JP2015088132A - Image processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image processor capable of highly accurately detecting a change area resulting in a detection object of a person or the like from an input image obtained by imaging a monitoring area.SOLUTION: The image processor includes an image acquisition par 20 for sequentially acquiring an input image, an image processing part 40 for detecting a detection object in the input image on the basis of a comparison result between the input image and a reference image, and a storage part 30 for storing background data comprising a plurality of pixel values generated by using past images acquired at past different times, and the image processing part 40 includes reference image generation means 42 for generating a reference image at a similar pixel value similar to the pixel value of the input image among pixel values of the background data.

Description

  The present invention relates to an image processing apparatus that sequentially processes images captured in a monitoring area and determines whether or not a detection target such as a person exists in the monitoring area.

  A technique for acquiring an input image obtained by capturing a monitoring area and detecting a detection target such as a person based on a change area extracted from the input image has been developed. In such a detection process, a change in brightness due to sunshine fluctuations may be extracted as a change region, and such a change region may lead to erroneous detection of a detection target.

  Therefore, a technique for preventing the change area from being extracted due to fluctuations in sunshine has been disclosed (see Patent Document 1). In this technique, an area where the difference between the input image and the reference image is a predetermined value or less is set as an area where a gradual change such as sunshine fluctuation occurs, and the pixel value of the input image is set as the pixel value of the new reference image for the area Update. In this way, the conventional technology generates a reference image so as to follow a gentle change in brightness in the monitoring area, and compares the input image with the reference image to extract the changed area, thereby eliminating the influence of sunshine fluctuations. It is what.

Japanese Patent Laid-Open No. 01-211083

  By the way, when a planting shake occurs due to wind or the like, the pixel value may be repeatedly changed at a time interval shorter than the sunshine fluctuation. In a reference image that follows a gradual change such as sunshine fluctuation, it has not been possible to prevent the planting fluctuation in which the pixel value changes quickly in this way from being extracted as a change area.

  Therefore, the present invention provides an image processing apparatus capable of detecting a detection target such as a person with high accuracy from an image including not only a gradual change such as sunshine fluctuation but also a temporal change such as planting shake. The purpose is to provide.

  One aspect of the present invention includes an image processing unit that sequentially acquires an input image, and an image processing unit that detects a detection target in the input image based on a comparison result between the input image and a reference image. A storage unit that stores background data composed of a plurality of pixel values generated using past images acquired at different times in the past, and the image processing unit includes pixel values of the background data Among these, the image processing apparatus includes a reference image generation unit that generates the reference image with a similar pixel value similar to the pixel value of the input image.

  Here, the image processing unit further includes a caution area setting unit that sets an image area included in the reference image as a caution area, and the reference image generation unit sets the similar pixel value to a pixel in the caution area. It is preferable to generate the reference image as a value.

  In addition, the reference image generation unit does not use the pixel values of the input image, but calculates the pixel values calculated using the pixel values of the plurality of background data in at least some of the regions other than the attention region. It is preferable to generate the reference image.

  In addition, it is preferable that the attention area setting unit sets the attention area according to a speed of temporal change of a pixel value of the input image.

  Further, it is preferable that the attention area setting means further sets the attention area according to an edge density of the input image.

  It is preferable to provide background data updating means for updating the oldest pixel value among a plurality of pixel values of the background data using the input image.

  ADVANTAGE OF THE INVENTION According to this invention, the image processing apparatus which can detect the change area | region resulting from detection targets, such as a person, from the input image which imaged the monitoring area | region with high accuracy can be provided.

It is a block diagram which shows the structure of the image processing apparatus in embodiment of this invention. It is a figure explaining the attention area and the production | generation method of a reference | standard image in this Embodiment. It is a flowchart of the image processing in this embodiment.

  Embodiments of the present invention will be described with reference to the drawings. The image processing apparatus according to the present embodiment processes an image obtained by capturing an outdoor monitoring area. The image processing apparatus according to the present embodiment detects a person who has entered the monitoring area from the outside of the monitoring area.

  In the present embodiment, an example will be described in which two types of background changes (background changes) such as planting fluctuations and sunshine fluctuations occur in the monitoring area.

  FIG. 1 is a diagram showing a configuration of an image processing apparatus 1 according to an embodiment of the present invention. The image processing apparatus 1 includes an image acquisition unit 20, a storage unit 30, and an image processing unit 40. In the present embodiment, each unit is described as the image processing apparatus 1, but the image acquisition unit 20, the storage unit 30, and the image processing unit 40 are separated from each other, and necessary control signals are made using communication technology. Etc. may be communicated. In the present embodiment, the imaging device 2 and the output unit 3 are separate from the image processing device 1, but may be incorporated in the housing of the image processing device 1. In addition, you may employ | adopt not only this but another housing structure.

  The imaging device 2 images the monitoring area. The imaging apparatus 2 includes an optical system, an imaging element such as a CCD element or a C-MOS element, an optical system component, an analog / digital converter, and the like. The imaging device 2 is fixedly installed and the field of view is kept constant. The imaging device 2 sequentially captures images at a constant time interval (for example, 1/5 second). The image is represented as a digital image having pixel values of 0 to 255, for example. Note that the imaging device 2 may generate a color image.

  The image acquisition unit 20 is an interface that sequentially acquires images captured by the imaging device 2. The acquired image is stored in the storage unit 30 as an input image. The input image is a processing target in the image processing unit 40. In the present embodiment, the image acquisition unit 20 is connected to the imaging device 2.

  In the present embodiment, the image acquisition unit 20 is connected to the imaging device 2 and sequentially acquires images in accordance with the imaging timing of the imaging device 2. However, the present invention is not limited to this, and the image acquisition unit 20 is connected to a storage device (not shown) provided outside the image processing device 1, and images captured by the imaging device 2 stored in the storage device are sequentially acquired. You may make it do.

  The storage unit 30 includes a semiconductor device such as a ROM (Read Only Memory) and a RAM (Random Access Memory), and a memory device such as a hard disk. The storage unit 30 can be accessed from the image processing unit 40. Main information stored in the storage unit 30 includes a previous input image 31, attention area data 32, a previous reference image 33, and background data 34. Although not shown, the storage unit 30 also stores various programs and parameters for realizing each process of the image processing apparatus 1.

  Here, the main information memorize | stored in the memory | storage part 30 is demonstrated easily. First, the previous input image 31 is an image used for calculating an inter-frame difference value from the current input image by the attention area setting unit 41 described later. In the present embodiment, the previous input image 31 is the input image one frame before the input image currently being processed, but it may be the input image before a predetermined number of frames. The attention area data 32 is data used when the attention area is set by the attention area setting means 41 described later. In the present embodiment, the value of the attention area counter set for each pixel is the attention area data 32. The attention area counter will be described later. The previous reference image 33 is a reference image generated one frame before by reference image generation means 42 described later. The previous reference image 33 is used when the calculation pixel value is calculated by the reference image generation means 42 described later. The background data 34 is used when a reference image is generated by reference image generation means 42 described later. The background data 34 is data including pixel values of images (past images) captured at a time point past the input image capturing time. Here, as the background data 34, pixel values of a plurality of past images having different imaging times for each pixel of the input image are stored. In the present embodiment, an example will be described in which pixel values of five past images are stored in association with each pixel of the input image. However, the present invention is not limited to this, and the number of pixel values of the past image may be changed. The background data 34 is appropriately updated by background data updating means 45 described later.

  The image processing unit 40 is configured by an arithmetic device such as a CPU (Central Processing Unit), a DSP (Digital Signal Processor), or an MCU (Micro Control Unit), and reads and executes various programs stored in the storage unit 30. The image processing unit 40 sequentially processes the input images acquired by the image acquisition unit 20. Note that the input image may be processed in order of shooting time for each frame, or may be processed every several frames.

  As shown in FIG. 1, the image processing unit 40 includes modules of an attention area setting unit 41, a reference image generation unit 42, a change area extraction unit 43, a person determination unit 44, and a background data update unit 45.

  The attention area setting means 41 sets, as the attention area, an area in which the pixel value changes rapidly in time such as planting shake from the input image acquired by the image acquisition unit 20. The attention area setting unit 41 outputs the set attention area to the reference image generation unit 42.

  Here, the region in which the pixel value changes rapidly in time means a region in which the pixel value of one pixel corresponding to each other changes more frequently than the sunshine fluctuation in a plurality of input images acquired at different times. To do. The sunshine fluctuation is a change in the pixel value caused when the monitoring area is exposed to sunlight or becomes cloudy.

  For example, the attention area setting unit 41 extracts, as the attention area, an area in which the pixel value changes due to the planting in the monitoring area being shaken by a wind or the like (planting fluctuation). More specifically, when planting sways in the wind, the leaves move repeatedly. For this reason, the area | region where the planting shake was imaged has the characteristic that the change of a temporal pixel value arises frequently on an image.

  The attention area setting unit 41 determines whether there is a temporal change in the pixel value for each pixel of the input pixel. Specifically, the attention area setting unit 41 calculates, for each pixel of the input image, an absolute difference value between the pixel value and the pixel value of the corresponding pixel of the previous input image 31 stored in the storage unit 30. Then, a pixel whose calculated difference absolute value is equal to or greater than a predetermined difference threshold is determined as “a pixel having a difference between frames”. For example, when the luminance range of the input image is 0 to 255, the difference threshold is preferably set to 8 or more. Note that the attention area setting unit 41 stores the input image in the storage unit 30 as the previous input image 31 and sequentially updates the previous input image 31 in order to execute this process. Specifically, after obtaining the inter-frame difference for the input image, the previous input image 31 stored in the storage unit 30 is updated with the input image.

  Then, the attention area setting unit 41 extracts pixels that frequently satisfy the condition that “the interframe difference is equal to or greater than a predetermined difference threshold” as the attention area. Specifically, the attention area setting unit 41 determines, for each pixel, whether or not “the interframe difference is equal to or greater than a predetermined difference threshold”, and sets the condition that “the interframe difference is equal to or greater than the predetermined difference threshold”. When it is satisfied, the attention area counter of the pixel is counted up by N (for example, 20). However, the attention area counter sets the maximum value N_MAX1 (for example, 100) as an upper limit, and limits the maximum value N_MAX1 when the maximum value N_MAX1 is exceeded by counting up. Further, when the pixel to be determined is “the difference between frames is less than a predetermined difference threshold”, the attention area setting unit 41 sets the attention area counter of the pixel to M (M <N, for example, 1) Count down. However, the attention area counter sets 0 as a lower limit, and limits it to 0 when it falls below 0 by counting down. Here, the attention area setting unit 41 stores the value of the attention area counter for each pixel in the storage unit 30 as the attention area data 32.

  The attention area setting unit 41 refers to the attention area data 32 in the storage unit 30 and determines whether or not the attention area counter is equal to or greater than a predetermined count threshold value for each pixel. For example, the count threshold is set to 40. The attention area setting unit 41 sets, as the attention area, a set of pixels for which it is determined that the attention area counter is equal to or greater than a predetermined count threshold value.

  According to the above processing, the attention area counter of the target pixel becomes equal to or greater than the count threshold when a state satisfying the condition of the attention area (interframe difference is equal to or larger than the predetermined difference threshold) occurs to some extent. Extracted as an attention area. That is, the attention area counter is counted up with a large count-up value when the condition of the attention area is satisfied, and is gradually decreased with a countdown value smaller than the count-up value when the condition of the attention area is not satisfied. As a result, the attention area setting unit 41 includes the attention pixel in the attention area when the condition of the attention area has been satisfied in the past even if the change in the pixel value of the attention pixel temporarily falls. Shall.

  For example, if the pixel value has changed due to planting fluctuations, the area that has continuously met the condition of the attention area in the past will remain for a while, even if the wind temporarily stops and the planting fluctuation temporarily stops Is extracted as the attention area. Thereby, the caution area | region setting means 41 is maintained as a caution area | region in the timing which a wind blows again and a planting sway occurs, when planting sway is settled temporarily.

  Note that the changes extracted as the attention area are not limited to planting shaking, but flag shaking (mainly the edge of the flag), shaking planting shadows, daylight leaking from shaking planting, and illumination. It may be a change in pixel value due to blinking or the like.

  Further, in the case of planting shaking, since a plurality of leaves are overlapped, there is a further feature that on the image, the overlapping leaf portions appear as a fine pattern. In view of this, it is possible to extract a region in which a pixel value frequently changes and a fine pattern appears as the attention region.

  In this case, the attention area setting unit 41 determines whether or not an image is captured as a fine pattern for each pixel of the input image. Specifically, the attention area setting unit 41 determines whether or not pixels having strong edges are dense (edge density is high) in the input image. First, the attention area setting unit 41 generates an edge image of an input image (hereinafter referred to as an input edge image). The input edge image can be obtained by binarizing the edge intensity calculated using a spatial differential filter such as a Sobel filter at a predetermined edge threshold value or more. For example, the edge threshold is preferably set to 10 or more. Then, the attention area setting unit 41 determines, for each pixel of the obtained input edge image, whether or not the average value of a predetermined range (for example, 5 × 5 pixels) of the pixel is equal to or greater than a predetermined edge density threshold. . The attention area setting unit 41 determines that the pixel whose average value is equal to or higher than the edge density threshold is “a pixel with high edge density”. For example, the edge density threshold is preferably set to 10 or more.

  Then, the attention area setting unit 41 extracts, as the attention area, a pixel that frequently satisfies the condition that “the difference between frames is equal to or greater than a predetermined difference threshold” and “the pixel having a high edge density”. Specifically, the attention area setting unit 41 determines, for each pixel, whether or not “the interframe difference is equal to or greater than a predetermined difference threshold” and “the edge density is high”. The attention area counter is counted up by K (for example, 20). However, the attention area counter sets the maximum value N_MAX2 (for example, 100) as an upper limit, and limits the maximum value N_MAX2 when the maximum value N_MAX2 is exceeded by counting up. In addition, the attention area setting unit 41, when the pixel to be determined does not satisfy the condition that “the difference between frames is equal to or greater than a predetermined difference threshold” or “the edge density is high”, the attention area counter of the pixel Is counted down by L (L <K, for example, 1). However, the attention area counter sets 0 as a lower limit, and limits it to 0 when it falls below 0 by counting down. Here, the attention area setting unit 41 stores the value of the attention area counter for each pixel in the storage unit 30 as the attention area data 32.

  The attention area setting unit 41 refers to the attention area data 32 in the storage unit 30 and determines whether or not the attention area counter is equal to or greater than a predetermined count threshold value for each pixel. For example, the count threshold is set to 40. The attention area setting unit 41 sets, as the attention area, a set of pixels for which it is determined that the attention area counter is equal to or greater than a predetermined count threshold value.

  According to the above processing, considering not only an image area in which pixel values frequently change but also an image area in which a fine pattern is reflected, an image area in which planting shake is imaged is used as an attention area. It can be extracted with high accuracy.

  In the present embodiment, the attention area setting unit 41 sets the attention area based on the image feature, but the attention area setting method is not limited to this. For example, the user may manually set the attention area. In this case, an operation input unit (not shown) is newly provided for the user to select an image area corresponding to the attention area in the input image. Then, the operation input unit outputs the area selected by the user to the attention area setting unit 41. The attention area setting unit 41 sets the area selected by the user as the attention area. Thereafter, the attention area setting unit 41 stores the set attention area in the storage unit 30. In this case, the reference image generation unit 42 described later may read the attention area from the storage unit 30. Further, in the present embodiment, the condition of the attention area is that “the pixel value frequently changes and a fine pattern is reflected”. However, the attention area is that “the fine pattern is reflected”. It is good also as conditions.

  The reference image generation unit 42 generates a reference image using the background data 34 stored in the storage unit 30. Specifically, the reference image generation unit 42 generates a reference image by applying similar pixel values within the attention region set by the attention region setting unit 41. That is, in the attention area, a reference image suitable for a fast change like a planting shake is generated. In addition, the reference image generation unit 42 generates a reference image using a calculation pixel value different from the similar pixel value for an area other than the attention area. That is, for regions other than the attention region, a reference image suitable for a gradual change such as sunshine fluctuation is generated.

  Hereinafter, a reference image generation method according to the present embodiment will be described. First, the reference image generation unit 42 reads the background data 34 stored in the storage unit 30. Then, the reference image generation unit 42 refers to the input image and the background data 34 and sets the pixel value of the corresponding input image among the pixel values of the plurality of past images corresponding to the pixel values of the input image. The closest pixel value is specified as the similar pixel value. Specifically, the reference image generation unit 42 obtains absolute differences between the pixel values of the plurality of past images and the pixel values of the corresponding input image, respectively. Then, the reference image generation unit 42 specifies the pixel value of the past image that minimizes the obtained difference absolute value as the similar pixel value. Here, the reference image generation unit 42 obtains similar pixel values for the pixels in the attention area set by the attention area setting unit 41. That is, the reference image generation means 42 specifies a pixel value that is as close as possible to the pixel value of the input image from the background data 34 for an attention region that is likely to change quickly in time, such as a planting shake.

  The similar pixel value may be a pixel value that is as close as possible to the pixel value of the input image, not the pixel value of the past image that minimizes the absolute difference between the pixel value of the past image and the pixel value of the input image. Also good. For example, the pixel value of the past image having the second smallest absolute difference between the pixel value of the past image and the pixel value of the input image may be used as the similar pixel value.

  Next, the reference image generation unit 42 refers to the background data 34 and calculates a calculation pixel value using the pixel value of the past image. The calculated pixel value is a pixel value that follows a gradual change such as sunshine fluctuation by calculating the pixel value of the past image without using the input image currently being processed.

Specifically, the reference image generation unit 42 calculates a value obtained by performing a recursive filter for each pixel of the input image as a calculation pixel value. The calculated pixel value by the recursive filter is obtained by the following formula.
Calculated pixel value = Pixel value of previous reference image × R + Pixel value of latest past image × (1-R)

  Here, the “pixel value of the previous reference image” used in this equation is the pixel value of the previous reference image 33 corresponding to the pixel position of the input image to be processed. For this reason, the reference image generation means 42 stores the previous reference image 33 in the storage unit 30 and further updates it in order to execute this process. Specifically, after generating the reference image, the reference image generating unit 42 updates the previous reference image 33 stored in the storage unit 30 with the reference image generated this time. The “pixel value of the latest past image” is the pixel value of the latest past image in the background data 34 associated with the pixel value of the input image to be processed.

  The reference image generation unit 42 calculates a calculation pixel value for a pixel outside the attention area set by the attention area setting unit 41. That is, the reference image generation unit 42 uses the background data 34 for a region where there is a high possibility that a gradual change such as a sunshine fluctuation is likely to occur instead of a rapid change such as a planting shake. Then, a pixel value that follows a gentle change is calculated as a calculation pixel value.

  In the present embodiment, the calculation pixel value is calculated by applying the recursive filter. However, the present invention is not limited to this, and is stored in the storage unit 30 as the background data 34 for each pixel of the input image. An average value of pixel values of a plurality of past images may be calculated, and the average value may be used as a calculation pixel value. If the change is gradual in time like sunshine fluctuation, the average value of the pixel values of a plurality of past images can be adopted as the pixel value that follows the change.

  Next, the reference image generation unit 42 generates a reference image using the similar pixel value and the calculated pixel value. Specifically, the reference image generation unit 42 sets the similar pixel value to the pixel value of the reference image in the attention area set by the attention area setting unit 41. Further, the reference image generation means 42 sets the calculated pixel value to the pixel value of the reference image outside the attention area set by the attention area setting means 41. The reference image generation means 42 generates a reference image by performing these processes for all pixels.

  In order to compare the input image with the reference image and extract the change area depending on the detection target with high accuracy, the reference image is extracted as a change in background change (planting fluctuation, sunshine fluctuation) that may occur in the input image. It is preferable that the image be prevented from being used. In order to generate such a reference image, the reference image generation means 42 identifies an area that is likely to cause a fast change such as planting shake as a caution area, and the inside of the caution area is temporally fast. A reference image that absorbs the change is generated. In addition, the reference image generation unit 42 generates a reference image that absorbs a gradual change such as sunshine fluctuation outside the attention area. This makes it possible to extract a change area depending on the detection target with a high degree of accuracy even for input images that include fast changes such as planting fluctuations and slow changes such as sunshine fluctuations. Can be generated.

  More specifically, planting fluctuation is a fast-changing change in time in which the position of the planting itself in the monitoring area does not move and the leaf part is repeatedly moved by the wind. There is a high possibility that the value is captured in the image. Therefore, it is possible to generate a reference image appropriate for planting fluctuation by using similar pixel values. On the other hand, since the sunshine fluctuation is a more gradual change with time, the change can be easily predicted by a time series filter such as a recursive filter. For this reason, it is possible to generate an appropriate reference image against a gradual change such as a sunshine fluctuation by using the calculated pixel value.

  In the present embodiment, the reference image is generated with the calculated pixel value for the pixels other than the attention area. However, at least a part of the pixels other than the attention area undergoes a gradual change such as sunshine fluctuation. Only a region to be used may be a reference image to which the calculated pixel value is applied. For example, for a region that is always shadowed by some object and hardly changes in time, the pixel value of the input image or background image may be used as the pixel value of the reference image as it is.

  In this embodiment, the reference image is generated by using the similar pixel value and the calculation pixel value separately in the attention area and the area other than the attention area. However, the reference image may be generated with the similar pixel value regardless of the attention area. .

  The change area extraction unit 43 compares the input image acquired by the image acquisition unit 20 with the reference image generated by the reference image generation unit 42 and extracts a change area. In the present embodiment, the change area extraction unit 43 calculates, for each pixel, the similarity between the input image and the reference image by the normalized correlation for the pixel value pattern of the pixel to be determined and the surrounding pixels. . And the change area extraction means 43 determines whether it is a pixel (change pixel) with a change based on the presence or absence of a correlation. Then, a set of pixels determined to be changed pixels is extracted as a changed area. Specifically, the change area extraction unit 43 calculates the normalized correlation value C between the input image and the reference image for 25 pixels in a range of 5 pixels × 5 pixels centering on the pixel to be determined. The normalized correlation value C is calculated as follows:

C = (Σ ((I (x) −Ia) × (B (x) −Ba))) / (Id × Bd)
Here, I (x) is the pixel value of the input image corresponding to the pixel represented by the index x out of 25 pixels, Ia is the average value of the pixel values of the input image of 25 pixels, and B (x) is 25 pixels. The pixel value of the reference image corresponding to the pixel represented by the index x, Ba is the average value of the pixel values of the 25-pixel reference image, Id is the standard deviation of the pixel values of the 25-pixel input image, and Bd is 25 pixels. This is the standard deviation of the pixel value of the reference image. The normalized correlation value C represents the degree of similarity between the 25-pixel pixel value pattern of the input image and the reference image, and takes a value of −1 to 1. The closer the normalized correlation value C is to 1, the more similar the pattern to be compared. The farther the normalized correlation value C is from 1, the more the pattern is different. When the normalized correlation value C is −1, the state is completely inverted.

  The change area extraction unit 43 determines that the pixel is a change pixel when the normalized correlation value C obtained for the pixel to be determined is lower than a predetermined correlation threshold. The change area extraction unit 43 determines that the pixel is not a change pixel when the normalized correlation value C obtained for the determination target pixel is higher than a predetermined correlation threshold. In the present embodiment, when the normalized correlation value C is lower than the correlation threshold, it means that a large change due to a person or the like has occurred in the input image. Further, when the normalized correlation value C is higher than the correlation threshold, it means that the input image has not changed, or a change smaller than a change by a person has occurred in the input image. In addition, the reference image used for extraction of the change pixel is generated so that changes due to planting fluctuations, sunshine fluctuations, and the like other than changes due to the detection target are not extracted as change regions. For this reason, the normalized correlation value C is higher than the correlation threshold even when a change caused by planting fluctuation or sunshine fluctuation occurs. For example, the correlation threshold is preferably set to 0.7.

  Similarly, the change area extraction unit 43 determines whether or not all the pixels are change pixels. A set of pixels determined to be changed pixels is set as a changed region. The change area extraction unit 43 outputs the extracted change area to the person determination unit 44 and the background data update unit 45.

  In the present embodiment, the change area extraction unit 43 uses the normalized correlation to extract the change area based on the similarity between the pixel value patterns of the pixel to be determined and the surrounding pixels. It was. However, the present invention is not limited to this, and the change area extraction unit 43 only needs to extract the change area by comparing the input image with the reference image. For example, the change area may be extracted by a process of binarizing the input image and the reference image. Moreover, the change area extraction means 43 may use both the extraction process by the normalized correlation and the extraction process by the difference binarization process.

  The person determination means 44 determines whether or not the change area extracted by the change area extraction means 43 is a person. The person determination unit 44 calculates a feature amount indicating the possibility of being a person for the change area.

  In the present embodiment, the feature quantity to be calculated is the “real area”, “long / short axis ratio”, and “long axis angle absolute value” of the change region. The “real area” of the change area is a feature amount indicating the validity of the change area as the size of the person. The person determination unit 44 obtains the height and width in the real space for the rectangle circumscribing the change area based on the position in the image and the parameters such as the installation height of the imaging device 2 and the depression angle information. Calculate the product as the actual area. The “long / short axis ratio” is a feature amount indicating the validity of the change area as the shape of the person. The person determination means 44 approximates the change region to an ellipse, and calculates the ratio between the short axis length and the long axis length (short axis length ÷ long axis length) of the ellipse as the long / short axis ratio. The “major axis angle absolute value” is a feature amount indicating the validity of the change area as the posture of the person. The person determination unit 44 calculates the absolute value of the inclination of the long axis as the long axis angle absolute value when the vertical direction (Y axis direction) is 0 degree when the change area is approximated to an ellipse.

  In the present embodiment, for simplicity of explanation, the description has been made with three feature amounts. However, the present invention is not limited to this, and various feature amounts may be used.

  Next, the person determination unit 44 determines whether or not the change area is a person based on the calculated feature amount. In the present embodiment, it is determined whether or not the values of “actual area”, “major axis ratio”, and “major axis angle absolute value”, which are the calculated three feature quantities, are appropriate values for a person. Specifically, if the “real area” is 6000 square centimeters or more, the person is determined to be a person. If not, the person is determined not to be a person. If the “long / short axis ratio” is 0.5 or less, the person is determined to be a person. If the “major axis angle absolute value” is 15 degrees or less, the person is determined to be a person, and if not, the person is determined not to be a person.

  Then, when it is determined that the change area is attributed to the person in all of the “real area”, “long / short axis ratio”, and “major axis angle absolute value”, the person determination unit 44 finally determines the change area. Is determined to be by a person. When it is determined that the change area is a person, the person determination unit 44 outputs a determination result to that effect to the output unit 3.

  In the present embodiment, an example has been described in which a person who has entered the monitoring area from the outside of the monitoring area is detected as a detection target. However, the present invention is not limited to this, and a detection target other than a person may be used. For example, the image processing apparatus 1 may detect an article brought into the monitoring area from the outside of the monitoring area as a detection target. The image processing apparatus 1 may detect a vehicle that has entered the monitoring area from the outside of the monitoring area as a detection target. In this case, instead of the person determination unit 44, a unit for determining the detection target using the feature amount representing the feature of each detection target may be provided.

  The background data update unit 45 stores and updates the background data 34 used by the reference image generation unit 42 in the storage unit 30. As described above in the description of the storage unit 30, the background data 34 is image data including pixel values of a past image captured at a time point past the input image capturing time.

  The background data update unit 45 first checks whether or not each pixel of the input image is a pixel determined to be a change pixel by the change region extraction unit 43. Then, the background data updating unit 45 stores the pixel value of the pixel determined not to be a changed pixel by the changed region extracting unit 43 in the storage unit 30 as the pixel value of the latest past image of the corresponding background data 34. . At this time, the pixel value of the oldest past image among the pixel values of the past image already stored in the storage unit 30 is deleted. Further, the background data update unit 45 does not update the background data 34 for the pixels determined by the changed region extraction unit 43 as changed pixels.

  That is, the background data update unit 45 does not update the background data 34 for an area where the pixel value has changed greatly due to a person or the like. In addition, the background data update unit 45 updates the background data 34 for an area where there is no change, an area where a change smaller than a change caused by a person occurs, or an area where planting shake or sunshine fluctuation occurs.

  Note that five past pixel values are not stored as background data 34 immediately after the image processing apparatus 1 is activated. For this reason, the background data update unit 45 starts the image processing apparatus 1 and stores the pixel values of the input image acquired by the image acquisition unit 20 until the pixel values of the past image are stored in each pixel by five. Are sequentially stored as the latest past pixel values.

  In the present embodiment, as the background data 34, pixel values of five past images are stored for each pixel. However, the number of pixel values of the past image may be plural.

  Further, in the present embodiment, the background data update unit 45 updates the background data 34 for an area that is determined not to be a change area by the change area extraction unit 43, but is not limited thereto. The background data update means 45 may update the background data 34 for areas other than the change area determined by the person determination means 44 as a person and not update the background data 34 for other areas. In this case, the person determination unit 44 outputs the change area determined as a person to the background data update unit 45 via the change area extraction unit 43. Then, the background data update unit 45 may update the background data 34 for each pixel of the input image with respect to a pixel that is not a change area determined as a person.

  When the signal indicating that the change area is a person is input from the person determination unit 44, the output unit 3 outputs an abnormal signal to an alarm unit (not shown), and an abnormality is detected in the surroundings by sounding a buzzer or displaying a warning light. Notify the occurrence. Further, it may be configured to notify the monitoring center of the occurrence of an abnormality by outputting an abnormality signal to a remote monitoring center (not shown) via a communication network such as the Internet.

  Next, image processing by the image processing apparatus 1 according to the present embodiment will be described with reference to FIGS. FIG. 2 is a diagram for explaining a method for generating the attention area and the reference image in the present embodiment. FIG. 3 is a flowchart of image processing in the present embodiment.

  In step S1, the image acquisition unit 20 acquires an image captured by the imaging device 2 as an input image.

  In step S <b> 2, the attention area setting unit 41 sets an attention area in the input image. In the example of FIG. 2, the attention area set by the attention area setting unit 41 is indicated by a white area 35.

  In steps S3 to S7, a reference image generating process is performed by the reference image generating means 42. The reference image generation unit 42 sequentially repeats the processes of steps S4, S5, and S6 for all the pixels included in the image, using the pixels included in the image as the target pixel, and calculates the pixel value of each pixel included in the reference image. decide. That is, the processes from step S3 to S7 are sequentially executed for all the pixels of the reference image.

  In step S4, the reference image generating means 42 determines whether or not the target pixel to be processed is within the attention area. If it is within the attention area, the process proceeds to step S5, and if it is outside the attention area, the process proceeds to step S6.

  In step S <b> 5, the reference image generation unit 42 obtains a similar pixel value using the input image and the background data 34. Then, the reference image generation unit 42 sets the similar pixel value as the pixel value of the pixel of the reference image that is the processing target. In the example of FIG. 2, a pixel a that is a pixel of the input image is a pixel in the attention area. Further, pixel values of five past images corresponding to the pixel a are stored as the background data 34a. The reference image generation unit 42 identifies a2 as the pixel value most similar to the pixel value of the pixel a in the background data 34a, and sets the pixel value as a similar pixel value.

  In step S <b> 6, the reference image generation unit 42 calculates a calculation pixel value using a plurality of pixel values included in the background data 34. In the present embodiment, a recursive filter is applied to a plurality of pixel values included in the background data 34 to calculate a calculation pixel value. Then, the reference image generation unit 42 sets the calculated calculation pixel value as the pixel value of the target pixel of the reference image. In the example of FIG. 2, the pixel b that is a pixel of the input image is a pixel outside the attention area. Further, pixel values of five past images corresponding to the pixel b are stored as the background data 34b. The reference image generation unit 42 calculates a calculation pixel value using a pixel corresponding to the target pixel of the background data 34b. Specifically, the pixel value b5 which is the latest pixel value among the pixel values of the past image and the pixel value b33 of the pixel 33b corresponding to the position of the pixel b in the previous reference image 33 are substituted into the recursive filter equation. To calculate the calculated pixel value.

  When the reference image is generated by the processes from step S3 to step S7, the process proceeds to step S8.

  In step S8, the change area extraction unit 43 compares the input image acquired in step S1 with the reference image, and determines whether each pixel is a change pixel. Then, the change area extraction unit 43 extracts a set of change pixels as a change area.

  In step S9, the person determination means 44 determines whether or not the change area extracted in step S8 is a person.

  In step S10 and step S11, when it determines with a person in step S9, a determination result is output to the external output part 3, and a process is transferred to step S12. If the person is not determined in step S9, the process proceeds to step S12 without outputting to the external output unit 3.

  In step S12, the background data updating unit 45 updates the pixel of the background data 34 corresponding to the pixel that has not been determined as the changed pixel in step S8. When the update process of the background data 34 is completed for all the pixels that are not determined to be changed pixels, the process returns to step S1, and the process for a new input image is started.

  As described above, according to the present embodiment, it is possible to detect a change area caused by a detection target such as a person with high accuracy from an input image obtained by imaging the monitoring area.

  DESCRIPTION OF SYMBOLS 1 Image processing apparatus, 2 Imaging device, 3 Output part, 20 Image acquisition part, 30 Storage part, 31 Previous input image, 32 Caution area data, 33 Previous reference image, 34 Background data, 35 area | region, 40 Image processing part, 41 Attention area setting means, 42 Reference image generation means, 43 Change area extraction means, 44 Person determination means, 45 Background data update means.

Claims (6)

An image processing apparatus comprising: an image acquisition unit that sequentially acquires an input image; and an image processing unit that detects a detection target in the input image based on a comparison result between the input image and a reference image,
A storage unit for storing background data composed of a plurality of pixel values generated using past images acquired at different times in the past;
The image processing unit
An image processing apparatus comprising: a reference image generation unit configured to generate the reference image with a similar pixel value similar to the pixel value of the input image among the pixel values of the background data. The image processing apparatus according to claim 1,
The image processing unit further includes attention area setting means for setting an image area included in the reference image as an attention area.
The image processing apparatus, wherein the reference image generation unit generates the reference image using the similar pixel value as a pixel value of the attention area. The image processing apparatus according to claim 2,
The reference image generation means does not use the pixel values of the input image, but sets the calculated pixel values calculated using the pixel values of the plurality of background data as pixel values of at least a part of the region other than the attention region. An image processing apparatus that generates the reference image. The image processing apparatus according to claim 2 or 3,
The image processing apparatus according to claim 1, wherein the attention area setting means sets the attention area in accordance with a temporal change speed of a pixel value of the input image. The image processing apparatus according to any one of claims 2 to 4,
The image processing apparatus, wherein the attention area setting means further sets the attention area according to an edge density of the input image. An image processing apparatus according to any one of claims 1 to 5,
An image processing apparatus, comprising: a background data updating unit configured to update an oldest pixel value among a plurality of pixel values of the background data using the input image.