JPH11203481A - Moving body identifying device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To exactly discriminate a car body and a human being by finding the distribution of the moving directions of an entire moving body from image data inside a moving object area, finding uniformity expressing the level of this distribution of moving directions, and identifying the class of the moving object based on this uniformity and the shape of the moving body area. SOLUTION: A moving body detector 4 is connected through an A/D converter 3 to the output terminal of a TV camera 1, difference image data are found by this moving body detector 4, and the moving body area including a moving body 2 is extracted from these difference image data and sent to a moving body shape detector 5 and a moving direction distribution detector 6. Then, the area of the moving body area and the aspect ratio of a rectangle circumscribed with the moving body area are found by the moving body shape detector 5, and uniformity to the direction distribution of respective moving vectors detected in the moving body area is found by the moving direction distribution detector 10. These found values are inputted to a moving body identifier 11 and the car body, human being and other can be identified.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving object identification apparatus for processing image data obtained by, for example, a television camera (hereinafter referred to as a TV camera) to identify the type of the moving object.

[0002]

2. Description of the Related Art FIG. 17 is a block diagram of a conventional moving object identification device. The outline of this identification device is as follows. A moving object 2 to be identified is imaged by a TV camera 1 and a sampling time Δt
, Two image data I (t) and I (t + Δt) obtained at two timing times t and t + Δt, respectively.
The difference image data | I (t + Δt) −I (t) | (1) is obtained.

A portion having a large value of the difference image data is an image portion of the moving body 2 and a portion having a small value of the difference image data is determined to be a stationary or background area. Extract only

[0004] Next, the area of the image portion of the moving body 2 is calculated, and only the area having a certain area or more is a region of the moving body 2. As shown in FIG. 17, the vehicle body can be distinguished from other moving objects such as humans and small animals (dogs, cats, etc.).

More specifically, the TV camera 1 captures an image of the moving body 2 that travels and outputs an image signal. This image signal is converted into digital image data by the A / D converter 3 and the quantized image data I at the time t.
It is sent to the moving object detector 4 as (t, i, j). here,
i and j are 1 ≦ i ≦ M and 1 ≦ j ≦ N, and M × N is the image size. FIG. 18A is a schematic diagram showing an example of the image data I (t, i, j) at time t.

After the sampling time Δt seconds,
The image signal output from the TV camera 1 is converted into digital image data by the A / D converter 3 and is converted into quantized image data I (t + Δt, i, j) by the moving object detector 4.
Sent to FIG. 18B is a schematic diagram showing an example of image data I (t + Δt, i, j) at time t + Δt.

The moving object detector 4 receives two image data I (t, i, j) and I (t + Δt, i, j) and executes an operation as shown in the following equation (2). And the difference image data J (t + Δt,
i, j).

J (t + Δt, i, j) = | I (t + Δt, i, j) −I (t, i, j) | (2) (1 ≦ i ≦ M, 1 ≦ j ≦ N) The moving object detector 4 calculates the difference image data J (t + Δt
, i, j) and a preset threshold value, and a pixel larger than the threshold value in the difference image data J (t + Δt, i, j), that is, a pixel whose brightness changes greatly, is set to a value “1”, and The pixel, that is, the pixel whose brightness does not change much is converted into a value “0” and sent to the moving object shape detector 5 as binary image data K (t + Δt, i, j).

FIG. 19 shows such binary image data K (t + Δt).
, i, j), where the pixel value of the area of the moving body 2 is represented as “1” and the pixel value of the surrounding area is represented as “0”.

Next, the moving object shape detector 5 extracts the value “1” of the binary image data K (t + Δt, i, j) as a moving object region 6 as one lump, The area is calculated in pixel units. The area value of the moving body region 6 is
Represents the size of.

Next, the moving body identifier 7 receives the area value of the moving body area 6 obtained by the moving body shape detector 5 and compares it with a preset threshold value. If so, the moving body area 6 is, for example, a vehicle body or a human, and if less than the threshold value, it is recognized as a small object that should not be detected, for example, a small animal such as a dog.

A moving object to be detected (vehicle or human) and an object not to be detected (for example, a small animal such as a dog or a cat) can be identified based on the size of the image area including the moving object 2 detected as described above.

Here, the moving body shape detector 5 is shown in FIG.
As shown in FIG. 0, the aspect ratio rr = w / h (3) of the vertical and horizontal lengths h and w of the rectangle circumscribing the moving object area 6 is calculated.

The moving object discriminator 7 has an aspect ratio r
Is compared with a preset threshold value (for example, 1). If the aspect ratio r is equal to or greater than the threshold value, the shape of the moving body 2 is determined to be horizontal and the vehicle body is determined. Is determined as a human, and the vehicle body and the human can be distinguished.

[0015]

However, when the vehicle body is observed from the front or when a human walks with both hands and legs wide open, the respective image data at that time are shown in FIGS. 21 (a) and 21 (b), respectively. An image as shown is obtained, and the values of these aspect ratios r are all close to 1, and if a vehicle body or a person is simply identified as being larger or smaller than 1, the possibility of an identification error increases.

As a threshold value, the aspect ratio r is r ≧ r _th1 ≧
When 1, the vehicle body, when r <r _th2 <1, human, when r _th2 ≦
When r <r _th1 , other than these are identified as
This simply increases the number of cases belonging to other than these, and when the aspect ratio r is close to 1, it is difficult to accurately distinguish a vehicle body from a person. Therefore, an object of the present invention is to provide a moving object identification device that can accurately identify a vehicle body and a human.

[0017]

According to the first aspect, difference image data of each image data having a predetermined sampling time difference is obtained from image data obtained when an image of a moving object is picked up by an imaging device. In a moving object identification device for identifying a type of a moving object based on a shape of a moving object region including the moving object, a distribution of a moving direction of the entire moving object is obtained from image data in the moving object region, and a distribution of the moving direction is obtained. The moving body identification device is provided with uniformity processing means for obtaining a degree of uniformity indicating the degree of the moving body, and identifies the type of the moving body based on the uniformity and the shape of the moving body region.

According to a second aspect of the present invention, in the moving body identification apparatus according to the first aspect, the uniformity processing means divides the moving body area into a plurality of small areas and detects each movement vector of these small areas. Further, it has a function as a moving direction distribution detecting means for obtaining a degree of uniformity indicating a degree of uniformity with respect to the direction distribution of these moving vectors.

According to a third aspect of the present invention, in the moving body identification apparatus according to the first aspect, the uniformity processing means detects an edge direction for each pixel of the moving body area, and detects uniformity in the direction distribution of the edge directions. It has a function as an edge direction distribution detecting means for obtaining the degree of uniformity representing the degree of the nature.

According to a fourth aspect of the present invention, in the moving object identification apparatus according to the first aspect, the uniformity processing means divides the moving object region into a plurality of small regions, and detects respective movement vectors of these small regions. A moving direction distribution detecting means for obtaining a degree of uniformity representing a degree of uniformity with respect to the direction distribution of the moving vectors; and an edge direction for each pixel of the moving body region, and the uniformity with respect to the direction distribution of these edge directions. Edge direction distribution detecting means for obtaining the degree of uniformity representing the degree.

[0021]

(1) Hereinafter, a first embodiment of the present invention will be described with reference to the drawings. The same parts as those in FIG. 17 are denoted by the same reference numerals. FIG. 1 is a configuration diagram of a moving object identification device.

A moving object detector 4 is connected to an output terminal of the TV camera 1 via an A / D converter 3. The moving object detector 4 obtains difference image data between two image data I (t) and I (t + Δt) obtained at two timing times t and t + Δt separated by an interval of the sampling time Δt. And a function of extracting a moving body region 6 including the moving body 2 from the difference image data and transmitting the extracted moving body region 6 to the moving body shape detector 5 and the moving direction distribution detector 10.

The moving body shape detector 5 receives the moving body area 6 extracted by the moving body detector 4 and inputs the area of the moving body area 6 to the aspect ratio of a rectangle circumscribing the moving body area 6. It has a function to find the value of r.

On the other hand, the moving direction distribution detector 10 divides the moving body area 6 into a plurality of small areas, detects each moving vector of these small areas, and determines the degree of uniformity of the moving vector with respect to the direction distribution. It has a function to find the uniformity f to be represented.

The moving body identifier 11 is a moving body shape detector 5
And the area of the moving body region 6 obtained by
The values of the aspect ratio r of the circumscribed rectangle and the uniformity obtained by the moving direction distribution detector 10 are input, and the area, aspect ratio r, and uniformity f of the moving body region 6 are input.
Has a function of identifying a vehicle body, a person, and others based on the information.

Next, the operation of the above-configured device will be described. The TV camera 1 captures an image of the moving object 2 that travels and outputs an image signal of the image. This image signal is A /
It is converted into digital image data by the D converter 3,
It is sent to the moving object detector 4 as quantized image data I (t, i, j) at time t as shown in FIG.

Subsequently, after a sampling time Δt seconds, the image signal output from the TV camera 1 is converted into digital image data by the A / D converter 3 and the above-mentioned FIG.
Quantized image data I (t + Δt, i, j) as shown in
Is sent to the moving object detector 4.

The moving object detector 4 receives two image data I (t, i, j) and I (t + Δt, i, j) and obtains the above equation (2).
And the difference image data J (t + Δt, i, j)
Get.

Next, the moving object detector 4 compares the difference image data J (t + Δt, i, j) with a preset threshold value.
In the difference image data J (t + Δt, i, j), a pixel larger than the threshold value, that is, a pixel whose brightness changes greatly, is converted into a value “1”, and a small pixel, ie, a pixel whose brightness does not change much, is converted into a value “0”. Then, the image data is sent to the moving object shape detector 5 and the moving direction distribution detector 10 as binary image data K (t + Δt, i, j) as shown in FIG.

The moving object shape detector 5 extracts the value “1” of the binary image data K (t + Δt, i, j) as one moving object region 6 as one lump. The area s is calculated for each pixel.

As shown in FIG. 20, the moving body shape detector 5 has an aspect ratio rr = w / h of each of the vertical and horizontal lengths h and w of the rectangle circumscribing the moving body area 6. Calculate (4).

On the other hand, the moving direction distribution detector 10 divides the moving body region 6 into a plurality of small regions as shown in FIG. 2, detects each moving vector of these small regions, and detects the direction distribution of these moving vectors. Is obtained, which represents the degree of uniformity with respect to.

The operation of the moving direction distribution detector 10 will be specifically described. As shown in FIG. 2, the moving body region 6 is divided into a plurality of square small regions _Rk . Where k =
.., K.

The size of one of the small regions R _k is such that the length of one piece is, for example, about 5 to 20 pixels. Then, as for the small region R _k near the boundary of the moving body region 6, the one having a non-square shape is excluded, or the moving body region 6 is enlarged so that the small region R _k has a square shape. .

That is, in the enlargement processing of the moving body area 6 as described above, all of the square areas that are included at least in part are selected and set as a new moving body area. A new moving object area is indicated by a dotted line shown in FIG.

Next, for each small region R _k , a movement vector (u _k , v _k ) for the image data contained therein
Ask for. The derivation of this movement vector (u _k , v _k )
It can be obtained by solving the following equation.

[0037]

(Equation 1) Here, Ix (i, j), Iy (i, j) and It (i, j) are the spatial differential image in the i direction, the spatial differential image in the j direction of the image I (t, i, j), respectively. And a time differential image of the image I (t, i, j).

The symbol Σ means that the sum of all the pixels included in the small area R _k is to be calculated. These differential images are calculated using the following formula (6).
To (8).

Ix (i, j) = I (i + 1, j) −I (i, j) (6) Iy (i, j) = I (i, j + 1) −I (i, j) ) ... (7) It (i , j) = I (t + Δt, i, j,) -I (t, i, j) ... (8) However the motion vector (u _k, v _k) is a small area Part of the moving object 2 included in R _k is shifted from time t to time t + Δt.
Distance spatially traveling in the i and j directions (pixels / Δ
t).

[0040] As a result, a set of motion vectors for all of the small region R _k as shown in FIG. _{3 (u k, v k)} (k =
1, 2, 3,..., K) are obtained. Next, the direction θ _k of the movement vector (u _k , v _k ) is calculated by the following equations (9) to (14). (a) When u _k > 0 and v _k ≧ 0, θ _k = tan ⁻¹ (v _k / u _k ) (9) (b) When u _k <0 and v _k ≧ 0, θ _k = ^{(π / 2) + tan -1} | v k / u k | ... (10) (c) when _{u k <0, v k <} 0, θ k = π + tan -1 | v k / u k | ... (11 ) (in the case of _{d) u k> 0, v} k <0, θ k = (3π / 2) + tan -1 | v k / u k | ... (12) (e) u k = 0, v k> 0 , Θ _k = π / 2 (13) (f) When u _k = 0, v _k <0, θ _k = 3π / 2 (14) (g) When u _k = v _k = 0 , Θ _k are not defined. The data of the small region R _k are excluded.

From these equations (9) to (14), each small area R _k
The direction θ _k of the motion vector (u _k , v _k ) has a value as shown in FIG. This movement vector (u
A moving direction histogram as shown in FIGS. 5 and 6 is generated from the value of the direction θ _k (k = 1, 2, 3,..., K) in the direction _k , v _k ). That is, the angle 0 to 2π is set to an appropriate width Δθ
, And the number of movement vectors is cumulatively added for each direction to create a movement direction vector.

A direction θ _M that gives the maximum value of the moving direction vector is obtained, and the total number of moving vectors falling within the range of the width θ _W around the direction θ _M is defined as L, and the uniformity f of the direction distribution of the moving vector is defined as L. It is defined by the following equation.

F = L / K (15) The uniformity f of the direction distribution of the movement vector indicates the following. As shown in FIG. 3, when the moving body 2 is a vehicle body, the moving body 2 moves in a certain direction during the sampling time Δt, but its shape is not changed.

Therefore, since the directions of the movement vectors are substantially in the same direction, the movement direction histogram has a high frequency in the direction θ _M giving the maximum value as shown in FIG. On the other hand, in the case of a human being as the moving body 2, when walking, as shown in FIG.
As a whole, they move in the same traveling direction, but if we look closely at each part of the body, they move in various directions when walking, and the moving direction histogram shows the maximum value of the direction θ _M as shown in FIG. Frequency is low.

Therefore, the uniformity f of the direction distribution of the movement vector with respect to the vehicle body is close to 1, and the uniformity f with respect to the human takes a small value. Next, the moving body discriminator 11 inputs the values of the area s and the aspect ratio r of the moving body region 6 obtained by the moving body shape detector 5 and the uniformity obtained by the moving direction distribution detector 10. f, and these moving body areas 6
The vehicle body, the person, and the others are identified based on the area s, the aspect ratio r, and the uniformity f.

The following distinction is made between the vehicle body, the human, and the others. (a) When s ≧ s _th , r ≧ r _th1 , and f ≧ f _th , it is identified as a vehicle body. (b) When s ≧ s _th , r <r _th2 , and f <f _th , it is identified as a person. ) In cases other than these (a) and (b), it is identified as a body other than a human body.

Here, s _th , r _th1 , r _th2 , and f _th are:
Each is a preset threshold. Note that r _th1 <
1, r _th2 > 1 can be set. Thus, the first
In this embodiment, the image data I (t) and I (t + Δt) are moved from the difference image data between two pieces of image data I (t) and I (t + Δt) obtained at two timing times t and t + Δt separated by an interval of the sampling time Δt. After extracting the moving body region 6 including the body 2, the moving moving body region 6 is divided into a plurality of small regions R _k to detect respective movement vectors of these small regions R _k , and to determine the direction distribution of these movement vectors. Since the moving direction distribution detector 10 for obtaining the degree of uniformity f representing the degree of uniformity is provided, the moving body 2 can be used, for example, when the vehicle body is observed from the front or when a person walks with both hands and both legs wide open. The body and the person can be accurately distinguished from the difference in the uniformity between the movement of the person and the movement of the person. (2) Next, a second embodiment of the present invention will be described with reference to the drawings. The same parts as those in FIG. 1 are denoted by the same reference numerals, and detailed description thereof will be omitted.

FIG. 8 is a configuration diagram of the moving object identification device. The edge direction distribution detector 12 detects an edge direction for each pixel of the moving body region 6 obtained by the moving body detector 4,
In addition, it has a function of obtaining a uniformity g representing a degree of uniformity with respect to the direction distribution of these edge directions.

The moving body discriminator 13 includes a moving body shape detector 5
And the value of the aspect ratio r of the rectangle circumscribing the moving body region 6 and the area s of the moving body region 6 obtained by
It has a function of inputting the uniformity g obtained by the edge direction distribution detector 12 and discriminating the vehicle body, the human, and the others based on the area s, the aspect ratio r, and the uniformity g of the moving body region 6. ing.

Next, the operation of the above-configured device will be described. The TV camera 1 captures an image of the traveling moving object 2 and digitizes the image data I (t, i, j) at time t (FIG. 18 (a)) in the same manner as described above. (t + Δt, i, j) is sent to the moving object detector 4 as {FIG. 18 (b)}.

The moving object detector 4 receives two image data I (t, i, j) and I (t + Δt, i, j) and obtains the above equation (2).
And the difference image data J (t + Δt, i, j)
Get.

Next, the moving object detector 4 compares the difference image data J (t + Δt, i, j) with a preset threshold value in the same manner as described above, and determines the difference image data J (t + Δt, i, j). 19), a pixel larger than the threshold value is converted into a value “1”, and a pixel smaller than the threshold value is converted into a value “0”, and the moving object shape detector 5 and the edge direction distribution detector 12 are converted into a binary image as shown in FIG. It is transmitted as data K (t + Δt, i, j).

The moving object shape detector 5 extracts the value “1” of the binary image data K (t + Δt, i, j) as one moving object region 6 as one lump, and extracts the area s of the moving object region 6 in pixel units. , And as shown in FIG. 20, the aspect ratio r of the lengths h and w of the rectangle circumscribing the moving body region 6 is calculated.

On the other hand, the edge direction distribution detector 12 receives the moving body region 6 obtained by the moving body detector 4, detects the edge direction for each pixel of the moving body region 6, and detects the edge direction. Is obtained, which represents the degree of uniformity with respect to the directional distribution.

The operation of the edge direction distribution detector 12 will be described in detail.
, A spatial differential image Ix (i, j) in the i direction and a spatial differential image Iy (i, j) in the j direction are calculated for the image data I (t, i, j), and the edge strength ( Ix ² + Iy ² ) ^1/2 (16) For a pixel _{equal to} or _larger than a certain threshold value Ith, the edge direction ψ (i,
j) is calculated by the following equations (17) to (20).

[0056]

(Equation 2)

That is, when a pixel whose edge strength is _{equal to} or greater than a certain threshold value Ith is extracted, a contour or a pattern of the moving object on the image is detected. The direction of the edge indicates the direction in which the change in brightness is greatest in each pixel of the edge. For example, in FIG. 9, a boundary where the brightness changes (in this case, a black portion and a white portion) is an edge, and the edge direction of the edge is 0 °.

From these equations (17) and (20), the above-mentioned FIG.
, The edge direction ψ (i,
j) is defined as the direction in which the change in brightness near the pixel is the largest as shown in FIG. That is, FIG.
As shown in (1), the boundary is a direction perpendicular to the edge direction with the boundary of the brightness change as an edge (0 ≦ ψ <π).

An edge direction histogram shown in FIG. 11 is generated from the value of the edge direction ψ (i, j). That is, angles 0 to π are quantized with an appropriate width Δψ, and the number of edges is cumulatively added in each direction to create an edge direction histogram.

[0060] determine the direction [psi _M giving the maximum value of the edge direction histogram, the following equation uniformity g direction distribution of the edge of the total number of pixels that fall within the scope of that direction [psi width _M around the [psi _W as N _e Defined in (21).

G = N _e / N (21) where N is the total number of pixels included in the moving body area 6.
The meaning of the uniformity g is as follows.

The vehicle body has a straight edge as shown in FIG. 21A, and the edge direction histogram is shown in FIG.
Tend to gather in a specific direction [psi _M and around which is as shown in, the value of the uniformity of g is increased.

On the other hand, in the case of a human, the edge direction is oriented in various directions as shown in FIG. 21B, and the edge direction histogram is not concentrated in a certain direction as shown in FIG. The value of the uniformity g decreases.

Next, the moving body discriminator 13 inputs the area s of the moving body area 6 obtained by the moving body shape detector 5 and the value of the aspect ratio r of the rectangle circumscribing the moving body area 6, and The uniformity g obtained by the edge direction distribution detector 12 is input, and a vehicle body, a person, and others are identified based on the area s, the aspect ratio r, and the uniformity g of the moving body region 6.

The distinction between the vehicle body, the person, and the others is made according to the following relationship. (a) When s ≧ s _th , r ≧ r _th1 , and g ≧ g _th , it is identified as a vehicle body. (b) When s ≧ s _th , r <r _th2 , and g <g _th , it is identified as a human (c) ) In cases other than these (a) and (b), it is identified as a body other than a human body.

As described above, in the second embodiment, the moving object 2 is included in the difference image data between the two image data I (t) and I (t + Δt) separated by the interval of the sampling time Δt. After extracting the moving body region 6, the edge direction is detected for each pixel of the moving body region 6, and the degree of uniformity g representing the degree of uniformity with respect to the direction distribution of the edge direction is obtained. Since the vehicle body, the person, and others are distinguished based on the area s, the aspect ratio r, and the uniformity g, for example, the vehicle body was observed from the front as the moving body 2 as in the first embodiment. Even in a case or when a human walks with both hands and both legs wide open, the vehicle body and the human can be accurately distinguished from the difference in uniformity between the movement of the vehicle and the human. (3) Next, a third embodiment of the present invention will be described with reference to the drawings. The same parts as those in FIG. 1 are denoted by the same reference numerals, and detailed description thereof will be omitted.

FIG. 13 is a configuration diagram of a moving object identification device.
Uniformity g found by edge direction distribution detector 12
Are sent to the moving object shape detector 5, the moving direction distribution detector 10, and the edge direction distribution detector 12.

The area s and the aspect ratio r of the moving body region 6 included in the binary image data obtained by the moving body shape detector 5, the uniformity f obtained by the moving direction distribution detector 10, The uniformity g obtained by the edge direction distribution detector 12 is sent to the discriminant function calculator 14.

The discriminant function calculator 14 calculates the moving object area 6
Has a function of calculating and calculating each value of the vehicle-likeness C and the human-likeness H based on the area s, the aspect ratio r, the uniformity f, and the uniformity g of the vehicle.

The moving object discriminator 13 includes a discriminant function calculator 14
Has the function of identifying a vehicle body, a person, and others based on the values of the vehicle-likeness C and the human-likeness H obtained by the above.

Next, the operation of the device configured as described above will be described. The TV camera 1 captures an image of the traveling moving object 2 and digitizes the image data I (t, i, j) at time t (FIG. 18 (a)) in the same manner as described above. (t + Δt, i, j) is sent to the moving object detector 4 as {FIG. 18 (b)}.

The moving object detector 4 inputs two image data I (t, i, j) and I (t + Δt, i, j) and obtains the above equation (2).
And the difference image data J (t + Δt, i, j)
Get.

Next, the moving object detector 4 compares the difference image data J (t + Δt, i, j) with a preset threshold value in the same manner as described above, and determines the difference image data J (t + Δt, i, j). ) Is converted to a value “1” and a pixel smaller than the threshold value is converted to a value “0”, and the moving object shape detector 5, the moving direction distribution detector 10, and the edge direction distribution detector 12 are used in FIG.
9 as binary image data K (t + Δt, i, j).

The moving object shape detector 5 extracts the value “1” of the binary image data K (t + Δt, i, j) as a single moving object region 6 and calculates the area s in units of pixels. , And as shown in FIG. 20, the aspect ratio r of the lengths h and w of the rectangle circumscribing the moving body region 6 is calculated.

At the same time, the moving direction distribution detector 10
As shown in FIG. 2, the moving body region 6 is divided into a plurality of small regions, each movement vector of these small regions is detected, and the degree of uniformity indicating the degree of uniformity of these movement vectors with respect to the direction distribution is obtained.

Further, the edge direction distribution detector 12 receives the moving object region 6 obtained by the moving object detector 4, detects the edge direction for each pixel of the moving object region 6, and detects the edge direction. Is obtained, which represents the degree of uniformity with respect to the directional distribution.

The area s and the aspect ratio r of the moving body region 6 included in the binary image data obtained by the moving body shape detector 5, the uniformity f obtained by the moving direction distribution detector 10, And the uniformity g obtained by the edge direction distribution detector 12 is sent to the discriminant function calculator 14.

The discriminant function calculator 14 calculates the moving object area 6
The values of the vehicle-likeness C and the human-likeness H are calculated and obtained based on the area s, the aspect ratio r, the uniformity f, and the uniformity g of. That is, (a) the vehicle-likeness C is (when S ≧ _Sth ) C = a ₁ · C ₁ (r) + a ₂ · C ₂ (f) + a ₃ · C ₃ (g) (22) (b The humanity H is (when S ≧ _Sth ) H = b ₁ · H ₁ (r) + b ₂ · H ₂ (f) + b ₃ · H ₃ (g) (23) (c) C = H = 0 (when s <s _th ) (24)

Here, a ₁ , a ₂ , a ₃ and b ₁ , b
₂ and b ₃ are weighting factors, and C ₁ , C ₂ and C ₃ are shown in FIG.
(a) to function as shown in FIG. _{(c), H 1, H} 2, H 3
Are functions as shown in FIGS. 15 (a) to 15 (c).

These functions are functions that output 1 or a value close to 1 when C ₁ , C ₂ , and C ₃ are images of the moving body area 6 of a vehicle body. For H ₁ , H ₂ , and H ₃ , Indicates a function that outputs a value of 1 or a value close thereto when the image of the moving body region 6 is a human.

Although a relatively simple function form has been shown here,
The same effect can be obtained with a function having the above properties. Therefore, as the aspect ratio r of the moving body region 6 is larger, the uniformity f of the moving direction distribution is closer to 1, and the uniformity g of the edge direction distribution is closer to 1, the value of the vehicle-likeness C becomes larger and the human-likeness H Becomes smaller.

Next, the moving object discriminator 15 discriminates the vehicle body, the human, and the others based on the values of the vehicle-likeness C and the human-likeness H obtained by the identification function calculator 14. The distinction between the vehicle body, the person, and the others is performed according to the following relationship, and is as shown in FIG. (a) Identified as a vehicle body when C ≧ C _th1 and H <H _th2 (b) Identified as a human body when C <C _th2 and H ≧ H _th1 (c) In cases other than these (a) and (b) Identify the vehicle as a non-human.

As described above, in the third embodiment, based on the area s and the aspect ratio r, the uniformity f, and the uniformity g of the moving body region 6, the values of the vehicle-likeness C and the human-likeness H are calculated. The vehicle body, the person, and the others are distinguished based on the values of the body-likeness C and the human-likeness H by calculation. Therefore, as in the first embodiment, for example, the vehicle 2 Even when observing from the front or when a human walks with both hands and legs wide open, the vehicle body and the human can be accurately distinguished from the difference in uniformity between the motion of the vehicle and the motion of the human.

The present invention is not limited to the first to third embodiments, but may be modified as follows. For example, in the above-described first to third embodiments, the vehicle body and the person are distinguished from each other, but it is needless to say that the present invention is not limited to these and can be applied to other moving bodies. .

[0085]

As described in detail above, according to the first to fourth aspects of the present invention, the moving object in the difference image data of each image data of a predetermined sampling time difference obtained when the moving object is imaged by the imaging device. When the type of the moving object is identified based on the shape of the moving object region included, the distribution of the moving direction of the entire moving object is obtained from the image data in the moving object region, and the uniformity representing the degree of the distribution in the moving direction is obtained. For example, dividing the moving object area into a plurality of small areas, detecting each movement vector of these small areas, and obtaining uniformity representing the degree of uniformity of the direction distribution of these movement vectors. Direction distribution detecting means for detecting an edge direction for each pixel in the moving object area, and detecting an edge direction distribution for obtaining a degree of uniformity indicating a degree of uniformity with respect to the direction distribution of the edge directions. Since with either or both of the means, can provide a mobile identification device capable of accurately identify the vehicle body and human.

[Brief description of the drawings]

FIG. 1 is a configuration diagram showing a first embodiment of a moving object identification device according to the present invention.

FIG. 2 is a schematic diagram showing an operation of dividing a moving body region into a plurality of small regions by a moving direction distribution detector in the same device.

FIG. 3 is a schematic diagram showing a set of movement vectors for all small areas of the vehicle body.

[4] subregion motion vector (u _k, v _k) direction θ
shows a _k.

FIG. 5 is a diagram showing a moving direction histogram for a vehicle body.

FIG. 6 is a diagram showing a movement direction histogram for a human.

FIG. 7 is a schematic diagram showing a set of motion vectors for all human small areas.

FIG. 8 is a configuration diagram showing a second embodiment of the moving object identification device according to the present invention.

FIG. 9 is a schematic diagram illustrating an edge direction of each pixel in a moving body region.

FIG. 10 is a vector diagram showing an edge direction.

FIG. 11 is a diagram showing a movement direction histogram for a vehicle body.

FIG. 12 is a diagram showing a movement direction histogram for a human.

FIG. 13 is a configuration diagram showing a third embodiment of the moving object identification device according to the present invention.

FIG. 14 is a diagram showing a function for outputting a value of 1 or a value close to 1 when an image of a moving body region is a vehicle body.

FIG. 15 is a diagram illustrating a function that outputs a value of 1 or a value close to 1 when an image of a moving body region is a human.

FIG. 16 is a schematic diagram showing identification of a vehicle body, a person, and others by a moving object identifier.

FIG. 17 is a configuration diagram of a conventional moving object identification device.

FIG. 18 is a schematic diagram of two image data separated by an interval of a sampling time Δt.

FIG. 19 is a schematic diagram of binary image data.

FIG. 20 is a diagram showing the aspect ratio of a rectangle circumscribing a moving body region by the moving body shape detector.

FIG. 21 is a schematic diagram of each image data when the vehicle body is observed from the front and when a human walks with both hands and feet wide open.

[Explanation of symbols]

 1: TV camera, 2: Moving object, 3: A / D converter, 4: Moving object detector, 5: Moving object shape detector, 10: Moving direction distribution detector, 11, 13: Moving object discriminator, 12: Edge direction distribution detector, 14: Discriminant function calculator, 15: Moving object discriminator.

Claims (4)

[Claims] 1. A method for obtaining difference image data of each image data having a predetermined sampling time difference from image data obtained when an image of a moving object is picked up by an imaging device, and determining a shape of a moving object region including the moving object in the difference image data. In the moving object identification device that identifies the type of the moving object based on the following, a distribution in the moving direction of the entire moving object is obtained from image data in the moving object region, and a uniformity representing a degree of the distribution in the moving direction is obtained. A moving object identification apparatus, comprising: a uniformity processing means for determining the type of the moving object based on the uniformity and the shape of the moving object region. 2. The uniformity processing means divides the moving body region into a plurality of small regions, detects respective movement vectors of the small regions, and indicates a degree of uniformity of the movement vectors with respect to a directional distribution. 2. The moving object identification device according to claim 1, wherein the moving object identification device has a function as a moving direction distribution detecting means for obtaining uniformity. 3. An edge direction distribution detecting means for detecting an edge direction for each pixel of the moving body area and obtaining a degree of uniformity indicating a degree of uniformity with respect to a direction distribution of the edge direction. The mobile object identification device according to claim 1, having a function as (1). 4. The uniformity processing means divides the moving body region into a plurality of small regions, detects respective movement vectors of the small regions, and indicates a degree of uniformity of the movement vectors with respect to a directional distribution. Moving direction distribution detecting means for obtaining uniformity; and edge direction distribution detecting means for detecting an edge direction for each pixel of the moving object area and obtaining uniformity representing a degree of uniformity with respect to the directional distribution of the edge directions. The moving object identification device according to claim 1, further comprising: