as the initial value IN _ij (step 8).

The initial value IN _ij set in either step 7 or 8 is stored in the environment storage unit 12 (step 9). Hereinafter, the counter i, j is being repeated processing of steps 4-9 is sequentially incremented, the initial value IN _ij height of each distance measuring sensor S _ij environmental storage unit 1
Stored in 2.

FIG. 9 shows a procedure for calculating the height H _ij when the apparatus is operating. Distance calculation unit 10 sets the measurement data SEN _ij from the distance measuring sensor S _ij and incorporates at predetermined intervals (10ms intervals), first similar to the procedure of FIG. 8, the counter i, j to "1" Then, after the measurement data SEN ₁₁ of the first distance measuring sensor S ₁₁ is fetched, the formula (1) is executed to calculate the height H ₁₁ (steps 1 to 4).

This height H ₁₁ is given to the comparison processing unit 13. The comparison processing unit 13 reads the initial value IN ₁₁ stored in the environment storage unit 12 and compares the given height H ₁₁ with this initial value IN ₁₁ (step 5). If it is determined by this comparison that the height H ₁₁ is larger than the initial value IN ₁₁ , step 5 becomes “YES”, and the height H ₁₁ is corrected to a value subtracted by the initial value IN ₁₁ ( Step 6). On the other hand, when it is determined that the height H ₁₁ is less than or equal to the initial value IN _11, it is determined that there is no object at this position, and the height H ₁₁ is set to “0” (step 7). .

When the value of the height H _ij is finally determined in step 6 or step 7, this height H _ij is output to the three-dimensional data storage area 26 of the RAM 25 (step 8). Hereinafter in step 9-12, the counter i, j is being sequentially incremented procedure repeated execution of step 3 to step 8, final height H _ij is three-dimensional data storage area for each distance measuring sensor S _ij 26.

FIG. 10 shows the procedure of grouping processing by the classification processing unit 14. In the following procedure, parameters such as the reference value NC _{0 of the} number of groups, the lower limit value N _{min of the} number of elements, the upper limit value V _{max of the} variance, and the minimum value CD _min of the distance between the group centers are appropriately used. These parameters are set as various determination criteria for performing group division processing and group fusion processing, and are stored in the ROM 24 in advance or read from the outside by a floppy disk device (not shown) or the like.

First, the classification processing unit 14 determines each distance measuring sensor S _ij.
Is sequentially read out to extract one to a plurality of data having the maximum z coordinate, and this data is registered in the group data storage area 27 in the RAM 25 as a temporary group center (step 1). .

FIG. 11 shows an example of a temporary group center extraction method. Here, for the sake of explanation, each data is represented in the shape of a rectangle on the xy plane, and the rectangle 34 marked with a circle represents the maximum z.
Data with coordinates. In this embodiment, z coordinates are compared in order from the upper left data, and when the maximum value is extracted, the comparison process for a predetermined number of data that follows is skipped. As a result, the temporary group centers 34a are extracted at positions that are dispersed to some extent.

Returning to FIG. 10, in the next step 2, the classification processing unit 14 causes the counter N for counting the number of effective groups.
After initializing C to 0, the distance from each group center set in step 1 is obtained for each three-dimensional data,
Each three-dimensional data is assigned as an element belonging to the center of the group at the shortest distance. Each assigned three-dimensional data is stored as the element data E _n of the corresponding area s _n in the group data storage area 27 (step 3).

Next, the classification processing unit 14 reads the element data E ₁ stored in the _first area s ₁ of the group data storage area 27, and the number N _{1 of} elements belonging to this group.
Is calculated (steps 4 and 5). If the number of elements N ₁ is equal to or more than the preset lower limit value N _{min of the} number of elements, it is determined as an effective group, and the effective group number counter NC is incremented (steps 6 and 7).

At the next step 8, the classification processing unit 14
The average value of the position coordinates of each element is calculated, and the group center coordinate G ₁ is reset to this average coordinate. In the subsequent step 9, the average distance A between the element and the center is calculated from the newly set coordinates G _{1 of the} group center and the element data E _1.
D ₁ is calculated and stored in the area s ₁ .

On the other hand, the number of elements N ₁ obtained in step 5
Is less than the lower limit value N _min , step 6 is “N
Becomes O "goes to step 10, the classification processing portion 14 sets the invalidation flag F ₁ in the region S _1.

When the above process is performed for all the set groups, step 12 becomes "YES",
In the next step 13, the classification processing unit 14 checks whether these processes have been performed a predetermined number of times. In this case, since this determination is “NO”, the classification processing unit 14 determines the effective group NC as the reference value NC _{0 in the} next steps 14 and 15.
Compare with.

This reference value NC ₀ is set in advance as a standard number of groups for classifying the groups in the monitoring area. In steps 14 to 18 below, the ISO-DATA method which is one of the clustering methods is used. Is used to check the number of effective groups and to divide and merge the groups based on the result.

That is, the number of effective groups NC is the reference value NC.
If it is less than half of _0, the number of groups is too small and the division processing is performed (steps 14 and 16). If the number of effective groups NC is twice the reference value NC ₀ or more, the number of groups is too large and the fusion processing is performed. Performed (steps 15,1)
7). When the number of effective groups is larger than one half of the reference value NC ₀ and smaller than twice, both steps 14 and 15 are “NO”, and the division processing and the fusion processing are sequentially performed (step 18).

FIG. 12 shows the detailed procedure of the division processing in steps 16 and 18 of FIG. First, in step 16-1, the counter n _A for counting the new group is set to “0” and the counter n for designating the area from which the group data is read is set to “1”. in step 16-2, reads each group data from area s _n of the group data storage area 27.

_If the invalidation flag F _n is not set in this area s _n , step 16-3 becomes "NO" and the process shifts to step 16-4, and the classification processing unit 14 determines the x-axis for each element. , Y-axis, z-axis for each direction,
The maximum value of the variance values in the three directions is set as the maximum variance value V _n .

If the maximum variance value V _n is larger than the upper limit value V _{max of the} variance value, step 16-5 is "YES".
Therefore, the classification processing unit 14 further checks whether or not this group can be divided (step 16-6 to step 16-8).

As a result, when the number NC of effective groups is less than half of the reference value NC ₀ , or when the average distance AD _n between the elements and the center is the lower limit value CD _{min of the} group center distance.
If the number of elements is equal to or more than the above, and the number of elements N _n is equal to or more than the lower limit value N _min , this group is divided in the axial direction in which the maximum variance value is obtained based on the group center (step 16-
9).

FIG. 13 shows an example of group division processing. In order to simplify the description, an example of two-dimensional data distribution in the x-axis and y-axis directions is shown.

In the figure, e represents each element, and c ₀ represents the center of the original group. In this example, the dispersion in the x-axis direction is maximum, and new group centers c ₁ and c ₂ are set at positions that are respectively moved by ± d in the x direction with reference to the group center c ₀ . The movement amount d of the group center is obtained by multiplying the maximum variance value V _n by a preset coefficient β (where 0 <β <1).

Of the group centers c ₁ and c ₂ newly set by the above method, the coordinates of one of the group centers is the area in which the original group center c ₀ is stored in the group data storage area 27. In s _n , the other coordinate is the area in which the invalidation flag is set or the empty area 2
9, and the new group number n _A is incremented (step 16-10).

Steps 16-5, 16-7, 16
If any of the determinations in -8 is "NO", the group division is not performed, the position data of the center of the group is held as it is, and the process proceeds to the next area s ₂ .
When the above process is performed for all groups, the determination in step 16-12 is “YES”, and step 1
In 6-13, the number of new groups n _A is added to the number of effective groups NC, and the division process ends.

FIG. 14 shows the detailed procedure of the fusion process in steps 17 and 18 of FIG. First step 17
After setting the counter n _D for counting the number of combinations of fusion candidates at −1 to “0” and the counter n for setting the read area of the group data to “1”, the classification processing unit 14 first Each group data is read from the area s _n (in this case, the area s ₁ ) of the group data storage area 27 (step 17-2).

If the invalidation flag F _n is not set in this area, step 17-3 becomes "NO" and the process moves to step 17-4, and the next area (in this case, the area s ₂ ) Group data is read.

[0079] If still set invalidation flag F ₂ in the region s _2, Step Step 17-5 17
In step -6, the classification processing unit 14 calculates the center-to-center distance CD from the center coordinates G ₁ and G ₂ of both groups, and compares this with the lower limit value CD _min of the center-to-group distance.

The calculated distance CD between the centers is the lower limit value CD.
If it is smaller than _min , the classification processing unit 14 determines that both groups are to be merged, and the areas (s ₁ and s _{2 in} this case) in which each group data is stored and the calculated distance CD.
Is stored in the fusion target table 35 (shown in FIG. 15) in the RAM 25, and the fusion candidate combination number n _D is incremented by 1 (steps 17-7 to 17-9).

Thereafter, the counter n is incremented to check the distances between the group centers for all combinations of groups in the group data storage area 27, the combinations of groups to be merged are extracted, and the combinations in the fusion target table 35 are extracted. Remember.

Next, the classification processing section 14 compares the distances CD between the centers of the respective combinations in the fusion target table 35, and carries out a sorting process for rearranging the data storage order in ascending order of distance (step). 17-1
3).

FIG. 15 shows an example of the data structure and sort processing of the fusion target table 35 in the RAM 25. In the fusion target table 35, first, the above step 17 is performed.
By the procedure of −2 to step 17-12, the address value of the storage area of the group data and the distance CD between the centers are stored in the order of extraction for each combination of the groups extracted as the fusion target (FIG. 1)). When the sort process of step 17-13 is performed after the data is stored, each data is rearranged in the ascending order of the center-to-center distance CD (FIG. 15 (2)).

When this sort processing is completed, the classification processing unit 14 compares the number of combinations n _{D of} fusion candidates with the reference value NC _{0 of the} number of groups in the next step 17-14. By this comparison, the number of combinations n _D is the reference value NC _0.
In the following cases, for all the extracted combinations,
If the number of combinations n _D is larger than the reference value NC ₀ , the fusion process is performed on each of the NC ₀ combinations in the ascending order of the center-to-center distance CD (step 1).
7-15, steps 17-17).

Group p extracted now as a fusion target
_{For 1} and p ₂ , the number of elements of each is S ₁ ,
Assuming S ₂ and center coordinates g ₁ and g ₂ , the group center coordinate g _{3 of the} group p ₃ obtained by fusing these two groups is given by the following equation (2).

[0086]

[Equation 2]

When the fusion process is completed, the classification processing unit 14
In step 17-16 or step 17-18, the number of merged groups is subtracted from the number of valid groups NC to update the number of valid groups NC, and the merge process ends.

Returning to FIG. 10, the processing of any of steps 16 to 18 is executed to reorganize each group, and the coordinates of the center of each group are reset. Division processing unit 14
Clears all the data in the group data other than the group center coordinates G _n (step 1).
9).

When each data is cleared in step 19, the process returns to step 3 and each three-dimensional data in the three-dimensional data storage area is again allocated to the center of one of the groups, and the same processing is repeated thereafter. It As a result, when the processing for the preset number of times is completed, step 13 becomes “YES”, and the process proceeds to the feature amount calculation process by the feature amount calculation unit 15.

FIG. 16 shows the procedure for calculating the characteristic amount by the characteristic amount calculation unit 15. In this process, the feature amount is sequentially calculated for each group classified by the above procedure, and each group data is first read from the first area s ₁ of the group data storage area 27 (steps 1 and 2). .

If the invalidation flag F ₁ is not set in this area s ₁ , step 3 becomes "NO", and the feature quantity calculation unit 15 calculates the number of elements of this group from the element data E ₁ and The first feature quantity S ₁ is stored in the area f ₁ of the feature quantity storage area 28 (step 4).

Next, the characteristic amount calculating section 15 extracts the x coordinate and the y coordinate from the position coordinate G _{1 of the} center of the group and stores them in the area f ₁ as the second and third characteristic amounts CX ₁ and CY ₁ , respectively. Then, the average value and the variance value of the z-coordinates of each element are calculated from the element data E ₁ and stored in the area f ₁ as the fourth and fifth characteristic quantities AZ ₁ and VZ _1. (Steps 6 and 7).

In the next step 8, the feature quantity calculation unit 15
Reads the coordinates of the centers of the other groups sequentially from the areas s _{2 to} s _N of the group data storage area 27 to obtain the area s ₁
The distance from the center G ₁ of the group is calculated, and the minimum value of these is determined as the distance DM ₁ to the shortest group and stored in the area f ₁ .

On the other hand, when the invalidation flag F ₁ is set in the area S ₁ , the group related to this area is regarded as invalid and the value of each feature quantity is stored in the area f ₁ of the feature quantity storage area 28. After "0" is given, the process shifts to the group data of the next area (steps 3, 9 to 1).
1).

Similarly, by performing the above steps 2 to 8 on the group data stored in the area s _n of the group data storage area 27, the number of elements S _n and the group center The x-coordinate CX _n , the y-coordinate CY _n , the z-coordinate average value AZ _n , the z-direction variance VZ _n , and the distance DM _{n from} the shortest group are calculated and stored in a predetermined area f _n of the feature amount storage area 28. It When the processing for all groups is completed, step 11 becomes "YES", and the determination processing by the attribute determination unit 16 is performed.

FIG. 17 shows the procedure of the attribute determination process. First, the attribute determination unit 16 determines S ₁ , CX ₁ , CY ₁ , AZ ₁ , V from the _first area f ₁ of the feature amount storage area 28.
Each feature quantity of Z ₁ and DM ₁ is read (step 1,
2).

As described above, the attribute determining unit 16 stores the pattern of the characteristic amount learned in advance using a predetermined model for each attribute such as a human being or an object in the form of a reference vector, and stores the characteristic amount storage area. The distance between each reference vector and the vector composed of the feature amount read from the 28 regions f ₁ is calculated, and the reference vector having the shortest distance among them is determined to be the closest to the group. As a result, the attribute of this reference vector is adopted as the attribute data Q ₁ of the group and stored in the area f ₁ (steps 3 and 4).

The attribute data Q _n is stored in each area f _n by sequentially performing the above processing for the data of the area f _n (n = 1 to N) in the feature amount storage area 28, and the route specifying processing is started. Transition.

FIG. 18 shows the procedure of the route specifying process. When the attribute determination unit 16 determines the attribute of each group, the route identification unit 18 initializes the counter m for the number of data stored in the extracted group storage unit 17 to 0, and then stores the data in the feature amount storage area 28. Each feature amount including the attribute data Q ₁ is read from the first area f ₁ (steps 1 and 2).

If the attribute data Q ₁ means an object other than a human being, step 3 becomes "NO", the following procedure is skipped, and the process proceeds to the next area f ₂ (step 6,). 7).

When step 3 is "YES", that is, when the attribute data Q ₁ means a human being, the route specifying unit 18 increments the counter m and reads out the respective feature quantities S ₁ , CX ₁ , CY ₁ , AZ ₁ , VZ ₁ ,
DM ₁ is extracted Each feature quantity NS ₁ , NX ₁ , NY ₁ , N of the area g _m (in this case area g ₁ ) of the group storage unit 17 is extracted.
It is stored as A ₁ , NV ₁ and ND ₁ (step 4,
5).

By sequentially performing the above procedure for all the data in the feature quantity storage area 28, groups having attributes corresponding to human beings are sequentially extracted, and the feature quantities are stored in the area NM of the extraction group storage unit 17. Stored in. At this time, step 7 becomes “YES”, and the route specifying unit 18 stores M as the number of data stored in the area MN and resets the counter m to 1 in step 8, and then extracts the group storage unit in step 9. The x, y coordinates NX _m , NY _m of the center of the group are read from the area g _{m of} 17 and are stored in the area q _m of the movement information storage unit 19.

Next, the route specifying unit 18 extracts the one located at the shortest distance from the group center of the g _m region from the position data of each group center stored in the region OM of the extraction group storage unit 17, and The shortest distance DS is compared with a predetermined threshold value th (steps 10 and 11).

This threshold value th corresponds to the maximum distance that a person can move within the above-mentioned predetermined sampling time, and is stored in the ROM 24 or the like in advance. If the minimum value DS of the distance is less than or equal to this threshold value, the determination in step 11 is “YES”, and the route specifying unit 18 sets the feature amount of the group at the shortest distance DS to the sample one sample before in the group of the region g _m . It is determined that the data is data, and the x and y coordinates thereof are stored in the area q _m as the data PX _m and PY _m of the movement information storage unit 19 (step 12).

On the other hand, the minimum distance D obtained in step 10
If S is larger than the threshold value, it is considered that the data corresponding to the group of the region g _m does not exist in the region OM, and “−1” is set as the data PX _m and PY _m of the movement information storage unit 19. It is set (step 13).

When the movement information for each group stored in the area NM is stored in the movement information storage section 19 by repeating the same procedure thereafter, step 15 returns "YE".
S ”, and the path specifying unit 18 moves each data in the area NM to the area OM and stores it as the data of one sample before (step 16).

FIG. 19 shows a counting processing procedure by the entrance / exit counting unit 20. In this counting procedure, step 1 is executed only when the device is initially activated, and the counter IN
And the counter OUT of the contestant is both initialized to "0". In the subsequent processing, step 1 is skipped and the values of the counters IN and OUT are updated sequentially.

The entrance / exit counting unit 20 sets the counter m to 1 in step 2, and then resets the flags FN and FO in the RAM 25 in step 3. This flag FN is for storing the latest data of the group, and the flag FO is for storing that the position of the group is on the entrance side for the data one sample before.

In step 4, the entrance / exit counting unit 20 reads the data in the area q _m of the movement information storage unit 19, and first checks the y-coordinate NY _m of the current group center by the check line LN (shown in FIG. 4). The y-coordinate Y _th is compared (step 5). When NY _m is larger than Y _th, step 5 becomes "YES", and the flag FN is set because this group is located on the entrance side (step 6).

Next, it is checked whether or not the y-coordinate PY _m of the group center one sample before takes a value of "-1" (step 7). If this judgment is "NO", the check line LN is further checked. Is compared with the y coordinate Y _th (step 8). If PY _m is larger than Y _{th as a} result, the flag FO is set (step 9).

FIG. 20 shows five groups P ₁ , P ₂ , P.
The routes in the monitoring area 40 for ₃ , P ₄ , and P ₅ are shown. In the figure, the black circle is the latest position of each group, and the white circle is 1.
The position before the sample is shown, and the route of each group is indicated by an arrow.

Returning to FIG. 19, the entrance / exit counting unit 20 performs steps 10 to 12 in which the flags FN and F in the RAM 25 are stored.
O is checked, and the presence / absence of an entrance / exit is determined based on the relationship between the flags. If the latest position of the center of the group is on the entrance side and the position one sample before is on the exit side (in the case of the group P ₁ ), step 10 is “YES”,
Step 12 becomes "NO", and the counter IN of the visitors
Is counted by 1 (step 13). When the latest position of the center of the group is on the exit side and the position one sample before is on the entrance side (in the case of group P ₂ ), step 10 is “NO”, step 11 is “YES”,
The exit counter counter 1 is counted one.

In both Step 10 and Step 12, "Y
ES ”, that is, the latest position of the center of the group and the position one sample before are both on the entrance side (in the case of group P ₃ ), and steps 10 and 11
Are both “NO”, that is, when the latest position at the center of the group and the position one sample before are both on the exit side (in the case of group P ₄ ), the group does not pass through the check line LN and the entrance side or It is judged that only the exit side is moving, and is not counted as the entry / exit count.

In the case of the group P ₅ or P ₆ in FIG. 20, the position of the center of the group one sample before (indicated by the dotted line in the figure) is outside the monitoring area 40 and is not detected.
This means that "-1" is stored as the data PX _m and PY _{m in the} movement information storage unit 19. Therefore, in these cases, the determination in step 7 is “YES”, and the following procedure is skipped.

When the above processing is performed for the data of all the areas in the movement information storage unit 19, step 16 becomes "YES" and the counting processing ends.

In this embodiment, the check line is x
Although it is provided parallel to the direction, as shown in FIG.
A plurality of check lines LN1 to L in each of the x and y directions
N3 may be provided, and the entrance and exit may be determined by checking the change in the x-coordinate and y-coordinate positions at the center of each group.

The determination process of the attribute determination unit 16 is LV.
Although the Q program is used to determine the attribute of each group, the present invention is not limited to this method, and the attribute of each group can be determined using, for example, a discriminant function method.

FIG. 22 shows the procedure of the attribute determination process using the discriminant function. In the attribute determination process of this embodiment, among the feature amounts calculated for each group, four features of the number of elements S _n , the average value AZ _{n in} the z direction, the variance value VZ _n , and the distance DN _{n to} the shortest group are included. RAM25 is used to determine whether or not this group corresponds to a human by using the amount.
Of the feature quantity S from the area f _n of the feature quantity storage area 27 in the
_n , AZ _n , VZ _n , and DN _n are sequentially read (steps 1 and 2).

At the next step 3, the attribute determining section 16
A discriminant function expression represented by the following expression (3) is executed using these feature amounts to calculate a function value F.

[0120]

[Equation 3]

This discriminant function formula determines whether or not the pattern represented by each feature amount is a pattern corresponding to a human, and the coefficients w ₁ , w ₂ , w ₃ , w ₄ and the offset value w are used. _{In 5} , a predetermined value is calculated and set in advance from data of a plurality of models.

Next, the attribute processing section 16 checks the value of the function value F calculated by the equation (3) (step 4). If the function value F takes a value of 0 or more, it is determined that the group to be discriminated is a human (step 5), and if F is 0 or less, the characteristic amount pattern of this group has a human characteristic amount. It is determined that it is not a pattern (step 6).

At the next step 7, the attribute determining section 16
The above determination result is stored in the area f _n of the characteristic value storing area 28 as attribute data Q _n, repeatedly executes the same processing for the following areas, to determine the attributes of each group (Step 7-9) .

Furthermore, this attribute determination process can be performed by fuzzy inference. In this case, the attribute determination unit 16 determines the relationship between the attributes of the four feature amounts, that is, the area (corresponding to the number of elements) of the group, the average value and variance of the height (z direction), and the distance to the shortest group. It is given as a fuzzy rule.

A specific example of the above fuzzy rule is shown below. (1) If the area is small, the group is not human. (2) If the height variance is small, the group is a simple shaped object. (3) If the area is large, the heights are widely dispersed, and the distance to the shortest group is large, the group is a human. (4) If the area is large, the height dispersion is large, the distance to the shortest group is small, and the average height is large, then the group is a baggage carried by humans. (5) If the area is large, the variance of the height is large, the distance to the shortest group is small, and the average height is small, the group is a human or a child located near an object.

23 (1) to 23 (4) show membership functions for each feature. In the figure, S is "small"
Is shown, and B is a fuzzy label B which means “large”. The horizontal axis represents the value of each feature amount and the vertical axis represents the goodness of fit.

FIG. 24 shows the above fuzzy rules (1) to
(5) is expressed in a tabular form. "Conclusion" in the figure
Column is a simplified representation of the consequent part of the fuzzy rule. If the attribute of the group to be judged is inferred to be human, “1” is inferred to have an attribute other than human. “0” is entered for each.

The attribute determination unit 16 reads out the feature quantities S _n , AZ _n , VZ _n , and DN _n from the feature quantity storage area 28 in the RAM 25 for each set group,
The membership function is used to obtain the goodness of fit for each feature quantity, and this is applied to the fuzzy rule to execute the inference operation. As a result, based on the conclusion shown in FIG.
A value of either “1” or “0” is determined as the group attribute data Q _n and stored in the feature amount storage area 28.

FIG. 25 shows a display example of the number of entering / leaving persons counted by the entering / leaving counting section 20. The number of persons entering / leaving is displayed in real time on the liquid crystal display 21a. There is.

In the above embodiment, the space under the gate 1 is used as the monitoring area to detect a person passing through the gate 1. However, the present invention is not limited to this. People who enter and exit the place may be detected. In this embodiment, the description has been limited to the device for detecting a human being, but it goes without saying that the detection target is not limited to a human being and can be applied to various objects and various applications.

[0131]

As described above, according to the present invention, in order to detect the presence or absence of a predetermined object in a specific monitoring area, a plurality of data obtained by measuring a plurality of three-dimensional data in the monitoring area are used. Since the presence or absence of a group having an attribute corresponding to the detection target is detected from among the groups, the three-dimensional characteristics of the object in the monitoring area can be grasped and the detection target can be accurately detected. be able to.

According to the seventh aspect of the present invention, the above process is repeated to detect the change in the position of the detection target and identify it as the movement path of this object. The movement path of the detection target in can be traced.

Further, according to the invention of claim 8, since it is detected that the detected movement route passes through the specific position, it is possible to accurately measure the number of times the detection target has passed through the specific position. is there.

According to the second aspect of the present invention, the distance measuring sensors are evenly arranged over the entire upper surface of the monitoring area.
Dimensional data can be easily obtained.

According to the third aspect of the present invention, since the group having the attribute corresponding to a human being is particularly detected, there is no possibility of erroneously detecting a complicated motion of a person or baggage, and accurate detection is performed.

In the inventions of claims 4, 5, and 6, since the attribute of the group is determined by using the neural network, the discriminant function, and the fuzzy inference, respectively, the accuracy of the attribute determination is improved and the object is accurately determined. Can be detected.

[Brief description of drawings]

FIG. 1 is a front view, a plan view, and a cross-sectional view showing the structure of a gate to which an entrance / exit detection device according to an embodiment of the present invention is applied.

FIG. 2 is an explanatory diagram showing a setting example of a distance measuring sensor and a light projection pattern generation unit.

FIG. 3 is an explanatory diagram showing a configuration of a light projection pattern generation unit.

FIG. 4 is an explanatory diagram showing three-dimensional data measured in a monitoring area.

FIG. 5 is a block diagram showing an electrical configuration of an entrance / exit detection device.

FIG. 6 is an explanatory diagram showing a relationship between measurement data of a distance measuring sensor and a height of an object.

FIG. 7 is an explanatory diagram showing a relationship between a data structure of a RAM, an extraction group storage unit, and a movement information storage unit and each processing unit.

FIG. 8 is a flowchart showing a procedure for setting an initial height value.

FIG. 9 is a flowchart showing a procedure for setting three-dimensional data from measurement data.

FIG. 10 is a flowchart showing a procedure of classification processing.

FIG. 11 is an explanatory diagram showing a temporary group center setting method.

12 is a flowchart showing a detailed procedure of the division processing of FIG.

FIG. 13 is an explanatory diagram showing a method of dividing a group.

14 is a flowchart showing a detailed procedure of the fusion process of FIG.

FIG. 15 is an explanatory diagram showing a data configuration of a fusion target table.

FIG. 16 is a flowchart illustrating a procedure of a feature amount calculation process.

FIG. 17 is a flowchart illustrating a procedure of attribute determination processing.

FIG. 18 is a flowchart showing a procedure of route identification processing.

FIG. 19 is a flowchart showing a procedure of entrance / exit counting processing.

FIG. 20 is an explanatory diagram showing a counting method of entering and leaving people.

FIG. 21 is an explanatory diagram showing another method of setting check lines.

FIG. 22 is a flowchart showing a procedure of attribute determination processing using a discriminant function.

FIG. 23 is an explanatory diagram showing a membership function used for attribute determination processing by fuzzy inference.

FIG. 24 is an explanatory diagram showing a fuzzy rule used for attribute determination processing.

FIG. 25 is an explanatory diagram illustrating a display example of a display unit.

[Explanation of symbols]

S _ij distance measurement sensor 10 distance calculation unit 14 classification processing unit 15 feature amount calculation unit 16 attribute determination unit 18 route identification unit 19 entrance / exit counting unit 25 RAM 40 monitoring area

Claims (8)

[Claims] 1. An object detection device for detecting the presence or absence of a predetermined object in a specific monitoring area, comprising: data measuring means for measuring a plurality of three-dimensional data in the monitoring area; Grouping means for classifying each three-dimensional data measured by the means into a plurality of groups, feature quantity calculating means for calculating the feature quantity for each group classified by the grouping means, and the feature quantity calculating means Attribute determining means for determining the attribute of each group based on the feature amount calculated by: and the presence or absence of a group having an attribute corresponding to the predetermined object in the monitoring area based on the determination result of the attribute determining means An object detection device comprising a detection means. 2. The object detecting device according to claim 1, wherein the data measuring unit includes a plurality of distance measuring sensors that are evenly arranged over the entire upper surface in the monitoring area and measure distances in the vertical direction. 3. The object detection device according to claim 1, wherein the detection means is means for determining the presence or absence of a group having an attribute corresponding to a human. 4. The object detection according to claim 1, wherein the attribute determination means is means for determining the attribute of each group by comparing the feature amount of each group with a reference vector learned by a neural network program. apparatus. 5. The attribute determining means is means for substituting a feature amount of each group into a predetermined discriminant function formula and determining whether each group has an attribute corresponding to the predetermined object. 1. The object detection device described in 1. 6. The object detecting apparatus according to claim 1, wherein the attribute determining unit is a unit that executes a fuzzy inference operation using the feature amount of each group to determine the attribute of each group. 7. An object detection device for detecting a movement path of a predetermined object within a specific monitoring area, comprising: data measuring means for measuring a plurality of three-dimensional data within the monitoring area; Grouping means for classifying each three-dimensional data measured by the measuring means into a plurality of groups, feature quantity calculating means for calculating a feature quantity for each group classified by the grouping means, and feature quantity calculation Attribute determining means for determining the attribute of each group based on the characteristic amount calculated by the means, and detecting the position of the group having the attribute corresponding to the predetermined object in the monitoring area based on the determination result of the attribute determining means Position detecting means, and a temporal change of the position of a group having an attribute corresponding to the predetermined object by repeatedly operating the respective means at predetermined times And a path specifying unit that specifies the change in the position as the movement path of the predetermined object. 8. An object detection device for detecting that a predetermined object passes through a specific position within a specific monitoring area, wherein a plurality of three-dimensional data within the monitoring area is measured. Data measuring means, grouping means for classifying each three-dimensional data measured by the data measuring means into a plurality of groups, and feature quantity for calculating the feature quantity for each group generated by the grouping means Calculating means, attribute determining means for determining the attribute of each group based on the characteristic amount calculated by the characteristic amount calculating means, and corresponding to the predetermined object in the monitoring area based on the determination result of the attribute determining means Position detecting means for detecting a position of a group having an attribute, and an attribute corresponding to the predetermined object by repeatedly operating each of the means at a predetermined time. A route specifying unit that detects a temporal change in the position of the group and specifies the change in the position as the moving route of the predetermined object, and the moving route specified by the route specifying unit is a specific position in the monitoring area. An object detecting device comprising: a specific position passage detecting means for detecting passage.