JP2011003144A - Device and program for detecting moving object area - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a moving object area detecting device and moving object detection program for detecting a moving object corresponding to complicated motion of a camera with high accuracy.SOLUTION: The moving object area detecting device is constituted so that a motion vector detecting part 11 expands an image by a spatial wavelet reconfiguration and detects a motion vector with high accuracy, a background area detecting part 13 detects a background area by movement vector clustering, a time base direction fluctuation area detecting part 14 detects a fluctuation area in a time base direction by temporal wavelet resolution, and a moving object detecting part 15 detects a moving object from the detected background area and the fluctuation area in the time base direction.

Description

  The present invention relates to a moving object region detection apparatus and a moving object region detection program for detecting a moving object region from a moving image. More specifically, the present invention relates to a moving object area detecting apparatus and a moving object area detecting program for detecting a moving object area from a moving image by using both a spatio-temporal resolution analysis and an optical flow.

  The technology for extracting moving object regions from moving images has been researched and applied in various fields, such as moving image recognition technology and object coding that encodes moving images and sounds of people and backgrounds as separate objects. Yes. For example, moving image recognition technology is used to extract cars on the road and monitor suspicious persons with a surveillance camera. In object coding, the moving object region is distinguished from the background, thereby increasing the video signal level. Efficient encoding is realized.

  A method for extracting a moving object region from a moving image includes (1) a method using difference processing such as background difference (see, for example, Patent Documents 1 and 2), and (2) an optical flow that represents the movement of an object as a vector. Can be roughly divided into two methods (for example, see Non-Patent Document 1).

  In the conventional moving object region extraction method based on background difference, since it is necessary to fix the camera, there has been a problem that it is not practical in shooting for broadcasting and the like. A method for solving the above-described camera fixing problem by extracting a moving object region from a video signal using the above camera is disclosed. In general, in a method using difference processing, a background difference method is used when a background image can be obtained relatively easily, and an inter-frame difference method is used in other cases.

  On the other hand, in the conventional method using optical flow, since the optical flow constraint equation defined in the coordinate system is for the case where the observation system (camera) does not move, it is not suitable when the camera moves. The technique of Non-Patent Document 1 discloses a method using an optical flow constraint equation considering camera motion.

JP 2003-179930 A JP 2002-252849 A

Taku Ebine and one other, "Detection of moving objects based on optical flow estimation considering the motion of the observation system", IEICE Transactions D-II, 2000, vol. 83, no. 6, p. 1498-1506

  However, in the methods of Patent Documents 1 and 2 using the background subtraction method, a plurality of cameras are required for detection of the moving object region. Is not practical. Further, the inter-frame difference method cannot sufficiently cope with a case where the object is temporarily stopped.

  On the other hand, in Non-Patent Document 1 using an optical flow, processing is performed on the assumption that the movement of the camera is known. However, there is no problem in the case of a CCD camera corresponding to the eyes of a robot vision. Cannot handle when is not known.

  As described above, various methods for moving object region detection of moving images have been proposed. However, it is not easy to realize moving object region detection corresponding to complicated camera movements such as panning and tilting.

  Accordingly, an object of the present invention is to provide a moving object area detecting apparatus and a moving object area detecting program for detecting a moving object area corresponding to a complicated movement of a camera with high accuracy from an image taken by a single television camera. It is in.

  In order to solve the above-described problem, a moving object region detection apparatus for detecting a moving object region from a frame image sequence including a plurality of frame images according to the present invention includes a motion vector detection unit that detects a motion vector for a processing target frame image. A background region detecting means for detecting a background region of a frame image by clustering the detected motion vectors, and a one-dimensional frequency for each of a plurality of sets of continuous frame sequences each consisting of frames of different time series A high-frequency component is extracted by performing conversion, and a time-axis direction fluctuation detecting unit that detects a time-axis direction fluctuation region in the frame image from a plurality of sets of the high-frequency components, the detected background region and the time-axis direction And a moving object area detecting means for detecting a final moving object area from the fluctuation area.

  In the moving object region detection device of the present invention, the motion vector detection means performs a two-dimensional first-order discrete wavelet transform on the processing target frame image to generate a two-dimensional frequency component, and the two-dimensional frequency component Substituting the component into the low-frequency component of the two-dimensional frequency component of the enlarged frame image having a size twice as large as that of the processing target frame image in the horizontal and vertical directions, and the two-dimensional frequency component of the enlarged frame image Means for substituting 0 for other components, means for performing two-dimensional first-order discrete wavelet reconstruction on the substituted two-dimensional frequency component of the enlarged frame image, and generating the enlarged frame image; Means for detecting a motion vector from the frame image; and reducing the detected motion vector to a motion vector corresponding to the frame image; And having a means for detecting a motion vector of the processing target.

  In the moving object region detection device of the present invention, the background region detection means clusters the motion vectors by a K-means method.

  Further, in the moving object region detection device of the present invention, the background region detection means compares the direction of the vector from the center of the frame image to each block position and the direction of the motion vector corresponding to each block, The image pickup apparatus further includes a zoom detection unit that detects whether the zoom-in or zoom-out is performed.

  Further, in the moving object region detection device of the present invention, the time axis direction fluctuation detection means may perform one-dimensional first floor for each set of four consecutive frame sequences each consisting of frames having different time series by the Debussy wavelet method. A high-frequency component is extracted by performing discrete wavelet transform.

  In the moving object area detecting device of the present invention, the moving object area detecting unit binarizes the detected signal component of the background area and performs a reversal process of 1 and 0 to generate a moving object area. The final moving object region is detected by calculating a logical product of the moving object region and the time axis direction variation region.

  The moving object region detection apparatus according to the present invention further includes motion vector smoothing means for smoothing the detected motion vector before executing the clustering.

  The moving object region detection program according to the present invention detects a motion vector for a frame image to be processed by a computer constituting a moving object region detection device that detects a moving object region from a frame image sequence composed of a plurality of frame images. A step of clustering the detected motion vectors to detect a background region of a frame image, and a one-dimensional frequency conversion for each set of a plurality of sets of continuous frame sequences each consisting of frames of different time series To extract a high-frequency component, detect a time-axis variation region in the frame image from a plurality of sets of the high-frequency components, and from the detected background region and time-axis variation region, And a step of detecting a typical moving object region.

  According to the present invention, it is possible to accurately detect a moving object region even when the background moves in a complicated manner due to panning or tilting.

It is a block diagram of the moving object area | region detection apparatus of one Example by this invention. It is a block diagram of the motion vector detection part of one Example by this invention. It is a block diagram of the background area | region detection part of one Example by this invention. FIG. 6 shows a process for enlarging an image by two-dimensional first-order discrete wavelet reconstruction in a motion vector detection unit according to an embodiment of the present invention. Fig. 6 shows a smoothing process of a movement vector for each block according to an embodiment of the present invention. 3 shows a processing method of one-dimensional first-order discrete wavelet decomposition in the time axis direction according to an embodiment of the present invention. 5 shows a fluctuation region extraction processing method in the time axis direction for a first high-frequency component (EvenH _t ) according to an embodiment of the present invention. 1 shows a processing method for one-dimensional first-order discrete wavelet decomposition in the time axis direction according to an embodiment of the present invention. 4 shows a processing method for extracting a fluctuation region in the time axis direction for the second high-frequency component (OddH _t ) according to an embodiment of the present invention. FIG. 6 shows a fluctuation region extraction processing method for the time axis direction of one frame according to an embodiment of the present invention. FIG. It is an operation | movement flowchart of the moving object area | region detection apparatus of one Example by this invention.

  The “moving object” in the present invention includes an arbitrary moving subject such as a person or an object. Hereinafter, a moving object region detection apparatus according to an embodiment of the present invention will be described in detail with reference to FIGS.

[Moving object area detector]
FIG. 1 shows a block diagram of a moving object area detecting apparatus 1 according to an embodiment of the present invention. The moving object region detection apparatus 1 according to the present embodiment includes a motion vector detection unit 11, a motion vector smoothing unit 12, a background region detection unit 13, a time axis direction variation region detection unit 14, and a moving object region detection unit 15. With. Data used for each process of the moving object area detection device 1 according to an embodiment of the present invention can be appropriately stored in a storage unit (not shown) included in the moving object area detection device 1.

  In this embodiment, the moving object region is detected in consideration of both the spatial region and the temporal region of the frame image.

  First, components for performing background area detection in the spatial area will be described.

(Background area detection in space area)
The motion vector detection unit 11 includes a frame image signal component value substitution unit 21, a frame image enlargement unit 22, a motion vector determination unit 23, and a motion vector reduction unit 24.

The motion vector detection unit 11 divides the processing target frame image into predetermined blocks, and detects motion vectors for all the blocks by the block matching method. In this embodiment, in order to detect a moving object region that is resistant to noise and the like with high accuracy, the block image is expanded twice in the horizontal and vertical directions, and a motion vector is determined with 1/2 pixel accuracy. . As shown in FIG. 4, the frame image signal component value substitution unit 21 performs a two-dimensional first-order discrete wavelet transform on the processing target frame image to generate a two-dimensional frequency component, and the two-dimensional frequency component is While substituting for the low-frequency component of the two-dimensional frequency component of the enlarged frame image having twice the size in the horizontal and vertical directions with respect to the frame image to be processed, and for other components of the two-dimensional frequency component of the enlarged frame image Substitute 0. Specifically, first, the frame image F (t) at time t is divided into M × N blocks, and each block is subjected to wavelet transform. In the following description, it is assumed that the block size is 8 × 8. In this embodiment, the horizontal low-frequency and vertical low-frequency (LL) components of the two-dimensional first-order discrete wavelet decomposition component of a block of 16 × 16 pixels obtained by doubling the obtained frequency component in the horizontal and vertical directions, The frequency components of the 8 × 8 pixel block thus obtained are substituted for the LL component in the 0th layer of wavelet decomposition into LL ⁰ (t). 0 is substituted for other horizontal low frequency and vertical high frequency (LH ⁰ ) components, horizontal high frequency and vertical low frequency (HL ⁰ ) components, and horizontal high frequency and vertical high frequency (HH ⁰ ) components. The two-dimensional frequency component for the obtained block of 16 × 16 pixels is output to the frame image enlargement unit 22.

  The frame image enlargement unit 22 performs two-dimensional first-order discrete wavelet reconstruction on the spatial domain two-dimensional frequency component of the assigned enlarged frame image, binarizes it, and generates an enlarged frame image F ′ (t) The result is output to the motion vector determination unit 23.

  Similarly, the frame image signal component value assigning unit 21 and the frame image enlarging unit 22 process the frame image F (t + 1) at time t + 1 to generate an enlarged frame image F ′ (t + 1), and a motion vector determining unit 23. Output to.

  The motion vector determination unit 23 uses the input frame image F ′ (t) as a reference frame and F ′ (t + 1) as a reference frame, determines a motion vector for a block expanded by the block matching method, and a motion vector reduction unit 24. Output to.

  The motion vector reduction unit 24 reduces the motion vector for the input enlarged image, generates a motion vector for the original frame image F (t), and outputs the motion vector to the motion vector smoothing unit 12.

  In this way, the motion vector detection unit 11 divides the input frame image into predetermined blocks, expands the image of each block in the horizontal and vertical directions, and generates the frame image F (t) at time t and the frame at time t + 1. A motion vector is calculated from the image F (t + 1) by a block matching method with a ½ pixel accuracy and output to the motion vector smoothing unit 12.

  Note that the processing in the frame image signal component value substitution unit 21 and the frame image enlargement unit 22 is repeated again to enlarge the image four times in the horizontal and vertical directions, and detect a motion vector with 1/4 pixel accuracy. You can also.

  The motion vector smoothing unit 12 calculates the median of motion vector components associated with the block and a predetermined number of blocks around the block for the block constituting the frame image, and calculates the median The motion vector is smoothed by replacing the motion vector component of the block. This smoothing of the motion vector is performed in order to avoid a detection error because a motion vector detection error occurs when the motion vector is detected when the boundary of the moving region substantially overlaps the block boundary.

  Here, the motion vector smoothing process will be described with reference to FIG.

  First, i and j are the block positions in the horizontal and vertical directions, and each block of 16 × 16 pixel size in the frame image F (t) at time t is B (i, j), and each B (i, j ) Is assigned the motion vector MV_B (i, j) obtained by the motion vector detection unit 12. Next, the median of motion vector components in the surrounding 3 × 3 region is calculated for all MV_B (i, j) by a 3 × 3 filter (not shown), and the obtained median Replace the motion vector component of MV_B (i, j). Thereby, the change of the motion vector between blocks is smoothed, and it becomes possible to avoid a motion vector detection error.

  The background region detection unit 13 includes a zoom detection unit 31, a zoom background region generation unit 32, and a motion vector clustering unit 33. The background area detection unit 13 detects the background area of the frame image by clustering the detected motion vectors.

  The zoom detection unit 31 compares the direction of the vector from the center of the frame image to each block position and the direction of the motion vector corresponding to each block, and detects whether or not the photographing apparatus is zooming in or zooming out. To do. When zooming in, the motion vector of the area corresponding to the background has a substantially constant size. When moving out from the center of the frame image and zooming out, the motion vector is a frame. It is expected to be directed to one point in the center of the image. On the other hand, the motion vector of the moving object region is expected to be in a different direction from the background region. Therefore, in the present embodiment, the line segments are extended in the vector direction with respect to all the motion vectors MV_B, and the line segments for more than half of all the motion vectors MV_B are within a predetermined small space area ZoomArea (x, y). It is determined that the camera is zooming in or out when crossing at. When it is determined that zoom-in or zoom-out is performed, background area information including block information corresponding to the intersected line segments is output to the zoom background area generation unit 32, and zoom-in or zoom-out is performed. If not determined, the motion vector is output to the motion vector clustering unit 33.

  The zoom background region generation unit 32 generates a background region BackGround (t) from the input background region information and outputs the background region BackGround (t) to the moving object region detection unit 15.

  The motion vector clustering unit 33 performs clustering on the input motion vector and detects a background region. When the photographing system is stationary, the motion vector component in the background area is zero because the background is stationary. Therefore, the background region and the moving object region can be divided by performing motion vector clustering. Further, when the photographing system is panning (or tilting), it is considered that the motion vector of the background area is oriented in a specific direction and has an arbitrary constant magnitude. Accordingly, by performing motion vector clustering in the same manner as when the background is stationary, the background moving by panning (or tilting) and the moving object region moving by panning (or tilting) and the object itself are divided. It becomes possible.

  The division process of the background area and the moving object area is performed by clustering by the K-means method as follows. First, the number of divisions (number of clusters) K and the initial value of the cluster center assigned to each cluster are set. In order to detect the moving object area, the frame image may be divided into two parts, that is, the background and the other moving object area. Therefore, the number K of clusters is set to a relatively small value of 2 or more. Next, all motion vectors are assigned to K clusters, the average value of vector components belonging to each cluster is calculated, and the cluster center is obtained again. The above motion vector assignment and cluster center calculation are repeated until the cluster center value does not change. When the value at the cluster center stops changing, the distance from the cluster center is calculated for all vectors, and the dispersion of the distance is obtained for each cluster. In this embodiment, the motion vector belonging to the cluster having the smallest variance of the obtained distance is regarded as the motion vector of the background region, and the background region BackGround (t) is generated using the block corresponding to the motion vector belonging to the cluster. And output to the moving object region detection unit 15.

  As described above, the background region detection unit 13 detects the background region from the motion vector input from the motion vector smoothing unit 12, generates the background region BackGround (t), and outputs it to the moving object region detection unit 15.

  As described above, it is possible to detect the moving object region in the frame image signal in the spatial region.

  Next, the fluctuation area detection with respect to the time axis direction will be described.

(Detection of fluctuation area in time axis direction)
The time-axis direction fluctuation region detection unit 14 performs one-dimensional frequency conversion on each set of a plurality of sets of continuous frame sequences each composed of frames having different time series, and extracts a high-frequency component. From the above, the fluctuation region in the time axis direction in the frame image is detected. In the present embodiment, an example will be described in which one-dimensional first-order discrete wavelet transform is performed in the one-dimensional frequency transform using, for example, a Debusie wavelet method having a wavelet length of 4.

First, as shown in FIG. 6, one-dimensional first-order discrete wavelet decomposition is performed with respect to the time axis direction of the continuous frame sequences F (t-3) to F (t) corresponding to the wavelet length. Thereby, the first high-frequency component EvenH _t (t ₀ ) in the time axis direction is obtained. The obtained first high-frequency component EvenH _t (t ₀ ) is binarized with a predetermined threshold Th, and a time-axis direction variation region frame image sequence is extracted from the frame image sequence. The obtained first high frequency component EvenH _t (t ₀ ) includes the second-order wavelet length of the debussy wavelet method, that is, the change of the fluctuation region over four frames.

Similarly, using the first high frequency component EvenH _{t (t 0)} the continuous frame sequence proceeds 2 frame to F (t-1) ~F the (t + 2), the first high frequency component EvenH _{t (t} 0 +1) Calculate

Subsequently, as shown in FIG. 7, by calculating the logical product of the first high-frequency component EvenH _t (t ₀ ) and the first high-frequency component EvenH _t (t ₀ +1), the frame image F (t−1) and A time axis direction fluctuation region with respect to F (t) is extracted. The obtained time axis direction fluctuation region includes a fluctuation region of two frames in the original image.

Next, as shown in FIG. 8, with respect to the frame sequence of F (t−2) to F (t + 1) shifted by one frame with respect to the first high-frequency component EvenH _t (t ₀ ), Perform one-dimensional first-order discrete wavelet decomposition. Thereby, the second high-frequency component OddH _t (t ₀ ) in the time axis direction is obtained. Next, the obtained second high-frequency component OddH _t (t ₀ ) is binarized with a predetermined threshold Th, and a time-axis direction variation region frame sequence is extracted from the frame image sequence. The resulting second high frequency component OddH _{t (t 0),} similar to the first high frequency component EvenH _{t (t 0),} 2-order wavelet length of Debussy wavelet method, i.e. the time axis direction variation region spreading four frames Contains.

Similarly, using the second high frequency component OddH _{t (t 0)} the continuous frame sequence proceeds 2 frame to F (t) to F a (t + 3), calculating a second high frequency component OddH _{t (t} 0 +1) To do.

Subsequently, as shown in FIG. 9, by calculating the logical product of the second high frequency component OddH _t (t ₀ ) and the second high frequency component OddH _t (t ₀ +1), the frames F (t) and F (t + 1) are calculated. ) In the time axis direction. The obtained time axis direction fluctuation region includes a fluctuation region of two frames in the original image.

Finally, as shown in FIG. 10, the first high-frequency component EvenH _t (t ₀ ), the first high-frequency component EvenH _t (t ₀ +1), the second high-frequency component OddH _t (t ₀ ), and the second high-frequency component OddH _t A logical product with (t ₀ +1) is calculated, and a time-axis direction variation region MovArea (t) for a certain frame image F (t) of the original image is generated and output to the moving object region detection unit 15.

  As described above, it is possible to extract the fluctuation region with respect to the time axis direction of the frame image signal.

  Thus, the final moving object region is detected from the background region and the time axis direction variation region. The moving object region detecting unit 15 binarizes the detected signal component of the background region, generates a moving object region by performing inversion processing of 1 and 0, and calculates the logic of the moving object region and the time axis direction variation region. The final moving object region is detected by calculating the product. Specifically, the final movement from BackGround (t) obtained by the background region detection unit 13 and the time axis direction variation region MovArea (t) for one frame obtained by the time axis direction variation detection unit 14. Detect the object area. Specifically, first, reverse processing of 1 and 0 (that is, NOT processing of logical operation) is performed on BackGround (t) to obtain a moving object region Foreground (t). Next, by calculating the logical product (AND of logical operation) of the Foreground (t) and the time axis direction change area for one frame, the overlapping area of the moving object area and the time axis direction change area in the space area Is determined as the final moving object region.

  As described above, the moving object region detection apparatus according to the embodiment of the present invention can accurately detect the moving object region even when the background moves in a complicated manner due to panning or tilting.

  Next, the operation of the moving object region detection apparatus according to an embodiment of the present invention will be described.

[Operation of moving object area detector]
FIG. 11 is a flowchart showing the operation of the moving object area detecting apparatus according to the embodiment of the present invention. For the description of the constituent elements, the reference numerals in FIG. 1 are used.

  First, in step S1101, it is determined whether or not to detect a moving object region in the spatial region.

  If it is determined in step S1101 that the moving object area is to be detected in the spatial area, a plurality of frame images are divided into predetermined blocks in step S1102, and motion vectors are detected for all blocks by the block matching method. To do. For example, the frame image F (t) at time t and the frame image F (t + 1) at time t + 1 advanced by one frame with respect to F (t) are divided into predetermined blocks, and after wavelet transform is performed, Binarization is performed using a threshold value to generate a spatial domain two-dimensional frequency component, and a motion vector at time t is detected for all blocks by a block matching method.

  Next, in step S1103, the motion vector smoothing unit 12 associates each detected motion vector with an arbitrary block in F (t) and a predetermined number of blocks around it. The median of the motion vector components is calculated, and the motion vector is smoothed by replacing the motion vector components with the median.

  Next, in step S1104, the background region detection unit 13 clusters the smoothed motion vectors, thereby detecting the background region of the frame image and generating the background region BackGround (t).

  Next, in step S1105, the time-axis direction fluctuation region detection unit 14 performs one-dimensional frequency conversion on four sets of continuous frame sequences each consisting of frames of different time series, and extracts high-frequency components. A time domain fluctuation region is detected by calculating a logical product for the set of high-frequency components. For example, among the frame images from time (t-3) to (t + 3), F (t-3) to F (t), F (t-2) to F (t + 1), and F (t-1) to F Each of the four consecutive frames of (t + 2) and F (t) to F (t + 3) is wavelet transformed and then binarized to generate a time-axis direction two-dimensional frequency component, and the time-axis direction two-dimensional frequency The horizontal and vertical high-frequency components are extracted from the components, and a logical product is calculated for the four sets of the extracted high-frequency components, thereby generating a time-axis direction fluctuation region MovArea (t) at time t.

  Finally, in step S1106, the moving object region detection unit 15 detects the final moving object region from the detected background region and time axis direction variation region. For example, a reversal process of 1 and 0 is performed on the generated signal component of the background region to generate a moving object region Foreground (t), and the obtained Foreground (t) and time axis direction variation region MovArea (t) To determine the final moving object region at time t.

  In this way, it is possible to detect the moving object region at time t with high accuracy.

  Furthermore, as one embodiment of the present invention, a computer that functions as the moving object region detection device 1 can be configured. A program for causing a computer to realize each of the above-described components is stored in a storage unit provided inside or outside each computer. Such a storage unit can be realized by an external storage device such as an external hard disk or an internal storage device such as ROM or RAM. The control unit provided in each computer can be realized by control of a central processing unit (CPU) or the like. In other words, the CPU can appropriately read from the storage unit a program in which the processing content for realizing the function of each component is described, and realize the function of each component on the computer. Here, the function of each component may be realized by all or part of the hardware.

  According to the present invention, it is possible to detect a moving object region corresponding to a complicated movement of a camera from an image captured by a single television camera with high accuracy. Useful for required automatic monitoring devices, intelligent traffic systems (ITS), object coding and decoding devices that require high-efficiency coding of video signals such as broadcast communications, high-resolution video devices, multi-view video devices, etc. is there.

DESCRIPTION OF SYMBOLS 1 Moving object area | region detection apparatus 11 Motion vector detection part 12 Motion vector smoothing part 13 Background area | region detection part 14 Time-axis direction fluctuation | variation detection part 15 Moving object area | region detection part 21 Frame image signal component value substitution part 22 Frame image expansion part 23 Motion Vector determination unit 24 Motion vector reduction unit 31 Zoom detection unit 32 Zoom background region generation unit 33 Motion vector clustering unit

Claims (8)

A moving object region detecting device for detecting a moving object region from a frame image sequence composed of a plurality of frame images,
Motion vector detecting means for detecting a motion vector for the frame image to be processed;
Background area detection means for detecting a background area of a frame image by clustering the detected motion vectors;
A high-frequency component is extracted by performing one-dimensional frequency conversion on each set of a plurality of sets of continuous frame sequences each composed of frames of different time series, and fluctuations in the time axis direction in the frame image from the plurality of sets of the high-frequency components A time axis direction variation detecting means for detecting a region;
A moving object area detecting means for detecting a final moving object area from the detected background area and the fluctuation area in the time axis direction;
A moving object region detection apparatus comprising: The motion vector detecting means is
A two-dimensional first-order discrete wavelet transform is performed on the processing target frame image to generate a two-dimensional frequency component, and the two-dimensional frequency component is doubled horizontally and vertically with respect to the processing target frame image. Means for substituting for the low-frequency component of the two-dimensional frequency component of the enlarged frame image having a size of
Means for performing a two-dimensional first-order discrete wavelet reconstruction on the substituted two-dimensional frequency component of the enlarged frame image to generate the enlarged frame image;
Means for detecting a motion vector from the enlarged frame image;
Means for reducing the detected motion vector to a motion vector corresponding to the frame image and detecting it as the motion vector to be processed;
The moving object region detection device according to claim 1, wherein   The moving object region detection apparatus according to claim 1, wherein the background region detection unit clusters the motion vectors by a K-means method.   Whether the background area detection means compares the direction of the vector from the center of the frame image to each block position with the direction of the motion vector corresponding to each block, and whether the photographing apparatus zooms in or out The moving object region detection apparatus according to claim 1, further comprising a zoom detection unit that detects whether or not.   The time-axis direction fluctuation detection means extracts a high-frequency component by performing a one-dimensional first-order discrete wavelet transform on each set of four consecutive frame sequences each composed of frames of different time series by the debusy wavelet method. The moving object area | region detection apparatus in any one of Claims 1-4 characterized by the above-mentioned.   The moving object area detecting means binarizes the detected signal component of the background area, performs a reversal process of 1 and 0, generates a moving object area, and the moving object area and the time axis direction fluctuation area 6. The moving object region detection apparatus according to claim 1, wherein a final moving object region is detected by calculating a logical product of the two.   The moving object region detection apparatus according to claim 1, further comprising a motion vector smoothing unit configured to smooth the detected motion vector before executing the clustering. In a computer constituting a moving object region detecting device for detecting a moving object region from a frame image sequence composed of a plurality of frame images,
Detecting a motion vector for the frame image to be processed;
Detecting a background region of a frame image by clustering the detected motion vectors;
A high-frequency component is extracted by performing one-dimensional frequency conversion on each set of a plurality of sets of continuous frame sequences each consisting of frames having different time series, and fluctuations in the time axis direction in the frame image from the plurality of sets of the high-frequency components Detecting a region;
A step of detecting a final moving object region from the detected background region and the fluctuation region in the time axis direction;
A moving object region detection program characterized by causing

Images