JP2005078195A - Background image generator - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a background image generator for generating background images rendered less vulnerable to noises. <P>SOLUTION: The background image generator includes a data accumulation means for generating background images based on the representative data of each pixel of an inputted animation resulting from the statistical processing of the input data of each pixel and for accumulating the data of each pixel of the inputted animation's frame for each pixel; a data deleting means for determining and deleting data not used in the extraction of the representative data from the accumulated input data of each pixel if the number of the process frames of the inputted animation exceeds a specified value; a representative data extraction means for extracting representative data from the data of each pixel accumulated in the data accumulation means, after the data have been deleted by the data deleting means; and an image generation means for generating images based on the representative data of each pixel. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a background image forming apparatus, and can be applied to, for example, an apparatus that generates a background image from which a moving moving object is removed based on video captured by a camera.

  For example, the data value of each pixel of the input image taken by a monitoring camera or the like that measures traffic flow is accumulated, and a representative value is selected from the accumulated data of each pixel, and an image in which no moving object exists (background As a technique for generating (image), there is a technique shown in Patent Document 1 below.

In Patent Document 1, an input image is accumulated for a certain period of time, and a class appearance frequency histogram corresponding to the luminance value of each pixel is created from the luminance of each pixel of the accumulated frame image. Then, based on the appearance frequency histogram, the luminance value having the highest frequency is extracted as a representative value of luminance, and the image is reconstructed using the representative value of luminance, thereby eliminating the background image that excludes the moving object. The technology of creating is described.
JP 2000-348184 A

  However, the above-described background image generation apparatus of Patent Document 1 simply and statistically obtains a luminance value (class) having the highest frequency (frequency) as a representative value based on the appearance frequency histogram of the luminance of each pixel. For example, if the background image update frequency is to be increased, the accumulation period of the input image is shortened, resulting in variations in the statistical representative value, and noise in the background image. There is a risk of impact.

  Further, in order to increase the update frequency of the background image, it is required to speed up the process of extracting the representative value from the appearance frequency histogram of the luminance of each pixel.

  Furthermore, it is desired to generate a color background image with high accuracy.

  Therefore, an object of the present invention is to provide a background image generation apparatus that solves at least one of such problems.

  In order to solve this problem, the background image generation apparatus of the present invention generates a background image based on representative data of each pixel obtained by statistically processing from input data of each pixel of the input moving image. In the generation apparatus, (1) data storage means for storing the input data of each pixel of the input moving image frame for each pixel, and (2) stored when the number of processing frames of the input moving image exceeds a specified value Data removal means for deciding data not used for extraction of representative data from input data of each pixel and removing the data; and (3) data accumulation means accumulated after data removal by the data removal means. It is characterized by comprising representative data extracting means for extracting representative data from the data of each pixel, and (4) image generating means for generating an image based on the representative data of each pixel.

  According to the background image generating apparatus of the present invention, the representative value is extracted based on the accumulated data obtained by removing one data from the data of each pixel. The background image can be made less susceptible to noise.

  Further, according to the background image generation apparatus of the present invention, the representative value extraction process can be speeded up.

  Further, according to the background image generation apparatus of the present invention, it is possible to generate a color background image with high accuracy.

  In the following, the following first to fourth embodiments are shown as the best mode for carrying out the background image generating apparatus according to the present invention.

  Hereinafter, a background image generating apparatus according to a first embodiment of the present invention will be described with reference to the drawings.

(1) Configuration of First Embodiment FIG. 1 is a block diagram illustrating an internal configuration of a background image generation apparatus according to the first embodiment.

  As illustrated in FIG. 1, the background image generation apparatus 1 includes an A / D conversion unit 101, a storage unit 102, a rejection unit 103, a counter unit 14, and a representative value extraction unit 105.

  The A / D conversion unit 101 converts an input image (analog signal) into image data for each frame. In the present embodiment, the image input by the A / D conversion unit 101 is, for example, a video taken by a camera or the like, and the video is converted into a grayscale image. The A / D conversion unit 101 gives the converted image for each frame to the storage unit 102 and the counter unit 104.

  The accumulation unit 102 receives the image for each frame from the A / D conversion unit 101 and creates and stores an appearance frequency histogram based on the luminance for each pixel of the image of one frame. The appearance frequency histogram stored in the storage unit 102 is provided for each pixel, and the class is a luminance value. Therefore, for example, when the image size of one frame is 320 × 240 pixels and the number of quantization steps is 256, the storage unit 102 has 320 × 240 appearance frequency histograms, and each histogram has The number of classes is 256.

  The exclusion unit 103 determines, for each pixel, a class to be excluded from each appearance frequency histogram stored in the storage unit 102 when the count value of the counter unit 104 described later exceeds the total frequency, and the determined class One frequency is rejected.

  The counter unit 104 counts the number of frames of the grayscale image generated by the A / D conversion unit 101. The count unit 104 is configured so that the total frequency of the appearance frequency histogram is set in advance, and when the count value exceeds the total frequency, the rejection unit 103 performs the rejection process. In this embodiment, a case where the total frequency is set as an odd number will be described.

  The representative value extraction unit 105 extracts the appearance frequency histogram of each pixel from the storage unit 102, and extracts the luminance value (this is called the median value) corresponding to the class to which the central frequency of the total frequency (called the median data) belongs as a representative value. Then, the representative value is given to background image generation means (not shown) to generate a background image. As described above, in the present embodiment, since the total frequency is set as an odd value, the median value can always be determined uniquely.

(2) Operation of Embodiment 1 Next, the operation of the background image generation apparatus 1 according to Embodiment 1 will be described with reference to FIG.

  First, the counter unit 104 and the storage unit 102 are initialized, the counter value of the counter unit 104 is reset to 0, and the appearance frequency histogram of each pixel of the storage unit 102 is reset (S101).

  When the initialization of the counter unit 104 and the storage unit 102 is completed, for example, a video (analog signal) captured by a camera or the like is input to the A / D conversion unit 101 and digitally converted, and an image for each frame is stored in the storage unit 102. And provided to the counter unit 104 (S102).

  In the storage unit 102, a luminance frequency histogram for each pixel is created and stored for all the pixels of the frame image from the A / D conversion unit 101 (S102).

  The appearance frequency histogram of the luminance of each pixel is created by detecting the luminance of each pixel when an image of one frame is input from the A / D conversion unit 101, and the appearance frequency histogram of each pixel having a class corresponding to the luminance value. , The frequency of the class corresponding to the detected luminance value of each pixel is incremented by 1, thereby creating an appearance frequency histogram for each pixel.

  Further, the counter unit 104 counts the number of frames of the image given from the A / D conversion unit 101 (S102).

  When the count value is less than or equal to the total frequency, the counter unit 104 returns to S102, and when the count value exceeds the total frequency, the counter unit 104 proceeds to the next step and causes the rejection unit 103 to perform the rejection process (S103).

  When the count value exceeds the total frequency, the rejection unit 103 refers to the appearance frequency histogram of each pixel stored in the storage unit 102, and selects from the appearance frequency histograms according to the counter value of the counter unit 104. The luminance value (namely, class) to be excluded is determined, and the frequency appearing in the determined luminance value is decreased by one (S104).

  Thereby, the total frequency on each appearance frequency histogram can be made equal to the set total frequency.

  Here, the exclusion process by the exclusion unit 103 determines the luminance value to be excluded according to the counter value of the counter unit 104. For example, when the counter value is an odd value, the minimum appearance frequency histogram of each pixel is determined. The frequency of the class is examined in order from the class, and the class having the frequency of 1 or more is found for the first time, and the found class is determined as the class to be excluded. If the counter value is an even number, the frequency of the class is examined in order from the largest class, and the class with the frequency of 1 or more is searched for and rejected only in the same manner as described above. Of course, the counter value may be searched from the maximum class when the counter value is an odd value, and may be searched from the minimum class when the counter value is an even value.

  As described above, when the exclusion process by the exclusion unit 103 is performed, the total frequency of the appearance frequency histogram of each pixel can be set to a preset setting value.

  The representative value extraction unit 105 refers to the appearance frequency histogram of each pixel after the exclusion process accumulated by the accumulation unit 102, calculates the luminance value (median value) to which the central frequency of the set total frequency belongs, and calculates the median value Extracted as a representative value (S105).

  This median value is detected by, for example, adding the frequencies of each class from the maximum or minimum class of the appearance frequency histogram, and setting the median frequency to the value when the half of the set total frequency is exceeded for the first time. And the brightness value of the median power is detected as the median value. In this embodiment, since the total frequency is an odd value, the median value can always be determined uniquely.

  The representative value extraction unit 105 detects the representative value of each pixel for all the pixels.

  The representative value extraction unit 105 gives representative values for all the extracted pixels to an image generation unit (not shown) in the subsequent stage, and generates a background image based on the representative values. When the representative value of each pixel is output from the representative value extraction unit 105, the process returns to S102, an input frame image is received from the A / D conversion unit 101, and the above-described processing can be repeated.

(3) Effects of Embodiment 1 As described above, according to the present embodiment, by including the rejection unit 103 that preferentially rejects data deviating from the representative value, it is simply based on the appearance frequency histogram as in the prior art. The background image can be made less susceptible to noise than when a statistically determined representative value is extracted.

  Further, according to the present embodiment, the frequency of the class rejected by the rejecting unit 103 is alternately determined from two directions according to the counter value, and therefore the same luminance value of each appearance frequency histogram is continuously input. Even if the value deviates from the representative value, it is not rejected as noise and can follow the change of the captured image.

  Further, according to the present embodiment, even after the counter value exceeds the total frequency, the background image can be generated based on the appearance frequency histogram considering all the frames accumulated by the accumulation unit 102.

  Next, a background image generating apparatus according to a second embodiment of the present invention will be described with reference to the drawings.

(1) Configuration of Embodiment 2 FIG. 3 shows the internal configuration of the background image generation apparatus of this embodiment. As illustrated in FIG. 3, the background image generation apparatus 2 according to the present exemplary embodiment includes an A / D conversion unit 101, a buffer unit 201, a storage unit 102, a rejection unit 103, a counter unit 104, a median value conversion detection unit 202, and parameter storage. Unit 203 and parameter extraction unit 204.

  Note that the A / D conversion unit 101, the storage unit 102, the rejection unit 103, and the counter unit 104 in FIG. 3 correspond to the respective components shown in FIG. A detailed description thereof is omitted here.

  The buffer unit 201 temporarily stores the luminance value of each pixel obtained by digital conversion of the A / D conversion unit 101.

  The parameter extraction unit 204 refers to the appearance frequency histogram of each pixel accumulated in the accumulation unit 102, and the luminance value (median value B) corresponding to the class to which the central data belongs, the frequency S of the median value, and The position M of the central data at the median is extracted as three parameters.

  Here, the parameters extracted by the parameter extraction unit 202 will be described with reference to FIG.

  FIG. 4A is an explanatory diagram showing a relationship between a class and the frequency of the class, and FIG. 4B is an appearance frequency histogram created based on FIG. In FIG. 4, for convenience of explanation, the class is 8 and the total frequency is 19. Therefore, in this case, the data corresponding to the tenth is the central data.

  The parameter extraction unit 204 obtains a class corresponding to the luminance value to which the central data (the tenth frequency in this embodiment) belongs based on the appearance frequency histogram of each pixel, and obtains the class B (the class in the present example). “4”). Next, the parameter extraction unit 204 extracts all frequencies belonging to the median value B as the median frequency S (in this embodiment, “5”), and the central data of the median frequency S A position M (corresponding to “4” in this embodiment) is obtained (see FIG. 4C).

  In this way, the parameter extraction unit 204 obtains the median value B, the median frequency S, and the median position M for each pixel.

  The parameter storage unit 203 stores the three parameters extracted by the parameter extraction unit 204 for each pixel.

  The median value change detection unit 202 detects whether or not the median value of each pixel has changed for all the pixels when the counter value exceeds the set total frequency. For each pixel, the median value change detection unit 202 includes a class to which the luminance value of the pixel stored in the buffer unit 202 belongs (this is referred to as an accumulated value I) and a class to which the frequency excluded by the rejection unit 103 belongs (this). Is referred to as a rejection value R), and it is detected whether the median value has changed based on the result obtained by comparing the median value stored in the parameter storage unit 203 described later.

(2) Operation of Second Embodiment Next, the operation of the background image generating apparatus 2 according to the second embodiment will be described with reference to the drawings.

  First, the counter unit 104 and the storage unit 102 are initialized, the counter value of the counter unit 104 is reset to 0, and the appearance frequency histogram of each pixel of the storage unit 102 is reset (S201).

  When the initialization of the counter unit 104 and the storage unit 102 is completed, for example, an input video (analog signal) photographed by a camera or the like is digitally converted by the A / D conversion unit 101, and an image for each frame is stored in the buffer unit 201. The data is given to the storage unit 102 and the counter unit 104 (S202).

  The buffer unit 201 temporarily stores the luminance value of each pixel for each frame input from the A / D conversion unit 101.

  Further, the accumulation unit 102 creates and stores each appearance frequency histogram of each pixel based on the luminance value of each pixel of the image for each frame input from the A / D conversion unit 101.

  The counter unit 104 receives an image for each frame from the A / D conversion unit 104 and counts the number of frames.

  In the counter unit 104, when the count value is less than the total frequency, the process returns to S202 and is repeated (S203).

  In the counter unit 104, if the count value is equal to the total frequency, the process proceeds to S204, and the parameter extraction unit 204 sets three parameters (median value B, frequency of median value) for each pixel for all pixels. S, the position M of the central data) is extracted, and the extracted three parameters are set in the parameter storage unit 203 (S204).

  Specifically, in S205, the parameter extraction unit 204 obtains central data based on each appearance frequency histogram of each pixel accumulated in the accumulation unit 102, and the class corresponding to the luminance value to which the central data belongs is Obtained as value B. Then, the frequency S appearing in the median value B obtained by the parameter extraction unit 204 and the position M of the central data are obtained (S205).

  After the parameter extraction unit 204 extracts the three parameters of each pixel, the parameter extraction unit 204 stores the extracted three parameters of each pixel in the storage area for each pixel of the parameter storage unit 203. Parameters are set (S206).

  The parameter extraction unit 204 confirms whether these three parameters are extracted and set for all the pixels. If not, all the parameters are set for all the pixels. It returns to S204 so that it calculates | requires, and it repeats and sets the parameter of all the pixels (S207).

  Returning to S203, if the counter value exceeds the total frequency, the rejection unit 103 extracts the appearance frequency histogram of each pixel from the storage unit 102, determines the class to be excluded according to the counter value, and determines the application frequency histogram. One frequency of the determined class is rejected so that the total frequency becomes equal to the set value (S209). The class for which the object of the exclusion process is determined by the exclusion unit 103 is defined as an exclusion value R.

  When the rejection process by the rejection unit 103 is performed, the median value change detection unit 202 updates the above-described parameters (median value B, frequency of median value S, position M of the central data) of each pixel (S210).

  Specifically, the median value change detection unit 202 extracts the luminance value of the pixel temporarily stored in the buffer unit 201 for a certain pixel among all the pixels of the image input this time, and calculates the luminance of the pixel. The class to which the value belongs in the appearance frequency histogram of the storage unit 102 is determined. The class to which the luminance value of the pixel belongs is defined as an accumulated value I.

  Next, the median value change detection unit 202 extracts three parameters (median value B, median frequency S, and median data position M) stored in the parameter storage unit 203. Each parameter extracted from the parameter storage unit 203 is a parameter set in the parameter storage unit 203 immediately before the image input this time.

  Next, the median value change detection unit 202 is a combination of the relationship between the rejection value R rejected by the rejection unit 103 and the median value B extracted from the parameter storage unit 203 and the relationship between the accumulated value I and the median value B. According to a predetermined rule, the value of the median frequency S and the value of the position M of the central data are stored in the parameter storage unit 203 and updated.

  Here, the predetermined rule regarding the update of each parameter by the median value change detection unit 202 is shown in FIG.

  As shown in FIG. 5, when the accumulated value I is less than the median value B, the median frequency S and the median data position M are set to the median according to the relationship between the median value B and the rejection value R as follows. The value change detection unit 202 is updated. When the rejection value R is less than the median value B, the frequency S of the median value and the position M of the median data are not updated. When the reject value R is equal to the median value B, the frequency S of the median value is decreased by 1, and the position M of the central data is decreased by 1. When the rejection value R exceeds the median value B, the median frequency S is not changed, and the position M of the median data is decreased by one.

  When the accumulated value I is equal to the median value B, the median value change detection unit 202 updates the median frequency S and the median data position M as follows. When the rejection value R is less than the median value B, the median frequency S is increased by 1, and the median position M is increased by 1. When the rejection value R is equal to the median value B, the median frequency S and the central data position M are not updated. When the rejection value R exceeds the median value B, the median frequency S is increased by 1, and the position M of the median data is not changed.

  When the accumulated value I exceeds the median value B, the median value change detection unit 202 updates the median frequency S and the median data position M as follows. When the rejection value R is less than the median value B, the median frequency S is not changed, and the position M of the median data is increased by one. When the rejection value R is equal to the median value B, the median frequency S is decreased by 1, and the position M of the median data is not changed. When the rejection value R exceeds the median value B, the frequency S of the median value and the position M of the median data are not updated.

  The median value change detection unit 202 performs updating by storing each parameter of the pixel in the storage area of the parameter storage unit 203 according to the above procedure.

  When the median value change detection unit 202 updates each parameter for the pixel, the median value change detection unit 202 indicates that the relationship between the position M of the median data and the frequency S of the median value is 1 ≦ M ≦ S. It is determined whether or not the condition is satisfied, and it is assumed that the median value B does not change when the condition is satisfied, and the median value changes when the condition is not satisfied (S211).

  When the median value change detection unit 203 determines in S211 that there is no change in the median value B, the process proceeds to S214, and the median value change detection unit 203 detects whether there is a change in the median value B for all pixels. (S214).

  When the median value change detection unit 203 determines that there is no change in the median value of each pixel, the median value of the pixel is obtained by using the parameter stored in the parameter storage unit 203 (that is, the median value of the previous frame). The median value is output (S215).

  Of course, when there is no change in the median value of all the pixels, the median value of all the pixels in the previous frame is output as it is, and it is determined that the median value of even one pixel among all the pixels has changed. In step S212, the parameter extraction unit 204 obtains a median value for the pixel and extracts a new median value.

  Returning to S211, when the median value change detection unit 203 detects a change in the median value B of the pixel, the parameter extraction unit 204 determines the median value B and median value based on the appearance frequency histogram of the pixel in the storage unit 102. The frequency S and the position M of the central data are extracted (S212).

  When the parameter extraction unit 204 extracts three parameters, the parameter extraction unit 204 sets the extracted three parameters in the storage area of the pixel in the parameter storage unit 203 (S213).

  The median value change detection unit 202 proceeds to S214 when the parameters of the pixel are set, and checks whether or not the process is completed for all the pixels. If the process is not completed for all the pixels, the process proceeds to S208. The operation is repeated until all the pixels are returned (S214).

  When the median values for all the pixels are extracted, the median value B of each pixel stored in the parameter storage unit 203 is output, and is supplied to image generation means (not shown) to generate an image (S215).

  As described above, the background image generation device 2 detects whether or not the median value of each pixel has changed, and for the pixel from which the median value has been detected, a new median value is extracted and its center value is extracted. A value is output as a representative value, and for a pixel for which a change in the median value has not been detected, the median value of the pixel output in the previous frame is output as a representative value, thereby speeding up the representative value detection process. it can.

(3) Effect of Embodiment 2 As described above, according to this embodiment, the same effect as described in Embodiment 1 can be obtained.

  Further, according to the present embodiment, the median value change detection unit 202 stores the accumulated value I of each pixel of the buffer unit 201 and the exclusion value R of the exclusion unit 103, respectively, stored in the parameter storage unit 203. Since it is possible to easily detect whether or not the median value B has changed based on the comparison with B, the representative value can be extracted at a higher speed.

(1) Configuration of Embodiment 3 Next, Embodiment 3 of the background image generation apparatus of the present invention will be described with reference to the drawings. In the present embodiment, a background image generation apparatus using RGB color system data using three types of density values of a red signal, a green signal, and a blue signal of an input color image will be described.

  FIG. 7 is a block diagram showing the internal configuration of the background image generation apparatus of this embodiment. As illustrated in FIG. 7, the background image generation apparatus 3 according to the third embodiment includes a color A / D conversion unit 301, a color storage unit 302, a color rejection unit 303, a counter unit 304, and a color representative value extraction unit. 305, a gravity center calculation unit 306, and a rejection data storage unit 307.

  The color A / D converter 301 digitally converts an input color video into a color image for each frame. The color A / D converter 301 provides the color storage unit 302 with three types of density values of red, green, and blue for each pixel. In the following, these three types of red, green, and blue are collectively referred to as color data. The color A / D converter 301 can use an existing A / D converter for color video.

  The color accumulating unit 302 has a data table for accumulating the color data of each pixel for each pixel, receives the digitized image from the color A / D converter 301, and receives the received image of one frame. The color data of each pixel is stored in the data table corresponding to each pixel.

  The color rejection unit 303 determines the color data to be rejected based on the data table stored in the color storage unit 302 when the counter value of the counter unit 304 described later exceeds the total frequency, and the determination The removed color data is removed from the corresponding data table.

  The counter unit 304 counts the number of color image frames from the color A / D conversion unit 301. The counter unit 304 has a total frequency set in advance, and causes the color analysis unit 303 to perform a rejection process when the counter value exceeds the total frequency.

  The color representative value extracting unit 305 extracts curler data for generating a background image as a representative value from the data table of each pixel accumulated in the color accumulating unit 302.

  The center-of-gravity calculation unit 306 takes out the color data in the data table of each pixel accumulated by the color accumulation unit 302, regards the three values of the color data as a three-dimensional vector, and in the three-dimensional coordinates of all the color data. The center of gravity position is calculated based on the distance.

  The rejection data storage unit 307 stores the color data of each pixel that is rejected by the color rejection unit 303 for each pixel. Note that the color data of each pixel stored in the exclusion data storage unit 307 is that of the previous frame related to background image generation.

(2) Operation of Embodiment 3 Next, the operation of the background image generation apparatus of the present embodiment will be described with reference to FIG.

  First, the counter unit 304 and the color storage unit 302 are initialized, the counter value of the counter unit 304 is reset to 0, and the contents of the storage unit 302 are also initialized (S301).

  The color data of each pixel of the image digitally converted by the color A / D conversion unit 301 is given to the color storage unit 302 and added to the data table of each pixel of the color storage unit 302 and stored ( S302).

  Further, the color data of one frame image from the color A / D conversion unit 301 is input to the counter unit 304, and the counter value of the counter unit 304 is incremented by one (S302).

  When the counter value of the counter unit 304 is equal to or less than the total frequency, the process returns to S302 and the operation of S302 is repeated (S303).

  When the counter value of the counter unit 304 exceeds the total frequency, the process proceeds to S304, where the color rejection unit 303 determines color data to be excluded from the data table of the color storage unit 302, and the color data is determined. The total number of data set in advance is made equal to the total number of all data (S304).

  Here, a method for determining color data to be rejected by the color rejecting unit 303 will be described.

  The color rejecting unit 303 calculates the center of gravity position of the three-dimensional space formed by the center of gravity calculating unit 306 based on all the color data of each data table of each pixel of the color storing unit 302 (this center of gravity is assumed to be G). Let As described above, the color data includes three values for classifying colors, and is a three-dimensional vector represented by these three values. The center-of-gravity calculation unit 306 calculates these center-of-gravity position coordinates based on the distance in the three-dimensional coordinates of all these color data.

  The color rejection unit 303 extracts color data previously rejected from the color data stored in the rejection data storage unit 307 as an exclusion value P, passes through the center of gravity G calculated by the center of gravity calculation unit 306, and uses the vector GP as a normal line. The three-dimensional space formed by the color data of each pixel is divided into two regions by the plane.

  The color rejection unit 303 excludes color data belonging to an area different from the previous rejection value P and distant from the barycentric position coordinates G from among the color data in the data table of the color storage unit 302. Determine as data.

  However, when there is no previous exclusion value P, color data at a position farthest from the center-of-gravity position coordinates G in the data may be simply selected.

  Whether or not it belongs to a region different from the previous rejection value P can be easily calculated by using the inner product of the vector GP and the vector GT if the coordinate to be examined is T. That is, when the inner product is 0 or less, the coordinate T exists in a region different from the previous rejection value P.

  After the color data rejecting process by the color rejecting unit 303, the color rejecting unit 3030 stores the rejected color data in the rejecting data storage unit 307 for each pixel (S304).

  After the color data to be rejected by the color rejecting unit 303 is stored and the rejected color data is stored, the color representative value extracting unit 305 obtains the barycentric position coordinates from all the data tables in the color storing unit 302. Color data at a position closest to G is extracted as a representative value (S305).

  The representative value extracted by the color representative value extraction unit 305 is given to a subsequent image generation unit (not shown) to generate a background image (S306). Then, after the color representative value extraction unit 305 extracts the representative value, the process returns to S302 and is repeated.

(3) Effects of Embodiment 3 As described above, according to the present embodiment, the same effects as described in Embodiment 1 can be achieved and a color background image can be created.

  In addition, according to the present embodiment, the color rejection unit 303 does not individually reject the three types of color signals (red, green, and blue), but rejects the color data for each pixel as one data. Thus, the hue of the image expressed in the background image can be expressed with a hue close to that at the time of shooting. This is because if the three types of color signals are individually rejected, the combined value of the color signals may be different from the hue at the time of shooting, but this embodiment avoids such a problem. Can do.

  Further, according to the third embodiment, when the color rejection unit 303 determines from the area away from the rejection value P previously rejected in the rejection data storage unit 307, the same value is continuously input. Even if the value is out of the range, the entire image is not rejected as noise, and can follow the change in the captured image.

  Next, examples other than the above-described first to third embodiments will be described.

(1) In the above-described first embodiment, the class (median value) to which the central data belongs is used as the representative value. However, the present invention is not limited to this. For example, the frequency average value or the mode value is used as the representative value. May be used.

(2) In the third embodiment described above, the color information is handled as three values of “red, green, and blue”, that is, the RGB color system is used. However, the present invention is not limited to this. For example, an XYZ color system may be used, or a uniform color space may be used. For example, an L * a * b * color space or an L * u * v * color space can be applied to the uniform color space.

(3) In the above-described third embodiment, the color data is a three-dimensional vector and the distance in the three-dimensional coordinates is used. However, the distance is not limited to this, and for example, the Mahalanobis distance in consideration of the dispersion of the color data May be used.

(4) Moreover, in Example 1 and Example 2, the exclusion part 103 reduces one frequency from the appearance frequency histogram of each pixel of the accumulation | storage part 102, when a counter value exceeds total frequency. It was decided to perform a waste treatment.

  However, the present invention is not limited to this. For example, in addition to the set total frequency, a rejection execution value greater than the total frequency is set, and when the counter value reaches the rejection execution value, the rejection unit 103 However, the frequency corresponding to the difference between the rejection execution value and the total frequency may be rejected at a time. In this case, the data to be rejected may be, for example, rejecting frequencies that are half of the frequencies related to rejection in order from the lowest class side and rejecting the remaining frequencies in order from the maximum class side.

(5) In the above-described second embodiment, the median value change detection unit 202 detects a change in the median value by combining the results of comparing the accumulated value I and the rejection value R with the median value B, respectively. explained.

  However, without being limited thereto, for example, the median value change detection unit 202 first determines whether or not the accumulated value I and the median value B are equal, and if they are equal, the median value of the pixel is determined. The median value change detection process described in the second embodiment may be performed only when the median value of the pixel in the previous frame is output without updating and the accumulated value I is different from the median value B.

(6) Furthermore, in the above-described third embodiment, by selecting the coordinate T where the inner product of the vector GP and the vector GT is 0 or less, the length of the TG is maximized from the region where the angle TGP is 90 degrees or more. However, the area for searching for the data to be excluded is not limited to this, and may be an area sufficiently separated from the previous rejection value P. For example, a region where the angle TGP is 30 degrees or more may be used, or a region where the distance between TGs is longer than the distance between GPs may be used.

1 is a block diagram illustrating an internal configuration of a background image generation apparatus according to Embodiment 1. FIG. 3 is a flowchart illustrating the operation of the background image generation apparatus according to the first embodiment. FIG. 10 is a block diagram illustrating an internal configuration of a background image generation apparatus according to a second embodiment. FIG. 9 is an explanatory diagram for explaining three parameters in the second embodiment. It is explanatory drawing for demonstrating the predetermined rule when the median value change detection part 302 in Example 2 updates a parameter. 10 is a flowchart illustrating the operation of the background image generation apparatus according to the second embodiment. FIG. 10 is a block diagram illustrating an internal configuration of a background image generation apparatus according to Embodiment 3. 10 is a flowchart illustrating the operation of the background image generation apparatus according to the third embodiment.

Explanation of symbols

1, 2, 3 ... background image generating device, 102 ... accumulating unit, 103 ... rejecting unit,
104, 304 ... counter unit, 105 ... representative value extracting unit,
302 ... Color storage unit, 303 ... Color rejection unit,
305... Color representative value extraction unit.

Claims (9)

In a background image generation device that generates a background image based on representative data of each pixel obtained by statistical processing from input data of each pixel of an input moving image,
Data storage means for storing the input data of each pixel of the input moving image frame for each pixel;
Data removal means for determining data not used for extraction of representative data from the accumulated input data of each pixel and removing the data when the number of processing frames of the input moving image exceeds a specified value When,
Representative data extracting means for extracting representative data from the data of each pixel accumulated by the data accumulating means after data removal by the data removing means;
A background image generation apparatus comprising: image generation means for generating an image based on representative data of each pixel.   The background image generation apparatus according to claim 1, wherein the data storage unit stores the luminance value of each pixel of the input moving image as an appearance frequency histogram.   The data removing means alternately removes from the appearance frequency histogram of each pixel as data for removing the data of the class first appearing from the minimum class or the data of the class first appearing from the maximum class. The background image generation apparatus according to claim 2. Representative data storage means for storing the representative data of the previous frame extracted by the representative data extraction means for each pixel;
A median value change detecting means for detecting a change between the median value of each pixel of the previous frame and the median value of each pixel of the frame, with the luminance value to which the median frequency of the total frequency of the appearance frequency histogram belongs. Prepared,
When the median value change detection means detects a change in the median value, the data extraction means extracts representative data from the data of each pixel of the data storage means to generate an image,
3. The method according to claim 2, wherein when the median value change detecting means does not detect a median value change, an image is generated based on the representative data of each pixel of the previous frame stored in the representative data storage means. 4. The background image generation apparatus according to 3. The representative data extracted by the data extraction means is the median value,
The median value detection means stores the luminance value of each pixel of the frame stored by the data storage means and the luminance value of the data of each pixel removed by the data removal means in the representative data storage means. 5. The background image generation apparatus according to claim 4, wherein the background image generation device is compared with the median value of the previous frame, and a change in the median value is detected from a combination of the comparison results.   2. The background image generation apparatus according to claim 1, wherein the input moving image is a color image, and the data storage unit stores each value obtained by classifying the color of each pixel into three as color data.   The data removal means calculates the centroid position from the color data of each pixel stored in the data storage means using the color data as a three-dimensional vector, and removes the color data having the maximum distance from the centroid position. The background image generation device according to claim 6, wherein the background image generation device is determined as data to be processed.   The data removing unit removes the color data that exists in an area sufficiently away from the previous removal data removed by the previous data removal processing in the processing of the pixel and has the maximum distance from the barycentric position. The background image generation apparatus according to claim 7, wherein the background image generation apparatus is determined as data. 9. The background image generation apparatus according to claim 7, wherein the data extraction unit uses the color data closest to the gravity center position as the representative data.

Images