JPH11266459A - Moving image coder, moving object detector and motion prediction device employed for the devices - Google Patents

Abstract

PROBLEM TO BE SOLVED: To correct a motion of an entire input image by predicting a global motion vector denoting a motion of the entire input image and using the predicted value. SOLUTION: A statistic calculation/prediction section 124 of the moving image coder applies storage processing to global motion vectors of each scene detected by a global motion vector detection section 122 and checks statistic properties as to a change in a motion of the entire image to discriminate the property of the change such as linearity and periodicity based on the statistic property. Then the statistic calculation/prediction section 124 predicts a global motion vector as to received images which follow present image, based on the statistic property and correction and performs the correction processing by using the predicted value. The calculation with respect to all blocks to obtain a motion of the entire image is omitted by predicting the entire motion of the succeeding input image based on the statistic property in this way, and the motion of the entire image is corrected by a small arithmetic operation.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving picture coding apparatus, a moving object detecting apparatus and a motion estimating apparatus, and more particularly to a moving picture coding apparatus used for a video monitoring system and the like.
It is another object of the present invention to provide a moving object detection device using the same, and a motion prediction device applied to these devices.

[0002]

2. Description of the Related Art In general, in a road traffic network monitoring system, a digital video camera is installed so as to capture an image of a certain area, and an image captured by this camera is monitored on a monitor screen in a remote monitoring room. Recording has been done. When performing such image monitoring,
Since it is necessary to send information from a large number of cameras installed on the road to a remote monitoring room, input screens from these cameras are usually compressed and coded to secure transmission bandwidth, and then these codes are coded. The data is multiplexed and transmitted. FIG. 11 shows a configuration of a general moving picture encoding device used in such a monitoring system.

As shown in FIG. 11, a picture taken by a digital video camera 70 is divided by a block dividing section 7.
The information is divided into a specific block (for example, 8 pixels × 8 lines) at 1, and the encoding process is performed by passing the information in block units through an orthogonal transform and quantization unit 72 and a variable length encoding unit 73. The orthogonally transformed and quantized block data is subjected to inverse processing in the inverse processing unit 74, returned to the original image information, and then stored and processed. The memory 75 is used as reference screen information for prediction between temporally different screens. Is accumulated in

When block data of a screen temporally different from the stored screen is input, the motion vector detecting section 76 compares the input block with the stored reference screen, and is referred to as a macro block. A motion vector is detected in units of a plurality of blocks (for example, 2 × 2 blocks), and a predicted image is created using the motion vector. By encoding only the difference between the input block data and the predicted image through the orthogonal transform and quantization unit 72 and the variable length encoding unit 73, inter-frame prediction encoding is performed, and the encoding efficiency is reduced. Enhanced.

[0005] Through the above processing, encoded data in units of blocks, motion vectors indicating motion in units of a plurality of blocks, and various parameters such as an encoding mode are transmitted as an encoded bit stream of a moving image signal. To restore the original data and display it on a monitor or the like.

By the way, in such a system, there is a problem that, for example, when the camera vibrates due to passage of a truck or the like, the whole screen is shaken, and the part which is actually still is recognized as moving. In this case, a large number of blocks having motion vectors are detected on the moving image encoding device side, which causes an increase in the amount of generated codes.

[0007]

Therefore, the motion of the entire screen is obtained and the motion of the entire input screen is corrected using the global motion vector as a global motion vector, thereby reducing the number of useless motion vectors and preventing an increase in the generated code amount. It is necessary. As a method of obtaining the global motion vector, a method of accumulating the motion vectors of all the blocks in the screen and examining the deviation of the root mean square value can be considered.

However, in this case, first, a motion vector including a motion due to shaking or the like is obtained for all the blocks, and then a global motion vector for the entire screen is calculated using the motion vector. Processing such as performing an operation for overall motion correction is required. Therefore, it is necessary to perform many calculations for each input screen.

SUMMARY OF THE INVENTION The present invention has been made in view of the above points, and enables a global motion vector indicating the motion of the entire screen of an input screen to be predicted, and uses the predicted value to correct the motion of the entire input screen. Provided are a moving image encoding device capable of greatly reducing the calculation of a global motion vector for each screen by performing the operation, a moving object detection device using the same, and a motion prediction device applied to these devices. The purpose is to:

[0010]

In order to solve the above-mentioned problems, a moving picture coding apparatus according to the present invention detects a motion vector for each predetermined image area of an input moving picture signal, and converts the detected motion vector. Coding means for performing inter-frame predictive coding using the same; detecting means for detecting a global motion vector indicating the motion of the entire screen of the input moving image signal; and accumulating and processing the global motion vector detected by the detecting means. A prediction unit that extracts a statistical property of a change in motion of the entire screen and predicts a global motion vector of a subsequent input screen based on the statistical property, and a predicted value of the global motion vector by the prediction unit. Motion correcting means for correcting the motion of the input moving image signal on the entire screen.

In this moving picture coding apparatus, the detected global motion vector is accumulated, and the statistical property of the change in the motion of the entire screen is examined. Is determined. Then, the value of the global motion vector is predicted for the input screens subsequent to the next screen based on this statistical property, and correction processing using the predicted value is performed. Therefore, arithmetic processing such as checking the deviation of the mean square value of the motion vectors of all blocks in the screen for each input screen need only be performed for the first few screens in order to detect the actual global motion vector. This eliminates the need to perform each screen individually. Therefore, it is possible to solve the problem of an increase in the generated code amount due to the movement of the entire screen with a small amount of calculation.

Further, according to the present invention, the global motion vector is re-detected at a predetermined time interval, and the re-detected global motion vector is stored and statistical properties are re-extracted, whereby the prediction means The method is characterized in that a predicted value of a global motion vector is corrected. In this way, by re-detecting the movement of the entire screen of the actual input screen and re-extracting the statistical properties, it is possible to perform the optimal prediction processing of the global motion vector that follows the change in the movement of the entire actual input screen. Become.

[0013] Further, when a picture whose global motion vector is known is present among a plurality of pictures on which inter-frame prediction coding is performed using the same reference picture, the prediction means sets the value of the global motion vector. And estimating a global motion vector of another of the plurality of screens based on the time difference from the reference screen between the plurality of screens.

In general, when several temporally consecutive screens use the same screen as a reference screen, a change in the motion of the entire screen between the screens has a linear characteristic. For this reason, if the global motion vector is known for any of the screens, the global motion vector value is increased or decreased by an amount corresponding to the time difference from the reference screen, so that the global A motion vector can be easily predicted.

The motion correcting means may correct a cut-out position of an image area for dividing the screen of the input moving image signal into a plurality of image areas using a predicted value of a global motion vector by the predicting means. Or a means for correcting the value of the motion vector detected for each image region using the predicted value of the global motion vector.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS (First Embodiment) Hereinafter, a moving picture coding apparatus according to a first embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a configuration example of a video surveillance system using the video encoding device according to the first embodiment. A video signal input from the digital video camera 11 is encoded by an encoder 121 of the video encoding device 12. After being digitally compressed and coded by inter-frame predictive coding using motion compensation, the data is transmitted to a decoding device 13 through a transmission path, where it is decoded and displayed on a monitor 14.

In order to realize a function for correcting the movement of the entire input screen due to the shaking of the digital video camera 11 and the like, the moving picture coding apparatus 12 has an arrangement shown in FIG. As described above, the global motion vector detection unit 122, the global motion vector storage memory 123, the statistic calculation / prediction unit 124, and the motion correction unit 125 are provided.

The global motion vector detection unit 122 accumulates the motion vectors of all the macroblocks in the screen using the motion vectors detected for each macroblock of the input screen by the encoder 121, and calculates the mean squared value. By examining the value deviation, a global motion vector indicating the motion of the entire input screen is detected. The value of this global motion vector (GMV) is obtained from the following equation (Equation 1).

[0019]

(Equation 1)

Here, N is the total number of macroblocks in one screen, and MVk is the motion vector of the kth macroblock. The global motion vectors detected by the global motion vector detecting unit 122 are continuously stored in the global motion vector storage memory 123 for several screens. In the statistic calculation / prediction unit 124, statistical properties such as periodicity and linearity of a change in motion of the entire screen are obtained based on the accumulated global motion vectors, and subsequent statistical properties are determined based on the statistical properties. The prediction of the global motion vector of the screen different in time is performed.

By using the predicted value, the motion correction unit 125 corrects a coordinate position at which a block is cut out from the input screen, or corrects a motion vector detected for each macroblock. As a result, the movement of the entire input screen is corrected.

FIG. 2 shows a specific configuration example of the moving picture coding apparatus 12 in the case where the block cutout position is corrected using the predicted value of the global motion vector. Here, a block cutout position correction unit 125a is provided as the motion correction unit 125 described above. The block cutout position correcting unit 125a calculates a coordinate value for correcting the block cutout position for each macroblock based on the predicted value of the global motion vector.

That is, the screen of the input moving image signal is first divided into macroblocks by the block dividing unit 201.
In this case, the coordinates of the cutout position of each block from the input screen are corrected by the coordinate information from the block cutout position correction unit 125a. Here, assuming that the coordinates of the original cutout position of each macroblock are (x, y) and the x, y components of the global motion vector are Gx, Gy, respectively, the corrected position coordinates (x ′, y ′) are , Given by:

X ′ = x−Gxy ′ = y−Gy The input video signal divided into macroblocks is input to a subtractor 202, where the difference from a predicted image signal generated as described later is obtained. Is taken to generate a prediction residual signal. This prediction residual signal is subjected to discrete cosine transform by DCT transform section 203. The DCT coefficient data obtained by the DCT transformer 203 is quantized by the quantizer 204, and then sent to the variable-length encoder 205 for variable-length encoding. The inverse quantizer 207 and the inverse DCT transform Part 20
8, where they are sequentially subjected to the processing opposite to the processing of the quantization section 204 and the DCT transform section 203, and then to the adder 209.
Is added to the predicted image signal to generate a local decoded signal. This local decoded signal is stored in the frame memory 2
10 and input to the motion detection unit 211.

The motion detector 211 detects a motion vector for each macroblock by performing motion detection from the correlation between the input moving image signal and the image of the previous frame stored in the frame memory 210. A predicted image signal is generated using the motion vector. The motion vector detected here is also sent to the global motion vector detection unit 122 for detecting the above-mentioned global motion vector.

In the variable length coding section 205, DCT coefficient information is coded together with the motion vector information, and the coded image data is sent to the transmission path. The function of the block cutout position correction unit 125a is the same as that of the block division unit 2
01 can also be incorporated.

FIG. 3 shows a specific configuration example of the moving picture coding apparatus 12 in the case where the motion vector of each macroblock is corrected using the predicted value of the global motion vector.

Here, a motion vector correction unit 125b is provided as the motion correction unit 125 in FIG. In the motion vector correction unit 125b, the motion vector is corrected for each macroblock based on the predicted value of the global motion vector, and the correction result is returned to the motion detection unit 211 and the variable length coding unit 205 Sent.

The x and y components of the motion vector of each macroblock detected by the motion detection unit 211 are respectively MVx,
MVy, and the x and y components of the global motion vector of the screen to which these macroblocks belong are Gx and Gy, respectively.
Then, the corrected motion vectors MVx ′ and MVy ′
Is given by the following equation.

MVx '= MVx-Gx MVy' = MVy-Gy In FIG. 3, the motion vector correcting section 125b is provided independently of the motion detecting section 211, but the function of the motion vector correcting section 125b is The motion vector correction unit 125b may be provided in the unit 211 and correct the motion vector value by itself using the predicted value of the global motion vector.

Next, the global motion vector prediction method according to the present embodiment will be specifically described. FIG.
This shows a distribution state of motion vectors for a certain screen.

In FIG. 4, the solid line shows the distribution state of the motion vector in one normal screen when there is no movement of the whole screen due to camera shake or the like. There will be a peak value. On the other hand, for example, when a camera placed on a bridge moves over the entire screen due to the passage of a track, the wind, and the like, the movement of the entire macro screen becomes zero, as indicated by the dashed line. The movement amount fluctuates in the plus or minus direction rather than the position.
Such a deviation of the motion of the entire screen is obtained by the above (Equation 1).

FIG. 5 shows an example of the statistic of the global motion vector value input to the statistic calculation / prediction unit 124. FIG. 5A shows a change in the vertical direction (V) of the global motion vector over several consecutive screens, that is, how the y component of the global motion vector changes. FIG. It shows how the global motion vector changes in the horizontal direction (H) over the screen, that is, how the x component of the global motion vector changes.

As can be seen from FIG. 5, changes in the movement of the entire screen often have a certain degree of periodicity. Therefore, the period of the vibration is obtained by obtaining the autocorrelation, and it is possible to predict the global motion vector for the next and subsequent screens based on the period. The cycle of the change in the motion of the entire screen is obtained as follows.

That is, assuming that the global motion vector of the input screen at time t is G (t) and the global motion vector of each screen from t = 0 to T is accumulated, the following equation (Equation 2) is obtained. At

[0036]

(Equation 2)

Τ at which the value of F (τ) becomes maximum is obtained. If the value of F (τ) at that time exceeds a certain reference value (for example, 25), there is periodicity in the motion between the stored pictures. And τ can be determined as a cycle. The predicted value of the global motion vector is obtained from the period τ thus obtained.

That is, as for the global motion vector G (t) of the input screen at a certain time t, the value of the global motion vector accumulated up to the time of calculating the cycle is G (T), and T is 0 ≦ T <τ. Then, it can be predicted by the following equation.

G (t) = G (T + Nτ), N = 0, 1, 2,... That is, an input for which a predicted value is to be obtained from the known global motion vector values G (T) within the period τ G (T) having the same phase as the screen is used as the predicted value of the global motion vector.

The relationship between the change in the motion of the input screen and the predicted value will be described below with reference to FIG. FIG. 6 shows a state in which vertical movement is repeated such that the entire screen moves upward by one macroblock in frame N + 1, and the entire screen moves downward by one macroblock in the subsequent frame N + 2. I have. Assuming that global motion vectors for frames N to N + 2 are accumulated, in frame N + 3, the entire screen is again shifted upward by 1
Since it is expected to move by a macroblock, the corresponding global motion vector of frame N + 1 is used as the predicted value of the global motion vector of frame N + 3. Similarly, for frame N + 4, the corresponding global motion vector of frame N + 2 is used as the predicted value.

It is preferable that the periodicity of the change in the motion of the entire screen is re-detected at regular time intervals, whereby the predicted value is periodically corrected based on the actual change in the screen. In this case, the motion correction of the entire screen for the input screen (for example, frames N + 8, N + 9,... In FIG. This is continuously performed by continuously using the accumulated data of the global motion vector for N + 2. When a new periodicity is detected, the periodicity used for prediction is corrected from the subsequent input screen. This enables more reliable prediction processing without impairing the real-time performance of the encoding processing.

Further, instead of continuously using the accumulated data of the global motion vectors for frames N to N + 2, the global motion vector detection unit 122 performs the motion correction of the entire screen for frames N + 8, N + 9,. Those frames N actually detected by
+8, N + 9,... Can be performed using respective global motion vector values.

Next, referring to the flowchart of FIG.
The procedure of the global motion vector detection / prediction process will be described. First, the global motion vector detection unit 122
, A global motion vector of the input screen is obtained (step S101), and the global motion vector is stored in the global motion vector storage memory 123 (step S102). Then, the statistic calculation / prediction unit 124
Thus, the auto-correlation of the global motion vector between several consecutive screens stored so far in the global motion vector storage memory 123 is calculated by the above-described equation (Equation 2) (step S103), and the correlation value is equal to or larger than the threshold value. It is determined whether or not there is a correlation of global motion vectors between screens (step S104).

If it is determined that there is no correlation between the global motion vectors between the screens, steps S101 to S102
, A global motion vector is detected and stored for a new input screen, and the presence or absence of a correlation between the global motion vectors is checked again.

When it is determined that there is a correlation between the global motion vectors between the screens, a cycle for the change in motion of the entire screen is obtained (step S105), and the input of the next and subsequent screens is performed using the cycle and the accumulated data. A predicted value of the global motion vector is obtained for each input of the screen (step S106). Then, when a predetermined time has elapsed since the cycle calculation was performed (step 107), step S
By re-executing the process from 101, re-detection of the periodicity and correction of the predicted value based on the re-detected period are performed.

In the present embodiment, the period of the change of the motion is obtained from the autocorrelation function. For example, a zero-cross point of the graph of the change of the global motion vector value as shown in FIG. Needless to say, the period may be obtained.

FIG. 8 shows another example of the prediction of the global motion vector. Generally, in motion prediction processing in image coding, an inter-frame prediction code is used in which a screen that is temporally separated from an input screen by two or more frames is used as a reference screen and a motion vector of the input screen is detected from the reference screen. In some cases. For example, the screen (41) shown in FIG.
This is a case where a motion vector of the screen (43) is searched from. Therefore, there are a plurality of screens to be subjected to inter-frame predictive coding using the same screen as a reference screen.

If both the screen (42) and the screen (43) in FIG. 8 are predicted from the screen (41), if the screen (4)
Assuming that the value of the global motion vector from 1) to the screen (42) has already been detected, the value of the global motion vector from the screen (41) to the screen (43) is
It can be predicted from the value of the global motion vector from 1) to the screen (42). That is, when the period of the movement of the entire screen is locally viewed, the change in the movement of the entire screen usually has a linear characteristic between screens that are temporally close to each other. If it is detected from the accumulated global motion vector that there is a period having such a linear characteristic, if there are several screens that use the same screen as a reference screen during that period, For example, between the screens, if the global motion vector is known for any of the screens, the value of the global motion vector is increased or decreased by the amount corresponding to the time difference from the reference screen, and the other It is possible to easily predict the global motion vector of the screen.

For example, in FIG. 8, the screen (43) is shifted from the screen (41) to the screen (42) because the time interval from the reference screen (41) is twice as long as the screen (42). If the value of the global motion vector from the screen (41) to the screen (43) has already been detected as V,
To 2V, which is twice the value V of the global motion vector. Conversely, if the value of the global motion vector on the screen (43) has already been detected,
For example, a value obtained by halving the value can be used as a predicted value of the global motion vector of the screen (42).

FIG. 9 shows a flow of processing when such a prediction processing is used. First, a global motion vector is calculated for the input screen, and the resultant is stored (step S201). Then, when the next screen is input, it is determined whether or not the input screen is a screen that is separated from the reference screen by two or more temporally (step S202).
If the input screen is a screen that is separated from the reference screen by two or more in time, the input screen and the reference screen
It is detected whether or not there is a screen that uses the same screen as a reference screen, and for which a global motion vector has already been detected and is already known (step S203).
If there is, a predicted value of the global motion vector of the input screen is obtained based on the value of the global motion vector for the detected screen and the time difference from the reference screen between the input screen and the detected screen (step). S20
4).

On the other hand, if the input screen is not two or more pages temporally apart from the reference screen, the same screen as the input screen is selected from among the screens two or more temporally ahead from the reference screen. Is used as a reference screen, and it is detected whether or not there is a screen whose global motion vector has already been detected and known (step S205). If there is, a predicted value of the global motion vector of the input screen is obtained based on the value of the global motion vector for the detected screen and the time difference from the reference screen between the input screen and the detected screen (step). S204).

As described above, according to the first embodiment, the nature of the change such as linearity and periodicity is extracted from the statistical property of the change in the motion of the entire screen, and the global motion of the next screen and thereafter is extracted based on the extracted property. Since the value of the vector is predicted, the arithmetic processing such as checking the deviation of the mean square value of the motion vectors of all the blocks in the screen for each input screen is performed only for the first few screens in order to detect the actual global motion vector. Only need to be performed, and it is not necessary to individually perform each input screen. Therefore, it is possible to solve the problem of an increase in the generated code amount due to the motion of the entire screen with a small amount of calculation and without impairing the real-time property of the encoding process.

In particular, since the predicted value is regularly corrected, it is possible to perform an optimal prediction process of the global motion vector that follows the actual change in the motion of the entire input screen. Although the above description has been made on the assumption that dedicated hardware is used, the encoder 121, the global motion vector detection unit 122, the global motion vector storage memory 123, the statistic calculation / prediction unit 124, and Part or all of the motion correction unit 125 can be realized by a computer program. In this case, the same function as that of the first embodiment can be realized only by installing the computer program from the recording medium to the computer and executing the computer program.

(Second Embodiment) Next, as a second embodiment of the present invention, a moving object detection device configured by applying the moving picture coding device of the first embodiment will be described.

FIG. 10 shows the system configuration of the whole moving object detecting device. This moving object detection device is used, for example, for automatically detecting an intruder or the like in a monitoring target area, and includes a moving image encoding device and a decoding device. The encoding side is realized by using the video encoding device 12 described in FIG. 1, and includes a camera 11, an encoder 121, a global motion vector detecting unit 122, a global motion vector storage memory 123, a statistic calculation / A prediction unit 124 and a motion correction unit 125 are provided. On the decoding side, a decoder 131, a motion vector extraction unit 132, a statistic calculation unit 133, a moving object detection unit 134, and a monitor 14 are provided as shown.

On the decoding side, the encoded data sent from the encoding side is decoded by the decoder, and the decoded image is displayed on the monitor 14. In addition, the motion vector extracting unit 132 extracts a motion vector from the encoded data, and detects a moving block. The statistic calculation unit 1 for this moving block
At 33, a comparison with the motion of the surrounding blocks is made,
A block group having the same property in the direction and magnitude of the motion is detected. Then, the moving object detection unit 134 determines whether or not the block group is the target moving object by comparing the size of the block group with the size of the monitoring target object. If the size of a group of blocks having the same motion characteristics is similar to the size of the target moving object, a warning is issued to notify the discovery of an intruder, and the movement of the corresponding block group is tracked. .

In such a moving object detection system, if a motion occurs on the entire screen due to the vibration of the monitoring camera 11 or the like, a motion component occurs in all blocks. Blocks with cannot be accurately extracted. However, in this example, since the encoding is performed in a state where the motion of the entire screen is corrected on the encoding side, the motion occurs on the entire screen due to vibration of the monitoring camera 11 or the like. On the decryption side,
A moving object can be detected accurately.

[0058]

As described above, according to the present invention,
It is possible to predict the whole movement of the next input screen from statistical properties, it is possible to omit the calculation for all blocks to obtain the movement of the whole screen, and it is possible to correct the movement of the whole screen with a small amount of calculation Becomes

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a video surveillance system using a video encoding device according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing an example of a specific connection relationship between a motion correction unit and an encoder provided in the video encoding device of the first embodiment.

FIG. 3 is a block diagram showing another example of a specific connection relationship between a motion correction unit and an encoder provided in the video encoding device of the first embodiment.

FIG. 4 is a view showing a distribution state of motion vectors on an input screen input to the video encoding device of the first embodiment.

FIG. 5 is a diagram showing an example of a statistic of a global motion vector value input to a statistic calculation / prediction unit provided in the video encoding device of the first embodiment.

FIG. 6 is an exemplary view for explaining a relationship between a change in motion of an input screen input to the video encoding device of the first embodiment and a predicted value;

FIG. 7 is an exemplary flowchart for explaining the procedure of global motion vector detection / prediction processing by the video encoding device of the first embodiment.

FIG. 8 is an exemplary view for explaining another example of prediction of a global motion vector used in the video encoding device of the first embodiment.

FIG. 9 is a flowchart for explaining another example of the global motion vector prediction processing operation by the video encoding device of the first embodiment.

FIG. 10 is a block diagram showing a configuration of a moving object detection device according to a second embodiment of the present invention.

FIG. 11 is a block diagram showing a configuration of a conventional video encoding device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... Camera 12 ... Video encoder 13 ... Decoder 14 ... Monitor 121 ... Encoder 122 ... Global motion vector detection part 123 ... Global motion vector accumulation memory 124 ... Statistic amount calculation / prediction part 125 ... Motion correction part 125a: block extraction position correction unit 125b: motion vector correction unit 201: block division unit 202: subtractor 203: DCT conversion unit 204: quantization unit 205: variable length coding unit 207: inverse quantization unit 208: inverse DCT conversion Unit 209 adder 210 frame memory 211 motion detector

Claims (8)

[Claims] An encoding means for detecting a motion vector for each predetermined image area of an input video signal and performing inter-frame predictive coding using the detected motion vector, and an entire screen of the input video signal Detecting means for detecting a global motion vector indicating the movement of the image, and accumulating and processing the global motion vector detected by the detecting means to extract a statistical property of a change in motion of the entire screen, and based on the statistical property. Prediction means for predicting a global motion vector of a subsequent input screen, and motion correction means for correcting the motion of the input moving image signal over the entire screen using the predicted value of the global motion vector by the prediction means. A video encoding device characterized by the above-mentioned. 2. The prediction means re-detects a global motion vector of an input screen at predetermined time intervals, accumulates the re-detected global motion vector, and re-extracts the statistical properties to obtain the prediction. 2. The moving picture coding apparatus according to claim 1, wherein the prediction value of the global motion vector by the means is dynamically corrected. 3. The method according to claim 2, wherein the predicting unit extracts, as the statistical property of the motion of the entire screen, the periodicity or linearity of a change in the motion of the entire screen from the accumulated global motion vector. The moving picture coding apparatus according to claim 1. 4. The method according to claim 1, wherein, when an input screen whose global motion vector is known is present among a plurality of input screens in which inter-frame prediction encoding is performed using the same reference screen, The moving picture according to claim 1, wherein a global motion vector of another screen in the plurality of input screens is predicted based on the value and a time difference from the reference screen between the plurality of input screens. Image coding device. 5. The motion correcting unit corrects a cutout position of each image region for dividing a screen of the input moving image signal into a plurality of image regions using a predicted value of a global motion vector by the prediction unit. 2. The moving picture coding apparatus according to claim 1, further comprising means for performing the moving picture coding. 6. The apparatus according to claim 1, wherein said motion correcting means includes means for correcting a value of a motion vector detected for each of said image areas using a predicted value of a global motion vector by said predicting means. Item 3. The video encoding device according to Item 1. 7. A moving object of each image area is extracted from encoded data of a moving image signal transmitted from a moving image encoding apparatus, and a moving object is extracted from the encoded data by analyzing the extracted motion vector. In the moving object detecting device for detecting, the moving image encoding device detects a motion vector for each predetermined image region of an input moving image signal, and performs inter-frame predictive encoding using the detected motion vector. Means for detecting a global motion vector indicating the motion of the entire screen of the input moving image signal; and storing and processing the global motion vector detected by the detecting means to statistically calculate a change in motion of the entire screen. Predicting means for extracting a property and predicting a global motion vector of a subsequent input screen based on the statistical property; Moving object detection apparatus characterized by comprising a motion compensation means for correcting the whole screen motion of the input moving image signal using the predicted value of the global motion vector by. 8. A detecting means for detecting a global motion vector indicating a motion of the entire screen of the input moving image signal, and a process for accumulating and processing a global motion vector detected by the detecting means to calculate a change in motion of the entire screen. A motion prediction device comprising: a prediction unit that extracts a statistical property and predicts a global motion vector of an input video signal based on the statistical property.