DE3036770T5 - TECHNIQUE FOR ESTIMATION OF DISPLACEMENT AND / OR VELOCITY OF OBJECTS IN VIDEO SCENES - Google Patents

Description

BLUMBACH · WESER. BERGEN · KRAMER

PATENT LAWYERS IN MUNICH AND WIESBADEN

4-

Patentconsult Redeckestraße 43 8000 Munich 60 Telephone (089) 883603/883604 Telex 05-212313 Telegrams Palenlconsuli Patentconsult Sonnenberger Straße 43 6200 Wiesbaden Telelon (06121) 562943/561998 Telex 04-186237 Telegrams Palentconsuli

PCT / US 80/00178 NETRAVALI 10-1

Western Electric Company Incorporated New York, N.Y. 10038, USA

Method for estimating the displacement and / or speed of objects in video scenes

Background of the invention Field of invention

The invention relates to a device for generating an estimated value for the displacement of an object in a Video scene between an earlier and a current image with circuits for calculating a first estimate for the shift using the image intensity difference between at least one picture element in the current picture and a spatially corresponding picture element in the previous picture.

Description of the prior art

A speed and displacement estimate has been found in proven to be useful in many applications, namely a

MUndien: R. Kramer Dipl.-Ing. · W. Weser Dipl.-Phys. Or. Tor. nat. . E. Hoffmann ulpl.-lng. Wiesbaden: P. G. Blumbacli Dipl.-Ing. · P. Bergen Prof. Or. Jur. Dipl.-Ing., Pat.-Ass., Pnt.-Anw. until 1979. G. Zwirner Dipl.-Ing. Dipl.-W.-Ing.

130616/0026

finally of tracking and monitoring systems, in the evaluation of photographic recordings that have been made from satellites, and in the effective coding of moving image sequences for transmission and / or storage. A method for measuring the speed is described in connection with a prediction video encoder ia of DE-PS 20 62 922. There, the intensity _: "of a specific image location in a current image; (frame) is correlated using a sequence of correlators with the intensity of a sequence of shifted image locations in a previous image. The correlator output signals are then used as interframe intensity correspondence between other moving points in the picture. This attempted solution for the speed determination has proven to be inconvenient, since many multiplications were required for each moving picture element. For example, if the speed was in the range between 0 and 4 pels (= picture elements of Picture Element) per frame (Pef by Picture Elements Per Frame) with an accuracy of 0.25 Pef, a range of motion correlation coefficient is required for each of the 1089 possible 1 / ^ element shifts Accuracy or a greater total speed range would require even larger processing capacities.

A significant reduction in the number of calculations required in the known method is due to

130616/0026

-H-

a later rate estimation method has been made possible. This later procedure is based on the knowledge that the size of the image difference signal (intensity difference of a certain image point between an image and the next image) produced by a moving object is a function of its speed and also the size of the object and the number of details, which contains the object. In addition, since the size and details are related to the size of the element or line difference signal (difference the intensity between two specific points in the same image = frame) are related, which is determined by the Object is generated, an estimate of the size and details essentially independent of the Preserved speed by normalizing the image difference with the element or line difference.

Mathematical is the displacement calculation described by Limb and Murphy (for the horizontal or x-direction) given by:

I (FDIF) ₁ . Sgn (EDIF) ₁

Vx = i i; ; (1),

wherein fDIF the image or frame difference (Dif frame of interference), EDIF is the difference element and the summation is performed over those elements in the image is for the fDIF above a threshold value, thereby indicating a moving area. To calculate Vy (displacement in the vertical direction), a corresponding calculation is made

130616/0026

calculation using the line difference LDIF instead the image difference EDIF. The speed can then be based on suitable vector combinations of Vx, Vy and Vt "" be calculated.

Although this later method leads to fewer calculation processes than the previous method, it is based on the simplification of a Taylor series expansion that an-.

assumes that Vx was small. However, when Vx is large, the Taylor approximation becomes inaccurate, thereby reducing the displacement estimate is severely affected. In addition, this method sometimes gives unsatisfactory results in Scenes with several moving objects.

In view of the foregoing, the invention generally has for its object, means and a method for more precise Determination of the displacement and / or speed of moving objects in video scenes or the like. The process must adapt to images with areas of different movement and thus become reality so that it doesn't get too complicated or expensive.

According to the invention, the problem arises with a device for generating an estimated value for the displacement of an object in a video scene between an earlier and a current image by the fact that the device an update circuit for repeatedly updating the first displacement estimated value

130616/0026

303 ^ 770

Use of the intensity difference between the at least one picture element in the current image and a location in the previous image versus the current picture element by the first displacement estimate is shifted. -

Summary of the Invention ".

According to the invention, the displacement of an ob- ^: -

object in a video scene is repeatedly determined using the previous displacement estimate to update the current displacement estimate. The repetition can be carried out for each picture element, for a group of picture elements or for a complete field or a complete picture (frame). Indeed, in certain cases, estimates can be made repeatedly for a given picture element. The iterative or recursive type according to the invention is characterized by a feedback path between the output of the system (which indicates the displacement) and the system itself, such that the error between each successive displacement estimate and the actual displacement ideally approaches zero.

According to the invention, the accuracy of the displacement estimate becomes much improved. This in turn makes the distinction between moving objects and the allows uncovered background in the processing image and can lead to an improved encoder design.

130616/0026

The recursive methods of the present invention can be readily implemented in a number of embodiments realize that are not overly complicated or burdensome. ■ '

Brief description of the drawings

The invention is based on the following, in detail; detailed description in conjunction with the drawings. Show it:

1 shows the block diagram of a device for repeatedly calculating an estimated value for the horizontal - zontal component of a shift in a

Number of images according to the invention; Fig. 2 shows the spatial locations that are used for the device according to Fig. 1 used method involved are;

3 shows the block diagram of a further exemplary embodiment for a device for calculating the displacement according to the invention.

Detailed description

The derivation of equation (1) given above is important for understanding the present invention. Takes assume that each frame of video information is along a plurality of generally parallel, horizontal Lines is scanned and that the position of a point within the image either in the form of the usual right-angled Coordinates (x, y) or in the form of a two-dimensional

130616/0026

Vector χ can be expressed, then the intensity of a certain point in the image can be represented as
l (x, t) at time t and I (x, tT) for the previous one
Specify image, 'where "E is the time between two images.
When an object moves translationally within the image and for the moment an uncovered background in the
Movement area is ignored, then every point in the "current image has an intensity which corresponds to the intensity:" - / ■ of a certain (usually different) point in the: previous image. Accordingly, the following applies:

i (x, t) = Kx-D, t-τ) (2),

where D is the translation vector for the object during the interval T is between the pictures.

Traditionally, the image difference signal FDIF is given at the point χ by:

FDIF (x) = I (x, t) - I (x, t-T) (3a),

which indicates that FDIF is the intensity difference at a fixed point χ for a time interval T. the
however, the same difference can be given with spatial expressions:

FDIF = I (x, t) - I (x + D, t) (3b),

since the intensity of the point χ + D is equal to the time t
the intensity at the point χ in the previous figure (time tT) is assumed.

Equation (3b) can be written for small D with the aid of a Taylor series expansion in the area of χ write as:

130616/0026

FDIF (x) = -D Vl (x, t) + higher order terms for D

where V is the gradient with reference to χ and the upper index T of a vector indicates its transposition. When the translational Movement of the object over the entire movement area (with the exception of the uncovered background) is constant and the higher order terms for D are neglected can, then both sides of equation (4) can be Sum the entire range of motion and you get one

Estimated value D for the translational movement is known as follows:

ZFDIP (x). Prefix (EDIF (x))

, Σ IEDIF (X) I

XfDIF (x). Prefix (LDIP (x))

D =

("O if z = O
where sign (z) = J

. • | in the other case (6).

Equation (5) is identical to one of the equations given by Limb and Murphy and indicates that Vl is a vector of Element and line differences (EDIF or LDIF) can be if the intensity of the image is available in a discrete grid is how this is true when the intensity is sampled. The upper term in equation (5) is related refers to the shift in the horizontal direction, while the lower term refers to the vertical direction.

As stated above, the estimate for D is only valid as long as D is small. When D increases, the quality will the Taylor series expansion is bad. According to the

130616/0026

- AO -

Finding this is overcome by a linearization of the intensity function in the range of an initial estimated value of D. Accordingly, for the ith image, the displacement estimate - <-

* i
value D obtained by linearizing the intensity function around the displacement estimate for the previous ((ii) th) image '. This procedure results in the following recursion: '.

a ¹ = ρ ¹ - ¹ ₊ u ¹ (?), o:

ai1 ^A ii

where D is an initial guess for JD and U of the updated value (update value) of

D, which makes the Yfert more accurate, i.e. an estimate

^A i-1
by DD.

* i 1

The size DFD (x, D) is now defined that the moved Displaced Frame Difference (DFD) is called and analogous to that in equations (3a) and (3b) used value FDIF (x) is:

DFD (x, D ¹ " ¹ ) s I (x, t) - Kx-D ¹ ^ ¹ Jt-T) (8).

From equation (8) it can be seen that DFD is defined on the basis of two quantities:

(a) the spatial point x> at which the evaluation takes place, and (b) the shift D "" with which the evaluation is carried out. Obviously becomes in the case of a two-dimensional lattice

discrete samples an interpolation process is used to evaluate! (xD ¹ " ¹ , tT) for non-integer values of D ¹ " ¹ . According to the definition, the value DFD has the egg

C zero convergence

130616/0026

* i property that it converges to zero when D becomes the actual

union displacement D of the object converges. In addition, DFD = FDIF when D ¹ " ¹ = 0.

DFD can be expressed in a similar way as above: DFD (X ^ D ¹ "* ¹ ) = I (xjt) - I (x + DD ^1-1 Kt) ==

- (DD ¹ " ¹ ) ¹ Vl (x, t) + higher order terms (9).

If higher-order expressions are disregarded and approximations similar to those for equation (5) are used, an estimated value D - tr results, which in combination with equation (7) gives:

JDFD (X, D ¹ " ¹ ). Sign. (EDIF (x))

"i" I ^

D ^x - D ^J

ZDFD (x, D ¹ "" ¹ ). Sign (LDIF (x))

(10). Sign (LDIF (x))

X | ldif (x) |

In equation (10), the summations are performed over the entire range of motion.

The upper term in equation (10) relates to the displacement in the horizontal direction and the lower term to the Shift in the vertical direction. From equation (10), according to the present invention, each Displacement estimate is an update of the previous estimate and the previous estimate for the Used to carry out the update. The beginning is for

* i-1 * i

D = 0, DFD = FDIF. After that, D becomes using D as well as a correction term which is a function of (a) DFD and EDIF for the x-direction and of

130616/002 6

(b) DFD and LDIF for the y direction. In other words, the Update is a function of the previous displacement estimate used to compute a displaced:. Image difference is used, which is then normalized to local intensity fluctuations between the processed Element and neighboring picture elements to be taken into account.

A first embodiment of the present invention which relates to the expression in equation (10) above realized only on the shift in the x-direction, is shown in the form of a block diagram in FIG. Input signal samples, which each represent the intensity of an image at a certain point (pel) within the image, are applied to an input line 101. These samples can be obtained by conventional scanning devices known to those skilled in the art, or any other means designed to provide intensity information for a sequence of pel sites. It will uses uniform formatting for intensity data derived from successive images (frames) are.

The samples on line 101 are sent to an image memory 102, an element delay circuit 103 and the first input of a pair of subtracting circuits 104 and 105 given. The image memory 102 is expediently a tapped delay line which has an output signal on the line 106 supplies, which is the intensity value of a

130616/0026

correct spatial position in the previous picture. The exact location is selected based on a control signal applied to line 107. According to the invention, this control signal is a function of the previous displacement '-'"; environment estimation value, and the selected location is the one that best matches the processed point in the current image (frame) The element delay circuit 103 provides an output signal to. a sampling interval -. vall is delayed with respect to its input signal. .--- ■ '

The intensity value output signal of the image memory 102 on the line 106 is applied to the second input of the subtraction circuit 104 and the intensity value output signal of the element delay circuit 103 to the second input of the subtracter circuit 105. Since the position selected in the previous image versus the one currently being processed Position is shifted by the earlier displacement estimate, the output of circuit 104 sets accordingly represents the shifted image difference DFD defined in equation (10). The output signal of the subtracting circuit 105 is the element difference EDIF also defined in equation (10).

The sign bit of the element difference EDIF is from a the output of the subtracter circuit 105 connected register 108 removed and used to selectively control the sign of DFD in a sign changing circuit 109, to which the output signal of the subtracting circuit 104 is fed. This sign signal on line 121

130616/0026

satisfies the "sign" function in equation (10). Subsequently the algebraic value of the product of DFD and the sign of EDIF is accumulated in a first accumulator 111, while the size of EDIF is summed in a second accumulator 112. However, since the calculation according to Equation (10) is only carried out in moving areas, the switches 113 and 114, which are inserted into the input line of the accumulators 111 and 112, respectively, only; "- ' closed at the same time when a corresponding control signal from the movement area segment circuit 120 occurs. The circuit 120, which can be constructed in a manner similar to the known device according to FIG. 2, closes the circuit Switches 113 and 114 when the image difference FDIF exceeds a preset threshold value.

The values contained in the accumulators 111 and 112, the are the numerator or denominator of the above expression in equation (10), are divided in a divider circuit 115 and at ' The end of each calculation interval of N pels is given to a first input of an adder 116. This input signal represents the update expression for the displacement in the x-direction and becomes the previous displacement

^A i-1

Estimated value D for the x-direction added. The adder output is fed back to the input of a delay circuit so that it becomes the next value to be updated is. The output of delay circuit 117 representing the previous displacement estimate, is also the input control signal to frame buffer 102 on line 107, as discussed above. The content of the

130616/0026

Accumulators 111 and 112 are cleared at the end of each interval of N pels.

The apparatus of FIG. 1 is designed to compute the displacement in the x direction. The postponement in the y-direction (corresponding to the lower expression in equation (10)) is calculated in a similar manner, where a line delay circuit instead of the element delayed-: circuit 103 is used. In this way, the.,. Line difference LDIF as an output signal from the subtracting circuit 105 instead of the element difference EDIF issued.

Computer simulations of the device shown in Figure 1 show that recursions using a field interval lead to a good estimate of the displacement. Of course the rate of convergence changes with the interval and with the type of image sequences used. Choosing one Field interval is typical, but a recursion can be performed for a smaller block if desired (e.g. 16 χ 16 pels).

The image memory 102 can be constructed in a simple manner if a further approximation is made in equation (10), in such a way that that the initial and subsequent estimates for the displacement D contain only integer components that are determined by The control signal fed back on line 107 can be rounded up or down. In this case it can the memory 102 together a tapped delay line

130616/0026

-At-

be with a data switch that selects appropriate taps. Alternatively, if the components of D are not integers the output of the tapped delay line are interpolated using methods that are known to the person skilled in the art.

The circuit arrangement according to Figure 1 can, as indicated above. give, use in many applications, but there will be ^. ^: certain amendments to the method of displacement. ... estimate recommended when the estimate facility is to be used in a predictive encoder. First, since a pel-by-pel based encoder operates, it is convenient to vary the displacement estimate for each range of motion pel. This has the additional advantage that it largely overcomes the problems that arise due to several moving objects or because different parts of an object show different displacements, since the recursion occurs more frequently. Of course, rapid convergence of the recursion equation is required.

Second, in a predictive coder it is desirable to to make the prediction error as small as possible so that the efficiency of the encoder can be increased when the dynamic Range of the error signal decreases. Since the postponed Image Difference (DFD) is a measure of the prediction error minimizing its value in the displacement estimator is expedient. Mathematically, if a pel should be at the

130616/0026

* ΐ-1
Place χ is predicted with the displacement D and the intensity I (x -D, tT) and leads to a prediction

M 1
error DFD (x, D) leads, the estimator tries to generate a new estimated value D such that

For that purpose it is

the following applies: IdFD (X ₃ , ^) <| dfd (χ ^, έ ¹ " ¹ )

desired, recursive [ÖFD (x, £> ) Minimize J for each range of motion element using a gradient approach. For example, using the usual procedures for the steepest drop, the new

M * i-1

Estimated value D are related to the old estimated value D in such a way that that applies:

^ = D ¹ " ¹ - (ε ^ νβί ^^, Α ¹ " ¹ )] ² ₍₁₁₎ ,

which can be pulled apart:

D ¹ = D ¹ " ¹ - EDFDQc ^ D ¹ - ¹ ) ^ ™ ^^ ¹ ") _{(12) #}

In both equations (11) and (12), "vL is the gradient with respect to the displacement D, and £ is a positive, scalar constant that is typically in the range 10" ^ <L < 10 " ^2. The gradient V ^ can be evaluated using the definition of DFD according to equation (9) and taking into account the following equation:

where V ^ is the gradient with respect to χ. This substitution leads to:

D ¹ = D ¹ " ¹ - £ dfd (x, d ¹ " ¹ ) Vi (x -Ετ ^, ΐ-τ) (14)

- * ~ "—a ~ —a

where DFD and Vl by interpolation for non-integer D - to be evaluated. A substantial reduction in

130616/0026

303677Q

Computational work for VI is accomplished by adding D to the Calculation of the gradient Yl to an integer value

C * i-1l
DJ a rounded or ] -

^A i-1 rounded value represents each of the components D, so the estimated value for equation (14) can be simplified

D ¹ = D ¹ " ¹ - fDFDCx ^ D ¹ " ¹ ) VI (X ₃ - [D ¹ " ¹ ], tT) (15)." -

Note that V _{D is} calculated using equation (9); could be evaluated, resulting in an estimated value in which Yl is evaluated at (x, t) instead of (x _& ~ Ö "» * -Ό. This alternative method includes an assumption for the linearity of 1, which then does not, could be valid when DD is ¹ ". Also, there is no difference in computational complexity if it is assumed that linear interpolation of I (x, tT) is used to compute DFD and the resulting shifted line and element differences can be used to calculate YI in equation (15).

When explaining the meaning of equation (15) it can be stated that with each approximation step to the old estimated value a vector quantity is added parallel to the direction of the spatial gradient of the image intensity, the size of which is proportional to the motion-compensated prediction error DFD (x, D ¹ "" ¹ ) is. With the help of equation (9) one can see

-a '

U-1

nen that - if the correction term (D-D ~) is perpendicular to the intensity gradient Vl is - the shifted image difference DFD Becomes zero, causing an update to zero for the

130616/0026

Recursion of equation (15) gives. That can occur, though the object actually moved. However, this is not an error in the motion estimation .. Werttechnik, but about the same case in which one intensity ramp is moved translationally and only one Movement parallel to the ramp direction (Yl) is perceptible. Movement perpendicular to the ramp direction cannot be detected and as such arbitrary. In other words, just because of "..". ' the appearance of edges with different orientations "

^A i

in real images there is a convergence of D on actual ones Value of D possible.

The meaning of the quantities in equation (15) is shown graphically in FIG. 2, in which the rectangles 201 and 202 represent the spatial boundaries of a past and a current image (frame). In picture 201 is a sequence of generally parallel scan lines C_? '4i-1' ^ n " ¹ ^ ^ n + 1 is shown ^{in a widely spread out form to} simplify the explanation. The corresponding scanning lines in figure 202 are labeled: ^ ' _n _ ₂ , / ¹ ^ ₁ , /' _n and / ^f _{n + 1} .

In the current image 202, the χ at 203 means the location of the picture element for which a displacement estimate is calculated, and the vector x _& indicates the location of the element 203 with respect to a fixed coordinate system within the image. In the previous figure 201, the χ at 204 indicates the same place at x_, and the vector 205 represents the

"" "" Cl

130616/0026

the displacement estimate -D.

* i — 1 According to equation (15), D is used to calculate VI;.

to the next integer value] _ £> J rounded up or down -; ... det. This is achieved by paying attention to the neighboring Places in the preceding, current and subsequent scanning line (in this example the lines ^ -? »

n-1 ^and ^ n) * ^by ^ ^es ^ ^plus ^ ^e ^ ^en t which discrete picture element-best represents D ". In this context, the position of a discrete picture element is determined by the point in time at which the image intensity signal was sampled. Accordingly, in Fig. 2 the! ^) Values denoted by b, c, d,, f indicate picture elements which surround the point indicated by the tip of the vector 205. In this example, the vector 206 of of point 204 to pel d is closest to the point designated by vector 204. Accordingly, Vl (χ -D, tT) is evaluated for pel d.

A simple scheme for estimating gradient VI is to calculate the mean vertical and horizontal intensity difference between pels surrounding the pel d and then vectorially combine these results. In this scheme, EDIF (horizontal component) = e ~ c and

I _b -I _f ^

LDIF (vertical component) = —s, where I _{b is} the intensity

at pel b, I is the intensity at pel c, and so on.

The other quantity in equation (15) that needs to be evaluated

^A i — 1 \
is DFD (x _a , D), namely the intensity difference between the

1 30616/0026

Pel 203 and the location indicated by vector 205. The intensity value of this last-mentioned point can be determined by interpolation in essentially the same way:. calculate as for Vl, namely by combining the correct one Fraction of EDIF and LDIF with the intensity values for the Pel d.

A block diagram of the device for calculating a displacement estimate according to equation (14) or (15) is " shown in Fig.3. The input video samples on line 301 are sent to a frame buffer 302 and an input a subtracting circuit 303 is given. Assume that

i-1 is the displacement estimate for the previous image D is stored in a delay circuit 304 so becomes it quantizes in the quantizer 305 and its integer number

Γ * ί-1Ί

part ID given to the control input of the image memory 302. This memory, which can be designed similarly to the memory 102 in Figure 1, provides output signals on a The number of lines 306 that represent the intensity of pels in the stored image that represent the location of the pel

C * i-1 ~ l
D [surrounded against the pel χ at the moment,

just processed image moved 1st. For the example according to FIG. 2, the intensity of the pels b, c, e and f is supplied.

The intensity value output signals of the memory 302 are sent to an interpolator 307 together with the output signal for the previous estimate from the delay element

304 given. The interpolator 307 has two functions, namely ^{to calculate the intensity gradient V1 and the intensity value at the point xD 1} ″. Both functions are easy to overlook and can be easily implemented by linear interpolations be used for this purpose.

* i 1 - The intensity value I (x -D) calculated in the interpolator 307 is given to the second input of the subtracting circuit 303, the output signal of which is the shifted image difference DFD which is defined in equation (8). This value is used as an input to a displacement error calculating circuit described below 309 given. The x and y components of the intensity gradient formed in the interpolator 307 are increased in a corresponding manner applied to inputs of the circuit 309 via lines 308.

The calculation circuit 309 is designed to perform the error term calculation defined in the equation (14) or (15) and may include appropriate multiplying and scaling circuitry. Another input, not shown, can be used to influence the fourth of £.

The output of circuit 309 is the displacement error or the update term, which by means of an adding circuit 310 to the previous one, in the delay element 304 stored displacement estimate is added. The update however, only in the area of movement of the image

13 0616/0026

and for this reason a switch 311 is inserted on the output line of the circuit 309. The desk is controlled under the control of the motion range segment circuit 312 which has an input signal on line 301 records. The segment circuit 312 can be corresponding In the same way as the segment circuit 120 in FIG. 1, so that the switch 311 is only closed when the. · -.- picture difference FDIP exceeds a preselected threshold The output signal of the estimated value circuit according to FIG. 3 can either be at the delay element 304 or at the output of the Adder 310, the output signal in the first case simply being a delayed version of the output signal for the second case is.

When performing simulations to evaluate the quality According to the displacement estimation method of the present invention, it has been found that equation (15) continues to be as can be simplified as follows:

D ¹ = D ¹ " ¹ - t sign. (DFD (x., [P ¹ " ¹ ])) · Before z. (YI (X ₁ , [d ¹ " ¹ ^))

(21),

where the sign of a vector quantity is the vector for the signs of its components. The sign function defined by equation (6) avoids the multiplication required to compute the update term. Instead, there are only three options for updating each shift component from one picture element to the next, namely 0 or + t

3 υ j ο /

The encoder using this simplification was apparently the same as one measured in coding bits / field. Encoder using an estimate circuit which Equation (15) is realized.

The shift estimated value circuits according to FIGS. 1 and 3 have in common the repeated updating;) each shift estimated value, where the previous estimate is used to calculate the correction term. In the circuit arrangement According to Fig.1, realizing the equation (10), the shifted image difference becomes based on the previous one Displacement estimate normalized and accumulated according to the image activity (measured by EDIF or LDIF) obtain the updated displacement estimate. The recursion can be performed over a field or a group of pels will. In the circuit arrangement according to FIG. 3, the equation (14) or (15) is realized, the intensity value and the intensity gradient are in the vicinity of the previous one Displacement estimate calculated, scaled and used to calculate the new estimate. The recursion is typically pel for pel.

It will be apparent to those skilled in the art that numerous changes can be made to the invention can be easily done. Accordingly, its scope is to be limited only by the following claims.

1 30616/0026

ORIGINAL INSPECTED

Claims (6)

BLUMBACH · WESER -BERGEN · KRAMER ZWIRNER HOFFMANN PATENT LAWYERS IN MUNICH AND WIESBADEN • 'Patenlconsult Radeckestrafio 43 8000 Munich 60 Tslolon (08?) 883603/883604 Telex 05-212313 Telegrams Patentconsull Patentconsult Sonnenberger Straße 43 6200 Wiesbaden Telephone (06121) 562943/561998 Telex 04-186237 Telegrams Palenlconsull PCT / US 80/00178 NETRAVAlLI 10-1 Western Electric Company Incorporated New York, N.Y. 10038, USA Claims : -: Gk Device for generating an estimated value for the displacement of an object in a video scene between an earlier and a current image, with circuits for calculating a first estimated value for the displacement using the image intensity differences.z between at least one picture element at the moment Image and a spatially corresponding image element in the previous image, characterized in that the device has an update circuit (Fig.1) for repeated updating of the first displacement estimate using the intensity difference between the at least one picture element in the current image and a location in the previous image that is against the current picture element by the first displacement estimate is shifted. Munich: R. Kremer Dlpl.-Ing. · W. Weser Dipl.-Phys. Or. Rer. nat. . E. Hoffmann Dipl.-Ing. Wiesbaden: P. G. Blumbadi Dipl.-Ing. · P. Bergen Pro !. Dr. jur. Dipl.-Ing., Püt.-Ar; see, Pal.-Anw. until 1979 G. Zwiinor Dipl.-Ing. Dipl.-W.-Ing. ". *> f * O * Ο ι fiC -2g- 2. Device according to claim 1, characterized in that the update circuit a memory and a subtracter (102, 104) for forming the intensity difference DFD (x, D) between the at least one picture element at the point χ and a point in the preceding one Image that is against the point χ by the first shift ^A i-1 - "" exercise estimate D is shifted, and a normalization .. ^: . ^: circuit (103, 105, 115) for normalizing DFD through a " Has measure of the local fluctuations in intensity that surround the point χ. 3. Device according to claim 2, characterized in that the normalization circuit is designed so that it the element difference EDIF between the Intensity at point χ and the intensity of the previous picture element are calculated. 4. Apparatus according to claim 2, characterized in that the normalization circuit is designed is that it is the line difference LDIF between the intensity at the point χ and the intensity of the vertical shifted picture element on the previous line is calculated. 5. Apparatus according to claim 2, characterized in that the update circuit further comprises an accumulator (111) for accumulating intensity differences for picture elements in a selected interval N 1 30616/0026 303677Q and a sign changing circuit (109) for incre orientation or decrement the accumulator as a function of the local intensity fluctuations. ■ 6. Apparatus according to claim 5, characterized in that a segment circuit (120) is also provided for recording only those picture elements in the selected interval N which have an image difference (B ¹ DIF) hs --- 'which exceeds a threshold value. 7 * Method 2.ur generating an estimated value for the displacement of an object in a video scene between an earlier and a current image according to device claims 1 to 6, characterized in that the following process steps are provided: Calculating a first estimate for the displacement using the image intensity difference between at least a picture element in the current picture and a spatially corresponding picture element in the previous picture; repeatedly updating the first displacement estimate using the intensity difference between the at least a picture element in the current picture and a point in the previous picture that opposes the corresponding Picture element is shifted by the first displacement estimate.