JP2004295416A - Image processing apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image processing apparatus that can precisely detect an object of detection in any motion whether the object of detection moves fast or slow or remains still. <P>SOLUTION: A first moving profile extracting means 51 extract part of profiles varied in a given period from profile images stored in a profile image storing means 4, and a second moving profile extracting means 52 extract part of profiles varied in a period longer than the given period from the profile images stored in the profile image storing means 4, to thereby create moving profile images in the detection times, respectively. First to third search area setting means 61 to 63 set first to third search areas, respectively, in the images created by the first and second moving profile extracting means 51 and 52 or a profile image created by profile extracting means 3. First to third similarity calculating means 71 to 73 calculate similarities with an object of detection in the first to third search areas, respectively. Detection determining means 8 resultantly determine whether the object of detection is present or not. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an image processing apparatus that detects the presence or absence and movement of a detection target such as a person or a vehicle in a detection area from an image of a predetermined detection area.
[0002]
[Prior art]
In a conventional image processing apparatus, a method such as a background subtraction method or an inter-frame difference method has been used to detect the presence or absence and movement of a detection target such as a person or a vehicle from an image in a predetermined detection area.
[0003]
In the method using the background subtraction method, an image in which no detection target exists is captured in advance, and this image is stored as a background image. After that, the image of the detection area is sequentially captured, a difference from the background image is extracted (this processing is called a background difference), and the presence or absence of the detection target is detected based on the feature of the region of the extracted part. Was. For example, the image of the region extracted by the background difference is binarized, and if the size and shape of the image and the template image of the detection target (for example, a person) are within a predetermined range, it is determined that the person is a person, If it is out of the range, it is determined that the person is not a person.
[0004]
In the method using the inter-frame difference method, a difference between an image captured at time (t-1) and an image captured at time t is extracted, and the presence or absence of a detection target is determined based on the feature of the extracted region. Had been detected. In addition, in the inter-frame difference method, the same process as the background difference method is performed to determine whether or not a frame is a detection target (for example, see Patent Document 1).
[0005]
[Patent Document 1]
JP-A-6-201715 (page 3 and FIG. 4)
[0006]
[Problems to be solved by the invention]
Among the image processing apparatuses using the background subtraction method, among the above-described image processing apparatuses, if the background image is an appropriate image, the area to be detected can be accurately detected, but an image without the detection object is captured and stored as the background image. And it is not easy to capture the background image. For example, when a background image is manually picked up, there is a problem that it is necessary to take an image at a timing when the detection target does not exist. In addition, when the brightness of the detection area changes due to an influence of a change in sunshine or the like, the change remains in the image obtained by the background difference. Therefore, in such an environment, every time the brightness of the detection area changes. It was necessary to update the background image, which could not be handled manually. Therefore, a method of automatically capturing a background image has also been proposed.For example, a method in which an average value of images captured for a long time is used as a background image, or updating is slow when a luminance change is large for each pixel in the image, When the size is small, there is a method of updating the background image sequentially for each pixel by updating it quickly.However, in any case, when a person is stationary for a long time in the detection area, the person is buried in the background image. As a result, there is a problem that a person cannot be detected.
[0007]
Further, in the method using the inter-frame difference method, the difference between two images captured at different times is obtained. Therefore, if the detection target moves greatly during that time, the detection target can be accurately extracted. When the movement of the target is small or when the target stops at the same place, there is a problem that only a part of the detection target can be extracted or the detection target cannot be extracted at all. Further, in order to accurately extract the detection target even when the detection target moves slowly, it is conceivable to increase the amount of movement of the detection target between frames by increasing the interval at which an image of the detection area is captured. However, in this case, there is a problem in that a search for a detection target becomes difficult because the position of the detection target changes greatly between frames in a detection target having a large motion.
[0008]
The present invention has been made in view of the above problems, and an object of the present invention is to detect a motion of a detection target regardless of whether the motion is fast, slow, or stationary. An object of the present invention is to provide an image processing device capable of detecting an object with high accuracy.
[0009]
[Means for Solving the Problems]
In order to achieve the above object, according to the first aspect of the present invention, in an image processing apparatus that detects the presence or absence and movement of a detection target such as a person or a vehicle from an image obtained by capturing a predetermined detection area, Image capturing means for capturing images at time intervals, contour extracting means for creating a contour image by extracting a contour from an image captured by the image capturing means, contour image storing means for storing the contour image created by the contour extracting means, and contour image storing First moving contour extracting means for extracting a part of the contour that has changed within a predetermined period based on one or more contour images stored in the means and creating a moving contour image at a detection target time; A second shift for extracting a portion of a contour that has changed within a period longer than the predetermined period based on a plurality of contour images stored in the image storage unit and creating a moving contour image at a detection target time. A first search range setting for setting, as a first search range, a range in which the detection target can be moved from a previous detection position of the detection target with respect to the image created by the contour extraction means and the first moving contour extraction means. Means, first similarity calculating means for calculating the similarity between the contour extracted by the first moving contour extracting means and the detection object in the first search range, and an image created by the second moving contour extracting means On the other hand, a second search range setting means for setting a range smaller than the first search range near the detection position of the previous detection target as the second search range, and a second moving contour in the second search range. A second similarity calculating means for calculating a similarity between the contour extracted by the extracting means and the detection target; Narrower than search range 2 Third search range setting means for setting the range as a third search range, and third similarity calculation means for calculating the similarity between the contour extracted by the current contour extraction means and the detection target in the third search range And detection determination means for determining the presence or absence of a detection target based on the similarities calculated by the first, second, and third similarity calculation means.
[0010]
According to a second aspect of the present invention, in the first aspect of the present invention, the first moving contour extracting means extracts a difference between the contour image at the time of the last two-time imaging stored in the contour image storage means and the contour image at the time of the previous imaging. A first contour image difference means for creating a first difference image, and a second contour image difference extracting means for extracting a difference between the contour image at the time of the previous imaging stored in the contour image storage means and the contour image created by the current contour extraction means. It is characterized by comprising a second contour image difference means for creating a difference image and a common contour extraction means for creating an image obtained by extracting a common part between the first difference image and the second difference image.
[0011]
According to a third aspect of the present invention, in the first aspect of the present invention, the second moving contour extracting means includes a difference between a contour image stored in the contour image storing means at the time of the last two-time imaging and a contour image created by the present contour extracting means. A first contour image difference unit for creating a first difference image from which the contour image has been extracted, and a difference between the contour image at the time of previous imaging stored in the contour image storage unit and the contour image created by the current contour extraction unit. It is characterized by comprising a second contour image differencing means for creating a second difference image, and a contour synthesizing means for creating an image obtained by combining the first difference image and the second difference image.
[0012]
According to a fourth aspect of the present invention, in the first aspect of the present invention, the first similarity calculating means includes a storing means for storing a template image representing a contour of a detection target, and an image created by the first moving contour extracting means. Counting means for scanning within the first search range as an image to be searched, and counting the number of pixels having an outline together with each pixel of the template image and a corresponding pixel of the image to be searched at each scanning point; Means for calculating the similarity by normalizing the count value of the counting means with the total number of pixels of the template image.
[0013]
According to a fifth aspect of the present invention, in the first aspect of the present invention, the second similarity calculating means includes a storing means for storing a template image representing a contour of a detection target, and an image created by the second moving contour extracting means. A first counting unit that scans the second search range as an image to be searched and counts the number of pixels having outlines at each scanning point between each pixel of the template image and a corresponding pixel of the image to be searched; And second counting means for counting the number of pixels having no contour between each pixel of the template image and the corresponding pixel of the image to be searched at each scanning point; and counting the counting result of the first counting means with the template image. Means for calculating the degree of similarity by adding the value normalized by the number of pixels having an outline and the value normalized by the number of pixels having no outline in the template image to the count result of the second counting means. This The features.
[0014]
According to a sixth aspect of the present invention, in the first aspect of the present invention, the third similarity calculating means includes a storage means for storing a template image representing an outline of the detection target, and an image created by the current contour extraction means for searching. A first counting unit that scans the third search range as an image and counts the number of pixels having an outline together with each pixel of the template image and a corresponding pixel of the image to be searched at each scanning point; The second counting means for counting the number of pixels having no contour between each pixel of the template image and the corresponding pixel of the image to be searched at the scanning point; A means for calculating the similarity by adding a value normalized by a certain number of pixels and a value normalized by the number of pixels having no outline of the template image to the count result of the second counting means. To.
[0015]
According to a seventh aspect of the present invention, in the first aspect of the present invention, if the third similarity calculated by the third similarity calculating means is equal to or larger than the first threshold, the detection determination unit stops. And when the third similarity is lower than the first threshold, and the second similarity calculated by the second similarity calculating means is equal to or greater than the second threshold. When it is determined that the detection target is moving slightly, the first similarity is lower than the first threshold and the second similarity is lower than the second threshold, the first similarity is determined. If the first similarity calculated by the similarity calculating means is equal to or more than the third threshold, it is determined that the detection target is moving more.
[0016]
BEST MODE FOR CARRYING OUT THE INVENTION
(Embodiment 1)
First Embodiment A first embodiment of the present invention will be described with reference to FIGS. The image processing apparatus according to the present embodiment is for detecting the presence or absence and movement of a detection target such as a person or a vehicle in a predetermined detection area. FIG. 1 is a block diagram showing a schematic configuration of an image processing apparatus. For example, an imaging unit 1 such as a TV camera installed on a ceiling surface and taking images of a room from the ceiling surface at predetermined time intervals, and an input from the imaging unit 1 Image input means 2 for taking in an image signal to be input, contour extraction means 3 for extracting a contour (edge) of an object based on the image signal input via the image input means to create a contour image, and contour extraction. A contour image storing means for storing an image of the contour extracted by the means; and a contour part which has changed within a predetermined period is extracted based on one or a plurality of contour images stored in the contour image storing means. Then, a first moving contour extracting means 51 for creating a moving contour image at the detection target time, and a range in which the detection target can be moved with respect to the image (moving contour image) created by the first moving contour extracting means 51, 1 search range The first search range setting means 61 to be set, the first similarity calculation means 71 for calculating the similarity between the contour within the first search range and the detection target (for example, a human body), and the contour image storage means 4 A second moving contour extracting means 52 for extracting a portion of the contour that has changed within a period longer than the predetermined period based on the stored plurality of contour images and creating a moving contour image at a detection target time; In the image (moving contour image) created by the second moving contour extracting means 52, a second search range that is smaller than the first search range in the vicinity of the previous detection target detection position is set. Search range setting means 62, second similarity calculation means 72 for calculating the similarity between the contour within the second search range and the detection target (human body), and the detection target time generated by the contour extraction means 3. For the contour image, the previous detection target Third search range setting means 63 for setting a range smaller than the second search range near the outgoing position as a third search range, and similarity between the contour in the third search range and the detection target (human body) Is calculated based on the similarity calculated by the first to third similarity calculators 71 to 73, and whether or not the detection target is present in the detection area is determined. In this case, it is constituted by the detection judging means 8 for detecting the existence position of the detection target. Note that the contour extracting means 3, the first and second moving contour extracting means 51 and 52, the first to third search range setting means 61 to 63, the first to third similarity calculating means 71 to 73, and The detection judging means 8 is constituted by, for example, an arithmetic function of a personal computer.
[0017]
The contour extracting means 3 extracts a contour using, for example, a SOBEL operator. The SOBEL operator is a 3 × 3 filter, and by using a line buffer for three lines, it is possible to extract an outline while inputting an image. A frame memory for storing image signals for one screen is provided. After the image signals for one screen input from the image input means 2 are stored in the frame memory, the contour extracting means 3 reads the necessary image from the frame memory. The signal may be read, the read image signal may be subjected to arithmetic processing using a SOBEL operator, and the extracted contour may be written in the contour image storage unit 4.
[0018]
The first moving contour extracting means 51 performs processing for extracting the contour with high accuracy when the detection target moves largely or the movement is large. For example, as shown in FIG. 2, the first moving contour extraction unit 51 compares the contour image B2 stored in the contour image storage unit 4 at the time of the previous imaging (time (t−1)) with the time of the current imaging (time t). By performing a difference process and a binarization process with the contour image B3, a portion of the contour that has changed within a predetermined period (between the previous imaging and the current imaging) is extracted, and the detection is performed in the detection target time. The moving contour image C1 is output. Incidentally, when the detection target is completely still, there is no difference between the contour image B1 at the time of the previous imaging and the contour image B2 of the present time, so that no contour is extracted from the moving contour image C1.
[0019]
Since the first search range setting means 61 does not know where on the screen the detection target appears before the detection target is detected as shown in FIG. Explore. That is, at the time of the initial operation, the detection target is searched for from all the contours moving in the image W0. Then, once the detection target is detected, the detection position of the detection target at the time of the previous imaging is stored, and the range in which the detection target can move around the detection position at the time of the previous imaging is set as the first search range W1. By limiting the search to the first search range W1, the influence of disturbance is reduced.
[0020]
The first similarity calculating means 71 calculates the similarity between the feature of the contour to be detected (template image) stored in advance in a memory (storage means) (not shown) and the feature of the contour X appearing in the moving contour image C1. Calculate the degree. The first similarity calculating means 71 performs a process for accurately detecting when the movement amount or the movement of the detection target is large. In this case, almost the entire contour of the detection target is extracted. Shape tends to change relatively large. Therefore, the first similarity calculating means 71 does not strictly evaluate the feature but makes a relatively rough evaluation. For example, the first similarity calculating means 71 previously stores the number of pixels of the contour portion of the template image to be detected as a reference feature amount, scans the first search range in the moving contour image C1, and performs each scan. At the point, a frame having the same size as the template image is set, the number of pixels of the contour existing in the frame is calculated, and the ratio between the calculated number of pixels and the reference feature amount is calculated as the similarity. If the ratio (similarity) of the number of pixels is closest to 1 in the first search range and is closer to 1 than a predetermined threshold value, it is determined that a detection target exists at the scanning point, and the position is detected. Output as the target location.
[0021]
The second moving contour extraction means 52 performs processing for extracting the contour with high accuracy when the detection target moves slightly or the movement is small. For example, as shown in FIG. 4, the second moving contour extracting unit 52 creates the contour image B1 stored in the contour image storing unit 4 at the time of the last two imagings (time (t−2)) and the current contour extracting unit 3 By performing a difference process and a binarization process with the contour image B3, a contour portion that has changed within a period longer than the above-described predetermined period (between the last two-time imaging and the current imaging) is extracted. Then, the moving contour image C2 at the detection target time is output. Since the sampling interval between the two contour images B1 and B2 for which the difference processing is performed is doubled as compared with the above-described first moving contour extracting means 51, it is suitable for extracting the contour Y of a slow-moving detection target. . Incidentally, when the detection target is stationary, there is no difference between the contour image B1 at the time of the last two-time imaging and the current contour image B3, so that no contour is extracted in the moving contour image C2.
[0022]
If the movement amount or the movement of the detection target is small, it is assumed that the detection target is present near the detection position at the time of the previous imaging, so the second search range setting unit 62 sets the position as shown in FIG. A range smaller than the first search range W1 near the detection position of the previous detection target is defined as a second search range W2, and the influence of disturbance is reduced by searching the detection target only in the second search range W2. It is hard to receive.
[0023]
The second similarity calculating means 72 calculates the similarity between the feature (template image) of the contour of the detection target stored in the memory in advance and the feature of the contour X appearing in the moving contour image C2. The second similarity calculating means 72 performs a process for accurately detecting when the movement amount or the movement of the detection target is small. In this case, the size and shape of the detection target tend not to change much. Therefore, the second similarity calculating means 72 relatively strictly evaluates the features of the outline, for example, scans the search range W2 in the moving outline image C2 and, at each scanning point, forms a frame of the same size as the template image. After setting, the outline in the frame is compared with the template image. If the corresponding pixel has both contours or no contour exists, counting is performed. If the count value is the highest in the search range W2 and higher than a predetermined threshold, the scanning point is determined. Is determined to exist, and the position is output as the position where the detection target exists.
[0024]
Further, the contour extracting means 3 performs processing for extracting the contour with high accuracy when the detection target is stationary. For example, as shown in FIG. 6, the contour extraction means 3 performs contour extraction and binarization processing on the image A input from the image input means 2 at time t, and outputs a contour image B3. All the contours of the detection target including the background are extracted from the contour image B3 regardless of whether the detection target is moving or stationary.
[0025]
When the detection target is stationary, it is assumed that the detection target exists very close to the detection position at the time of the previous imaging, so the third search range setting means 63 sets the last detection target as shown in FIG. The range W3 or W4 that is at or near the detection position of the detection target and is smaller than the second search range is set as the third search range, and the detection target is searched for only in the third search range W3 or W4. This makes it less susceptible to disturbances. Since the contour image B3 extracted by the contour extracting means 3 contains many contours of an object (such as a background) other than the detection target, it is desirable to narrow the third search range as much as possible. W3 in FIG. 7 is a search range when the range is limited to the previous detection position of the detection target, and W4 is a search range when the range is limited to the vicinity of the previous detection position.
[0026]
The third similarity calculating means 73 calculates the similarity between the contour feature (template image) of the detection target previously stored in the memory and the contour feature of the contour image B3 created by the contour extracting means 3 this time. calculate. Here, consideration is given so that detection can be performed with high accuracy when the detection target is stationary. In this case, it is assumed that the size and shape of the detection target hardly change. Therefore, the third similarity calculating means 73 strictly evaluates the features of the outline, for example, scans a search range in the outline image B3 and sets a frame of the same size as the template image at each scanning point, The outline in this frame is compared with the template image. If the corresponding pixel has a contour or both have no contour, the count is incremented. If the count value is the highest in the search range W3 and is higher than a predetermined threshold, the detection is performed at the scanning point. It is determined that the target exists, and the position is output as the detection target existing position.
[0027]
The detection determining means 8 determines the presence or absence of a detection target based on the calculation results of the first to third similarity calculating means 71 to 73. The result is given the highest priority, and the calculation result of the second similarity calculation means 72 and the calculation result of the third similarity calculation means 73 are weighted in order, and the position of the detection target is output from any of the similarity calculation means 71 to 73. Is output, it is determined that there is a detection target (for example, a person), and if there is no output, there is no human.
[0028]
As described above, in the present embodiment, the first moving contour extraction means 51 that can accurately detect the detection target when the movement amount or the movement of the detection target is large, and the detection target when the movement amount or the movement is relatively small. Is provided with a second moving contour extracting means 52 capable of detecting the object accurately, and a contour extracting means 3 capable of accurately detecting the object to be detected when the object to be detected is stationary. Sets the search range wide, sets the second search range near the detection position at the time of previous imaging when the movement amount or movement of the detection object is relatively small, and sets the previous imaging range when the detection object is stationary. Since a range smaller than the second search range near the detection position at the time is set as the third search range, the detection target can be accurately and reliably detected regardless of the behavior of the detection target. be able to.
[0029]
(Embodiment 2)
Embodiment 2 of the present invention will be described with reference to FIGS. In the present embodiment, in the first embodiment described above, the first moving contour extracting means 51 is used to determine whether the first moving contour extracting means 51 and the contour image B1 stored in the contour image storing means 4 (time (t-2)) before and after the previous capturing. By taking a difference from the contour image B2 at (time (t-1)), a first difference image D1 in which the contour of a portion that has changed within a predetermined period (between the last two-time imaging and the previous imaging) is extracted. The first contour image difference means 51a for creating the contour image and the difference between the contour image B2 stored in the contour image storage means 4 and the contour image B3 taken at the current time (time t) are calculated. A second contour image difference unit 51b for creating a second difference image D2 that extracts a contour of a portion that has changed during the period (between the previous imaging and the current imaging); Extraction of the common part with the second difference image D2 The constitutes in common contour extracting means 51c for outputting an image (moving outline image) in the detection target time. Since the configuration other than the first moving contour extraction means 51 is the same as that of the first embodiment, the same reference numerals are given to the same components, and the description thereof will be omitted.
[0030]
The operation of the first moving contour extracting means 51 will be described below with reference to FIG. The imaging unit 1 captures an image of the detection area at predetermined time intervals, and the contour extraction unit 3 extracts a contour based on an image signal of the imaging unit 1 input via the image input unit 2. Are created, and the created images are sequentially stored in the contour image storage means 4. Then, in the first moving contour extracting means 51, the first contour image differentiating means 51a performs a difference process between the contour image B1 obtained at the time of the previous image capturing and the contour image B2 obtained at the previous image capturing, and performs the image processing at the time of the image capturing before the second time and the image capturing at the previous The contour changed between the times is extracted to create a difference image D1. Further, the second contour image difference means 51b performs a difference process between the contour image B2 at the time of the previous imaging and the current contour image B3, and extracts the outline changed between the previous imaging and the current imaging. A difference image D2 is created. Then, the common contour extraction unit 51c creates an image D1 obtained by extracting a common part of the difference images D1 and D2 output from the first and second contour image difference units 51a and 51b.
[0031]
For example, when the detection target moves greatly from the last two imagings to the current imaging, the first outline image difference means 51a calculates the difference between the outline images B1 and B2, and thereby the outline X of the detection object is extracted in the difference image D1. However, since the outline of the background portion hidden by the shadow of the detection target at the time of the imaging two times before appears in the outline image B2 of the previous imaging, this portion is also extracted as the difference image D1. Similarly, the contour X to be detected is extracted in the difference image D2 by the second contour image difference means 51b calculating the difference between the contour images B2 and B3. Since the outline of the part appears in the current outline image B3, this part is also extracted in the difference image D2. Then, the common contour extracting means 51c extracts a contour common to the difference images D1 and D2 (MIN processing), performs a binarization process to create a moving contour image C1, and generates a moving contour image C1 for each of the difference images D1 and D2. Since the outline of the appearing background part exists only in one image, the outline of the background part can be removed when extracting the common part, and only the outline X of the detection target at the detection target time can be accurately extracted. it can.
[0032]
Since the second similarity calculating means 71 searches for the presence or absence of a detection target from the moving contour image C1 obtained by the above-described processing, the influence of the background is large even when the movement amount or movement of the detection target is large. The detection target can be detected with high accuracy without receiving.
[0033]
(Embodiment 3)
Second Embodiment A second embodiment of the present invention will be described with reference to FIGS. In the present embodiment, in the first embodiment described above, the second moving contour extracting means 52 is used to determine whether the second moving contour extracting means 52 is stored in the contour image storing means 4 at the time of the previous image pickup (time (t-2)) and the current image pickup time. The difference from the contour image B3 at (time t) is obtained, and the first difference image difference means 52a for creating a first difference image D3 in which the contour of the portion changed during this time is extracted, and stored in the contour image storage means 4. A second difference image D4 is created in which the difference between the contour image B2 at the time of the previous imaging (time (t-1)) and the current contour image B3 is obtained, and the contour of a portion changed during this time is extracted. And an outline synthesizing unit 52c that outputs an image C2 obtained by synthesizing the first differential image D3 and the second differential image D4. In this way, the difference between the contour images B1 and B3 at the time of the last two-time imaging and the current imaging is obtained, and the outline image of the portion changed during this time is extracted, and the outline images B2 and B3 at the previous imaging and the current imaging are obtained. Is obtained, and the outline image of the part changed during this period is extracted and synthesized, so that the outline of the part changed during a period longer than the predetermined period is extracted. Since the configuration other than the second moving contour extraction means 52 is the same as that of the first embodiment, the same components are denoted by the same reference numerals and the description thereof will be omitted.
[0034]
Hereinafter, the procedure of image processing by the second moving contour extraction means 52 will be described with reference to FIG. The imaging unit 1 captures an image of the detection area at predetermined time intervals, and the contour extraction unit 3 extracts a contour based on an image signal of the imaging unit 1 input via the image input unit 2. Are created, and the created images are sequentially stored in the contour image storage means 4. In the second moving contour extracting means 52, the first contour image differentiating means 52a performs a difference process between the contour image B1 obtained at the time of preceding image capturing and the current contour image B3, and performs the difference processing between the image captured two times before and the current image capturing. The changed outline is extracted to create a difference image D3. Further, the second contour image difference means 52b performs a difference process between the contour image B2 at the time of the previous imaging and the contour image B3 at the time of the current imaging, and extracts a portion of the outline that has changed between the previous imaging and the current imaging. To create a difference image D4. Then, the contour combining means 52c combines the first and second difference images D3 and D4, performs a binarization process, and creates a moving contour extracted image C2.
[0035]
As described above, the second moving contour extraction unit 52 extracts both the movement between the time before the last imaging and the current imaging and the movement between the previous imaging and the current imaging, and combines the results. Therefore, for example, considering a case in which the detection target person moves his / her hand to a certain position between the time of the last imaging and the time of the previous imaging, and then returns the hand to the position where the imaging was performed twice before the current imaging, Since there is no human motion between the time of the imaging and the current imaging, it is not possible to extract the difference between the contour images B1 and B3 as a moving outline, but between the previous imaging and the current imaging, the motion of the human is Therefore, if a difference between the outline images B2 and B3 is obtained, it can be extracted as a moving outline. In addition, when the person to be detected moves his / her hand to a certain position between the last two imagings and the previous imaging, and then has a hand at the same position as the previous imaging at the current imaging (that is, from the previous imaging to the current imaging). Considering the case where the hand is stopped in between), since there is no movement of the person between the previous imaging and the current imaging, the difference between the outline images B2 and B3 cannot be extracted as a moving outline, but is taken two times before. Since there is a human motion between the time of imaging and the time of current imaging, if a difference between the outline images B1 and B3 is obtained, it can be extracted as a moving outline.
[0036]
When the person to be detected moves as described above, in this embodiment, the moving outline can be extracted by one of the first and second outline image difference units 52a and 52b. Since the outputs of the first and second contour image difference means 52a and 52b are combined, the moving contour can be extracted in any case, and the detection target can be extracted from the moving contour image C2 thus obtained. Since the presence / absence of the search is searched, the detection target can be detected with high accuracy even when the movement amount or the movement of the detection target or a part thereof is small, or when the detection target moves or stops.
[0037]
(Embodiment 4)
Embodiment 4 of the present invention will be described with reference to FIGS. This embodiment is the same as the first embodiment except for the similarity calculation processing by the first similarity calculation means 71. Therefore, common components are denoted by the same reference numerals, and description thereof is omitted. Only the features of the embodiment will be described.
[0038]
The first similarity calculating means 71 scans the first search range W1 set by the first search range setting means 61 in the image W0 to be detected as shown in FIG. A frame having the same size as the detected template image T to be detected is set, and a matching process is performed between the image E in this frame and the template image T. The first similarity calculating unit 71 performs a less strict similarity calculation than the second and third similarity calculating units 72 and 73 as described in the first embodiment, and the pixel in the template image T And the pixels of the image E corresponding to this pixel are counted, and the count value is normalized by the size (total number of pixels) of the template image T.
[0039]
Here, the calculation procedure of the similarity by the first similarity calculating means 71 will be described with reference to the images of FIGS. FIG. 13A shows an example of a template image T, which is a 7 × 8 pixel binary image (total number of pixels: 56 pixels). Each square indicates a pixel, white pixels are portions having an outline, and hatched pixels are portions without an outline. FIGS. 13B to 13D respectively show images E1, E2, and E3 in a frame set by three scanning points in the first search range W1, and the image E1 is an image having relatively many contours. , The number of pixels having an outline together with the pixel of the image E1 and the corresponding pixel of the template image T is 17, and the number of pixels having no outline is 21. Further, the image E2 is an image having a few outlines, and the number of pixels having an outline is 11 among the pixels of the image E2 and the corresponding pixel of the template image T, and the number of pixels having no outline is 35. Further, the image E3 has the smallest outline among the three images E1 to E3, and the number of pixels having the outline both in the pixels of the image E3 and the corresponding pixels of the template image T is 3, and the number of the pixels having no outline in both. Is 35. The first similarity calculating means 71 counts the number of pixels having a contour together with the pixels in the template image T and the pixels of the search target images E1 to E3 corresponding to the pixels, and the result is referred to as the template image. The similarity (coincidence) is obtained by normalizing with the size of T (total number of pixels). In the case of the image E1, the number of pixels having an outline is 17 and the similarity is 17/56 = 0.304, since the number of pixels having an outline is 11 in the case of the image E2, the similarity is 11/56 = 0.196, and the number of pixels having the outline is 3 in the case of the image E3. , The similarity becomes 3/56 = 0.0536, and the image E1 has the highest similarity among the three images E1 to E3.
[0040]
As described above, the first similarity calculating means 71 sets a frame of the same size as the template image T at all the scanning points in the first search range W1, and determines the similarity between the image in the frame and the template image T. If the maximum value of the similarity is higher than a predetermined threshold value, it is determined that the detection target exists in the image having the highest similarity, and the scanning point is output as the position where the detection target exists. The feature of the calculation according to the present embodiment is that the similarity is high at a position having a relatively large number of contours, because it is assumed that the movement amount or the movement of the detection target is large. Further, since the similarity is obtained by normalizing the size of the template image T (the total number of pixels), there is an advantage that the similarity can be used on the same scale regardless of the size of the template image T.
[0041]
(Embodiment 5)
Embodiment 5 of the present invention will be described with reference to FIGS. This embodiment is the same as Embodiment 1 except for the similarity calculation processing by the second similarity calculation means 72, so that the same reference numerals are given to the same components, and the description thereof will be omitted. Only the features of the embodiment will be described.
[0042]
The second similarity calculating means 72 scans the second search range W2 set by the second search range setting means 62 in the image W0 to be detected as shown in FIG. 14, and stores in advance at each scanning point. A frame having the same size as the detected template image T to be detected is set, and a matching process is performed between the image F in the frame and the template image T. The second similarity calculating unit 72 performs a relatively strict similarity calculation as compared with the first similarity calculating unit 71 as described in the first embodiment, and the pixel in the template image T The number of pixels having outlines together with the pixels in the image F corresponding to the pixels and the pixels having no outline are separately counted, and the similarity is calculated by normalizing the ratio of pixels having outlines to pixels having no outline. Since it is obtained, the similarity of the image that matches the template image in a well-balanced manner with respect to both the portion having the contour and the portion having no contour increases, and the conversely, the similarity decreases.
[0043]
Here, the procedure of calculating the similarity by the second similarity calculating means 72 will be described with reference to the images of FIGS. FIG. 15A shows an example of a template image T, which is a 7 × 8 pixel binary image (total number of pixels: 56 pixels). Each square indicates a pixel, and the number of pixels with outlines (white pixels) is 21, and the number of pixels without outlines (pixels in oblique lines) is 35. FIGS. 15B to 15D respectively show images F1, F2, and F3 in a frame set by three scanning points in the second search range W2, and the image F1 is an image having relatively many contours. , The number of pixels having an outline both in the pixels of the image F1 and the corresponding pixels in the template image T is 21, and the number of pixels having no outline is 21. Further, the image F2 is an image having a few outlines. The number of pixels having an outline in both the pixels of the image F2 and the corresponding pixels of the template image T is 11, and the number of pixels having no outline is 35. Further, the image F3 has the smallest outline among the three images F1 to F3, and the number of pixels having the outline both in the pixels of the image F3 and the corresponding pixels of the template image T is 3, and the number of the pixels having no outline in both. Is 35.
[0044]
The second similarity calculation means 72 calculates the number of pixels having outlines together with the pixels in the template image T and the pixels of the search target images F1 to F3 corresponding to the pixels, and the number of pixels having no outlines together. Are separately calculated, and each count result is normalized by the number of pixels having an outline and the number of pixels having no outline in the template image T to obtain a similarity (coincidence). The total similarity is obtained by adding the similarity of the pixel and the similarity of the pixel having no contour. That is, in the case of the image F1, the number of pixels having both contours is 17 and the number of pixels having no contour is 21. Therefore, the similarity of a portion having both contours is 17/21 = 0.810, and both of the contours have contours. The similarity of the missing part is 21/35 = 0.60, and the total similarity (sum of similarities) is 1.410. In the case of the image F2, the number of pixels having both contours is 11 and the number of pixels having no contour is 35. Therefore, the similarity of a portion having both contours is 11/21 = 0.524, and both of the contours have contours. The similarity of the missing part is 35/35 = 1.0, and the total similarity is 1.524. Further, in the case of the image F3, the number of pixels having both contours is 3 and the number of pixels having no contour is 35, so that the similarity of the portion having both contours is 3/21 = 0.143, and the contour is both The similarity of the missing part is 35/35 = 1.0, and the total similarity is 1.143. Thus, the image F2 has the highest similarity among the three images F1 to F3.
[0045]
As described above, the second similarity calculating means 72 sets a frame of the same size as the template image T at all the scanning points in the second search range W2, and determines the similarity between the image in the frame and the template image T. If the maximum value of the similarity is higher than a predetermined threshold value, it is determined that the detection target exists in the image having the maximum similarity, and the scanning point is output as the position where the detection target exists. The feature of the calculation of the present embodiment is that the similarity increases when both the contoured portion and the non-contoured portion match the template image with good balance. This is because a small case is assumed.
[0046]
(Embodiment 6)
Embodiment 6 of the present invention will be described with reference to FIG. This embodiment is the same as the first embodiment except for the similarity calculation processing by the third similarity calculation means 73. Therefore, common components are denoted by the same reference numerals, and description thereof is omitted. Only the characteristic portions of the embodiment will be described.
[0047]
The third similarity calculating unit 73 scans the third search range W3 or W4 set by the third search range setting unit 63 in the image W0 to be detected. A frame having the same size as the template image T to be detected stored in advance only at the detection position is set, and a matching process is performed between the image G in this frame and the template image T. Here, the third similarity calculating unit 73 obtains the similarity by performing the same calculation processing as the calculation processing by the second similarity calculating unit 72 described in the fifth embodiment, and calculates the similarity. Is higher than a predetermined threshold value, it is determined that the detection target exists in the image G, and the scanning point is output as the detection target existence position. The feature of the similarity calculation according to the present embodiment is that the similarity increases when both the contoured portion and the non-contoured portion match the template image T in a well-balanced manner. This is because it is assumed that When the degree of stillness is made stricter, the threshold value of the similarity may be increased.
[0048]
(Embodiment 7)
Embodiment 7 of the present invention will be described with reference to FIG. This embodiment is the same as Embodiment 1 except for the judgment processing by the detection judging means 8. Therefore, the same components are denoted by the same reference numerals, and the description thereof will be omitted. Only the features of the present embodiment will be described. explain.
[0049]
The detection determining means 8 first determines whether or not the maximum value of the similarity (third similarity) obtained by the third similarity calculating means 73 is equal to or greater than a first threshold (S1). If the maximum value of the similarity is equal to or greater than the first threshold value, it is determined that the detection target is stationary (S2). Also, as a result of the determination in S1, if the maximum value of the third similarity is lower than the first threshold, the detection determination means 8 determines the similarity (the second similarity) determined by the second similarity calculation means 72. It is determined whether or not the maximum value of the similarity is equal to or greater than a second threshold (S3). If the maximum value of the second similarity is equal to or greater than the second threshold, the detection target performs a slight movement. It is determined that it is performed (S4). Further, as a result of the determination in S3, when the maximum value of the second similarity is lower than the second threshold, the detection determination means 8 determines the similarity (the first similarity) determined by the first similarity calculation means 71. It is determined whether or not the maximum value of the similarity is equal to or greater than a third threshold (S5). If the maximum value of the first similarity is equal to or greater than the third threshold, the detection target moves greatly. Is determined (S6). Also, as a result of the determination in S5, if the maximum value of the third similarity is lower than the third threshold, the detection determination means 8 determines that there is no detection target.
[0050]
As described above, the detection determination unit 8 first determines whether the detection target is stationary, and then determines whether the detection target is moving slightly, and then determines whether the detection target is moving largely. Therefore, the detection target is less likely to be affected by disturbance and can be accurately detected. The first to third search ranges in which the first to third similarity calculation means 71 to 73 have determined the similarities are narrower in the second search range than in the first search range, and Since the third search range is narrower than the second search range, even when there are a plurality of detection targets, the search is performed from the vicinity of each detection target, so that the plurality of detection targets can be identified. Therefore, there is an advantage that tracking of each detection target can be easily performed.
[0051]
【The invention's effect】
As described above, the invention according to claim 1 is an image processing apparatus that detects the presence or absence and movement of a detection target such as a person or a vehicle from an image obtained by capturing a predetermined detection area. Image capturing means for capturing an image, contour extracting means for creating a contour image obtained by extracting a contour from an image captured by the image capturing means, contour image storing means for storing the contour image created by the contour extracting means, and storing in the contour image storing means First moving contour extracting means for extracting a part of the contour changed within a predetermined period based on the obtained one or more contour images and creating a moving contour image at a detection target time, and a contour image storing means Moving contour extracting means for extracting a portion of a contour that has changed within a period longer than the predetermined period based on a plurality of contour images stored in the storage device, and creating a moving contour image at a detection target time. A first search range setting unit that sets, as a first search range, a range in which the detection target can move from a detection position of a previous detection target with respect to the image created by the first moving contour extraction unit; The first similarity calculating means for calculating the similarity between the contour extracted by the first moving contour extracting means and the detection target in one search range, and the image created by the second moving contour extracting means, A second search range setting unit that sets a range narrower than the first search range near the detection position of the target to be detected as the second search range, and a second moving contour extraction unit that extracts the second search range in the second search range. A second similarity calculating means for calculating the similarity between the detected contour and the detection target, and a second search range near the detection position of the previous detection target with respect to the contour image at the detection target time created by the contour extraction means. A smaller area than the third A third search range setting unit that sets a search range, a third similarity calculation unit that calculates a similarity between the contour extracted by the current contour extraction unit and the detection target in the third search range, The second and third similarity calculating means are provided with a detection judging means for judging the presence or absence of a detection target based on the similarity calculated respectively.
[0052]
Thus, the second moving contour extracting means creates an image in which a contour portion that has changed within a period longer than the sampling interval (the predetermined period) of the first moving contour extracting means is extracted. Therefore, by making the sampling time interval longer than that of the first moving contour extraction means, the contour of the detection target can be reliably extracted even when the movement amount or the movement of the detection target is small, and furthermore, the detection target contour can be extracted. When the movement amount or the movement is small, it is considered that the detection target exists near the detection position at the time of the previous imaging, so that the second search range setting means sets a range smaller than the first search range near the previous detection position. Since the second similarity calculating means obtains the similarity with the detection target within the second search range, the second similarity calculation unit is less likely to be affected by contours other than the detection target. Accurate small detection target It can be known. Also, the first moving contour extracting means creates an image in which a portion of the contour which has changed within a predetermined period shorter than the sampling interval of the second moving contour extracting means is extracted, so that the moving amount of the detection target is And even when the movement is relatively large, the contour of the detection target can be reliably extracted. In addition, the first search range setting means sets the range in which the detection target can move as the first search range, and sets the first search range Since the first similarity calculating means calculates the similarity with the detection target in the above, the detection target can be accurately detected even when the movement amount or the movement of the detection target is relatively large. Still further, the third search range setting means sets the third search range smaller than the second search range in the vicinity of the detection position of the detection target at the time of the previous imaging with respect to the contour image extracted by the current contour image extraction means. As the range, the third similarity calculating means obtains the similarity with the detection target within the third search range, and all the contours including the background appear in the contour image extracted by the contour image extracting means this time. Therefore, even when the detected object is stationary, its outline can be accurately extracted, and when the detected object is stationary, it is considered that it is near the previous detection position. By setting the range smaller than the second search range as the third search range, it is less likely to be affected by contours other than the detection target, and the stationary detection target can be detected with high accuracy. The detection determining means determines the presence or absence and movement of the detected object based on the calculation results of the first, second, and third similarity calculating means. In the first similarity calculating means, a detected object having a small moving amount or movement is detected by the second similarity calculating means, and a stationary detected object is detected by the third similarity calculating means. By judging the presence or absence and the movement of the detection target based on the calculation result of the degree calculation means, there is an effect that the detection target can be detected accurately and reliably regardless of the behavior of the detection object.
[0053]
According to a second aspect of the present invention, in the first aspect of the present invention, the first moving contour extracting means extracts a difference between the contour image at the time of the last imaging stored in the contour image storage means and the contour image at the previous imaging. A first contour image difference means for creating a first difference image, and a second contour image difference extracting means for extracting a difference between the contour image at the time of the previous imaging stored in the contour image storage means and the contour image created by the current contour extraction means. A second contour image difference means for creating a difference image; and a common contour extraction means for creating an image obtained by extracting a common part between the first difference image and the second difference image. If the time interval between the imaging, the previous imaging, and the current imaging is relatively long (that is, the moving amount or movement of the detection target is large), the first difference image is hidden behind the shadow of the detection target at the time of imaging two times before and the last imaging. The outline of the background that appears sometimes is extracted However, if the outline of the background appears in both the outline image at the time of the previous imaging and the outline image of this time, the outline does not appear in the second difference image. By extracting the common part with the image, the outline of the background can be removed, and there is an effect that the outline of the detection target can be accurately extracted without being affected by the background.
[0054]
According to a third aspect of the present invention, in the first aspect of the present invention, the second moving contour extracting means includes a difference between the contour image stored in the contour image storing means at the time of the last two-time imaging and the contour image created by the current contour extracting means. A first contour image difference unit for creating a first difference image from which the contour image has been extracted, and a difference between the contour image at the time of previous imaging stored in the contour image storage unit and the contour image created by the current contour extraction unit. A second contour image difference means for creating a second difference image; and a contour synthesizing means for creating an image obtained by combining the first difference image and the second difference image. The contour image subtracting means creates a first difference image by taking the difference between the contour image obtained two times before and the contour image taken this time, and the second contour image differencing means uses the contour image Find the difference from the contour image Since the second difference image is generated by the contour synthesizing unit and the first difference image is synthesized with the second difference image, the second difference image is added to the portion that has changed between the time of the last two-time imaging and the current time. Since a portion that has changed between the time of the previous imaging and the current time can also be extracted, there is an effect that the contour of the detection target can be accurately extracted even when the movement amount or movement of the detection target is small.
[0055]
According to a fourth aspect of the present invention, in the first aspect, the first similarity calculating means includes a storage means for storing a template image representing an outline of a detection target, and an image created by the first moving outline extracting means. Counting means for scanning within the first search range as an image to be searched, and counting the number of pixels having an outline together with each pixel of the template image and a corresponding pixel of the image to be searched at each scanning point; Means for calculating a similarity by normalizing the count value of the counting means with the total number of pixels of the template image, wherein each pixel of the template image corresponds to an image to be searched at each scanning point. Since the number of pixels that have both contours is counted, the effect of being less susceptible to changes in the shape and size of the detection target is obtained. Because they were-normalized, there is an effect that the same scale regardless of the size of the detection object can be evaluated similarity.
[0056]
According to a fifth aspect of the present invention, in the first aspect of the present invention, the second similarity calculating means includes a storing means for storing a template image representing an outline of a detection target, and an image created by the second moving outline extracting means. A first counting unit that scans the second search range as an image to be searched and counts the number of pixels having outlines at each scanning point between each pixel of the template image and a corresponding pixel of the image to be searched; And second counting means for counting the number of pixels having no contour between each pixel of the template image and the corresponding pixel of the image to be searched at each scanning point; and counting the counting result of the first counting means with the template image. Means for calculating the degree of similarity by adding the value normalized by the number of pixels having an outline and the value normalized by the number of pixels having no outline in the template image to the count result of the second counting means. thing As a feature, the first and second counting means count the number of pixels having an outline and the number of pixels having no outline at each pixel of the template image and corresponding pixels of the image to be searched. Therefore, it is possible to accurately detect a detection target having a small amount of movement or a small amount of movement, and furthermore, the counting results of the first and second counting means are normalized by the number of pixels having an outline of the template image and the number of pixels having no outline. Therefore, there is an effect that the similarity can be evaluated using the same scale regardless of the extraction degree and the size of the outline of the detection target.
[0057]
According to a sixth aspect of the present invention, in the first aspect of the present invention, the third similarity calculating means includes a storage means for storing a template image representing an outline of a detection target, and an image created by the current contour extraction means for searching. A first counting unit that scans the third search range as an image and counts the number of pixels having an outline together with each pixel of the template image and a corresponding pixel of the image to be searched at each scanning point; The second counting means for counting the number of pixels having no contour between each pixel of the template image and the corresponding pixel of the image to be searched at the scanning point; Means for calculating a similarity by adding a value normalized by a certain number of pixels and a value normalized by the number of pixels having no contour of the template image to the count result of the second counting means. Since the first and second counting means count the number of pixels having an outline and the number of pixels having no outline in each pixel of the template image and the corresponding pixel of the image to be searched, respectively. In addition, it is possible to accurately detect a detection target having a small amount of movement or movement, and further, the count results of the first and second counting means are normalized by the number of pixels having an outline and the number of pixels having no outline in the template image, respectively. Therefore, there is an effect that the similarity can be evaluated using the same scale regardless of the extraction degree and the size of the outline of the detection target.
[0058]
According to a seventh aspect of the present invention, in the first aspect of the present invention, the detection determination unit determines that the detection target is stationary if the third similarity calculated by the third similarity calculation unit is equal to or greater than the first threshold. And when the third similarity is lower than the first threshold, and the second similarity calculated by the second similarity calculating means is equal to or greater than the second threshold. When it is determined that the detection target is making a slight movement, and the third similarity is lower than the first threshold and the second similarity is lower than the second threshold, the first similarity is determined. If the first similarity calculated by the similarity calculating means is equal to or more than a third threshold, it is determined that the detection target is moving more. By comparing the level of the third similarity calculated by the third similarity calculating means with the first threshold for the search range, It is determined whether or not the detection target is stationary, and then the second similarity calculated by the second similarity calculating means for the second search range having the second narrowest search range and the second threshold are calculated. After determining whether the detection target is moving relatively small by comparing the height of the first search range, the first similarity calculation unit calculates the first search range calculated by the first similarity calculation unit for the first search range having the widest search range. By comparing the level of the similarity with the third threshold to determine whether the detection target is moving relatively large, it is difficult to be affected by disturbance and the detection target is accurately determined. It can be detected well. Further, since the second search range is narrower than the first search range and the third search range is narrower than the second search range, even if there are a plurality of detection targets, the detection can be performed. Since the search is performed from the vicinity of the detection position of the target at the time of the previous imaging, identification between a plurality of detection targets becomes possible, and there is an effect that individual detection targets can be easily tracked.
[Brief description of the drawings]
FIG. 1 is a block diagram of a schematic configuration of an image processing apparatus according to a first embodiment.
FIG. 2 is an explanatory diagram illustrating a procedure of the image processing according to the first embodiment.
FIG. 3 is an explanatory diagram of a first search range according to the first embodiment.
FIG. 4 is an explanatory diagram illustrating another image processing procedure according to the first embodiment.
FIG. 5 is an explanatory diagram of a second search range according to the embodiment.
FIG. 6 is an explanatory diagram illustrating another image processing procedure according to the first embodiment.
FIG. 7 is an explanatory diagram of a third search range according to the embodiment.
FIG. 8 is a block diagram of a main part of an image processing apparatus according to a second embodiment.
FIG. 9 is an explanatory diagram illustrating a procedure of the image processing according to the embodiment.
FIG. 10 is a block diagram of a main part of an image processing apparatus according to a third embodiment.
FIG. 11 is an explanatory diagram illustrating a procedure of the image processing described above.
FIG. 12 is a diagram illustrating image processing performed by the image processing apparatus according to the fourth embodiment.
FIG. 13A is an example of a template image, and FIGS. 13B to 13D are examples of a search target image.
FIG. 14 is an explanatory diagram of image processing by the image processing apparatus according to the fifth embodiment.
FIG. 15A is an example of a template image, and FIGS. 15B to 15D are examples of images to be searched.
FIG. 16 is an explanatory diagram of image processing performed by the image processing apparatus according to the sixth embodiment.
FIG. 17 is a flowchart of image processing performed by the image processing apparatus according to the seventh embodiment.
[Explanation of symbols]
3 Contour extraction means
4 Contour image storage means
8 Detection and judgment means
51,52 First and second moving contour extracting means
61 to 63 first to third search range setting means
71-73 First to Third Similarity Calculation Means

Claims (7)

In an image processing apparatus for detecting the presence or absence and movement of a detection target such as a person or a vehicle from an image obtained by imaging a predetermined detection area, an imaging unit for imaging an image of the detection area at predetermined time intervals, and an image captured by the imaging unit Contour extracting means for creating a contour image obtained by extracting a contour from an image, contour image storing means for storing the contour image created by the contour extracting means, and one or more contour images stored in the contour image storing means. A first moving contour extracting means for extracting a contour part which has changed within a predetermined period and creating a moving contour image at a detection target time; and a plurality of contour images stored in a contour image storing means. The second moving contour extracting means for extracting a part of the contour that has changed within a period longer than the predetermined period and creating a moving contour image at the detection target time, and the first moving contour extracting means First search range setting means for setting, as a first search range, a range in which the detection target can move from the detection position of the previous detection target with respect to the image, and a first moving contour in the first search range. The first similarity calculating means for calculating the similarity between the contour extracted by the extracting means and the detection target, and the first similarity calculating means for calculating the similarity between the image generated by the second moving contour extracting means and the first detecting means near the detection position of the previous detection target. A second search range setting unit that sets a range narrower than the search range of the second search range, and a similarity between the contour extracted by the second moving contour extraction unit and the detection target in the second search range. For the contour image at the detection target time created by the second similarity calculation means to be calculated and the contour extraction means, a range narrower than the second search range near the detection position of the previous detection target is set to the third search range. Third search set as Surrounding setting means; third similarity calculating means for calculating the similarity between the contour extracted by the current contour extracting means in the third search range and the detection target; first, second and third similarity calculating means An image processing apparatus comprising: a detection determination unit configured to determine the presence or absence of a detection target based on the similarity calculated by each of the units. The first moving contour extracting means generates a first difference image by extracting a difference between a contour image at the time of the last two-time imaging stored in the contour image storing means and a contour image at the previous imaging. Means, and second contour image difference means for creating a second difference image by extracting a difference between the contour image at the time of previous imaging stored in the contour image storage means and the contour image created by the current contour extraction means, 2. The image processing apparatus according to claim 1, further comprising common outline extracting means for generating an image in which a common part between the first difference image and the second difference image is extracted. The second moving contour extracting means generates a first difference image by extracting a difference between a contour image stored in the contour image storing means and a contour image created by the present contour extracting means at the time of the last two-time imaging. A contour image difference unit, and a second contour image difference for creating a second difference image which is obtained by extracting a difference between the contour image at the time of the previous imaging stored in the contour image storage unit and the contour image created by the current contour extraction unit. 2. An image processing apparatus according to claim 1, wherein said image processing apparatus comprises: means for generating an image obtained by synthesizing the first difference image and the second difference image. The first similarity calculation means scans the first search range with a storage means for storing a template image representing an outline of a detection target and an image created by the first moving outline extraction means as an image to be searched. Counting means for counting the number of pixels having an outline at each scanning point with each pixel of the template image and the corresponding pixel of the image to be searched; and counting the counting means by the total number of pixels of the template image. 2. The image processing apparatus according to claim 1, wherein the image processing apparatus is configured to calculate the degree of similarity by performing the conversion. The second similarity calculation means scans the second search range using the image created by the second moving outline extraction means as a search target image, with the storage means storing a template image representing the outline of the detection target. First counting means for counting the number of pixels having a contour together with each pixel of the template image and the corresponding pixel of the image to be searched at each scanning point, and each pixel of the template image at each scanning point and Second counting means for counting the number of pixels having no contour together with the corresponding pixels of the image of the image, a value obtained by normalizing the count result of the first counting means with the number of pixels having the contour of the template image, and a second 2. The image processing apparatus according to claim 1, further comprising means for calculating a similarity by adding a count result of said counting means to a value normalized by the number of pixels having no contour of the template image. The third similarity calculating means scans the third search range with the storage means for storing a template image representing the contour of the detection target and the image created by the contour extraction means this time as the search target image. First counting means for counting the number of pixels having outlines at each point of each pixel of the template image and the corresponding pixel of the image to be searched; and at each scanning point, each pixel of the template image and the image of the image to be searched. A second counting means for counting the number of pixels having no contour with the corresponding pixel; a value obtained by normalizing the count result of the first counting means with the number of pixels having the contour of the template image; 2. The image processing apparatus according to claim 1, further comprising means for calculating a similarity by adding a result obtained by counting the number of pixels and a value normalized by the number of pixels having no contour of the template image. If the third similarity calculated by the third similarity calculating means is equal to or greater than the first threshold, the detection determining means determines that the detection target is stationary, and determines that the third similarity is the third similarity. If the second similarity calculated by the second similarity calculating means is lower than the first threshold and the second similarity calculated by the second similarity calculating means is equal to or more than the second threshold, it is determined that the detection target is making a slight movement, When the third similarity is lower than the first threshold and the second similarity is lower than the second threshold, the first similarity calculated by the first similarity calculating means is equal to the first similarity. 2. The image processing apparatus according to claim 1, wherein if the threshold is equal to or greater than 3, the detection target is determined to be moving more.

Images