JP4167469B2 - Motion vector detection device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a motion vector detection device that detects the amount and direction of movement of a moving object in a video signal.
[0002]
[Prior art]
A technique called a block matching method is known as a method for detecting a motion vector. Such a technique is applied to an image encoding technique such as MPEG and put into practical use.
[0003]
However, since this block matching method divides the screen into a plurality of blocks and detects a motion vector for each block, there is a problem that consistency of detection results between blocks may be disturbed.
[0004]
In contrast to the method of detecting a motion vector for each block as in the block matching method, there is a method called a gradient method as a method of detecting a motion vector for each pixel. The principle of the gradient method is described in detail in Patent Document 1, Non-Patent Document 1, and the like, but will be briefly described below.
[0005]
First, assume that an image g (x, y) at t = 0 (t is time) has moved to t = n by α = (a, b) and becomes g (x−a, y−b). .
[0006]
If this function g (xa, yb) is Taylor-expanded and only its first order term is taken,
g (x−a, y−b) = g (x, y) −a · ∂ / ∂x · g (x, y) −b · ∂ / ∂y · g (x, y) (1) )
It becomes.
[0007]
If the inter-frame difference signal is d (x, y) = g (x−a, y−b) −g (x, y),
a · ∂ / ∂x · g (x, y) + b · ∂ / ∂y · g (x, y) = − d (x, y) (2)
It becomes.
[0008]
Here, if i and j are basic unit vectors in the x and y directions,
(I · a + j · b) · (i · ∂ / ∂x + j · ∂ / ∂y) · g (x, y) = − d (x, y) (3)
And rewriting this,
α · gradg (x, y) = − d (x, y) (4)
It becomes. From this fourth equation, the motion vector can be obtained if the gradient of the image and the frame time are known.
[0009]
[Patent Document 1]
JP 60-158786 A [Non-patent Document 1]
IEICE Technical Report No. IE78-67 “Measurement of moving amount and speed of moving object by image signal”
[0010]
[Problems to be solved by the invention]
Since the gradient method detects motion for each pixel, there is no problem caused by detecting a motion vector for each block unlike the block matching method.
[0011]
However, since the gradient method uses only the first-order term of the Taylor-expanded polynomial in its principle, there is a problem that the error of the detection result becomes large when the motion is intense.
[0012]
The gradient method detects a motion vector from the gradient of the screen, and this gradient is caused by the accumulation effect and afterimage effect of the camera. However, in recent years, CCDs are often used as imaging means of cameras, and such CCDs have very little afterimages and the accumulation effect is small. Therefore, when a motion vector is detected by applying a gradient method to an image photographed by a camera using a CCD as an imaging means, there is a problem that it is difficult to detect the motion vector with high accuracy.
[0013]
SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and an object of the present invention is to provide a motion vector detection device that solves this problem.
[0014]
[Means for Solving the Problems]
In order to solve the above problems, the present invention has taken the following measures.
[0016]
That is, the motion vector detection apparatus according to the present invention includes a recursive filter, a subtracting unit that calculates a difference between an input signal to the recursive filter and an output signal from the recursive filter, and a differential signal for the subtracted signal. Differentiating means, conversion means for converting the amplitude of the differentiated signal into a motion amount, and sign detection means for detecting the sign of the differentiated signal.
[0017]
The motion vector detection device according to the present invention includes a recursive filter, subtracting means for calculating a difference between an input signal to the recursive filter and an output signal from the recursive filter, and an output of the subtracted signal. Normalizing means for adjusting the signal to a constant amplitude level, differentiating means for differentiating the normalized signal, conversion means for converting the amplitude of the differentiated signal into an amount of motion, and the sign of the differentiated signal It is preferable to comprise a code detecting means for detecting.
[0018]
Furthermore, the motion vector detection device according to the present invention includes a recursive filter, subtracting means for calculating a difference between an input signal to the recursive filter and an output signal from the recursive filter, and the subtracted signal to the subtracted signal. Correction means for multiplying the inverse function of the transfer function of the recursive filter, normalization means for adjusting the output of the correction means to a constant amplitude level, differentiation means for differentiating the normalized signal, and this differentiated It is preferable to include conversion means for converting the amplitude of the signal into the amount of motion and sign detection means for detecting the sign of the differentiated signal.
[0019]
Here, it is preferable that the normalization means includes a peak detection means for detecting a peak level of the input signal, and a division means for dividing the input signal by the detected peak level.
[0020]
Moreover, it is preferable that the said conversion means calculates the reciprocal number of the amplitude level of the differentiated signal.
[0021]
DETAILED DESCRIPTION OF THE INVENTION
First, the principle of the present invention will be described.
[0022]
FIG. 1 is a block diagram of a well-known recursive (IIR) filter. In FIG. 1, the video signal input from the input terminal 101 is multiplied by a predetermined coefficient (K) by a first coefficient multiplier 102. The video signal multiplied by the coefficient (K) is delayed by a predetermined time in the frame memory 105 and further added by the adder circuit 103 with the signal multiplied by the predetermined coefficient (1-K) by the second coefficient multiplier 104. Is done. The added video signal is input to the frame memory 105 and output from the output terminal 106. Note that the value of the coefficient K is normally set to about 0.8.
[0023]
FIG. 2 shows input / output waveforms of such a cyclic filter. In the figure, the horizontal axis represents the position on the screen, and the vertical axis represents the amplitude level although omitted.
[0024]
FIG. 2A shows an input signal to the input terminal 101 in FIG. In FIG. 2A, reference numeral 201 denotes a moving object. A state is shown in which the moving object 201 at the position P (T0) at the time T0 is moving to the position P (Tn) at the time Tn.
[0025]
FIG. 2B shows a signal waveform output from the output terminal 106 of FIG. 1 when such a signal is input.
[0026]
As shown in FIG. 6B, the output waveform is accompanied by a tailing waveform such as 202 because the data from time T0 to Tn is cumulatively added while being sequentially multiplied by the coefficient. Here, the tailing waveform 202 is actually a waveform represented by an exponential function. However, for the sake of simplicity, it is assumed that the tailing waveform 202 is approximated by a straight line as shown in FIG.
[0027]
Next, output waveforms when a fast-moving moving object and a slow-moving moving object are input to the cyclic filter of FIG. 1 are compared.
[0028]
FIG. 3 shows this state. In the figure, the horizontal axis represents the position on the screen, and the vertical axis represents the amplitude level although omitted.
[0029]
FIG. 4A shows an input waveform to the cyclic filter of FIG. 1, and FIG. A waveform 301 with a black circle represents a slow-moving animal body, and a waveform 302 with a white circle represents a fast-moving animal body. That is, the slow-moving moving object 301 moves from the position Ps (T0) to Ps (Tn) between the times T0 and Tn, and the fast-moving moving object 302 moves between the positions T0 and Tn at the position Pf ( It represents that the movement from T0) to Pf (Tn).
[0030]
Here, when focusing on the angle 303 formed by the tail waveform 304 of the output waveform shown in FIG. 5B with the horizontal axis, the angle increases in the slow motion, and the angle decreases in the fast motion. Understandable. That is, it can be seen that the angle 303 formed by the trailing waveform 304 with the horizontal axis is inversely proportional to the speed of movement of the moving object. The present invention intends to detect a motion vector using this relationship. Therefore, unlike the conventional motion vector detection device based on the gradient method, it is possible to detect a motion vector without depending on the camera characteristics (afterimage characteristics / accumulation effect) and the like.
[0031]
Hereinafter, an embodiment of a motion vector detection device according to the present invention will be described.
[0032]
FIG. 4 is a block diagram showing a first embodiment of a motion vector detection device according to the present invention.
[0033]
First, the flow of signals in this block diagram will be described. A video signal input from the input terminal 401 is input to the recursive filter 402 shown in FIG. The difference between the input signal and the output signal of the recursive filter 402 is calculated by the subtraction circuit 403, and the amplitude is adjusted by the normalization circuit 404 that follows the result. The amplitude adjusted signal is differentiated by the differentiation circuit 405, and the resulting amplitude is detected by the amplitude detection circuit 406. The amplitude value is converted by the movement amount conversion circuit 407 and output from the output terminal 409 as the movement amount of the moving object. On the other hand, the sign resulting from the differential operation is detected by the sign detection circuit 410, and this sign is output from the output terminal 410 as the moving direction of the moving object.
[0034]
With the configuration of this block diagram, the moving amount and moving direction of the moving object are obtained, and the motion vector is detected.
[0035]
Next, the operation of this block diagram will be described.
[0036]
FIG. 5 is a waveform diagram at each point in the block diagram of FIG. In the figure, the horizontal axis represents the position on the screen, and the vertical axis represents the amplitude level although omitted. Reference numeral 501 denotes a moving object, 502 a stationary object, and 503 a tail waveform.
[0037]
Now, it is assumed that a moving object 501 moving from a position P (T0) to P (Tn) and a stationary object 502 are input as input signals between times T0 and Tn. In this case, at point A in FIG. 4, a signal in which the trailing waveform 503 is generated in the moving object 501 is observed. Here, when the signal at point A and the input signal are subtracted, a signal like point B is obtained. That is, the moving object 501 itself and the stationary object 503 are deleted, and only the trailing waveform 503 remains.
[0038]
By the way, this waveform is next input to the normalization circuit 404. Here, for simplicity of explanation, the normalization circuit 404 is ignored and input to the differentiation circuit 405. The operation of the normalization circuit 404 will be described later.
[0039]
Therefore, next, it is necessary to detect the angle formed with the horizontal axis P of the tailing waveform 503. This angle is the inclination angle of the tailing waveform. I know it ’s good. In other words, when the slope of the trailing waveform 503 is a and y = ax + b, the differentiation is dy / dx = a.
[0040]
The amplitude of the differentiated signal at point D is a value that is inversely proportional to the amount of motion. That is, since the amplitude of the signal at point D is a value that is inversely proportional to the amount of motion, it must be converted to a proportional value. This conversion is performed by the motion amount conversion circuit 407 in FIG. The conversion method may be a method of calculating the reciprocal number, calculating the reciprocal number and multiplying by an appropriate coefficient, or preparing a conversion table.
[0041]
FIG. 6 shows the above operation in comparison with a case where a fast-moving moving object and a slow-moving moving object are input. In the figure, the horizontal axis represents the position on the screen, and the vertical axis represents the amplitude level although omitted. FIG. 4A shows a case where a fast-moving moving object is input, and FIG. 5B shows a case where a slow-moving moving object is input. In this way, there is a difference in the amplitude level of the waveform of point D, that is, the differentially calculated waveform, between the fast-moving moving object and the slow-moving moving object, and this difference gives the difference in moving speed of the moving object. Can be detected.
[0042]
Next, the operation of the code detection circuit in the block diagram of FIG. 1 will be described.
[0043]
FIG. 7 shows a waveform diagram at each point in the block diagram of FIG. In the figure, the horizontal axis represents the position on the screen, and the vertical axis represents the amplitude level although omitted. FIG. 4A shows a case where the moving object is moving from left to right, and FIG. 4B shows a case where the moving object is moving from right to left. As shown in the figure, the inclination of the tailing waveform is inverted depending on the moving direction of the moving object, and as a result, the sign of the differential value (point D) is inverted. That is, when the moving object is moving from left to right (a), the sign of the differential value (point D) is positive, but the moving object is moving from right to left (b) ), The sign of the differential value (D point) is negative.
[0044]
Therefore, if the sign of this differential value (D point) is detected, the moving direction of the moving object can be detected.
[0045]
Next, the normalization circuit 404 of FIG. 1 whose description has been omitted will be described.
[0046]
In the above description, all the amplitude levels of the moving object have been considered to be the same. However, in reality, it is normal that animals having various amplitude levels are mixed. In such a case, if the normalization circuit 404 is not provided, the following adverse effects occur.
[0047]
FIG. 8 is a diagram for explaining how the adverse effect occurs. In the figure, the horizontal axis represents the position on the screen, and the vertical axis represents the amplitude level although omitted.
[0048]
This figure shows an moving object that moves from position P (T0) to P (Tn) between times T0 and Tn as an input signal. FIG. 5B shows the amplitude level of moving object 802 (a ) Is smaller than the amplitude level of the moving object 802, and the moving speed of the moving object in (b) is set to be slower than the moving speed of the moving object in (a). In such a situation, the slope of the trailing waveform 803 at the points A and B is approximately the same, and the amplitude of the waveform at the point D, which is a differential value thereof, is approximately the same and no difference is produced. Therefore, in reality, although the moving body of (a) and the moving body of (b) have a difference in moving speed, the detected amount of movement is almost the same.
[0049]
Such an adverse effect is solved by aligning the amplitude level of the moving object to a constant value. Specifically, the amplitude level of the signal as shown in FIG. 9 may be adjusted. That is, the amplitude level may be increased or decreased so that the peak value of the waveform at point B becomes a predetermined specified value.
[0050]
Therefore, the circuit for aligning the amplitude level of the moving object to a constant value is the normalization circuit 404 in FIG. 1, and the configuration of this normalization circuit 404 is shown in FIG.
[0051]
In FIG. 10, the peak value (X) of the signal input from the input terminal 1001, that is, the signal (Y) at point B in FIG. The input signal (Y) is divided by this peak value (X), and the result is output from the output terminal 1004. With such a configuration, the output is output with the amplitude adjusted so that the peak value becomes “1”.
[0052]
Next, a second embodiment of the motion vector detection device according to the present invention will be described.
[0053]
In the first embodiment, since the tailing waveform output from the recursive filter has been considered to approximate a straight line, an error due to approximation occurs. Here, in order to detect a motion vector with higher accuracy, it is preferable to correct the trailing waveform to a straight line.
[0054]
The state of this correction is shown in FIG. In FIG. 11, reference numeral 1101 denotes a trailing waveform output from the recursive filter. By multiplying this trailing waveform 1101 by a waveform 1102 having an inverse function with this trailing waveform 1101, a linear trailing waveform 1103 is obtained. That is, since the tailing waveform 1101 is a waveform obtained from the transfer function of the recursive filter, the inverse function of the transfer function may be multiplied.
[0055]
FIG. 12 shows a block diagram of a second embodiment of the motion vector detection apparatus according to the present invention. The same components as those in FIG. 4 are denoted by the same reference numerals. In the present embodiment, a correction circuit 1201 is newly added between the subtraction circuit 403 and the normalization circuit 404. By performing the correction as shown in FIG. 11 by this correction circuit, it is possible to detect a motion vector with higher accuracy. In other words, unlike the conventional motion vector detection device based on the gradient method, it is possible to detect a motion vector with higher accuracy because it does not perform an approximate process such as using only a first-order term from a Taylor-expanded polynomial. .
[0056]
Even if the positions of the normalization circuit 404 and the correction circuit 1201 are switched, the same purpose can be achieved and the same effect can be obtained. In the above description, the description has been made in one dimension only for horizontal movement, but it is easy to expand this to two dimensions in the horizontal and vertical directions. In the above description, the signals have been treated as continuous signals, but in television signals and the like, the signals are sampled in the time direction. For this reason, although each waveform demonstrated above becomes discrete, the above theory is applicable also to a discrete signal.
[0057]
【The invention's effect】
According to the motion vector detection device of the present invention, it is possible to detect a motion vector with high accuracy with a relatively simple configuration.
[0058]
In addition, a motion vector can be detected with high accuracy even for a fast motion.
[0059]
Furthermore, the motion vector detection accuracy is not affected by the characteristics of the imaging device such as a camera.
[Brief description of the drawings]
FIG. 1 is a block diagram of a recursive filter.
FIG. 2 is an input / output waveform diagram of a recursive filter.
FIG. 3 is a principle diagram of a motion vector detection device according to the present invention.
FIG. 4 is a first embodiment of a motion vector detection device according to the present invention.
FIG. 5 is a waveform diagram of each part in the first embodiment of the motion vector detection apparatus according to the present invention;
FIG. 6 is a waveform diagram for explaining detection of a motion amount in the motion vector detection device according to the present invention.
FIG. 7 is a waveform diagram for explaining the detection of the direction of motion in the motion vector detection apparatus according to the present invention.
FIG. 8 is a waveform diagram for explaining the necessity of a normalization circuit in the motion vector detection device according to the present invention.
FIG. 9 is a waveform diagram for explaining the operation of a normalization circuit in the motion vector detection device according to the present invention.
FIG. 10 is a block diagram of a normalization circuit in the motion vector detection device according to the present invention.
FIG. 11 is a waveform diagram for explaining the operation of the correction circuit in the motion vector detection device according to the present invention.
FIG. 12 is a second embodiment of the motion vector detection device according to the present invention.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 101, 401, 1001 ... Input terminal, 102 ... 1st coefficient multiplier, 103 ... Adder circuit, 104 ... 2nd coefficient multiplier, 105 ... Frame memory, 106, 409 , 410, 1004... Output terminal, 201, 301, 302, 501, 801, 802 ... moving object, 202, 1101 ... tail waveform, 203, 304, 503, 803 ... linear approximation Tail waveform, 303... Angle between tail waveform and horizontal axis, 402... Cyclic filter, 403... Subtraction circuit, 404... Normalization circuit, 405. 406 ... Amplitude detection circuit, 407 ... Motion amount conversion circuit, 408 ... Signal detection circuit, 502 ... Still object, 1002 ... Peak detection circuit, 1003 ... Division circuit, 1102.・ Ohiki Inverse function waveform of the waveform, 1103 ... corrected tailing waveform, 1201 ... correction circuit

Claims (5)

A recursive filter,
Subtracting means for calculating a difference between an input signal to the cyclic filter and an output signal from the cyclic filter;
Differentiating means for differentiating the subtracted signal;
A conversion means for converting the amplitude of the differentiated signal into a motion amount;
Code detection means for detecting the sign of the differentiated signal;
A motion vector detection apparatus comprising: A recursive filter,
Subtracting means for calculating a difference between an input signal to the cyclic filter and an output signal from the cyclic filter;
Normalizing means for adjusting the output of the subtracted signal to a constant amplitude level;
Differentiating means for differentiating the normalized signal;
A conversion means for converting the amplitude of the differentiated signal into a motion amount;
Code detection means for detecting the sign of the differentiated signal;
A motion vector detection apparatus comprising: A recursive filter,
Subtracting means for calculating a difference between an input signal to the cyclic filter and an output signal from the cyclic filter;
Correction means for multiplying the subtracted signal by the inverse function of the transfer function of the cyclic filter;
Normalizing means for adjusting the output of the correcting means to a constant amplitude level;
Differentiating means for differentiating the normalized signal;
A conversion means for converting the amplitude of the differentiated signal into a motion amount;
Code detection means for detecting the sign of the differentiated signal;
A motion vector detection apparatus comprising: In the motion vector detection apparatus according to claim 3 or 4,
  The normalizing means includes
Peak detection means for detecting the peak level of the input signal;
Dividing means for dividing the input signal by the detected peak level;
A motion vector detection apparatus comprising: The motion vector detection device according to any one of claims 2 to 5,
  The said conversion means calculates the reciprocal number of the amplitude level of the differentiated signal, The motion vector detection apparatus characterized by the above-mentioned.

Images