JP4954110B2 - Movement information similarity determination device - Google Patents

Description

  The present invention discriminates a movement scenario similar to an input movement scenario from a plurality of movement scenarios accumulated in advance (information indicating the positions of a plurality of targets at preset times). The present invention relates to a movement information similarity determination device.

For example, in the conventional movement information similarity determination device disclosed in Patent Document 1 below, it is assumed that the number of targets included in the movement scenario is one.
In the conventional movement information similarity determination device, the similarity between a plurality of movement scenarios already stored in a database and a newly input movement scenario is defined as follows.
That is, the center of gravity of the movement scenario is matched to obtain the Euclidean distance between the target position vectors at each time, and the similarity is defined by the square root value of the square sum of the Euclidean distance.
In the conventional movement information similarity determination device, as the square root value is smaller, the similarity between movement scenarios is higher. In particular, if the square root value is zero, it is determined that the movement scenarios to be compared match.

Japanese Patent No. 3984134 (paragraph numbers [0053] to [0081], FIG. 1)

  Since the conventional movement information similarity determination apparatus is configured as described above, the number of targets included in the movement scenario is limited to one. For this reason, when two or more targets are included in the movement scenario, it cannot be determined whether or not the newly input movement scenario is similar to the movement scenario stored in advance. There was a problem.

  The present invention has been made in order to solve the above-described problems, and even when a movement scenario includes two or more targets, a newly input movement scenario is accumulated in advance. It is an object of the present invention to obtain a movement information similarity determination device that can determine whether or not the information is similar.

  In the movement information similarity determination device according to the present invention, for each scenario time, an input scenario that is a newly input movement scenario and a retention scenario that is a movement scenario that is held in advance are set by the projection space setting means. A spatio-temporal distribution generating means for generating a spatio-temporal distribution indicating a target number of spatial distributions for each scenario time by projecting onto a projective space, and a spatio-temporal distribution and a holding scenario relating to an input scenario generated by the spatiotemporal distribution generating means A spatio-temporal difference distribution generating means for generating a spatio-temporal difference distribution indicating a difference between spatio-temporal distributions, and a similarity evaluation value between the input scenario and the holding scenario from the spatio-temporal difference distribution generated by the spatio-temporal difference distribution generating means. A similar evaluation value calculation means for calculating, and a similar scenario discriminating means refers to the similarity evaluation value calculated by the similarity evaluation value calculation means and holds the input scenario Nario is that so as to determine whether or not similar.

  According to the present invention, at each scenario time, an input scenario that is a newly input movement scenario and a stored scenario that is a movement scenario that is stored in advance are projected onto the projection space set by the projection space setting means. A spatio-temporal distribution generating means for generating a spatio-temporal distribution indicating a spatial distribution of a target number for each scenario time; a spatio-temporal distribution relating to an input scenario generated by the spatio-temporal distribution generating means and a spatio-temporal distribution relating to a holding scenario; Spatio-temporal difference distribution generation means for generating a spatio-temporal difference distribution indicating a difference, and similarity evaluation value calculation for calculating a similarity evaluation value between the input scenario and the holding scenario from the spatio-temporal difference distribution generated by the spatio-temporal difference distribution generation means And the similar scenario discriminating means refers to the similarity evaluation value calculated by the similarity evaluation value calculating means, and the input scenario and the holding scenario are similar. Because it is configured to determine whether or not, even if the movement scenario contains two or more targets, whether or not the newly entered movement scenario is similar to the movement scenario that has been accumulated in advance There is an effect that can be determined.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing a movement information similarity determination apparatus according to Embodiment 1 of the present invention. In the figure, an input scenario group accumulator 1 is a newly input movement scenario (at each scenario time set in advance). This is a memory for storing one or more input scenarios that are information indicating the positions of a plurality of targets.
The spatial shift method group setting unit 2 uses a parallel shift amount when the input scenario is translated as a spatial shift amount of the input scenario stored in the input scenario group accumulator 1 or a rotational shift of the input scenario. Processing for setting values such as the position, orientation, and rotation amount of the rotation axis is performed.
The space shift method group setting device 2 constitutes a space shift amount setting means.

The space shifter 3 reads the input scenario stored in the input scenario group accumulator 1 and spatially shifts the input scenario by the shift amount set by the space shift method group setting unit 2 (parallel movement, rotation movement). Then, a process of outputting the shifted input scenario to the spatiotemporal distribution generator 7 is performed.
The space shifter 3 constitutes a space shift means.
The retention scenario group accumulator 4 is a memory that accumulates one or more retention scenarios that are movement scenarios to be compared with the input scenario.
The holding scenario group reader 5 performs a process of reading the holding scenario from the holding scenario group accumulator 4.

The projection space setting unit 6 sets a projection space for projecting an input scenario and a holding scenario, which are movement scenarios, at each scenario time, and executes a process of setting a step size of the projection space for counting the target number.
The projection space setting unit 6 constitutes a projection space setting means.
The spatio-temporal distribution generator 7 projects the input scenario output from the space shifter 3 to the projection space set by the projection space setting unit 6 at each scenario time, and shows the spatial distribution of the target number for each scenario time. A process for generating a spatiotemporal distribution is performed.
The spatio-temporal distribution generator 8 projects the holding scenario read by the holding scenario group reader 5 onto the projection space set by the projection space setting unit 6 for each scenario time, and a target number of spaces for each scenario time. A process of generating a spatiotemporal distribution indicating the distribution is performed.
The spatiotemporal distribution generators 7 and 8 constitute spatiotemporal distribution generation means.

The time shifter 9 sets various temporal shift amounts (amount to shift in the time direction) with respect to the spatiotemporal distribution related to the input scenario generated by the spatiotemporal distribution generator 7, and the spatiotemporal distribution is changed by various shift amounts. A process of shifting in the time axis direction and outputting one or more shifted spatiotemporal distributions to the spatiotemporal difference distribution generator 10 is performed. The time shifter 9 constitutes a time shift means.
The spatio-temporal difference distribution generator 10 shows a spatio-temporal difference distribution indicating a difference between the spatio-temporal distribution related to the input scenario output from the time shifter 9 and the spatio-temporal distribution related to the holding scenario generated by the spatio-temporal distribution generator 8. The process to generate is performed. The spatiotemporal difference distribution generator 10 constitutes a spatiotemporal difference distribution generating means.

The spatial filter setting unit 11 performs a process of setting a spatial filter that is a weight distribution having a spread in the spatial axis direction. The spatial filter setting unit 11 constitutes a spatial filter setting unit.
The spatial filter convolution unit 12 convolves the spatial filter set by the spatial filter setting unit 11 with the spatio-temporal difference distribution generated by the spatio-temporal difference distribution generator 10, and calculates a similar evaluation value for the spatio-temporal difference distribution after the spatial filter convolution. The process of outputting to the device 13 is performed. The spatial filter convolution unit 12 constitutes a spatial filter convolution means.

The similarity evaluation value calculator 13 performs a process of calculating a similarity evaluation value between the input scenario and the holding scenario from the spatiotemporal difference distribution in which the spatial filter is convolved by the spatial filter convolution unit 12. The similarity evaluation value calculator 13 constitutes a similarity evaluation value calculation unit.
The similar evaluation value group accumulator 14 is a memory for accumulating the similarity evaluation values between the input scenario calculated by the similarity evaluation value calculator 13 and the holding scenario.

The discrimination criterion setting unit 15 performs processing for setting a discrimination criterion for a holding scenario similar to the input scenario.
The similar scenario discriminator 16 reads the similarity evaluation value between the input scenario and the holding scenario from the similarity evaluation value group accumulator 14, collates the similarity evaluation value with the discrimination criterion set by the discrimination criterion setting unit 15, and inputs the similarity evaluation value. It is determined whether or not the scenario and the retention scenario are similar. If the input scenario and the retention scenario are similar, a process of outputting the retention scenario to the similar scenario accumulator 17 is performed.
The similar scenario accumulator 17 is a memory for accumulating a retained scenario determined by the similar scenario discriminator 16 as being similar to the input scenario.
The discrimination criterion setting unit 15 and the similar scenario discriminator 16 constitute a similar scenario discrimination means.

FIG. 2 is an explanatory diagram showing the translation of the input scenario in the space shifter 3. In FIG. 2, L1 indicates the target trajectory included in the input scenario before the translation, and L2 is the input after the translation. The trajectory of the target included in the scenario is shown.
FIG. 3 is an explanatory diagram showing the rotational movement of the input scenario in the space shifter 3. In FIG. 3, L3 shows the locus of the target included in the input scenario before the rotational movement, and L4 is the input after the rotational movement. The trajectory of the target included in the scenario is shown.

FIG. 4 is an explanatory diagram showing a projection space set by the projection space setting unit 6 and a spatio-temporal distribution that is a spatial distribution of a target number for each scenario time.
4A shows an example of the spatial distribution of the target number at the scenario time 0, FIG. 4B shows an example of the spatial distribution of the target number at the scenario time 1, and FIG. 4C shows the scenario time. 2 shows an example of the target number spatial distribution in FIG.
FIG. 5 is a flowchart showing the processing contents of the projective space setting unit 6 and the spatiotemporal distribution generators 7 and 8.

FIG. 6 is an explanatory diagram showing the processing contents of the time shifter 9.
In particular, FIG. 6A shows an example of the original spatiotemporal distribution.
FIG. 6 (b) shows an example of the spatiotemporal distribution after shifting the original spatiotemporal distribution by +1 time number, and FIG. 6 (c) shows the original spatiotemporal distribution after shifting by +2 time number. An example of a spatiotemporal distribution is shown.

FIG. 7 is a principle diagram for explaining the processing contents of the spatial filter convolution unit 12 and the similarity evaluation value calculator 13.
In FIG. 7, A shows an example of the spatiotemporal difference distribution before the spatial filter convolution, B shows an example of the spatial filter, and C shows an example of the spatiotemporal difference distribution after the spatial filter convolution.
D shows an example of the calculation result of the similarity evaluation value when it is assumed that the spatial filter convolution is not performed by the spatial filter convolution unit 12, and E is the case where the spatial filter convolution is performed by the spatial filter convolution unit 12. An example of the calculation result of the similarity evaluation value is shown.

FIG. 8 is an explanatory diagram showing an example of a spatial filter set by the spatial filter setting unit 11.
In particular, FIG. 8A shows an example of a filter at the time of three-dimensional space projection, and FIG. 8B shows an example of a filter at the time of two-dimensional plane projection.
FIG. 8C shows an example of a filter at the time of one-dimensional axis projection, and FIG. 8D shows an example of a filter at the time of azimuth axis projection.

Next, the operation will be described.
First, the movement scenario (input scenario, holding scenario) accumulated by the input scenario group accumulator 1 and the holding scenario group accumulator 4 will be described.
The movement scenario indicates the position of each target at each time t (n_t).
Here, n_t is a time number, and n_t = 0, 1, 2,..., N_t−1. N_t is the total number of times.
Hereinafter, t (n_t) is referred to as the n_t scenario time, and t (n_t) is handled at equal intervals of time step width Δt as shown in the following equation (1).
t (n_t) = n_t · Δt (1)
However, the intervals of the scenario times are not limited to being equal intervals, and may be unequal intervals.

The target position can be expressed in various forms such as an azimuth elevation angle altitude and a north reference orthogonal coordinate system.
Between these, conversion formulas have already been established, and in the following, on the premise of using these conversion formulas, the positions of multiple targets are expressed in a three-dimensional orthogonal coordinate system without loss of generality. I will do it.

In the first embodiment, the target number is represented by k (k = 0, 1,..., K−1) on the assumption that K targets are included in the movement scenario.
The three-dimensional orthogonal coordinate axes are represented by x_0, x_1, and x_2, and the position of the kth target in the x_j axis (j = 0, 1, 2) direction at the n_t scenario time is represented by R (n_t, k, j). Like that.
Thereby, the three-dimensional position of each target at each scenario time can be specified.

In the first embodiment, an effective flag U (n_t, k) indicating whether or not the information indicating the three-dimensional position of the kth target at the n_t scenario time is effective is introduced.
When the information indicating the three-dimensional position of the kth target at the n_t scenario time is valid → U (n_t, k) = 1
When the information indicating the three-dimensional position of the kth target at the n_t scenario time is invalid → U (n_t, k) = 0
Therefore, when the valid flag U (n_t, k) is “1”, the three-dimensional position of the kth target at the n_t scenario time is used, and the valid flag U (n_t, k) is “0”. In this case, the three-dimensional position of the kth target at the n_t scenario time is not used.

Hereinafter, the number of input scenarios accumulated in the input scenario group accumulator 1 is represented by Min, and the number of holding scenarios accumulated in the holding scenario group accumulator 4 is represented by Mho.
Further, the target position in the m-th input scenario (m = 0, 1,..., Min−1) stored in the input scenario group storage 1 is represented by Rin_m, the valid flag is represented by Uin_m, and the target number is represented by kin_m. .
Similarly, the target position in the m-th holding scenario (m = 0, 1,..., Mho-1) stored in the holding scenario group storage 4 is represented by Rho_m, the valid flag is represented by Uho_m, and the target number is represented by kho_m. To do.

Next, the processing content of the movement information similarity determination apparatus of FIG. 1 is demonstrated.
First, the spatial shift method group setting unit 2 rotates the input scenario as the spatial shift amount of the input scenario stored in the input scenario group storage unit 1 or the input scenario when the input scenario is translated. Set values such as the position, orientation, and amount of rotation of the rotating shaft.
Hereinafter, the rotational movement of the input scenario in the space and the parallel movement of the input scenario are collectively referred to as a space shift.
Depending on the application, the input scenario may be treated the same whether it is spatially shifted or not spatially shifted. In some cases, it may be handled approximately the same.
Here, the spatial shift is performed by determining the degree of similarity between the movement scenarios in the later stage, the input scenario stored in the input scenario group storage 1, and the holding scenario stored in the holding scenario group storage 4. Are compared with correction processing for eliminating the positional deviation and orientation deviation of the target so that the target position and orientation are matched as much as possible.

  When the spatial shift method group setting unit 2 sets the shift amount (parallel movement amount, rotation amount, etc.), the spatial shifter 3 is selected from the Min input scenarios stored in the input scenario group storage unit 1. The m input scenario (m = 0, 1,..., Min−1) is read out, and the mth input scenario is spatially shifted (parallel movement, rotational movement) by the shift amount set by the spatial shift method group setting unit 2, and the spatial shift is performed. The subsequent m-th input scenario is output to the spatiotemporal distribution generator 7.

The processing contents of the space shift method group setting device 2 and the space shift device 3 will be specifically described below.
First, the spatial shift method group setting unit 2 sets a column vector dd indicating the parallel movement of the input scenario as a spatial shift amount of the input scenario.
FIG. 2 shows an image of translation of the input scenario by the space shifter 3.
In the example of FIG. 2, the target locus L1 included in the input scenario before the parallel movement is translated in the direction of the column vector dd by the space shifter 3, and the target locus changes to L2. Is shown.

Next, the space shift method group setting device 2 sets values such as the position, orientation, and rotation amount of the rotation axis when the input scenario is rotated and moved as the spatial shift amount of the input scenario.
That is, as shown in FIG. 3, the space shift method group setting unit 2 sets a position vector ppc which is a column vector indicating the position of one point on the rotation axis in order to determine the position of the rotation axis.
The space shift method group setting unit 2 sets a unit vector uu in the direction of the rotation axis, which is a column vector in the direction along the rotation axis. However, the x_0, x_1, and x_2 axis components of the unit vector uu are u0, u1, and u2.
Furthermore, the space shift method group setting unit 2 sets the rotation angle θ in the right-hand screw direction with reference to the unit vector uu as the rotation angle (rotation amount) around the rotation axis.
FIG. 3 shows an image of the rotational movement of the input scenario by the space shifter 3.
In the example of FIG. 3, the target locus L3 included in the input scenario before the rotational movement is rotated and moved by the space shifter 3, and the target locus is changed to L4.

Here, in general, the point on the position vector pp _old is moved to the position of the position vector pp _new shown in the following equation (2) by a spatial shift combining the above-described parallel movement and rotational movement.
pp _new = Arot (uu, θ) (pp _old −ppc) + ppc + dd
(2)
In the equation (2), Arot (uu, θ) is a rotation matrix and is given by the following equation (3).
Arot (uu, θ)
= S1 (uu) + (I ₃₃ -S1 (uu)) cos θ + S2 (uu) sin θ
(3)

Ｉ₃₃は３行３列の単位行列であり、Ｔｒａｎｓｐｏｓｅ（ａａ）は行列又はベクトルａａの転置を得るオペレータである。 However,
S1 (uu) = uu · Transpose (uu) (4)

I ₃₃ is a unit matrix of 3 rows and 3 columns, and Transpose (aa) is an operator that obtains transpose of a matrix or vector aa.

In the above equation (2), if the vectors dd, ppc, uu and the rotation angle θ are changed, different position vectors pp _new can be obtained for the same position vector pp _old .
The space shift method group setting unit 2 sets ranges for changing the vectors dd, ppc, uu, and the rotation angle θ, and generates a combination in which each is changed. This combination of values is referred to as a spatial shift amount.
The total number of spatial shift amounts is represented as H_p.

In the space shifter 3, the values of j = 0, 1, and 2 of the kth target position Rin_m (n_t, k, j) at the n_t scenario time included in the mth input scenario are set to the position vector pp _old . 0 to the second component.
The spatial shifter 3 uses the vectors dd, ppc, uu, and the rotation angle θ for the combination of the position vector pp _old and the h_p-th spatial shift (h_p = 0, 1,..., H_p−1). The position vector pp _new is obtained from the equation (2).
In the space shifter 3, the 0th to 2nd elements of the position vector pp _new are stored in j = 0, 1, 2 of Rinp_m (n_t, k, h_p, j).

When the column vector dd indicating the parallel movement of the input scenario is a zero vector and the rotation angle θ is zero, the position vector pp _old and the position vector pp _new match.
Setting only one type of spatial shift amount that satisfies this condition is equivalent to omitting the spatial shift method group setting unit 2 and the spatial shifter 3 from the movement information similarity determination device of FIG.

Next, the projecting space setting unit 6 sets a projecting space in which the movement scenario (input scenario, holding scenario) is projected for each scenario time by the subsequent spatiotemporal distribution generators 7 and 8, and totalizes the target number. Sets the step size of the projection space.
FIG. 4 shows a projection space set by the projection space setting unit 6 and a spatio-temporal distribution which is a spatial distribution of the target number for each scenario time.
In the example of FIG. 4, the three-dimensional positions of the two targets (target 0, target 1) are changing with time.
That is, the three-dimensional position of the target 0 at time 0 is indicated by ◯, and the three-dimensional position of the target 1 is indicated by ●.
Further, the three-dimensional position of the target 0 at time 1 is indicated by ◇, and the three-dimensional position of the target 1 is indicated by ◆.
Further, the three-dimensional position of the target 0 at time 2 is indicated by □, and the three-dimensional position of the target 1 is indicated by ■.

FIG. 4 shows an example in which the projection space is a two-dimensional plane.
This two-dimensional plane has its origin at the position of the position vector rrcy in the x_0-x_1-x_2 space.
The two-dimensional plane is defined as a plane that includes two unit vectors iiy_0 and iiy_1 that pass through the origin and are orthogonal to each other.
In this two-dimensional plane, a grid having components parallel to each unit vector iiy_j (j = 0, 1) is drawn as representing the step size of the projection space for counting the target number. The lattice spacing in each unit vector direction is Δy_j (j = 0, 1).

When the projection space setting unit 6 sets the projection space and the step size, the spatiotemporal distribution generator 7 projects the input scenario output from the space shifter 3 onto the projection space for each scenario time, and for each scenario time. A spatiotemporal distribution indicating the spatial distribution of the target number is generated.
In addition, when the projection space setting unit 6 sets the projection space and the step size, the spatiotemporal distribution generator 8 projects the holding scenario read by the holding scenario group reading unit 5 onto the projection space at each scenario time. Thus, a spatio-temporal distribution indicating the spatial distribution of the target number for each scenario time is generated.

In the example of FIG. 4, the spatiotemporal distribution generators 7 and 8 place the positions of two targets (target 0, target 1) at each time (time 0, time 1, time 2) into a projection space that is a two-dimensional plane. Orthographic projection is made.
In FIG. 4, the positions of the two targets orthogonally projected by the spatio-temporal distribution generators 7 and 8 are indicated at the tip of an arrow marked from the point before projection toward the two-dimensional plane (dotted line graphic ( Round, diamond, square)).
The spatio-temporal distribution generators 7 and 8 count the number of targets projected on each grid at each time (time 0, time 1 and time 2) to obtain the spatial distribution of the target numbers in the projection space. Yes.
4A shows the spatial distribution of the target number at scenario time 0, FIG. 4B shows the spatial distribution of the target number at scenario time 1, and FIG. 4C shows the target number space at scenario time 2. Distribution is shown.

FIG. 4 shows an example in which the projection space is a two-dimensional plane, but various projection spaces other than the two-dimensional plane are conceivable.
The following is an example of a typical projection space. However, it is not limited to the projection space specified here.

In preparation for defining the projection space, three unit column vectors iiy_0, iiy_1, and iiy_2 that are orthogonal to each other are defined in the x_0-x_1-x_2 orthogonal coordinate system so as to satisfy the relationship of the following expression.
iiy_j · iiy_j = 1 (j = 0, 1, 2) (6)
iiy_0 × iiy_1 = iiy_2
iiy_1 × iiy_2 = iiy_0
iiy_2 × iiy_0 = iiy_1
(7)

A space defined by these three unit column vector direction axes is particularly referred to as a “projection original space”.
Further, the position vector of the origin of the projection original space in the x — 0-x — 1-x — 2 orthogonal coordinate system is assumed to be rrcy.
In the following, it is assumed that the position vector of the projected point is generally represented by ppx (m, n_t, k, h_p).

Next, the projection position zy_j (m, n_t, k, h_p) with respect to the axis in the direction of each unit vector iiy_j (j = 0, 1, 2) in the projection original space is defined by the following equation (8).
zy_j (m, n_t, k, h_p)
= (Ppx (m, n_t, k, h_p) -rrcy) · iiy_j (8)
However, j = 0, 1, and 2.
Aa1 and aa2 are operators that calculate the inner product of the vector aa1 and the vector aa2.

Next, the lattice width in each axial direction is Δy_j (j = 0, 1, 2), and the discretized position iy_j (m, n_t, k, h_p) in the iiy_j axial direction is defined by the following (9).
iy_j (m, n_t, k, h_p)
= Round (zy_j (m, n_t, k, h_p) / Δy_j) (9)
However, j = 0, 1, and 2.
Also, round (a) is a function that rounds off the decimal point of the real number a.

Next, the azimuth angle az (m, n_t, k, h_p) and the elevation angle el (m, n_t, k, h_p) are expressed by the following expressions (10) and (11) with the origin position in the projection original space as a reference. Define in. The unit is radians.
az (m, n_t, k, h_p)
= Atan2 (zy_1 (m, n_t, k, h_p), zy_0 (m, n_t, k, h_p))
(10)

ただし、ｓｉｇｎ（ａ）は実数ａ＜０で“−１”、ａ≧０で“＋１”を返すオペレータである。 However, atan2 is obtained by extending the range of the arctangent function to a range of “−π to π”, and is normally implemented in the c language or the like.
The contents of atan2 (a2, a1) can be expressed as in the following formula (12), for example.

However, sign (a) is an operator that returns “−1” when the real number a <0, and “+1” when a ≧ 0.

Finally, assuming that the azimuth increment is Δaz and the elevation angle increment is Δel, the discretized positions iaz (m, n_t, k, h_p), iel () in each angle axis direction are expressed by the following formulas (13) and (14). m, n_t, k, h_p).
iaz (m, n_t, k, h_p)
= Round (az (m, n_t, k, h_p) / Δaz) (13)
iel (m, n_t, k, h_p)
= Round (el (m, n_t, k, h_p) / Δel) (14)

As the projection space, for example, the following can be considered.
(A) Three-dimensional space that is a projection original space (b) Two-dimensional space (plane) defined by two of the three axes constituting the projection original space
(C) One-dimensional space (straight line) defined by one of the three axes constituting the projected original space
(D) A two-dimensional space having the azimuth and elevation angles of the projected original space as two axes. (E) A one-dimensional space having the azimuth or elevation angle of the projected original space as one axis.

Here, one projection source space is defined and the projection space is set from the orthogonal three-dimensional space. However, a plurality of projection source spaces are defined, and a projection space is set from each of the plurality of projection source spaces. May be.
For example, it is possible to define a plurality of projection original spaces in which the directions of the origin and the respective axes are different, set projection spaces from the plurality of projection original spaces, and combine the plurality of projection spaces.

Next, with reference to FIG. 5, general-purpose processing contents in which the projective space setting unit 6 and the spatiotemporal distribution generators 7 and 8 obtain the target number of spatiotemporal distributions from the movement scenario will be described.
First, the projection space setting unit 6 sets a projection method as an initial setting (step ST1).
That is, the projective space setting unit 6 includes the number of scenarios M, the number of times N_t, the number of space shifts H_p, the number of projective spaces H_c, the dimension F of the projective space, and the number of lattices in the f-th axial direction J_f (f = f = 0, 1, ..., F-1) are set.
Further, the projective space setting unit 6 represents the spatiotemporal distribution of the target number by an array TSD (m, n_t, h_p, h_c, j_0, j_1,..., J_ {F-1}).
However, m, n_t, h_p, h_c, and j_f take the following values.
m = 0, ..., M-1
n_t = 0,..., N_t−1
h_p = 0,..., H_p−1
h_c = 0,..., H_c-1
j_f = 0,..., J_f-1 (f = 0,..., F-1)
Note that the initial value of each cell in the array TSD is set to 0, and the value of each counter (scenario number m, target number k, time number n_t, space shift number h_p, projection space number h_c) is set to 0.
In the spatial distribution of the holding scenarios stored in the holding scenario group storage 4, H_p = 1.

When the projection space setting unit 6 sets the projection method, the spatiotemporal distribution generators 7 and 8 calculate the discretized positions in the respective axial directions according to the projection method (step ST2).
However, when an orthogonal coordinate system such as the projection spaces (a), (b), and (c) described above is used as the projection space, the iiy_j-axis direction discrete is obtained using the equations (2) to (9). The conversion position iy_j (m, n_t, k, h_p) is calculated.
Further, when an angular coordinate system such as the projection spaces of (d) and (e) described above is used as the projection space, the discretization positions in the respective angle axis directions are used using the equations (14) and (15). iaz (m, n_t, k, h_p) and iel (m, n_t, k, h_p) are calculated.
By combining these, the position in the u-th axial direction of the projection space is ij_f (f = 0, 1,..., F−1).
For example, when azimuth and elevation are used, the following is performed.
ij — 0 = iaz (m, n_t, k, h_p)
ij_1 = iel (m, n_t, k, h_p)

Next, the spatiotemporal distribution generators 7 and 8 calculate the following expression (15) based on the uth axial position ij_f (f = 0, 1,..., F−1) of the projection space. The spatiotemporal distribution TSD is updated (step ST3).
TSD (m, n_t, h_p, h_c, ij_0, ij_1, ij_2, ..., ij_ {F-1})
= TSD (m, n_t, h_p, h_c, ij_0, ij_1, ij_2, ..., ij_ {F-1}) + U_m (n_t, k)
(15)
However, when the spatial distribution related to the input scenario stored in the input scenario group storage 1 is generated, the above-described valid flag Uin_m is used as U_m (n_t, k).
Further, when generating the spatial distribution related to the holding scenario stored in the holding scenario group storage 4, the above-described valid flag Uho_m is used as U_m (n_t, k).

After updating the spatiotemporal distribution TSD, the spatiotemporal distribution generators 7 and 8 determine whether or not the current projection space number h_c has reached H_c−1 (step ST3), and h_c <H_c−1. For example, “1” is added to the current projection space number h_c (step ST5), and the process returns to step ST2.
On the other hand, if h_c = H_c-1, the current projection space number h_c is set to "0" (step ST6), and the process proceeds to step ST7.

When the current projection space number h_c reaches H_c−1 and the current projection space number h_c is set to 0, the spatio-temporal distribution generators 7 and 8 reach the current space shift number h_p to H_p−1. (Step ST7), and if h_p <H_p-1, add "1" to the current space shift number h_p (step ST8), and return to the process of step ST2.
On the other hand, if h_p = H_p−1, the current space shift number h_p is set to “0” (step ST9), and the process proceeds to step ST10.

When the current space shift number h_p reaches H_p−1 and the current space shift number h_p is set to 0, the spatiotemporal distribution generators 7 and 8 reach the time number N_t−1. (Step ST10), if n_t <N_t-1, add "1" to the current number of times n_t (step ST11), and return to the processing of step ST2.
On the other hand, if n_t = N_t−1, the current time number n_t is set to “0” (step ST12), and the process proceeds to step ST13.

When the current time number n_t reaches the time number N_t−1 and the current time number n_t is set to 0, the spatio-temporal distribution generators 7 and 8 reach the scenario number M−1. (Step ST13) If m <M−1, “1” is added to the current scenario number m (Step ST14), and the process returns to Step ST2.
On the other hand, if m = M−1, the current scenario number m is set to “0” (step ST15), and the series of processing ends.

  Although FIG. 5 shows the case where the loop is turned in the order of the projection space number h_c, the space shift number h_p, the time number n_t, and the scenario number m, the order of turning the loop is not particularly limited, and the order of turning the loop is appropriately changed. May be.

When the spatio-temporal distribution generator 7 generates the spatio-temporal distribution related to the input scenario, the temporal shifter 9 generates a spatio-temporal difference distribution generator 10 of the latter stage to the spatio-temporal distribution related to the input scenario and the spatio-temporal distribution related to the holding scenario In order to be able to compare the distributions of the same time when calculating the difference of time, various temporal shift amounts (amounts shifted in the time direction) with respect to the spatio-temporal distribution related to the input scenario are set and various shifts are made. The spatiotemporal distribution is shifted in the time axis direction according to the amount, and one or more shifted spatiotemporal distributions are output to the spatiotemporal difference distribution generator 10.
Here, between the input scenario and the holding scenario, the start time of each trajectory based on the shape of the included trajectory and a certain time (for example, the time at which one of the trajectories included in the scenario becomes valid first) is similar. However, if the reference time is shifted, it cannot be determined that the spatio-temporal distributions of both are similar. Therefore, the purpose of the time shifter 9 is to make it possible to compare the spatiotemporal distributions by matching the reference time as much as possible.

FIG. 6 shows the processing contents of the time shifter 9.
FIG. 6B is an example in which the time shifter 9 is shifted by +1 time number with respect to the original spatiotemporal distribution generated by the spatiotemporal distribution generator 7 (see FIG. 6A). Yes, comparing the spatiotemporal distribution of FIG. 6A with the spatiotemporal distribution of FIG. 6B, the distribution of time number i (i = 0, 1, 2) of FIG. 6A and FIG. The distribution of time number i + 1 in (b) matches.
6C, the time shifter 9 is shifted by +2 time number with respect to the original spatiotemporal distribution generated by the spatiotemporal distribution generator 7 (see FIG. 6A). For example, comparing the spatiotemporal distribution of FIG. 6A with the spatiotemporal distribution of FIG. 6C, the distribution of time number i (i = 0, 1, 2) of FIG. The distribution of time number i + 2 in FIG. 6C matches.

Time number 3 and time number 4 in the spatiotemporal distribution in FIG. 6A, time number 0 and time number 4 in the spatiotemporal distribution in FIG. 6B, time number 0 in the spatiotemporal distribution in FIG. The distribution of time number 1 is a blank portion generated by time shift.
The blank part is expanded so that all elements are “0”.

Although FIG. 6 shows the case where the time shift direction is a positive direction, the time shift direction may be a negative direction.
In this case, the time number may be negative. Even if the time number becomes negative, there is no particular problem in essence, but the management of the sequence may be complicated.
In addition, the management of the array may become complicated due to the change in the size of the array. Based on this, for each scenario, a time interval in which there is no effective scenario around the start time and end time of the scenario is set sufficiently wide as will be described later.

The time shifter 9 shifts only the segmented array of the spatio-temporal distribution from the start time at which a valid scenario exists to the end time at which a valid scenario exists.
A time interval in which there is no valid scenario around the start time and end time of each scenario is ensured so that the result of shifting this segmented array does not exceed the time range of the original array.

Hereinafter, the processing content of the time shifter 9 is demonstrated concretely.
First, the time shifter 9 defines the shift amount of the spatiotemporal distribution with the shift definition array Sft (b).
However, b (b = 0, 1,..., B−1) is an array number of the shift definition array, and B is the number of types of shift amounts.
For example, Sft (0) = − 10, Sft (1) = − 9, Sft (2) = − 8,..., Sft (20) = 10.

Next, the time shifter 9 secures an array TSD_Tin for storing the spatiotemporal distribution after the time shift.
This array is represented as TSD_Tin (m, n_t, h_p, b, h_c, j_0, j_1, j_2,..., J_ {f−1}).
However, m, n_t, h_p, b, h_c, j_0, and j_f take the following values.
m = 0, 1,..., M−1
n_t = 0, 1,..., N_t−1
h_p = 0, 1,..., H_p−1
b = 0, 1,..., B-1
h_c = 0, 1,..., H_c-1
j_f = 0, 1,..., J_f-1 (f = 0, 1,..., F-1)

In addition, the time shifter 9 has “0” in any cell in the spatio-temporal distribution TSD_Tin (m, n_t, h_p, b, h_c, j_0, j_1, j_2,..., J_ {f−1}) of the input scenario. The first time number n_t in which a value other than "" is entered is n_tstt, and the last time number n_t in which a value other than "0" is entered in any cell is n_tend.
Then, for each b (b = 0, 1,..., B−1), the effective partial array of TSD_in is shifted in the time direction by the following equation (16) and stored in TSD_Tin.
TSD_Tin (:, n_tst + Sft (b): n_tend + Sft (b),:, b,:,:,:,:, ..., :)
= TSD_in (:, n_ttt: n_tend,:,:,:,:,:, ..., :)
(16)
However, “:” is an operator that selects all the element numbers of the dimension of interest in the array.
Also, “k1: k2 (individually, k1 <k2)” is a parameter for selecting only element numbers of k1 to k2 among the element numbers of the dimension of interest in the array.

As a result, a partial array in which the element numbers n_ttt and n_tend of the n_t dimension in the spatio-temporal distribution TSD_Tin of the input scenario are selected, and the element numbers b of the dimension b of TSD_Tin are selected. From element number n_tstt + Sft (b) to n_tend + Sft (b), all element numbers of other dimensions can be copied to the selected partial array.
When only “0” is set as the time shift, it is equivalent to the case where the time shifter 9 is omitted from the movement information similarity determination device of FIG.

When the time shifter 9 shifts the spatiotemporal distribution related to the input scenario in the time direction, the spatiotemporal difference distribution generator 10 is generated by the spatiotemporal distribution related to the input scenario after the time shift and the spatiotemporal distribution generator 8. A spatiotemporal difference distribution that is a difference distribution from the spatiotemporal distribution related to the holding scenario is generated.
Here, the array of the spatiotemporal difference distribution is represented by DSD (m_in, m_ho, n_t, h_p, b, h_c, j_0, j_1, j_2,..., J_ {f−1}).
However, m_in is an input scenario number (m_in = 0, 1,..., M_in−1), m_ho is a holding scenario number (m_ho = 0, 1,..., M_ho−1), M_in is the number of input scenarios, and M_ho is a holding scenario. Is a number.

Based on this, by changing the element number h_p (h_p = 0, 1,..., H_p−1) related to the spatial shift and the element number b (b = 0, 1,..., B−1) related to the time shift, respectively. The spatio-temporal difference distribution generator 10 obtains the spatio-temporal difference distribution DSD by the following formula (17) for the combinations of B · H_p types.
DSD (m_in, m_ho,:, h_p, b,:,:,:,:, ..., :)
= TSD_Tin (m_in,:, h_p, b,:,:,:,:, ..., :)
-TSD_ho (m_ho,:, 0,:,:,:,:, ..., :)
(17)

Next, the spatial filter setting unit 11 sets a spatial filter that is a weight distribution having a spread in the spatial axis direction.
When the spatial filter setting unit 11 sets a spatial filter, the spatial filter convolution unit 12 convolves the spatial filter with the spatio-temporal difference distribution generated by the spatio-temporal difference distribution generator 10, and the spatio-temporal difference distribution after the spatial filter convolution is performed. Is output to the similarity evaluation value calculator 13.
Hereinafter, the processing contents of the spatial filter convolution unit 12 and the similarity evaluation value calculator 13 will be described in detail with reference to FIG.

Here, for simplification of explanation, a one-dimensional array is considered as a spatial distribution.
It is assumed that the spatial distribution regarding the input scenario is Gin (k) (k = 0, 1,..., K−1), and this spatial distribution Gin (k) is filtered (see A in FIG. 7).
However, as Gin, as shown in FIG. 7, a simple spatial distribution is considered in which “1” is set for one element number k — 0 and “0” is set for other element numbers.
On the other hand, three types of spatial distributions Gho — 0 (k), Gho — 1 (k), and Gho — 2 (k) are prepared as spatial distributions related to the holding scenario.
Among these spatial distributions, Gho_0 (k) is a spatial distribution in which k_0 is “1” and other than this is “0”, and has the same form as Gin (k).
Gho_1 (k) is a spatial distribution that is “1” for k_1 different from k_0 and “0” otherwise.
Gho_2 (k) is a spatial distribution that does not coincide with k_0 and is “1” at k_2 farther from k_0 than k_1, and “0” otherwise.

Here, it is assumed that the spatiotemporal difference distribution Dg_m generated by the spatiotemporal difference distribution generator 10 is obtained from the following equation (18).

As apparent from Expression (18), if the spatio-temporal distribution related to the input scenario matches the spatio-temporal distribution related to the holding scenario, the spatio-temporal difference distribution Dg_m becomes “0”. The difference distribution Dg_m is “2”.
This spatio-temporal difference distribution Dg_m is shown in the calculation result example of the similar evaluation value when it is assumed that the spatial filter convolution unit 12 does not perform the spatial filter convolution (see D in FIG. 7).
In FIG. 7D, the horizontal axis represents k_m when the spatial distribution related to the holding scenario is limited to only a distribution that is “1” with only one element number k_m and “0” otherwise. The vertical axis represents the value of Dg_m.

The spatial distribution of each retention scenario corresponds to the translation of the spatial distribution of the input scenario, so that the above-mentioned spatio-temporal difference distribution Dg_m has a parallel translation amount of “0” (the input scenario and the retention scenario Except for the case where the spatial distributions match, the value is constant regardless of the amount of parallel movement.
However, in general, there are many applications where it is desired to determine that the difference between the two increases as the amount of parallel movement increases (determines that they are not similar).

Therefore, in the first embodiment, the spatial filter waveform as shown in FIG. 7B is convolved with the original spatial distribution Gin (k) or Gho_m (k), and the spatial filter as shown in C of FIG. The spatial distribution Cin (k) and Cho_m (k) (m = 0, 1, 2) after convolution are obtained.
Using the spatial distributions Cin (k) and Cho_m (k) after the spatial filter convolution, the spatiotemporal difference distribution Dc_m is obtained by the following equation (19).

The spatio-temporal difference distribution Dc_m is shown in the calculation result example of the similarity evaluation value when the spatial filter convolution is performed by the spatial filter convolution unit 12 (see E in FIG. 7).
When the spatial filter setting unit 11 appropriately sets the shape of the spatial filter, as shown in FIG. 7, it is possible to obtain a tendency that the difference increases as the parallel movement amount increases.
Further, other tendencies can be obtained by changing the shape of the spatial filter.
As the shape of the spatial filter set by the spatial filter setting unit 11, for example, from a general shape such as a Gaussian shape, a triangular shape, a rectangular shape, a Hanning weight, a Hanning weight, etc., to a special one set according to some request. Can be used.
Here, for simplification, the spatial distribution is shown as a one-dimensional array and the spatial filter is also shown as a one-dimensional distribution. However, the dimensions can be two or more.

FIG. 8 shows an example of the spatial filter set by the spatial filter setting unit 11.
FIG. 8A shows an example of a filter at the time of three-dimensional space projection, and FIG. 8B shows an example of a filter at the time of two-dimensional plane projection.
FIG. 8C shows an example of a filter at the time of one-dimensional axis projection, and FIG. 8D shows an example of a filter at the time of azimuth axis projection.
What is necessary is just to select suitably from these spatial filters according to the dimension F of the space of a spatio-temporal vector.

In general, the convolution operation of the three-dimensional functions f (tx, ty, tz) and f (tx, ty, tz) can be expressed by the following equation.

The shape of the spatial filter is represented as Wflt (ij — 0, ij — 1,..., Ij_ {F−1}). However, Wflt satisfies the following equation.

As a special example of the weight filter, the value of one cell including the origin is set to “1”, and other values are set to “0”.
Using such weights is equivalent to omitting the spatial filter setting unit 11 and the spatial filter convolution unit 12 from the movement information similarity determination apparatus of FIG.

Here, although what was directly calculated from Formula (20) was shown, it can also be obtained using the nature of convolution.
That is, the multidimensional Fourier transform of the spatiotemporal difference distribution f (tx, ty, tz) generated by the spatiotemporal difference distribution generator 10 and the spatial filter g (tx, ty, tz) set by the spatial filter setting unit 11 ) Multi-dimensional Fourier transform.
Then, the inverse Fourier transform of the product of both multidimensional Fourier transform results is performed, and the inverse Fourier transform result is obtained as a spatio-temporal difference distribution after the spatial filter convolution.

In FIG. 7, the difference is calculated after convolving the spatial filter into the spatial distributions Gin (k) and Gho_m (k) (in the absolute value of the equation (19)).
However, due to the linearity of the convolution operation, it is equivalent to convolve the spatial filter after taking the difference first.
In the first embodiment, based on this property, after the difference is calculated by the spatiotemporal difference distribution generator 10, the subsequent spatial filter convolution unit 12 convolves the spatial filter.
Specifically, for each m_in, m_ho, n_t, h_p, b, h_c of DSD (m_in, m_ho, n_t, h_p, b, h_c, j_0, j_1, j_2, ..., j_ {f-1}), J_0, j_1, j_2,..., j_ {f-1} is convolved with the spatial distribution of dimensions, and the result is arrayed DSD_flt (m_in, m_ho, n_t, h_p, b, h_c, j_0, j_1, j_2,..., j_ {F-1}).

ただし、Ｄｆｌｔ０（ｍ＿ｉｎ，ｍ＿ｈｏ，ｈ＿ｐ， ｂ）は下記の式（２３）により得られる。

When the spatial filter convolution unit 12 convolves the spatial filter with the spatiotemporal difference distribution, the similarity evaluation value calculator 13 calculates a similarity evaluation value between the input scenario and the holding scenario from the spatiotemporal difference distribution after the spatial filter convolution.
That is, the similarity evaluation value calculator 13 calculates the similarity evaluation value Dflt (m_in, m_ho) between the m_in input scenario and the m_ho holding scenario by the following equation (22).

However, Dflt0 (m_in, m_ho, h_p, b) is obtained by the following equation (23).

Here, as expressed in Expression (22), the minimum value when h_p related to the spatial shift and b related to the time shift are applied is output as the similarity evaluation value.
The similarity evaluation value calculated by the similarity evaluation value calculator 13 is accumulated in the similarity evaluation value group accumulator 14.

Next, the discrimination criterion setting unit 15 sets a discrimination criterion for a holding scenario similar to the input scenario.
For example, the following discrimination criteria are set.
(A) Discrimination criterion for selecting the holding scenario with the smallest similarity evaluation value (b) Setting the threshold Th, and the determination criterion for selecting the holding scenario having the similarity evaluation value of the threshold Th

The similar scenario discriminator 16 reads the similarity evaluation value between the input scenario and the holding scenario from the similar evaluation value group accumulator 14 and acquires the discrimination criterion for the holding scenario from the discrimination criterion setting unit 15.
Next, the similar scenario discriminator 16 collates the similarity evaluation value with the discrimination criterion to discriminate whether or not the input scenario and the holding scenario are similar. If they are similar, the holding scenario group reader The stored scenario is acquired from 5, and the stored scenario is stored in the similar scenario accumulator 17.
Hereinafter, the processing content of the similar scenario discriminator 16 will be specifically described.

When the discrimination criterion (a) is used, the similar scenario discriminator 16 reads out the similar evaluation values Dflt (m_in, m_ho) stored in the similar evaluation value group accumulator 14, and those similar evaluation values Dflt (m_in) , M_ho), a holding scenario m_ho corresponding to the minimum similarity evaluation value is determined for each input scenario m_in.
When the similar scenario discriminator 16 determines the holding scenario m_ho corresponding to the minimum similarity evaluation value for each input scenario m_in, the similar scenario discriminator 16 acquires the holding scenario m_ho from the holding scenario group reader 5 and makes the holding scenario m_ho similar. Store in the scenario accumulator 17.

When the discrimination criterion (b) is used, the similar scenario discriminator 16 reads out the similar evaluation values Dflt (m_in, m_ho) stored in the similar evaluation value group accumulator 14, and those similar evaluation values Dflt (m_in) , M_ho), a retention scenario m_ho corresponding to a similarity evaluation value equal to or less than the threshold Th is selected for each input scenario m_in.
When the similar scenario discriminator 16 selects the holding scenario m_ho corresponding to the similar evaluation value equal to or less than the threshold Th for each input scenario m_in, the similar scenario discriminator 16 acquires the holding scenario m_ho from the holding scenario group reader 5 and determines the holding scenario m_ho. Stored in the similar scenario accumulator 17.
However, when the discrimination criterion (b) is used, a plurality of similar holding scenarios may be selected.

  As is apparent from the above, according to the first embodiment, the newly input scenario and the previously held scenario are stored in the projection space set by the projection space setting unit 6 at each scenario time. The spatiotemporal distribution generators 7 and 8 that project and generate a spatiotemporal distribution indicating the spatial distribution of the target number for each scenario time, and the spatiotemporal distribution related to the input scenario generated by the spatiotemporal distribution generators 7 and 8 And a spatio-temporal difference distribution generator 10 that generates a spatio-temporal difference distribution indicating a difference between spatio-temporal distributions related to the holding scenario, and between the input scenario and the holding scenario based on the spatio-temporal difference distribution generated by the spatio-temporal difference distribution generator 10. The similarity evaluation value calculator 13 for calculating the similarity evaluation value is provided, and the similarity scenario discriminator 16 refers to the similarity evaluation value calculated by the similarity evaluation value calculator 13 to determine the input scenario and the holding scenario. Since it is configured to determine whether or not they are similar, it is possible to determine whether or not the input scenario and the retention scenario are similar even if the input scenario or the retention scenario includes two or more targets. There is an effect that can.

  Further, according to the first embodiment, the projective space setting unit 6 determines an orthogonal three-dimensional space from the origin position vector and unit vectors representing the directions of three axes orthogonal to each other, and the orthogonal three-dimensional space. Is set in the projection space, so that the frequency distribution of the target number for each scenario time can be easily generated.

  Further, according to the first embodiment, the projective space setting unit 6 determines an orthogonal three-dimensional space from the origin position vector and unit vectors representing the directions of three axes orthogonal to each other, and the orthogonal three-dimensional space. The two-dimensional space defined by two axes among the three axes that constitute, or the one-dimensional space defined by one axis among the three axes that constitute the orthogonal three-dimensional space is set as the projection space. Therefore, the frequency distribution of the target number for each scenario time can be easily generated, and the processing load can be reduced.

  Further, according to the first embodiment, the projective space setting unit 6 is a two-dimensional space with two axes of azimuth and elevation in an orthogonal three-dimensional space, or a one-dimensional space with azimuth or elevation as one axis. Is set in the projection space, the frequency distribution of the target number for each scenario time can be easily generated, and the processing load can be reduced. Furthermore, there is an effect capable of dealing with a case where the frequency distribution for each direction around a certain position becomes a similar index.

  Further, according to the first embodiment, the projective space setting unit 6 determines a plurality of orthogonal three-dimensional spaces from the origin position vector and unit vectors representing the directions of three axes orthogonal to each other, and a plurality of orthogonal Since the three-dimensional space is set as the projection space, the frequency distribution of the target number for each scenario time can be easily generated, and the processing load can be reduced. Moreover, there exists an effect which can raise the similar determination precision further.

  According to the first embodiment, the spatial shift method group setting unit 2 that sets the spatial shift amount of the input scenario, and the input scenario is spatially shifted by the shift amount set by the spatial shift method group setting unit 2. In addition, since the spatial shifter 3 that outputs the shifted input scenario to the spatio-temporal distribution generator 7 is provided, it is possible to correct the overall positional deviation and orientation deviation between the input scenario and the holding scenario. There exists an effect which can distinguish similarity.

  Further, according to the first embodiment, a temporal shift amount with respect to the spatiotemporal distribution related to the input scenario generated by the spatiotemporal distribution generator 7 is set, and the spatiotemporal distribution is temporally converted by the shift amount. Since it is configured to provide a time shifter 9 that shifts and outputs the spatiotemporal distribution after the shift to the spatiotemporal difference distribution generator 10, similarity is determined while correcting the time lag between the input scenario and the holding scenario. The effect which can be done is produced.

  Further, according to the first embodiment, the spatial filter setting unit 11 that sets a spatial filter that is a distribution of weights spreading in the spatial axis direction, and the spatial filter that is set by the spatial filter setting unit 11 are converted into a spatio-temporal difference. Since the spatial filter convolution unit 12 that convolves the spatiotemporal difference distribution generated by the distribution generator 10 and outputs the spatiotemporal difference distribution after the spatial filter convolution to the similarity evaluation value calculator 13 is provided, the spatial filter By adjusting the shape, it becomes possible to set the tolerance of misalignment for each target included in the movement scenario, and the characteristics of the change in the degree of similarity to the magnitude of misalignment. There is an effect that the degree of freedom and accuracy of similarity determination can be increased.

  Further, according to the first embodiment, the spatio-temporal distribution generators 7 and 8 each have a valid flag indicating whether or not the position is valid information for each target position at each scenario time. Is included in the movement scenario because only the position of the target indicating that the validity flag is valid information is projected onto the projection space set by the projection space setting unit 6. Among the target positions at each scenario time, the similarity can be determined only for the target position of the part of interest.

  Furthermore, according to the first embodiment, the spatial shift method group setting unit 2 sets a column vector indicating the parallel movement of the input scenario as the spatial shift amount of the input scenario, Prepare a column vector indicating the position, a column vector along the rotation axis, and a rotation angle around the rotation axis, and rotate the input scenario from the column vector along the rotation axis and the rotation angle around the rotation axis. Rotation matrix 3 is set, and space shifter 3 divides the column vector indicating the position of the point on the rotation axis from the column vector indicating the position of a plurality of targets at each scenario time, and rotates to the column vector after the division Multiply the matrix from the left, add the column vector indicating the position of the point on the rotation axis to the column vector after multiplication, and add the column vector indicating the parallel movement to calculate the position vector after the spatial shift Since it is configured to so that an effect that can determine the similarity while correcting the overall positional deviation and direction of the deviation between the input scenarios and holding scenario.

  According to the first embodiment, the spatial filter convolution unit 12 includes a multi-dimensional Fourier transform of the spatio-temporal difference distribution generated by the spatio-temporal difference distribution generator 10 and the spatial filter set by the spatial filter setting unit 11. Multi-dimensional Fourier transform is performed, inverse Fourier transform of the product of both multi-dimensional Fourier transform results is performed, and the inverse Fourier transform result is transmitted to the similar evaluation value calculator 13 as a spatio-temporal difference distribution after spatial filter convolution. Since it was comprised so that it might output, there exists an effect which can raise calculation speed.

It is a block diagram which shows the movement information similarity determination apparatus by Embodiment 1 of this invention. It is explanatory drawing which shows the parallel movement of the input scenario in the space shifter. It is explanatory drawing which shows the rotational movement of the input scenario in the space shifter. It is explanatory drawing which shows the spatiotemporal distribution which is the projection space set by the projection space setting device 6, and the spatial distribution of the target number for every scenario time. It is a flowchart which shows the processing content of the projection space setting device 6 and the spatiotemporal distribution generators 7 and 8. FIG. It is explanatory drawing which shows the processing content of the time shifter. It is a principle figure explaining the processing content of the spatial filter convolution device 12 and the similar evaluation value calculation device 13. It is explanatory drawing which shows an example of the spatial filter set by the spatial filter setting device.

Explanation of symbols

  1 input scenario group accumulator, 2 space shift method group setter (space shift amount setting means), 3 space shifter (space shift means), 4 holding scenario group accumulator, 5 holding scenario group reader, 6 projective space setting (Projection space setting means), 7, 8 spatiotemporal distribution generator (spatiotemporal distribution generation means), 9 time shifter (time shift means), 10 spatiotemporal difference distribution generator (spatiotemporal difference distribution generation means), DESCRIPTION OF SYMBOLS 11 Spatial filter setting device (spatial filter setting means), 12 Spatial filter convolution device (spatial filter convolution means), 13 Similar evaluation value calculator (similar evaluation value calculation means), 14 Similar evaluation value group accumulator, 15 Discrimination standard setting (Similar scenario discriminator), 16 similar scenario discriminator (similar scenario discriminator), 17 similar scenario accumulator.

Claims (12)

  Projection space setting means for setting a projection space for projecting a movement scenario indicating the positions of a plurality of targets at each preset scenario time at each scenario time, and an input that is a newly input movement scenario at each scenario time A space-time that generates a spatio-temporal distribution indicating a spatial distribution of a target number for each scenario time by projecting a scenario and a retained scenario that is a movement scenario that is retained in advance to the projection space set by the projection space setting means. A distribution generation means, a spatiotemporal difference distribution generation means for generating a spatiotemporal difference distribution indicating a difference between the spatiotemporal distribution related to the input scenario generated by the spatiotemporal distribution generation means and the spatiotemporal distribution related to the holding scenario; Similarity evaluation value for calculating the similarity evaluation value between the input scenario and the holding scenario from the spatiotemporal difference distribution generated by the spatiotemporal difference distribution generation means Similar to the movement information having the output means and the similar scenario determination means for determining whether or not the input scenario and the retained scenario are similar with reference to the similarity evaluation value calculated by the similarity evaluation value calculation means Degree determination device.   The projective space setting means determines an orthogonal three-dimensional space from an origin position vector and unit vectors representing directions of three axes orthogonal to each other, and sets the orthogonal three-dimensional space as a projection space. Item 2. The movement information similarity determination device according to Item 1.   The projective space setting means defines an orthogonal three-dimensional space from the origin position vector and unit vectors representing the directions of three axes orthogonal to each other, and is defined by two of the three axes constituting the orthogonal three-dimensional space. 2. The movement information similarity determination device according to claim 1, wherein a one-dimensional space defined by one of the three axes constituting the orthogonal three-dimensional space is set as a projection space. .   The projective space setting means sets a two-dimensional space having two axes of an azimuth angle and an elevation angle in an orthogonal three-dimensional space, or a one-dimensional space having the azimuth angle or the elevation angle as one axis as a projection space. The movement information similarity determination apparatus according to claim 3.   Projection space setting means defines a plurality of orthogonal three-dimensional spaces from a position vector of the origin and unit vectors representing directions of three axes orthogonal to each other, and sets the plurality of orthogonal three-dimensional spaces as projection spaces. The movement information similarity determination device according to claim 1.   Spatial shift amount setting means for setting the spatial shift amount of the input scenario, and the input scenario is spatially shifted by the shift amount set by the spatial shift amount setting means, and a spatio-temporal distribution is generated for the input scenario after the shift. The movement information similarity determination apparatus according to any one of claims 1 to 5, further comprising space shift means for outputting to the means.   A temporal shift amount is set with respect to the spatiotemporal distribution related to the input scenario generated by the spatiotemporal distribution generation means, the spatiotemporal distribution is temporally shifted by the shift amount, and the shifted spatiotemporal distribution is changed to the time. The movement information similarity determination apparatus according to any one of claims 1 to 6, further comprising time shift means for outputting to the spatial difference distribution generation means.   Spatial filter setting means for setting a spatial filter that is a weight distribution having a spread in the spatial axis direction, and the spatial filter set by the spatial filter setting means to the spatio-temporal difference distribution generated by the spatio-temporal difference distribution generating means The movement according to any one of claims 1 to 7, further comprising: a spatial filter convolution unit that outputs a space-time difference distribution after the convolution and the spatial filter convolution to the similarity evaluation value calculation unit. Information similarity determination device.   The spatio-temporal distribution generation means, for each target position at each scenario time, if the flag indicating whether the position is valid information is included in the movement scenario, the flag is valid information 9. The movement information similarity determination device according to claim 1, wherein only a target position indicating the effect is projected onto a projection space set by the projection space setting means. .   The spatial shift amount setting means sets a column vector indicating the parallel movement of the input scenario as a spatial shift amount of the input scenario, a column vector indicating the position of the point on the rotation axis, and a direction along the rotation axis. And a rotation matrix that indicates the rotation of the input scenario from the column vector in the direction along the rotation axis and the rotation angle around the rotation axis. The shift means divides a column vector indicating the position of the point on the rotation axis from a column vector indicating the position of a plurality of targets at each scenario time, and multiplies the column matrix after division from the left by the rotation matrix, The column vector indicating the position of the point on the rotation axis is added to the column vector after multiplication, and the column vector indicating the parallel movement is added to calculate the position vector after the space shift. Movement information similarity determination apparatus according to any one of claims 9 claim 6, wherein.   The spatial filter convolution means performs the multidimensional Fourier transform of the spatiotemporal difference distribution generated by the spatiotemporal difference distribution generation means and the multidimensional Fourier transform of the spatial filter set by the spatial filter setting means, 9. The inverse Fourier transform of the product of the multidimensional Fourier transform results is performed, and the inverse Fourier transform results are output to the similarity evaluation value calculation means as a spatiotemporal difference distribution after spatial filter convolution. The movement information similarity determination device according to any one of 10.   The similar scenario discriminating means includes a discriminant setting unit for setting a discriminant criterion for holding scenarios similar to the input scenario, a discriminant set by the discriminant criterion setting unit, and a similarity calculated by the similarity evaluation value calculating unit. 12. A similar scenario discriminator that collates an evaluation value and discriminates whether or not the input scenario and the retained scenario are similar to each other. The movement information similarity determination apparatus according to any one of the preceding claims.

Images