JPH0793556A - Mobile object detector - Google Patents

Abstract

PURPOSE:To provide a mobile object detector which is capable of always and stably detecting the changed area of an image by only a mobile object by reducing the influence of the change of an inputted image due to the vibration of an image input system. CONSTITUTION:An intra-small block change calculation part 3 calculates the changes of the picture elements between two images at the inside of each small block and calculates the average value of the change of the picture element every block in plural small blocks set within an inputted image. A changed value statistical calculation part 4 calculates the appearance frequency distribution of the changed value of the picture element within specified time which is longer than the time width of a picture element change calculation for this calculated inter-block change average value and determines the probability density function of the changed value based on the parameter determined from the appearance frequency distribution of this calculated changed value of the picture element. An object changed value calculation part 5 calculates the changed amount caused by only an object by using the probability density and the changed value of the the picture element determined by the changed value statistical calculation part 4. An area extraction part 12 extracts the changed area to be an object from the change detection image calculated in an object changed value calculation part 5.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving object detecting apparatus for automatically detecting, for example, when a moving object other than a train has entered an intrusion prohibited area (monitoring area) near a track in a station. .

[0002]

2. Description of the Related Art Conventionally, in a moving object detecting apparatus of this type, a plurality of detection regions are provided in a monitoring image as a process for detecting a moving object to be detected from continuously input images. , There is a method of calculating a correlation value in a predetermined shift in each detection region, obtaining a motion vector for each detection region by evaluating the value, and further calculating a motion vector of the entire image based on this. .

Regarding this, a method has been proposed in which the reliability of the calculated motion vector in each window (reference area) is examined to improve the calculation accuracy of the motion vector as a whole. Vectors are always sought after uniformly, which makes them less adaptable to the environment.

[0004]

The above-mentioned prior art is based on the correlation calculation in a fixed window. Therefore, the processing cost increases when there is little image change such as vibration. . In addition, since the image change that does not depend on the movement of the target object can be detected only as the change of the entire image, there is a drawback that the entire image is easily affected by the change depending on the position in the image.

Therefore, the present invention provides a moving object detecting apparatus which can reduce the influence of a change in an input image due to vibration of an image input system and the like, and can always stably detect a change area of an image due to only a moving object. The purpose is to do.

[0006]

A moving object detecting apparatus according to the present invention comprises image input means for continuously inputting images in a surveillance area, and a plurality of small images set in the images input by the image input means. In the block area, the pixel change between the two images is calculated inside each small block, and the change calculation means within the small block calculates the average value of the pixel change for each block, and the change calculation means within the small block. For the average change value within a block, the appearance frequency distribution of pixel change values within a specific time longer than the time width of pixel change calculation is calculated, and the change is calculated based on the parameters obtained from the calculated appearance frequency distribution of pixel change values. Change value statistical calculation means for obtaining a probability density function of values, target change value calculation means for calculating a change amount due to only the target object using the pixel change value and the probability density, And it includes a region extracting means for extracting a changed region of interest from the target change-value determining means change detection image calculated by.

Further, the moving object detecting apparatus of the present invention comprises: an image input means for continuously inputting images in the surveillance area; and a plurality of small block areas set in the image input by the image input means. Intra-small-block change calculation means for calculating the pixel change between two images inside each small block and calculating the average value of pixel change for each block, and intra-block change average calculated by this small-block change calculation means For the values, the appearance frequency distribution of pixel change values within a specific time longer than the time width of pixel change calculation is calculated, and the probability density of change values is calculated based on the parameters obtained from the calculated appearance frequency distribution of pixel change values. Expected value of change value obtained by the change value statistical calculation means for obtaining a function and the appearance frequency distribution of pixel change values for each block calculated by this change value statistical calculation means An image change expected value calculation means for obtaining an average value for all blocks to obtain an image change expected value, and if the image change expected value obtained by the image change expected value calculation means is larger than a predetermined threshold value, A set of a plurality of reference areas in which a plurality of blocks are set, a movement vector calculation means for obtaining a movement vector inside the set reference area, and a set of movement vectors in each of the set reference areas An image movement vector calculation means for calculating the entire movement vector, and an image obtained by moving the currently focused one of the two images subject to change detection in the opposite direction of the image movement vector and the other image. Image change recalculating means for recalculating the image change of, and area extraction for extracting a target change area from the change detection image calculated by the image change recalculating means It is and a stage.

Further, the moving object detecting apparatus of the present invention comprises: an image input means for continuously inputting images in the monitoring area; and a plurality of small block areas set in the image input by the image input means. Intra-small-block change calculation means for calculating the pixel change between two images inside each small block and calculating the average value of pixel change for each block, and intra-block change average calculated by this small-block change calculation means For the values, the appearance frequency distribution of pixel change values within a specific time longer than the time width of pixel change calculation is calculated, and the probability density of change values is calculated based on the parameters obtained from the calculated appearance frequency distribution of pixel change values. A predetermined reference area is selected according to a variation value statistical calculation means for obtaining a function and an appearance frequency distribution of pixel variation values for each block calculated by the variation value statistical calculation means. Subregion selection means, movement vector calculation means for obtaining a movement vector inside the reference area selected by the reference area selection means, and movement vectors in each of the set reference areas are collected together to move the entire image. An image movement vector calculation means for calculating a vector,
An image change recalculation means for recalculating the image change between the image obtained by moving the current one of the two images subject to change detection in the opposite direction of the image movement vector and the other image, A region extracting unit that extracts a target change region from the change detection image calculated by the image change recalculating unit is provided.

Further, in the moving object detecting apparatus of the present invention, the image input means for continuously inputting the images in the monitoring area and the plurality of small block areas set in the image input by the image input means are 2 inside each small block
Pixel change calculation means for calculating pixel change between images and calculating an average value of pixel change for each block, and pixel change calculation for the average change value in block calculated by this small block change calculation means Change value statistical calculation that calculates the frequency distribution of pixel change values within a specific time longer than the time width, and calculates the probability density function of the change values based on the parameters calculated from the calculated frequency distribution of pixel change values And the expected value of the change value obtained by the distribution frequency distribution of pixel change values for each block calculated by the change value statistical calculation means If the expected value calculation means and the expected image change value obtained by the expected image change value calculation means are larger than a predetermined threshold value, a plurality of the small blocks are combined. Set the number of reference regions, a movement vector calculation means for obtaining a movement vector in the interior of the set reference region, wherein the motion vector in each reference area, obtains the dispersion value of the direction and magnitude,
When the calculated variance value is less than or equal to a predetermined threshold value, the average value of the movement vectors calculated for each of the reference regions is selected as the image movement vector, and in other cases, weighted interpolation is performed by the position from each of the reference regions. A moving vector interpolating means for calculating a moving vector corresponding to the position in the image, and the two images to be detected for change, which are currently focused, are moved in the opposite direction of the interpolation moving vector. The image change recalculating means for recalculating the image change between the image and the other image, and the area extracting means for extracting the target change area from the change detection image calculated by the image change recalculating means are provided. There is.

[0010]

According to the present invention, when the change amount of the entire image is small, the frequency distribution of the change values between the images within a certain time is calculated for each small block, and the change amount caused by only the target object is calculated. Relatively fast, and
It is less susceptible to the effects of disturbance.

When the amount of change in the entire image is large, the moving vector of the entire image is obtained by detecting the moving vector in some windows (reference areas) in the image, and the image used in the change detection calculation is corrected. However, since those windows that are stationary in time are selected, it is efficient to reduce image changes due to factors other than the target object.

Further, since the image correction method used for the change detection calculation is changed according to the variation of the movement vector for each window, it is possible to realize the processing speed and the correction effect adapted to the image pickup environment.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 2 schematically shows the configuration of, for example, a railroad track monitoring device to which the moving object detection device according to the present invention is applied. That is, for example, the ITV camera 21 as an image input means captures an image of a forbidden area of the railway track 22 in the station as a monitoring area and converts it into an electric signal. The image captured by the ITV camera 21 is the processing device 2
Sent to 3. The processing device 23 continuously captures images in the monitoring area captured by the ITV camera 21, and performs image processing and determination for detecting a moving object (intruding object). As a result, when a moving object is detected, a moving object detection display is displayed on the display unit of the monitoring display device 24, and an alarm sound is output by the alarm unit.

This railroad track monitoring device is a system for automatically detecting when an object other than the train has entered an intrusion prohibited area near the track in the station yard, and notifying the staff. As shown in FIG. 2, an image obtained by obliquely capturing the vicinity of the railroad track inside the station yard is input, and if an object other than the train entering the intrusion prohibition area exists, the staff is notified. In a station, a heavy object called a train frequently moves near the image input system, and the input image vibrates due to the vibration.

In the present embodiment, it is possible to reduce the influence of the image change due to such vibration and detect the changed region of the image due to only the intruding object. Figure 1
Shows the configuration of the processing device 23 in detail.
The processing device 23 includes an image storage unit 2 that stores an image from the image input unit 1 (ITV camera 21 in FIG. 2), a small block change calculation unit 3, a change value statistical calculation unit 4, and a target change value calculation unit 5. , An image change expected value calculation unit 6, a movement vector calculation unit 7, an image movement vector calculation unit 8, an image change recalculation unit 9,
The window selection unit 10, the movement vector interpolation unit 11, the area extraction unit 12, the object determination unit 13, and the operation control unit 14 that controls the operation of each unit. The object determination unit 13 includes a display unit 15 of the monitoring display device 24.
And the alarm unit 16 is connected. FIG. 3 shows the flow of the overall processing of this embodiment, and the processing operation will be described below with reference to this flowchart.

(1) Input of moving image The image in the station yard is input by the image input unit 1 at a constant interval and accumulated in the image accumulation unit 2 in time series. The input image here is assumed to be a grayscale image in which one pixel is represented by 8 bits.

(2) Temporal statistical calculation of difference value The intra-small-block change calculation unit 3 divides an image into a plurality of small block regions, calculates the pixel difference between two images in each small block, and calculates each pixel for each block. Calculate the average value of the pixel differences of. Subsequently, the change value statistical calculation unit 4 calculates the appearance frequency distribution of the difference value within a certain time longer than the time width of the pixel difference for the average value of the difference between blocks, and then calculates the difference average value and the difference variance with respect to time. Find the value. Furthermore, the probability density function of the difference value is obtained from it.

This is because the difference value of the image in the block j at each time point is xi, the appearance frequency distribution of the difference value is H (xji),
Assuming that the time average value is Ej and the time variance value is Vj, and the number of processed images in the fixed time is n,

[0019]

[Equation 1]

[0020]

[Equation 2] Becomes However, Sj is

[0021]

[Equation 3]

It is Here, a normal distribution is assumed as the form of the probability density function φj (xji) of this difference value,

[0023]

[Equation 4]

It is assumed that However, j = 1 to m × m, that is, a small block divides one image into m vertically and horizontally. In addition, the image change expected value calculation unit 6 is used for later processing.
Then, the Ej of the above equation 1 is averaged over all blocks, and the image difference expected value E is obtained as follows.

[0025]

[Equation 5]

If this is not larger than the predetermined threshold value EE of the image difference expected value, the processing shown in (4) described later is performed, and if not, it is shown in (3) described below. Perform processing. The flow of the above processing is shown in the flowcharts of FIGS.

(3) Correction of Difference Value by Time Statistic Next, in the target change value calculation unit 5, the output difference value (target difference) which is due to the influence of the target object is calculated for the difference image in each block as follows. Call the value). That is, when the number of difference pixels in a certain block j is p, 1
If one difference pixel value is dk and the target difference value is ddk, ddk = (1−φj (dk)) * dk (1) Therefore, in each of the m blocks in the image, the above equation (1) is calculated for all the difference pixels k = 1 to p. The flow of processing at this time is shown in the flowchart of FIG.

(4) Pixel correlation statistical calculation This processing is performed by the movement vector calculation unit 7 in a plurality of small blocks, for example, in r reference regions (windows) in which q units are arranged in each of vertical and horizontal units. Then, the moving direction and the distance of the small block to the position where the image correlation is maximized in the continuous image are obtained, and the window moving vector is calculated.

Next, the image movement vector calculation unit 8 calculates the average value of the window movement vectors in each window w as an image movement vector candidate. That is,

[0030]

[Equation 6]

However, the maximum value of rr is r = (m × m) /
(Q × q). The flow of the process of calculating the image movement vector by such window averaging is shown in the flowchart of FIG. Also, the variance value for each of these windows w is calculated by the following equation 7.

[0032]

[Equation 7]

In the above, it is considered that the position where the sum of squares of the differences in pixel values between consecutive images is the minimum in each window is the maximum image correlation, using the block region as a mask. Then, as shown in FIG. 3, when the magnitude | MV | of the variance value obtained in Equation 7 is not larger than the predetermined threshold MVT, M in Equation 6 is used as the following image movement vector, If not, the value described in (5) described later is used.

Next, the image change recalculation unit 9 selects one of the two continuous images to be subtracted, which is currently focused, and which is translated in the opposite direction of the above-mentioned image movement vector. The target difference value is obtained by recalculating the image difference from the image.

In the above, the value Vw obtained by averaging Vj of Equation 2 for m × m blocks in each window is the threshold value VT of the time dispersion value.
Rr smaller windows w are selected and used by the movement vector calculation unit 7. That is, w = {s | VWs <VT |} (2)

(5) Image movement vector calculation by position interpolation The movement vector interpolating unit 11 obtains the variance of the direction and size of the movement vector in each window. As described above, instead of calculating the movement vector of the entire image, weighted interpolation is performed according to the position from each window, and the movement vector according to the position in the image is calculated.

That is, the barycentric coordinates of each window are (x
w, yw), the movement vector of the window is (Mxw, M
yw) and the pixel coordinates of interest are (x, y), the movement vector at this point

[0038]

[Equation 8] Is calculated by the following equations 9 and 10.

[0039]

[Equation 9]

[0040]

[Equation 10]

The flow of the image movement vector calculation processing by such position interpolation is shown in the flowchart of FIG.

(6) Extraction of object area The difference image detected by the processing of (3) or (4) is binarized by the area extraction unit 12 and the circumscribed rectangle of the object is detected. A candidate area of the target object is extracted.

Further, the object determination unit 13 determines whether or not the candidate area is the target by using the shape of the circumscribed rectangle. At this time, an object obtained experimentally in advance is used as an object smaller than the train and larger than the noise in the image. If any of the above is suitable as the target object, the alarm unit 16 issues an alarm to the surveillance staff, and the display unit 15 further displays, for example, a portion where the target object (moving object) is detected in the image. Enclose and display in a rectangular frame.

[0044]

As described above in detail, according to the moving object detecting apparatus of the present invention, when the change amount of the entire image is small, the frequency distribution of the change value between images within a certain time is calculated for each small block. However, since the amount of change caused by only the target object is obtained by this, it is relatively fast and is less susceptible to the influence of disturbance.

When the amount of change in the entire image is large, the moving vector of the entire image is obtained by detecting the moving vector in some windows (reference areas) in the image, and the image used for the change detection calculation is corrected. However, since those windows that are stationary in time are selected, it is efficient to reduce image changes due to factors other than the target object.

Furthermore, since the image correction method used for the change detection calculation is changed according to the variation of the movement vector for each window, it is possible to realize the processing speed and the correction effect adapted to the image pickup environment.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a processing device in a railway track monitoring device to which a moving object detection device according to an embodiment of the present invention is applied.

FIG. 2 is a diagram schematically showing the configuration of a railway track monitoring device.

FIG. 3 is a flowchart illustrating an overall processing flow.

FIG. 4 is a flowchart illustrating a flow of difference value time statistical calculation processing.

FIG. 5 is a flowchart illustrating the flow of difference value frequency calculation processing in the time direction for blocks at each time point.

FIG. 6 is a flowchart illustrating the flow of difference average and variance calculation processing in the time direction for blocks at each time point.

FIG. 7 is a flowchart illustrating the flow of image difference expected value calculation processing.

FIG. 8 is a flowchart illustrating the flow of difference value correction processing based on time statistics.

FIG. 9 is a flowchart illustrating the flow of image movement vector calculation processing based on window averaging.

FIG. 10 is a flowchart illustrating the flow of image movement vector calculation processing by position interpolation.

[Explanation of symbols]

21 ... ITV camera, 22 ... railroad tracks inside the station, 2
3 ... Processing device, 24 ... Monitoring display device, 1 ... Image input unit, 2 ... Image storage unit, 3 ... Small block change calculation unit, 4 ... Change value statistical calculation unit, 5 ... Target change value calculation unit, 6 ... Image change expected value calculation unit, 7 ... Movement vector calculation unit, 8 ... Image movement vector calculation unit, 9 ... Image change recalculation unit, 10 ... Window selection unit, 11 ...... Movement vector interpolation unit, 12 ...... Region extraction unit, 13 ...... Object determination unit, 14 ...... Motion control unit, 15 ...... Display unit, 16 ......
Alarm section.

Claims (4)

[Claims] 1. An image input unit for continuously inputting images in a monitoring region, and a plurality of small block regions set in the images input by the image input unit, between two images in each small block. Pixel change calculation means for calculating the average value of pixel change for each block, and the time width of pixel change calculation for the average change value in block calculated by this small block change calculation means. A change value statistical calculation means for calculating an appearance frequency distribution of pixel change values within a longer specific time, and for obtaining a probability density function of change values based on a parameter obtained from the calculated appearance frequency distribution of pixel change values; A target change value calculating means for calculating a change amount caused only by the target object using the pixel change value and the probability density, and a change detecting value calculated by the target change value calculating means. Moving object detection apparatus being characterized in that comprises a region extracting means for extracting a changed region of interest from the image. 2. An image input means for continuously inputting images in the monitoring area, and a plurality of small block areas set in the image input by the image input means, between two images in each small block. Pixel change calculation means for calculating the average value of pixel change for each block, and the time width of pixel change calculation for the average change value in block calculated by this small block change calculation means. A change value statistical calculation means for calculating an appearance frequency distribution of pixel change values within a longer specific time, and for obtaining a probability density function of change values based on a parameter obtained from the calculated appearance frequency distribution of pixel change values; , The average value for all blocks of the expected value of the change value obtained by the appearance frequency distribution of the pixel change value for each block calculated by this change value statistical calculation means And an image change expected value calculating means for obtaining the image change expected value and an image change expected value calculated by the image change expected value calculating means is larger than a predetermined threshold value. The reference vector is set, and the movement vector calculating means for obtaining the movement vector inside the set reference area, and the movement vector in each of the set reference areas are collected to calculate the movement vector of the entire image. Image movement vector calculation means, and of the two images subject to change detection, the image change between the image obtained by moving the currently focused one in the opposite direction of the image movement vector and the other image is recalculated. Image change recalculation means, and area extraction means for extracting a target change area from the change detection image calculated by the image change recalculation means. Moving object detection apparatus. 3. An image input means for continuously inputting images in a monitoring area, and a plurality of small block areas set in the image input by the image input means, between two images in each small block. Pixel change calculation means for calculating the average value of pixel change for each block, and the time width of pixel change calculation for the average change value in block calculated by this small block change calculation means. A change value statistical calculation means for calculating an appearance frequency distribution of pixel change values within a longer specific time, and for obtaining a probability density function of change values based on a parameter obtained from the calculated appearance frequency distribution of pixel change values; A reference area selecting means for selecting a predetermined reference area according to the appearance frequency distribution of pixel change values for each block calculated by the change value statistical calculating means; A movement vector calculation unit that obtains a movement vector inside the reference region selected by the selection unit, and an image movement vector calculation unit that collectively calculates the movement vector in each of the set reference regions. And an image change recalculating means for recalculating the image change between the image obtained by moving the current one of the two images subject to change detection in the opposite direction of the image movement vector and the other image. And a region extraction unit that extracts a target change region from the change detection image calculated by the image change recalculation unit, and a moving object detection apparatus. 4. An image input means for continuously inputting images in a monitoring area, and a plurality of small block areas set in the image input by the image input means, between two images inside each small block. Pixel change calculation means for calculating the average value of pixel change for each block, and the time width of pixel change calculation for the average change value in block calculated by this small block change calculation means. A change value statistical calculation means for calculating an appearance frequency distribution of pixel change values within a longer specific time, and for obtaining a probability density function of change values based on a parameter obtained from the calculated appearance frequency distribution of pixel change values; , The average value of all the expected values of the change values obtained by the appearance frequency distribution of the pixel change values for each block calculated by this change value statistical calculation means And an image change expected value calculating means for obtaining the image change expected value and an image change expected value calculated by the image change expected value calculating means is larger than a predetermined threshold value. And a movement vector calculation means for obtaining a movement vector inside the set reference area, and for the movement vector in each of the reference areas, a variance value with respect to its direction and size is obtained, and the obtained When the variance value is less than or equal to a predetermined threshold value, the average value of the movement vector calculated for each of the reference regions is selected as the image movement vector, and at other times, the weighted interpolation is performed by the position from each of the reference regions, The movement vector interpolator that calculates a movement vector according to the position in the image, and the two images that are the targets of change detection are The image change recalculating means for recalculating the image change between the image moved in the opposite direction of the interpolation movement vector and the other image, and the change area targeted from the change detected image calculated by the image change recalculating means. A moving object detecting device, comprising: