JP2015069412A - Moving region detection device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a moving region detection device capable of properly detecting a region of a moving object even when the region of the moving object has uniform pixel values.SOLUTION: A binarization part 20 converts a target frame, a previous frame, and a subsequent frame into binary data of a label 1 and a label 0. A distance conversion part 30 converts a pixel inside a region of the label 1 into a distance value indicating a shortest distance to a region of the label 0. A skeleton detection part 40 detects a pixel having a maximum value of the distance value as a skeleton point inside the region converted into the distance value. An arithmetic part 60 calculates a difference between the target frame and the previous frame, and a difference between the target frame and the subsequent frame, and calculates AND of difference images of two sides. A moving region restoration part 70 restores a region of a moving object of the target frame on the basis of a skeleton of an AND image.

Description

  The present invention relates to a moving region detection device that detects a region of a moving object included in an image.

  There is an image quality improvement method called multi-sheet super-resolution processing that obtains a lot of error information from a plurality of images and generates a high-resolution high-resolution image. In addition to the multi-image super-resolution processing, random noise component reduction processing, dynamic range expansion processing, and the like are known by superimposing a plurality of images.

  In order to obtain a plurality of images, a method of photographing a plurality of images at the same time using a plurality of photographing systems whose positions are slightly shifted from each other, and a plurality of images while sequentially shifting the positions using one photographing system. There is a method of shooting the images in time series. Since the scale of the photographing system is large in the former method, the latter method with a small scale of the photographing system is used relatively frequently.

International Publication No. 2010/084902

  A problem that arises when shooting a plurality of images in time series while sequentially shifting the position of the shooting system is the moving object existing in the image sequence. When super-resolution processing is performed on an image sequence including a moving object, it is necessary to extract the moving object in the image sequence in order to prevent the moving object from being multiplexed on the generated high-resolution image. .

  A general method for extracting a moving object in an image sequence is a method of taking a difference in units of pixels between images and extracting a region having a large difference value as a moving region in which the moving object exists. This extraction method is often used in various types of image processing.

  However, the above extraction method has the following problems. In a monotone image or an animation image with a monotonous color, an area of a moving object may be formed by pixels having uniform pixel values. If the moving amount of the moving object is small and the moving object overlaps between the image sequences, the difference value of the overlapped portion may be small and may not be extracted as a moving region.

  As described above, the conventional moving region detection apparatus may not be able to accurately detect the region of the moving object.

  The present invention has been made in view of such problems, and an object thereof is to provide a moving region detection device that can accurately detect a region of a moving object even when the region of the moving object has a uniform pixel value. .

  In order to solve the above-described problems of the related art, the present invention provides a target frame that is a target for detecting an area of a moving object in an image signal, a previous frame before the target frame, and a rear frame after the target frame. For each of the subsequent frames, each luminance value from the lowest luminance to the highest luminance of the image signal is set as a reference value, an area having a luminance value as a reference value is labeled 1, and an area having a luminance value other than the reference value Is converted to binary data with label 0 as the label 0, and the pixel in the area of label 1 is set to the shortest distance to the area of label 0 in each of the target frame, the previous frame, and the rear frame. An image having a maximum value of the distance value in the area converted into the distance value in each of the target frame, the previous frame, and the rear frame. A first difference image is generated by calculating a difference between a skeleton detection unit that detects skeleton points as a skeleton point, a skeleton that is a set of skeleton points in the target frame, and a skeleton that is a set of skeleton points in the previous frame. A first difference calculation unit, a second difference calculation unit that generates a second difference image by calculating a difference between a skeleton in the target frame and a skeleton formed by a set of skeleton points in the subsequent frame, A logical product calculation unit that calculates a logical product of the first difference image and the second difference image to generate a logical product image, and the distance conversion unit converts each skeleton point of the logical product image A moving region restoring unit that restores the region of the moving object in the target frame by arranging a circle having a radius of the distance value thus determined. Providing motion area detection apparatus according to claim.

  According to the moving region detection apparatus of the present invention, it is possible to accurately detect a moving object region even when the moving object region has a uniform pixel value.

It is a block diagram which shows the moving region detection apparatus of one Embodiment. It is a figure which shows the example of the image of 1 frame in the image signal Sin input into the moving region detection apparatus shown in FIG. It is a figure for demonstrating the operation | movement of the binarization by the binarization part 20 in FIG. It is a figure for demonstrating the outline of the distance conversion by the distance conversion part 30 in FIG. It is a figure which shows the distance conversion image which carried out distance conversion by the distance conversion part 30 in FIG. It is a figure which shows the skeleton detected by the skeleton detection part 40 in FIG. It is a figure for demonstrating the movement area detection operation | movement by the movement area detection apparatus of one Embodiment. It is a figure for demonstrating the movement area detection operation | movement by the movement area detection apparatus of one Embodiment. It is a flowchart for demonstrating the moving region detection operation | movement by the moving region detection apparatus of one Embodiment. It is a flowchart for demonstrating the moving region detection operation | movement by the moving region detection apparatus of one Embodiment. It is a flowchart for demonstrating the moving region detection operation | movement by the moving region detection apparatus of one Embodiment. It is a figure which shows the moving region detection operation | movement by a comparative example.

  Hereinafter, a moving region detection apparatus according to an embodiment will be described with reference to the accompanying drawings. In FIG. 1, an image signal Sin that is a moving image signal is input to the storage unit 10. The image signal Sin is, for example, 8-bit digital data. The storage unit 10 stores images for at least three frames.

  FIG. 2 shows a predetermined one-frame F image in the image signal Sin stored in the storage unit 10. The image of the frame F includes the object OB. For convenience, it is assumed that the object OB and the background BG other than the object OB have uniform luminance values. Let the luminance value of the object OB be Lob and the luminance value of the background BG be Lbg.

  The binarization unit 20 binarizes each frame image stored in the storage unit 10 as follows. The binarization operation in the binarization unit 20 will be described taking the case of the image shown in FIG. 2 as an example.

  The binarization unit 20 converts the image of the frame F into binary data using each luminance value from the luminance value 0 (lowest luminance) to the luminance value 255 (maximum luminance) as a reference value. If the luminance value as the reference value is L, the binarization unit 20 sets the pixel having the luminance value L as label 1 and the pixels other than the luminance value L as label 0.

  The luminance value Lob of the object OB is set to 100, for example, and the luminance value Lbg of the background BG is set to 200, for example. When the luminance value L as the reference value is 0 to 99, 101 to 199, 201 to 255, the entire frame F is converted into binary data of label 0.

  When the luminance value L is 100, as shown in FIG. 3A, the frame F is converted into binary data in the region R1 where the object OB is a label 1 and the region R0 where the background BG is a label 0. Is done. When the luminance value L is 200, as shown in FIG. 3B, the frame F is converted into binary data in the region R0 where the object OB is the label 0 and the region BG where the background BG is the label 1 Is done. In FIGS. 3A and 3B, for convenience, label 1 is shown in black and label 0 is shown in white.

  The binarization unit 20 converts the image of the frame F into binary data using 256 luminance values L from the luminance value 0 to the luminance value 255 as reference values, respectively. Value data is output. 256 pieces of binary data are input to the distance converter 30.

  The distance conversion unit 30 performs distance conversion on the region R1 of the label 1 in each of 256 binary data. The distance used in the distance conversion unit 30 is not the Euclidean distance, but is, for example, a 4-neighbor distance. As described above, the binary data other than the binary data shown in FIGS. 3A and 3B does not have the region 1 of the label 1, so that in the case of the image of FIG. Only the respective areas R1 of b) are distance-converted.

  The distance conversion unit 30 converts the region R1 of the label 1 in FIGS. 3A and 3B using a 4-neighbor distance. An outline of distance conversion will be described with reference to FIGS. As shown in FIG. 4A, the region R1 of the label 1 having the shape shown in the figure exists with the region R0 of the label 0 as the background. The distance conversion is a conversion process that gives each pixel in the region R1 the shortest distance to the background of the region R0.

  When the shortest distance to the background is given to each pixel in the region R1 in FIG. 4A, the distance value shown in FIG. 2B is obtained. Here, the distance value between the pixels adjacent to the background is set to zero. There is also a distance conversion method in which the distance value between pixels adjacent to the background is 1.

  The distance conversion unit 30 performs distance conversion on the region R1 in FIGS. 3A and 3B by the method described in FIGS. 4A and 4B. FIG. 5 shows a distance-converted image obtained by distance-converting the region R1 in FIG. In FIG. 5, the distance value is represented by gray shades. The distance value 0 is represented in white.

  Although the illustration of the distance-converted image obtained by distance-converting the region R1 in FIG. 3B is omitted, the background BG portion is represented by gray shades as in FIG.

  Returning to FIG. 1, the distance conversion image is input to the skeleton detection unit 40. The skeleton detection unit 40 sets a pixel having the maximum value of the distance value as a skeleton point in the distance conversion image. Since the distance conversion unit 30 performs distance conversion using the four neighborhood distances, the skeleton detection unit 40 determines that the pixel of interest is the largest among the five pixels of the pixel of interest and the four neighboring pixels adjacent to the top and bottom and left and right of the pixel of interest. If it has a value, the pixel of interest is a skeleton point.

  In the distance conversion unit 30, eight neighboring pixels may be used instead of the four neighboring pixels. When eight neighboring pixels are used, the skeleton detection unit 40 sets the target pixel as a skeleton point if the target pixel has the maximum value among nine pixels of the target pixel and the eight neighboring pixels around the target pixel. .

  A set of skeleton points detected by the skeleton detection unit 40 is referred to as a skeleton. FIG. 6A shows a skeleton SKLob generated corresponding to the region of the object OB. FIG. 6B shows a skeleton SKLbg generated corresponding to the background BG region. Skeletons SKLob and SKLbg are thinned images composed of lines having a width of one pixel.

  The image of the frame F shown in FIG. 2 is a frame F having skeletons SKLob and SKLbg by binarization by the binarization unit 20 described above, distance conversion by the distance conversion unit 30, and detection of skeleton points by the skeleton detection unit 40. It becomes. From the skeleton detection unit 40, a frame F having skeletons SKLob and SKLbg shown in (c) obtained by adding (a) and (b) in FIG. 6 is output.

  The skeleton detection unit 40 sets, for each pixel of the distance conversion image, as a skeleton flag, a label 1 if it is a skeleton point and a label 0 if it is not a skeleton point. Each pixel in the frame F has a unique value including a luminance value, a skeleton flag, and a distance value.

  The luminance value as the eigenvalue here is a reference value when the image of the frame F is converted into binary data by the binarization unit 20.

  The binarization unit 20 to the skeleton detection unit 40 perform the above processes on the images of the respective frames in the image signal Sin. The storage unit 50 stores frames for three frames in which each area in the frame is converted into a skeleton.

  The processing after the storage unit 50 in FIG. 1 will be described with reference to FIGS. 7A and 7B. In FIG. 7A and FIG. 7B, to simplify the description, only the process for the object OB will be described, and the description of the process for the background BG will be omitted.

  In FIG. 7A, frames F0, F1, and F2 are three consecutive frames in the image signal Sin. The object OB in the frames F0, F1, and F2 is moving from the left side to the right side. The frame F1 is a target frame (target frame) for detecting the area of the moving object. In order to detect the object OB in the frame F1, a frame F0 (front frame) and a frame F2 (back frame) before and after the frame F1 are used.

  As shown in FIG. 7A, when the above-described binarization, distance conversion, and skeleton point detection are performed on the frames F0, F1, and F2, skeletons SKLob0, SKLob1, and SKLob2 are generated. Is done. Frames F0, F1, and F2 having skeletons SKLob0, SKLob1, and SKLob2 are stored in storage unit 50.

  A difference calculation unit (first difference calculation unit) 601 in the calculation unit 60 calculates a difference between the frame F1 having the skeleton SKLob1 and the frame F0 having the skeleton SKLob0. By calculating the difference between the frame F1 and the frame F0, the frame F10 having the difference image Diff_10_skl (first difference image) is obtained.

  A difference calculation unit (second difference calculation unit) 602 in the calculation unit 60 calculates a difference between the frame F1 having the skeleton SKLob1 and the frame F2 having the skeleton SKLob2. By calculating the difference between the frame F1 and the frame F2, the frame F12 having the difference image Diff_12_skl (second difference image) is obtained.

  In the example shown in FIG. 7A, since the object OB moves horizontally, the skeletons SKLob0, SKLob1, and SKLob2 are slightly displaced in the horizontal direction on the same horizontal line. The right end of the skeleton SKLob0 and the left end of the skeleton SKLob1 overlap. The right end of skeleton SKLob1 and the left end of skeleton SKLob2 overlap. Therefore, the difference images Diff_10_skl and Diff_12_skl are in a state of being divided into left and right when the difference becomes 0 at the central portion in the horizontal direction.

  Skeletons generated based on the same object have the property of being connected. When the difference images Diff_10_skl and Diff_12_skl are divided to the left and right on the same horizontal line, the connecting units 604 and 605 connect the left and right images to obtain the difference images Diff_10c_skl and Diff_12c_skl.

  In this example, since the skeleton of the object OB is extracted as one horizontal line, when the object OB moves in an oblique direction, the skeletons SKLob0 and SKLob1 do not overlap each other, and the skeletons SKLob1 and SKLob2 do not overlap each other. When the skeletons are displaced at the same horizontal position as in the example illustrated in FIG. 7A, the difference image of the skeleton is divided. In such a case, the divided difference images may be connected based on the connectivity of the skeleton.

  The frame F10 having the difference image Diff_10c_skl and the frame F12 having the difference image Diff_12c_skl are input to the AND operation unit 603 in the operation unit 60. As shown in FIG. 7B, the logical product operation unit 603 calculates the logical product of the frame F10 having the difference image Diff_10c_skl and the frame F12 having the difference image Diff_12c_skl.

  A frame obtained by calculating a logical product of the frames F10 and F12 is defined as a frame F1012. When a logical product of the difference images Diff_10c_skl and Diff_12c_skl is taken, a logical product image A1012_skl is obtained. The position of the logical product image A1012_skl in the frame F1012 coincides with the position of the skeleton SKLob1. The frame F1012 corresponds to the frame F1.

  The moving area restoration unit 70 restores the object OB that is the area of the moving object in the frame F1 based on the logical product image A1012_skl. Each skeleton point constituting the logical product image A1012_skl has a luminance value and a distance value. The moving area restoration unit 70 arranges a circle Cd having a radius as a distance value for each skeleton point of the logical product image A1012_skl. In FIG. 7B, only five circles Cd are shown for simplification.

  If the moving area restoration unit 70 arranges the circle Cd having the radius as the distance value for all the skeleton points, the position and shape of the object OB in the frame F1 can be restored.

  The moving area restoration unit 70 fills the circles Cd arranged at all the skeleton points with the luminance values of the skeleton points. Thereby, the brightness of the object OB can also be restored.

  Since the same processing is performed on the background BG, the moving region restoration unit 70 can also restore the background BG of the image F1.

  The operation of the moving region detection apparatus shown in FIG. 1 will be described again using the flowcharts shown in FIGS. 8A to 8C. In FIG. 8A, the binarization unit 20 sets n in the frame Fn to 0 in step S1. The frame Fn is one of the frames F0 to F2 in FIG. 7A. By setting n to 0, the frame F0 is selected.

  In step S2, the binarization unit 20 sets the luminance value L, which is a binarization reference value, to 0. In step S3, the binarization unit 20 binarizes the frame F0 with the luminance value L (luminance value 0).

  In step S4, the distance conversion unit 30 performs distance conversion on the region 1 of the label 1 of the binary data binarized in step S3. The skeleton detection unit 40 detects the skeleton of the distance conversion image that has been subjected to the distance conversion in step S4.

  In step S6, the binarization unit 20 determines whether or not the luminance value L is 255. If the luminance value L is not 255 (NO), the binarization unit 20 moves the process to step S7. The binarization unit 20 increments the luminance value L by 1 in step S7, and returns the process to step S3.

  The binarization unit 20 repeats the processing of steps S3 to S7. If the luminance value L is 255 in step S6 (YES), the frame F0 is binarized with all the luminance values L of 0 to 255. Become.

  In step S8, the binarization unit 20 determines whether n is 2. If n is not 2 (NO), the binarizing unit 20 moves the process to step S9.

  In step S9, the binarization unit 20 increments n by 1, and returns the process to step S2. Next, by incrementing n by 1, the frame F1 is selected. When the binarization unit 20 to the skeleton detection unit 40 repeat the processes of steps S2 to S7 for the frame F1, the binarization, distance conversion, and skeleton detection of the frame F1 are completed.

  The binarization unit 20 determines that n is not 2 in step S8, increments n by 1 in step S9, and returns the process to step S2. By incrementing n by 1, the next frame F2 is selected. When the binarization unit 20 to the skeleton detection unit 40 repeat the processes of steps S2 to S7 for the frame F2, the binarization, distance conversion, and skeleton detection of the frame F2 are completed.

  If n is 2 in step S8 (YES), the binarization unit 20 has completed all binarization, distance conversion, and skeleton detection of the frames F0 to F2.

  8B, the calculation unit 60 sets n in the frame Fn to 0 in step S10, and sets the pixel position p to 0 in step S11. Thereby, first, for example, the pixel at the upper left end of the frame F0 is selected.

  In step S12, the calculation unit 60 confirms whether the skeleton flag at the pixel position p (the pixel at the upper left corner) of the frame F1 is the label 1 or the label 0, and the pixel position p of the frame F1 is the skeleton point. It is determined whether or not there is.

  If it is a skeleton point (YES), the calculation unit 60 determines in step S13 whether or not the pixel position p of the frame Fn is a skeleton point. If it is not a skeleton point (NO), the arithmetic unit 60 determines whether or not the pixel position p of the frame Fn is a skeleton point in step S14.

  If the skeleton point is determined in step S13 (YES), the calculation unit 60 determines whether the luminance difference between the pixel position p of the frame F1 and the pixel position p of the frame Fn is larger than a predetermined threshold value in step S15. To do. If the luminance difference is larger than the predetermined threshold (YES), the calculation unit 60 shifts the process to step S18, and if not (NO), the calculation unit 60 shifts the process to step S19.

  If there is no skeleton point in step S13 (NO), the arithmetic unit 60 shifts the process to step S18.

  If it is a skeleton point in step S14 (YES), the calculating part 60 will transfer a process to step S17, and if it is not a skeleton point (NO), the calculating part 60 will transfer a process to step S16.

  In step S16, the calculation unit 60 sets the difference point flag at the pixel position p of the frame F1n to label 0. The frame F1n here is the frame F10. In step S17, the calculation unit 60 sets the difference point flag at the pixel position p of the frame F1n as the label 1 and sets the distance value of the pixel position p of the frame Fn.

  In step S18, the calculation unit 60 sets the difference point flag at the pixel position p of the frame F1n as the label 1 and sets the distance value of the pixel position p of the frame F1. In step S19, the calculation unit 60 sets the difference point flag at the pixel position p of the frame F1n to label 0.

  After steps S16 to S19, the calculation unit 60 determines whether or not the pixel position p is the last pixel in step S20. For example, the pixel at the lower right corner is the final pixel. If it is not the last pixel (NO), the calculation unit 60 increments the pixel position p by 1 in step S21, returns the process to step S12, and repeats steps S12 to S21.

  If it is the last pixel in step S20 (YES), it means that the difference calculation between the image F1 and the image F0 is completed. The calculation unit 60 shifts the process to step S22. The calculation unit 60 determines that n is not 2 (NO) in step S22, increments n by 1 in step S23, and returns the process to step S11. By incrementing n by 1, the difference calculation between the frame F1 and the frame F2 is executed next.

  When n = 2 and the processes in steps S12 to S21 are repeated, the difference calculation between the frame F1 and the frame F2 is completed.

  If n is 2 in step S22 (YES), the calculation unit 60 has completed the difference calculation between the frames F1 and F0 and the difference calculation between the frames F1 and F2.

  By the processing shown in FIG. 8B, the difference images Diff_10_skl and Diff_12_skl of the frame F10 shown in FIG. 7A are obtained. A set of pixels with the difference point flag of label 1 is the difference images Diff_10_skl and Diff_12_skl. Although not particularly illustrated in FIG. 8B, as described above, the difference images Diff_10_skl and Diff_12_skl are connected as necessary to form the difference images Diff_10c_skl and Diff_12c_skl.

  Next, in FIG. 8C, the moving region restoration unit 70 initializes the entire frame with the label 0 in step S24. The moving region restoration unit 70 sets the pixel position p to 0 in step S25. In step S26, the moving area restoration unit 70 determines whether or not the difference point flags of the pixel positions p of the frames F10 and F12 are both 1.

  If the difference point flags at the pixel positions p of the frames F10 and F12 are both 1 (YES), the moving region restoration unit 70 determines the distance value at the pixel position p of the frame F10 around the pixel position p at step S27. The process proceeds to step S28 with the range of the circle as label 1. Instead of the distance value at the pixel position p of the frame F10, the distance value at the pixel position p of the frame F12 may be used.

  If any of the difference point flags at the pixel positions p of the frames F10 and F12 is not 1 (NO), the moving area restoring unit 70 shifts the process to step S28.

  In step S28, the moving area restoring unit 70 determines whether the pixel position p is the last pixel. If it is not the last pixel (NO), the moving region restoration unit 70 increments the pixel position p by 1 in step S29, returns the process to step S26, and repeats steps S26 to S29.

  If it is the last pixel in step S28 (YES), the object OB is restored as shown in FIG. 7B, and the process ends.

  For comparison, a case where the calculation unit 60 calculates the frames F0 to F2 as they are without performing binarization, distance conversion, and skeleton conversion will be described with reference to FIG. In FIG. 9, it is assumed that the object OB in the frames F0 to F2 has a small amount of movement, the objects OB in the frames F1 and F0 are in an overlapping positional relationship, and the objects OB in the frames F1 and F2 are in an overlapping positional relationship.

  When the difference calculation unit 601 calculates the difference between the frame F1 and the frame F0, the frame F10 having the difference image Diff_10 is obtained. When the difference calculation unit 602 calculates the difference between the frame F1 and the frame F2, the frame F12 having the difference image Diff_12 is obtained.

  When the logical product operation unit 603 calculates the logical product of the frame F10 having the difference image Diff_10 and the frame F12 having the difference image Diff_12, the logical product image A1012 has left and right parts of the object OB as shown in FIG. The image is missing. The portions where the object OB overlaps in the frames F1 and F0 and the portions where the object OB overlaps in the frames F1 and F2 are areas that cannot be detected.

  As described above, according to the moving region detection apparatus of the present embodiment, even if the object OB overlaps between adjacent frames, the object OB can be overlapped by converting it into a skeleton that is a thinned image. Can be greatly reduced. Therefore, the object OB can be easily detected.

  As described with reference to FIG. 7A, there are rare cases where skeletons overlap. However, it is possible to accurately detect the object OB by connecting the difference images divided based on the connectivity of the skeleton.

  According to the moving region detection device of this embodiment, even when the moving object region has a uniform pixel value, the moving object region can be accurately detected regardless of the moving amount of the moving object.

  If the moving region detection method by the moving region detection apparatus of this embodiment is used for, for example, a plurality of super-resolution processes, random noise components can be reduced and the dynamic range can be expanded. The moving area detecting method by the moving area detecting apparatus of the present embodiment can be used for noise reduction other than the multi-sheet super-resolution processing, and the noise reduction performance can be improved.

  The present invention is not limited to the embodiment described above, and various modifications can be made without departing from the scope of the present invention. The moving area detection apparatus according to the present embodiment may be configured by hardware or software (computer program). The moving area detection program may cause the computer to execute the steps shown in FIGS. 8A to 8C.

20 binarization unit 30 distance conversion unit 40 skeleton detection unit 60 calculation unit 70 dynamic region restoration unit 601 602 difference calculation unit 603 logical product calculation unit 604 605 connection unit

Claims (4)

A target frame that is a target for detecting a region of a moving object in an image signal, a previous frame before the target frame, and a rear frame after the target frame are set to the highest from the lowest luminance of the image signal. A binarization unit that converts each luminance value up to luminance as a reference value, converts an area having a luminance value that is a reference value as a label 1, and converts an area having a luminance value other than the reference value as a label 0 into binary data;
A distance converter that converts pixels in the region of label 1 into a distance value indicating the shortest distance to the region of label 0 in each of the target frame, the previous frame, and the rear frame;
In each of the target frame, the previous frame, and the rear frame, a skeleton detection unit that detects a pixel having a maximum value of a distance value as a skeleton point in a region converted into a distance value;
A first difference calculation unit that generates a first difference image by calculating a difference between a skeleton formed by a set of skeleton points in the target frame and a skeleton formed by a set of skeleton points in the previous frame;
A second difference calculation unit that calculates a difference between a skeleton in the target frame and a skeleton including a set of skeleton points in the subsequent frame to generate a second difference image;
An AND operation unit that calculates an AND of the first difference image and the second difference image to generate an AND image;
A moving region restoration unit that restores a region of the moving object in the target frame by arranging a circle having a radius of the distance value transformed by the distance transformation unit for each skeleton point of the logical product image;
A moving area detecting apparatus comprising: The distance conversion unit converts the pixels in the area of the label 1 into distance values using a 4-neighbor distance or an 8-neighbor distance,
When the distance conversion unit uses a 4-neighbor distance, the skeleton detection unit is configured such that the pixel of interest is the largest among the five pixels of the pixel of interest and four neighboring pixels located above and below and on the left and right of the pixel of interest. If the pixel of interest is a skeleton point, and the distance converter uses an 8-neighbor distance, the pixel of interest is the largest among the 9 pixels of the 8-neighbor pixels around the pixel of interest. The moving region detection device according to claim 1, wherein the pixel of interest is a skeleton point if it has a value.   The moving region restoration unit restores the region of the moving object using a luminance value of a circle whose radius is the distance value converted by the distance conversion unit as a luminance value which is a reference value in the binarization unit. The moving area detecting device according to claim 1 or 2, characterized in that   When the skeleton in the target frame and the skeleton in the previous frame, or the skeleton in the target frame partially overlaps with the skeleton in the subsequent frame, and the first or second difference image is in a divided state, The moving region detection device according to claim 1, further comprising a connecting unit that connects the divided first or second difference images.