JP2006172248A - Spot image detecting method and spot image detecting program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To detect, by easy processing, a spot image showing a vehicle with a desired vehicle occupancy range and a desired existence range of a specific portion, such as a roof, from captured images with low resolution including a plurality of spot images showing the vehicle respectively. <P>SOLUTION: From pixels whose density is the maximum value or the minimum value compared with densities of peripheral pixels in a captured image, at least one pixel of the extreme value point with density of predetermined one of the maximum value and the minimum value is detected. A group of monotone changing pixels whose density has a monotone changing relation to the density of the detected extreme value point pixel is detected from pixels within the maximum occupancy range specified beforehand for an detection object. Depending upon whether an extreme value area including the detected extreme value point pixel and the monotone changing pixel group detected for the extreme value point pixel satisfies specific conditions or not, it is determined whether the extreme value area is an area representing the spot image for the detection object. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention obtains a spot image of a desired detection target from a low-resolution captured image including a plurality of spot images each representing one of the plurality of objects, which is obtained by imaging a space where the plurality of objects are located. The present invention relates to a spot image detection method and a spot image detection program suitable for detection, and in particular, from a satellite image having a low resolution obtained by imaging a plurality of automobiles on a road, a desired vehicle size and a desired roof presence range are obtained. The present invention relates to a spot image detection method and a spot image detection program suitable for detecting a spot image for a car having the same.

  As a method of automatically measuring traffic on the road, sensors such as TV cameras, loop detectors, ultrasonic detectors, and optical detectors are installed along the road, A method of detecting a car is used. However, with this method, the cost of equipment necessary for measurement becomes very large. For this reason, recently, a method for automatically detecting the number and positions of automobiles traveling on a road using remote sensing images such as aerial photographs and satellite images has been studied. The advantage of this method is that the field of view of remote sensing images is several tens of kilometers square, and it is expected that the position and number of automobiles in a wide area can be observed from the same image. Secondly, the traffic observation system using remote sensing images is mainly composed of image data receiving facilities and computers, and requires much smaller facilities than conventional systems using sensors, so the cost is low. That is. Third, the capacity of remote sensing images is at most several tens of megabytes, and can be processed at high speed by current computers.

  For example, in Patent Document 1, after applying image enhancement processing to a grayscale image obtained by shooting a road, threshold processing is performed with a certain threshold to detect a plurality of spot images representing an automobile. Here, the spot image refers to an image with a small number of pixels and a low resolution that is present at the position where the detection target exists but does not clearly represent the outer shape of the detection target. In Non-Patent Document 1, first, a background image is generated by removing a spot image of a car from a plurality of road gray images, and then a gray image including a car spot image and a background image are subtracted to mainly perform a car. A grayscale image including the spot image is acquired, and the grayscale image is subjected to threshold processing with a certain threshold to detect a spot image of a car.

Patent Document 1 or Non-Patent Document 1 uses an image density threshold, but Non-Patent Document 2 does not use an image density threshold and uses a spot image representing an automobile as follows. The position of is detected. Hereinafter, the pixel p is represented as a pixel (i, j) using the two-dimensional coordinates (i, j) of the pixel. For a certain pixel p, a square frame m (d) having a radius d is defined by the following equation.
m (d) = {(i + x, j + y) | max (| x |, | y |) = d}
The maximum value of the density of the pixels on the frame m (d) is defined as fmax (m (d)), and the minimum value of fmax (m (d)) in the range of the radius d from 1 to the maximum value D is defined as F. . That is,
F = min {fmax (m (d)) | 1 ≦ d ≦ D}

If the condition that the density of the pixel p is much larger than the minimum value F is satisfied, a bright spot image exists at the position of the pixel p. As the density of the pixel p, the density of the pixel p and the minimum value F are The density difference is set. When the condition that the density of the pixel p is much larger than the minimum value F is not satisfied, 0 is set as the density value of the pixel p. After executing such processing for all the pixels, a plurality of pixels determined to have a spot image are grouped by a symbol inference system, and each vehicle has a substantially branch-like shape intersected in an X shape. An image composed of two pixel columns is generated as an image representative of one automobile. The intersection position and direction of these two pixel columns represent the position and direction of the automobile. This automobile detection method uses an infrared image and makes it possible to distinguish between a moving car and a stopped car depending on the intensity of the infrared light.
JP 2003-030649 A SP Hoogendoorn, HJ Van Zuylen, M. Schreuder, B. Gorte, and G. Vosselman, "Microscopic Traffic Data Collection by Remote Sensing", Transporation Research Board (TRB) 82nd Annual Meeting, Washington DC, January 12-16, 2003, [online], [Search February 1, 2004], Internet <URL: http://www.geo.tudelft.nl/frs/papers/2003/Microscopic%20traffic%20Ben%20GeorgeV.pdf> U. Stilla and E. Michaelsen, "Estimating Vehicle Activity Using Thermal Image Sequences and Maps", ISPRS Symposium 2002, "Geospatial Theory, Processing and Applications", from 9 to 12 July, 2002 in Ottawa, Canada, [online], [ Search on February 1, 2004], Internet <URL: http://www.isprs.org/commission4/proceedings/pdfpapers/399.pdf>

  In a vehicle detection method such as Patent Document 1 or Non-Patent Document 1, a threshold value of density is determined in advance, and whether the density of a certain part in a captured image obtained by imaging a region where the vehicle travels is equal to or greater than this threshold value. It is determined whether or not the part represents a car depending on whether or not. In such a method, the detection result differs depending on how much the threshold is set. That is, the density of the automobile in the captured image is usually different from the density of the background, and the density of the automobile is higher (white) than the density of the background or vice versa. However, even if the density of the automobile is whiter than the background (the density is higher), the density varies depending on the automobile. That is, the density of the automobile in the captured image varies depending on the color and reflectance of the surface of the automobile, and also varies depending on the intensity of sunlight and the irradiation direction irradiated on the automobile. Furthermore, the density of the road serving as the background in the captured image varies depending on the color of the road surface and the reflectance.

  Therefore, there is a problem that it is difficult to appropriately determine the threshold value in the method for detecting a part representing an automobile from the captured image using the threshold value as described above. That is, if the threshold value is set to a bright value, a dark car cannot be detected. Conversely, if the threshold value is set to a dark value, an object in the background unrelated to the car is also detected.

  On the other hand, as in Non-Patent Document 2, there is no such problem in a method that does not use an image density threshold for detecting the position of an automobile. However, in the method disclosed in Non-Patent Document 2, after detecting a plurality of pixel candidates included in a spot image, the position and the like of the spot image using the symbol inference system using the density and position of those pixels is used. Since an image composed of two substantially linear pixel rows intersecting in an approximate X-shape representing the vehicle is generated as an image representative of one automobile, the process of generating the automobile representative image is complicated.

  In an image such as a satellite image or an aerial photograph, the resolution of the image is low, and the position and shape of the car itself are usually not expressed so accurately. The resolution of a geographical image obtained by actually photographing the ground from an artificial satellite is, for example, about 50 to 60 cm, and an image representing one car is not clear. It is represented by a small white spot image.

  Even if such a low-resolution image is used to detect the position or existence area of a small area such as a car on a road, it is difficult to increase the detection accuracy of those positions or existence areas. However, it is not always necessary to increase the detection accuracy of the position and the existence area of the vehicle for the application of detecting the congestion state of the road by detecting the position and number of the cars on the road. The process for detecting the position and the existing area is preferably a simple process that can be executed at high speed.

  By the way, the image obtained by photographing the road surface includes a spot image representing something other than the automobile. For example, where the road surface has a high reflectance, a wide part having a relatively high density is detected, and such a part may be detected as a spot image having a large size or a bright image part widely distributed. Furthermore, if there is a foreign object having a high reflectance other than an automobile on the road, the foreign object may be detected as a spot image having a small size. Furthermore, road signs are drawn on the road surface, and some of them may be detected as small spot images. Therefore, from the viewpoint of detecting a spot image representing an automobile, such a spot image representing an object other than an automobile or a bright image portion having a density distribution close thereto is also referred to as background noise. It is necessary to detect the image separately from such background noise.

  Since detection objects such as automobiles vary in size, the spread of the spot image varies from automobile to automobile. For example, a small passenger car has a small number of pixels of a spot image, for example, about 3 × 6 pixels, and a large bus has about 6 × 20 pixels. Furthermore, spot images are generated in certain parts of automobiles, such as roofs, where the density may be different from other parts on the detection target when imaged. is there. In addition, although the density | concentration when a specific part (for example, roof) is normally imaged by receiving irradiation light becomes large, the density when it image | photographs the peripheral part according to the direction of the surface does not become large. Sometimes. for that reason. Even when an automobile is imaged, the spread of the image portion having a high density corresponding to the existing range of the roof is usually smaller than the existing range of the roof.

  The size of a specific part such as a roof is different if the size of the automobile is different. Furthermore, even if the size of the automobile is the same, the size and position of the specific part may differ depending on the structure of the automobile. For example, in the case of a truck, a roof exists only in the driver's seat, and the rest has no roof and a cargo bed. For this truck, a spot image located in the front part of the vehicle is generated for the roof covering the driver's seat, but a high-density part is not generated for the part of the loading platform behind the vehicle. Sometimes. On the other hand, in the case of a bus, there is a common roof for the driver's seat and other parts. For such a bus, a spot image that covers almost the entire automobile is generated. In addition, there are cars that have roofs on the driver's seat and the cooler, respectively, such as trucks with a cooler. For such a truck, different spot images may be generated separately on the roof of the driver's seat and the roof of the cold storage. Therefore, when detecting the position of the automobile on the road, it is not easy to detect spot images of various automobiles at once without considering the occupation area of the automobile and the occupation area of the roof.

  Therefore, an object of the present invention is to obtain a desired image from a low-resolution captured image including a plurality of spot images each representing one of the plurality of objects, which is obtained by imaging a space where a plurality of objects such as an automobile are located. A spot image detection method suitable for detecting spot images representing a desired detection object, such as an automobile, having a vehicle occupation range and a desired existence range of a specific part such as a roof, and a simple process A spot image detection program is provided.

  An outline of the principle of the spot image detection method according to the present invention for achieving the above object will be described first. The density of the spot image representing the detection target (for example, an automobile) in the captured image is different from the density of the background. That is, the density is greater than the background density (ie, higher or white) or vice versa. The concentration distribution of an automobile or the like is complicated and changes depending on the direction of the surface of the automobile or the irradiation direction of sunlight. The background density also changes depending on the color of the road surface and the reflectance. As a result, the difference between the density of the spot image of the automobile having a higher density than the background and the density of the background changes depending on the automobile or the intensity of sunlight. Therefore, it is generally not easy to distinguish a background from a spot image of a detection object such as an automobile using a threshold value.

  However, when the density of the spot image of the automobile is larger than the background, the present inventor shows that the density of a certain pixel (maximum point pixel) is a maximum in the spot image of the automobile as a density distribution portion that does not depend on the automobile. I noticed that there are multiple pixels (monotonously decreasing pixel group) whose density is monotonously decreasing around the pixel. An area where the maximal point pixel and the monotonously decreasing pixel group exist may be referred to as a maximal area. Using this fact, in order to detect a spot image showing a car with a density higher than that of the background, it has been found that it is only necessary to detect a local maximum area where the local maximum pixel and the surrounding monotonically decreasing pixel group exist. . Similarly, when the density of a car is darker than the background (the density is lower), the density of a certain pixel (minimum point pixel) has a minimum value in the car image as a density distribution part that does not depend on the car. Then, he noticed that there are a plurality of pixels (monotonically increasing pixel group) whose density increases monotonously around them. Hereinafter, a region where the minimum point pixel and the monotonically increasing pixel group exist may be referred to as a minimum region. In order to detect a spot image indicating a car having a darker (darker) density than the background using this fact, it is only necessary to detect a minimum area where a minimum point pixel and a surrounding monotonically increasing pixel group exist.

  In the following, detection of the maximum area including the maximum point pixel will be mainly described. However, when detecting a spot image representing an automobile having a density lower than that of the background, it is only necessary to detect a minimum area including a minimum point pixel and a surrounding monotonically increasing pixel group. It can be similarly applied if it is read in consideration of the difference between the value and the minimum value and the difference between the monotonically decreasing pixel group and the monotonically increasing pixel group. Further, in the following, the maximum value and the minimum value are sometimes referred to as extreme values without being distinguished, and the maximum point pixel and the minimum point pixel are sometimes referred to as extreme value pixels without being distinguished, and the monotonously decreasing pixel group And the monotonically increasing pixel group may be referred to as a monotonically changing pixel group without distinction, and the maximum region and the minimum region may be referred to as extreme value regions without distinction.

  As already described, the existence range and position of a spot image obtained when a detection object such as an automobile is imaged depend on the spread and position of a specific part such as a roof, and The existence range and position vary depending on the size and structure of the vehicle. Therefore, in order to detect a spot image representing a car on a road from a low-resolution image such as a satellite image obtained by photographing a road, a large number of vehicles for all vehicle types having different vehicle sizes and roof sizes and positions are used. Rather than detecting spot images at once, specify the size of the car, specify the extent and position of a specific part such as a roof, and to such a specified car size, specified range and position A spot image for a car having a specific part such as a roof is detected, and such detection is performed by changing the specification of the vehicle size and the specification of the range and position of the specific part such as the roof. It is expected that it is simpler to detect a large number of spot images for a large number of automobiles by repeating the detection.

Therefore, the present invention makes it possible to detect a spot image for a desired vehicle having a desired vehicle size and satisfying the condition that the range and position of a specific part such as a roof are desired. In order to do so, the following four condition values are designated for the spot image of the desired detection object.
(1) The central minimum occupation range designated in advance for the detection target object as the minimum occupation range of the central portion of the spot image that should represent the detection target object (2) The spot image representing the detection target object is the maximum The maximum occupied range specified in advance for the detection target as the range on the captured image that can be limitedly occupied (the maximum occupied range is specified as a range when viewed from the center position of the central minimum occupied range, for example) )
(3) Allowable maximum density difference regarding density difference between density of pixel at center of spot image and density of plural pixels within the central minimum occupation range of spot image (4) Density of pixel at center of spot image And the minimum allowable density difference related to the density difference between the spot image and the density of the surrounding pixels within the maximum occupied range of the spot image

  Here, it is assumed that the center of the spot image is a point having the highest density. This center is assumed to be substantially the same as the geometric center of the distribution range of the spot image at the same time. Even if the latter is deviated from the former, it is considered that there is no problem with the present invention even if the deviation is ignored. The central portion of the spot image with respect to the detection target is a region having a high density (that is, a bright area) composed of pixels having a density close to the density of the center pixel of the spot image. The range in which the central portion of the spot image corresponding to the detection target object also varies depending on the detection target object. For example, when the roof of an automobile becomes wider, the range occupied by the bright portion at the center of the spot image becomes wider. Therefore, in order to detect the spot image corresponding to the detection target object, it is desirable to designate the central minimum occupation range as the range that the central portion of the spot image of the detection target object should be occupied at the minimum. When the detection target is an automobile, the central region of the spot image is a central portion of a specific portion such as the roof of the automobile. Therefore, it is desirable to designate the existence range of the central portion of the specific portion, which is narrower than the existence range of the specific portion such as the roof of the car, as the central minimum occupation range of the spot image for the object to be detected.

  The maximum occupation range designated as a range that can be occupied by the spot image corresponding to the detection object is an area including the entire detection object and a part of its shadow. This maximum occupation range increases with the size of the occupation range of the detection target. Therefore, in order to detect a spot image for the detection target, the detection is performed as the maximum occupation range that the spot image can occupy. It is desirable to specify a range having a width and position corresponding to the width and position of the occupation range of the object. Specifically, the maximum occupation range may be specified such that, for example, when the detection target is an automobile on the road, the maximum occupation range is the same as or slightly wider than the existence range of the vehicle. . By doing so, it is possible to prevent detection omission of pixels belonging to the monotonically decreasing pixel group with respect to the maximum point pixel among the pixels within the range where the automobile is actually present.

  Furthermore, the maximum occupation range is designated as a range located at a position viewed from the center position of the central minimum occupation range. For example, when the center of the maximum occupation range coincides with the center of the central minimum occupation range, the maximum occupation range in the state where the center of the maximum occupation range is located at the center of the central minimum occupation range is specified. Good. However, when the center of the maximum occupancy range does not match the center of the central minimum occupancy range, the maximum occupancy range in a state where the center of the maximum occupancy range is located at a position shifted from the center of the central minimum occupancy range. Specify the existence range of. Thus, by designating the maximum occupation range as viewed from the center of the central minimum occupation range, the present invention can be applied even when the centers of the two do not coincide.

  As already described, a spot image for an automobile or the like is distributed at least in a maximum region composed of a maximum point pixel and a surrounding monotonously decreasing region. There are several methods for defining the extent of the spread of the spot image based on the density distribution on the spot image. In the present invention, the maximum region for the automobile is considered to indicate at least the main part of the existing range of the spot image. It is considered that there is not necessarily a direct relationship between the range where the spot image is spread on the image and the maximum region. That is, a portion having a large density that can be seen as belonging to the spot image on the image may not belong to the maximum region because it is not a monotonously decreasing pixel. Conversely, a low-density portion that cannot be clearly seen as belonging to the spot image on the image is a pixel that satisfies the monotonous decrease relationship and may belong to the maximum region.

  The center of the spot image with respect to the detection target object may not coincide with the center of the maximum occupation range of the detection target object. The spot image is located in the existence range of a specific part such as a roof in the detection target. Typically, the center of the spot image is located at the center of the automobile roof. When the roof exists over the entire range of the automobile, the range of the automobile and the range of the roof match, and the center of the spot image and the center of the range of the automobile (maximum occupied range) substantially match. However, these centers may not match. For example, when the roof of an automobile is located in a part of the automobile, for example, in the front part, the center of the range where the automobile exists is not coincident with the center of the roof. Therefore, in such a vehicle, the center of the spot image for the vehicle is deviated from the center of the range (maximum occupied range) of the vehicle. Since the center of the central minimum occupation range described above is considered to be the center of the corresponding spot image, if the maximum occupation range as seen from the center of the central minimum occupation range is specified, the presence of the vehicle as described above Even when the center of the range is deviated from the center of the spot image, it is possible to detect a pixel group having a monotonous change relationship with respect to the detected extreme point pixel from the maximum occupied range.

  The central minimum occupancy range described above should be considered as a range near the center of the spot image where the density difference between the pixel density within that range and the density of the center pixel of the spot image is below a certain upper limit. You can also. Since this upper limit changes depending on the detection object, it is desirable to specify the upper limit of the density difference as the allowable maximum density difference when specifying the central minimum occupation range according to the detection object. Similarly, it is considered that the density of the peripheral portion of the spot image with respect to the detection target should be sufficiently changed compared to the density of the center pixel of the spot image. Since the lower limit of the density difference allowed as the density difference between the density of the center pixel of the spot image and the density of the plurality of pixels in the peripheral portion of the spot image differs depending on the detection object, It is desirable to designate the lower limit as the above-mentioned allowable minimum density difference.

  As is clear from the definition of the above four condition values, the density distribution of the spot image detected from the captured image is determined in advance related to these four condition values specified in relation to the desired detection target. If the condition is satisfied, it should be possible to determine that the spot image is for the detection object. Based on such an idea, the present invention detects a spot image for a detection target object from a captured image using the above four condition values.

  More specifically, the spot image detection method according to claim 1 includes an extreme point pixel detection step, a monotonous change pixel group detection step, and an extreme value region determination step. In the extreme point pixel detection step, the density is obtained from a captured image including a plurality of spot images each representing one of the plurality of objects, which is obtained by imaging a space where the plurality of objects including the detection target are located. At least one extreme point pixel having a density that is a predetermined extreme value of a maximum value or a minimum value with respect to the density of surrounding pixels is detected. In the monotonically changing pixel group detection step, the range on the captured image that can be occupied by the spot image representing the detection target is detected from pixels within the maximum occupation range designated in advance for the detection target. A group of monotonically changing pixels whose density has a monotonic change relationship with respect to the density of the extreme point pixel is detected. In the extreme value region determination step, depending on whether or not an extreme value region including the detected extreme value pixel and the monotonically changing pixel group detected for the extreme value pixel satisfies a predetermined condition Then, it is determined whether or not the extreme value region is a region representing a spot image with respect to the detection target.

  The predetermined condition is: (a) a center specified in advance for the detection target as a range that should be occupied by a central portion of the spot image of the detection target in the detected monotonically changing pixel group; A plurality of pixels located within the minimum occupation range all belong to the monotonically changing pixel group, and (b) a density of the detected extreme point pixel and a density of the plurality of pixels within the central minimum occupation range Is equal to or smaller than an allowable maximum density difference specified in advance with respect to the central minimum occupation range with respect to the detection target for any pixel, and (c) the detected extreme point pixel The density difference between the density and the density of a plurality of pixels located in the peripheral portion of the detected monotonically changing pixel group is designated in advance for the maximum occupied range of the detection target for any pixel. Was is that tolerance is the minimum density difference or more.

  Here, in the extreme point pixel detection step, the extreme point pixel is detected as a pixel candidate at the center of the spot image. When detecting the maximum point pixel as the extreme point pixel in the extreme point pixel detection step, the density of the pixel p selected from the captured image is the maximum point. Is not smaller than any of the density values of the eight neighboring pixels surrounding the pixel. In this definition, even if the density of the selected pixel p has the same density as any of the density of the eight neighboring pixels, is the density of any of the other pixels in the eight neighboring pixels the same as the density of the selected pixel p? If it is smaller, the density of the selected pixel p is regarded as a maximum value. Extremely, even if the density of the selected pixel p and the density of all the 8 neighboring pixels are the same, when the local maximum pixel is detected, the selected pixel p is regarded as the local maximum pixel. However, when the density of the selected pixel p is lower than any one of the densities of the eight neighboring pixels, the density of the selected pixel p is not a maximum value. It should be noted that the density of the selected pixel p is not compared with the density of a pixel farther than eight neighboring pixels of the selected pixel p. The minimum point pixel can be similarly defined.

  The other pixels that are detected in the monotonic change pixel group detection step and have a monotonic change relationship with respect to the density of the extreme point pixel are assumed to be the maximum point pixel from the maximum point pixel. As a path to the pixel, there is a path that satisfies the condition that the density of the pixel monotonously decreases. A monotonically decreasing density is when the density is smaller or equal. The range in which the monotonically changing pixel group is detected is the designated maximum occupation range. Therefore, the maximum area is detected from the maximum occupied range. As a result, even if a pixel that is in a monotonous change relationship with respect to a certain maximal point pixel exists at a position that exceeds the maximum occupied range, the pixel is not detected as a monotonous change pixel. Thus, the local maximum region including the local maximum pixel is detected from the maximum occupied range. On the other hand, when a monotonically changing pixel group around a certain maximal point pixel exists only in a part of the maximum occupied range, these monotonically changing pixel groups are all detected. In addition, the plurality of pixels located in the periphery of the detected monotonically changing pixel group whose density is determined in the extreme value region determining step are the monotonic changing pixel group in the monotonic changing pixel group. It refers to a plurality of pixels located adjacent to pixels that do not belong. As described above, the extreme point pixel is detected as a candidate for the center of the spot image by the extreme point pixel detection step and the monotonically changing pixel group detection step, and is detected for the extreme point pixel and the extreme point pixel. The extreme value region including the monotonically changing pixel group thus detected is detected from the maximum occupation range designated for the detection target.

  Further, in the extreme value region determination step, the first condition relating to the central minimum occupation range indicating the central region of the spot image designated for the detection object, the density of the detected extreme point pixel and the central minimum The second condition relating to the density difference with the density of the plurality of pixels within the occupied range, the density of the detected extreme point pixel, and the plurality of pixels located in the peripheral part of the detected monotonically changing pixel group When any of the third conditions relating to the density difference from the above-described density is satisfied, the detected extreme value region is determined as a spot image for the detection target.

  In the spot image detection method according to claim 2, the detection target is a desired existence range in which a specific part having a density different from that of another part among a plurality of automobiles on the road when imaged is a desired existence range. Yes, the vehicle occupation range is a vehicle having a desired area and a desired position with respect to the center of the specific portion, and the maximum occupation range is the desired vehicle occupation range of the detection target vehicle. The central minimum occupation range is specified to be a value narrower than the existence range of the specific portion of the vehicle.

  According to the spot image detection method and the spot image detection program according to the present invention, a spot image that should represent a detection target from a low-resolution captured image obtained by imaging a space where a plurality of objects such as an automobile are located. The central minimum occupation range that the central portion should occupy at least, the maximum occupation range that the spot image can occupy, and the density difference between the density of the center pixel of the spot image and the density of a plurality of pixels in the central area And an allowable minimum density difference relating to a density difference between a density of a pixel at the center of the spot image and a density of a pixel at a peripheral portion of the spot image. It is possible to detect a spot image that satisfies the defined condition as a spot image for the detection target.

Embodiments of a spot image detection method and a spot image detection program according to the present invention will be described below in detail with reference to the drawings.
FIG. 1 is a schematic block diagram of one embodiment of a spot image detection apparatus according to the present invention. Reference numeral 1 denotes the whole of one embodiment of the apparatus. The apparatus 1 includes, for example, a processing device 10 realized by a personal computer or a workstation, and a RAM (Random Access Memory) (not shown) used as a main memory and an auxiliary storage device (not shown) such as a magnetic disk storage device. And the input / output device 30. The input / output device 30 includes an input device 31 including a keyboard and a pointing device such as a mouse, a display device 32 such as a CRT display device, or an output device such as a printer 33. The input device 31 is used for inputting parameters and starting commands. The display device 32 or the printer 33 displays or prints an image such as a captured image obtained by imaging a plurality of detection objects, an image with a spot image marker generated by the device 1, or data generated by the device 1. Used for. When data is stored in the storage device 20, it is determined in advance for each piece of data whether the data is stored in a RAM (not shown) or an auxiliary storage device (not shown). The processing device 10 incorporates a spot image detection program 40 for detecting each detection target from a captured image obtained by imaging a region including a plurality of detection targets.

  Before further describing the embodiment of the spot image detection method and spot image detection program according to the present invention, the principle of the processing executed in this embodiment will be described. In the following, the case of detecting a maximum point pixel will be described, but the same applies to the case of detecting a minimum value pixel. The captured image is a geographic image obtained by photographing the ground surface from an artificial satellite or an aircraft, and the detection target is an automobile traveling on the road or stopping on the road.

  FIG. 2 is a diagram for explaining search condition data specified when spot images are detected according to the present invention. FIG. 1A is a schematic plan view for one automobile. In the figure, S1 indicates the occupation range of the vehicle of the automobile, and the vehicle occupation range S1 is assumed to be rectangular in the figure. Point C1 indicates the center of the vehicle occupation range S1. S2 shows the existence range of the roof provided in the said motor vehicle. The roof presence range S2 is an example of a specific portion having a density different from that of other portions when imaged, and the spot image has a higher density in the roof presence range. In the example of FIG. 5A, the roof presence range S2 is provided so as to extend over almost the entire automobile. However, in the figure, the roof existence range S2 is shown smaller than the vehicle occupation range S1 for the sake of clarity. In the example of FIG. 5A, it is assumed that the center C2 of the roof existence range S2 coincides with the center C1 of the vehicle occupation range S1.

  In the present invention, when a spot image for a detection target is detected, a maximum occupation range that can be occupied by the spot image is designated. For the automobile in FIG. 5A, the maximum occupation range R1 of the spot image is designated as the maximum occupation area R1 of the spot image so as to be the same as or slightly larger than the vehicle occupation area S1. In the example of FIG. 6A, the maximum occupation range R1 is rectangular, as is the vehicle occupation range S1, and the center thereof also coincides with the center C1 of the vehicle occupation range S1. The vertical and horizontal sides of the maximum occupation range R1 may be set to be, for example, 10 to 20% larger than the vertical and horizontal sides of the vehicle occupation range S1. Furthermore, in the present invention, a central minimum occupation range R2 that should be minimally occupied by the central portion of the spot image with respect to the detection target is also designated. The existence range of the portion having a high density in the spot image is a central portion of the roof existence range S2, and is smaller than the roof existence range S2. Therefore, in the example of FIG. 5A, an area slightly narrower than the roof existence range S2 is designated as the central minimum occupation range R2. The lengths of the vertical and horizontal sides of the central minimum occupation range R2 may be set to be, for example, 10 to 20% smaller than the lengths of the vertical and horizontal sides of the roof existence range S2.

  In the case of FIG. 9A, the center C1 of the maximum occupation range R1 is not shifted from the center C2 of the center minimum occupation range R2. However, in the present embodiment, these centers may be shifted. When the center C1 of the maximum occupation range R1 is deviated from the center C2 of the central minimum occupation range R2, the maximum occupation range R1 is designated as a range that can be seen when viewed from the center C2 of the central minimum occupation range R2. For example, the amount of shift between these centers and the coordinates of the four vertices of the maximum occupied range R1 when viewed from the center C1 of the maximum occupied range R1 are designated. Alternatively, the coordinates of the four vertices of the maximum occupation range R1 when viewed from the center C2 of the central minimum occupation range R2 may be designated.

  In the present invention, the allowable minimum density difference T1 is designated with respect to the difference between the peripheral pixels in the maximum occupation range R1 and the density of the center pixel of the spot image with respect to the detection target. That is, it is required that the density of the pixels in the peripheral part of the spot image is sufficiently changed with respect to the center of the spot image. Furthermore, in the present invention, the allowable maximum density difference T2 is specified with respect to the difference between the pixel in the central minimum occupation range R2 and the density of the pixel at the center of the spot image with respect to the detection target. That is, it is required that the density of the pixel within the center minimum occupation range R2 of the spot image does not change much with respect to the density of the pixel at the center of the spot image. The allowable minimum density difference T1 and the allowable maximum density difference T2 may use predetermined values irrespective of the occupation range of the vehicle to be detected and the existence range of the roof portion.

  FIG. 2B is a diagram showing the search condition data described above. The search condition data 25 includes spot image maximum occupation range data 251, spot image central minimum occupation range data 252, allowable minimum density difference (T 1) 253, and allowable maximum density difference (T 2) 254 as shown in the figure. It is out. Note that the maximum occupation range data 251 of the spot image is data specifying the maximum occupation range R1 when viewed from the center C2 of the central minimum occupation range R2, as already described. The central minimum occupation range data 252 is also data that specifies the central minimum occupation range when viewed from the center C2, and includes, for example, the coordinates of four vertices of the range.

  FIG. 2C is a diagram showing a relationship between an example of the density distribution of the spot image and the search condition data described above. In the figure, a curve f (p) is an example schematically showing a density distribution on a straight line passing through the center of the spot image, specifically, a horizontal straight line in FIG. 2 (a). Assume that the point p is the center of the spot image. As shown in the drawing, the maximum occupied range R1 and the central minimum occupied range R2 of the spot image include the local maximum point p as the center. As will be described later, the maximum region P is composed of a group of monotonically changing pixels having a monotonously decreasing relationship with respect to the maximum point p located around the maximum point p according to the present invention. In the example shown in the figure, the pixel A is the pixel with the minimum density on the left side in the maximum region P at the maximum point p, and the pixel B is the pixel with the minimum density on the right side in the maximum region P at the maximum point p. And In the drawing, it is assumed that the pixel B is the pixel having the maximum density among the pixels in the peripheral portion of the maximum region P. In order for a spot image including the pixel A, the maximum point p, and the pixel B to be a pixel of the detection target, the density f (p) of the maximum point p and the maximum density of the peripheral portion within the maximum occupied range R1 of the spot image ( Under the present assumption, it is necessary to satisfy the condition that the density difference with respect to the density of the pixel B is equal to or larger than the allowable minimum density difference (T1). Furthermore, all the pixels in the central minimum occupation range R2 are located in the maximum area P, and the allowable minimum density is obtained for any pixel in which the density difference between the density of the maximum point p and the pixel in the central minimum occupation range R2 is any. The condition that the difference is equal to or smaller than (T2) needs to be satisfied.

  FIG. 2D is a diagram illustrating another example of the roof presence range S2 and the central minimum occupation range R2. Unlike the figure (a), the roof presence range S2 is biased to the left of the vehicle occupation range S1. Such a roof presence range S2 exists in the case of a truck in which a roof is provided so as to cover the driver's seat part and a loadless platform without a roof is provided in the remaining part. The center C1 of the vehicle occupation range S1 and the center C2 of the roof presence range S2 are displaced as shown in the figure. In an image obtained by photographing such a car, the spot image for the car has a center at the center C2 of the roof existence range S2. In the case of this automobile, the center C1 of the maximum occupation range R1 is shifted as shown in the figure as compared with the center C2 of the central minimum occupation range R2. The present invention can also be applied to a case in which a detection object is an automobile in which the roof existence range is a small part of the vehicle occupation range. In order to make such a car an object to be detected, the maximum occupancy range R1 slightly larger than the vehicle occupancy range S1 is designated as the maximum occupancy range of the spot image, as in FIG. As the minimum occupation range, a central minimum occupation range R2 slightly smaller than the roof existence range S2 may be specified. In the following, the processing in one embodiment of the present invention will be described for the case of the automobile in FIG. 2A, but the following processing can also be applied to the case of the automobile in FIG.

The maximum region used in the present embodiment can be detected by the following method, for example. The maximum point pixel p is detected from the captured image, and the maximum region M (p) with respect to the detected maximum point pixel p can be obtained inductively as follows. Let M0 (p) be the 0th-order maximum region composed of the maximum point pixels p. Starting from the 0th order maximum region M0 (p) of the local maximum pixel p, the region growth of only the (n + 1) th order increment region Bn + 1 (p) is repeatedly determined for the nth order maximum region Mn (p). Then, the (n + 1) th order maximum region Mn + 1 (p) is obtained by the following equation 1.
Mn + 1 (p) = Mn (p) ∪Bn + 1 (p) (1)

Here, the (n + 1) th order increment region Bn + 1 (p) has a condition that the pixel q is adjacent to the nth order maximum region Mn (p) and the density of the pixel q is the nth order maximum region Mn (p). Is a region including a pixel q that satisfies the condition that the pixel q is not larger than the density of at least one adjacent pixel (has a monotonous decrease relationship). The following equation 2 is established for the smallest m in which the (m + 1) th order increment region Bm + 1 (p) is an empty set.
M (p) = Mm (p) (2)
Therefore, when the (m + 1) th increment region Bm + 1 (p) first becomes an empty set, the mth order maximum region Mm (p) obtained so far becomes the maximum region M (p).

  The above repeated process will be described in detail. The first-order incremental region B1 (p) is obtained for the 0th-order maximum region M0 (p) consisting only of the maximum point pixel p. That is, the first increment region B1 (p) including all the pixels q that satisfy the condition that the pixel q is adjacent to the pixel p and the density of the pixel q is not higher than the density of the pixel p (monotonic decrease relationship) is obtained. . At this stage, the only pixels on the path from the maximal point pixel p to the arbitrary pixel q in the first incremental region B1 (p) are these two pixels p and q, and the density monotonously decreases between these pixels. The condition to do is satisfied. Furthermore, the primary maximum region M1 (p) is determined by the logical sum of the original pixel p and the newly obtained primary incremental region B1 (p) according to Equation 1.

  Thereafter, a second incremental region B2 (p) is obtained. That is, the condition that the pixel q is adjacent to any pixel in the first maximum region M1 (p) and the density of the pixel q is at least one adjacent to the pixel q in the first maximum region M1 (p). The region of all the pixels q that satisfy the condition that the density is not larger than the density of one pixel (monotonic decrease relationship) is determined as the second incremental region B2 (p). Further, the second maximum region M2 (p) is determined by the logical sum of the first maximum region M1 (p) and the newly obtained second increment region B2 (p) according to Equation 1. At this stage, the pixel on the path from the maximum point pixel p to any pixel q in the second increment region B2 (p) is the pixel p and any of the first increment regions B1 (p) in the increment first increment region B1 (p). 1 pixel and the second pixel in the second incremental region B2 (p) adjacent to the first pixel, and the density between the first pixel and the second pixel is increased from the pixel p. The condition of monotonically decreasing is established.

  Similarly, the maximum region Mn (p) (n = 1, 2,...) Can be obtained by repeating the above processing. When any pixel belonging to the (m + 1) th increment region Bm + 1 (p) cannot be detected for the first time for any integer m, the processing is terminated, and the calculation is performed up to that time according to Equation 2. The m-th maximum region Mm (p) may be used as the maximum region M (p).

  In order to detect the size of the maximum region M (p) only from the pixels within the maximum occupation range including the pixel p as a center, for example, any one of the increment regions ((n + 1) th increment region Bn + 1 (p)) is determined by sequentially selecting all the pixels within the maximum occupation range including the pixel p as a center, and sequentially determining all the pixels within the same range when determining the next increment region. It is only necessary to repeat selecting all the pixels within the same range sequentially so as to select them. When the incremental regions are sequentially obtained in this way to obtain the maximum region M (p), an arbitrary pixel q belonging to the maximum region M (p) has a pixel density on the route as a route from the pixel p to the pixel q. The condition that there is a monotonically decreasing path is satisfied. This can be proved. In this way, a monotonically changing pixel group is detected, and as a result, a maximum region including the maximum point and the monotonically changing pixel group is detected.

  After obtaining the maximum region M (p) in this way, the first condition regarding the central minimum occupation range indicating the central region of the spot image designated for the detection object, the detected extreme value pixel density and The second condition regarding the density difference with the density of the plurality of pixels within the central minimum occupation range, the density of the detected extreme point pixel, and the plurality of pixels located in the peripheral part of the detected monotonically changing pixel group When any of the third conditions relating to the density difference from the pixel density of the pixel is satisfied, the detected extreme value region is determined as a spot image for the detection target.

Hereinafter, the details of the embodiments of the spot image detection method and the spot image detection program according to the present invention will be further described.
Returning to FIG. 1, the spot image detection program 40 is stored in the storage device 20 and executed by the processing device 10. However, for the sake of clarity, the spot image detection program 40 and a plurality of modules constituting the spot image detection program 40 are illustrated in FIG. It is described inside a block showing the processing apparatus 10. The spot image detection program 40 includes modules of an initial value setting unit 100, an extreme point pixel search unit 200, an extreme value region growth unit 300, an extreme value region determination unit 400, and a spot image related data generation unit 500. . Furthermore, the spot image detection program 40 includes an extreme point pixel search success / failure determination step 250 and an extreme value region pass / fail determination step 450. The extreme point pixel search success / failure determination step 250 is executed after execution of the extreme point pixel search unit 200, and determines whether or not the extreme point pixel search unit 200 has successfully searched for the extreme point pixel. The extreme value region acceptance / rejection determination step 450 is executed after the execution of the extreme value region determination unit 400 and determines whether or not the determination result of the extreme value region determination unit 400 for the detected extreme value region is acceptable.

  The plurality of modules 100, 200, 300, 400, and 500 are sequentially executed in this order. After the execution of the spot image related data generation unit 500, the processing moves to the initial value setting unit 100, and the processing after this processing is repeated. However, the extreme value region growing unit 300 is executed when the extreme value pixel search is determined to be successful in the extreme value pixel search success / failure determination step 250, and the spot image related data generation unit 500 is detected. It is executed when it is determined in the extreme value region pass / fail determination step 450 that the extreme value region is acceptable. When it is determined that the extreme value region detected in the extreme value region acceptance / rejection determination step 450 is unacceptable, the processing moves to the initial value setting unit 100. If it is determined in the extreme point pixel search success / failure determination step 250 that the extreme point pixel search is unsuccessful, the processing of the spot image detection program 40 ends.

  In the processing apparatus 10, the initial value setting unit 100, the extreme point pixel search unit 200, the extreme value region growing unit 300, the extreme value region determining unit 400, and the spot image related data generating unit 500 in the spot image detection program 40 are executed. Function block for setting initial values, function block for searching for extreme point pixels, function block for growing an extreme region starting from the searched extreme point, and extreme value obtained by region growth A functional block for determining whether or not an area is an extremum area belonging to a spot image of a detection object, and when it is determined that the extremum area belongs to a spot image representing the detection object, the spot image It operates as a plurality of functional blocks called functional blocks for generating related spot image related data. Therefore, a plurality of functional blocks corresponding to each module is realized by these modules of the processing device 10, the storage device 20, the input / output device 30, and the spot image detection program 40. Therefore, the processing device 10, the storage device 20, the input / output device 30, and the spot image detection program 40 realize one embodiment of the spot image detection device according to the present invention.

  The spot image detection program 40 implements one embodiment of the spot image detection program according to the present invention, and is recorded on a recording medium (not shown) or stored in the storage device 20 via a network (not shown). It is executed by the processing device 10. The spot image detection program 40 is recorded on a recording medium (not shown) or can be sold via a network (not shown). The procedure in which the processing apparatus 10 executes the spot image detection program 40 to detect the object from the captured image realizes one embodiment of the spot image detection method according to the present invention.

  The storage device 20 stores the captured image data 21 in advance before the spot image detection program 40 is executed. The captured image data 21 is data representing an image obtained by capturing an area including the detection target. The region attribute data 22 holds region attribute values related to whether or not each pixel within the maximum occupation range described later belongs to an extreme value region including an extreme point pixel detected from the captured image data 21. It is data, is set to an initial value when the execution of the spot image detection program 40 is started, and is updated to the spot image detection program 40 during the execution.

  The extreme value region data 23 and the image data 24 with the spot image marker are data generated as a result of execution of the spot image detection program 40. The extreme value region data 23 includes data related to each of a plurality of extreme value regions that are detected by the spot image detection program 40 and are determined to be included in the spot image representing the detection target. The image data 24 with the spot image marker is obtained by superimposing a predetermined spot image marker indicating a spot image of the detection target on the spot image including the detected extreme value region in the captured image data 21. Data representing an image. The search condition data 25 is data that has already been shown in FIG. The search condition data 25 is data specified by the user in order to specify the search object before the spot image detection program 40 is executed. A plurality of data included in the search condition data 25 may be collectively stored in the storage device 20 as one data, but is specified by the user before the search and incorporated in the spot image detection program 40. May be. Instead of the user specifying the data related to the maximum occupied range R1 and the data related to the central minimum occupied range R2 included in the search condition data 25, the vehicle occupied range S1 and the roof existing range S2 from which the respective data are determined are determined. The spot image detection program 40 uses the spot image detection program 40 to multiply the data related to the maximum occupied range R1 and the data related to the central minimum occupied range R2 by multiplying the values of the vehicle occupied range S1 and the roof existing range S2 by a certain ratio. You may make it determine using methods, such as dividing by a fixed ratio. Also in this case, as far as the present invention is concerned, it is considered that the user has designated the maximum occupation range R1 and the central minimum occupation range R2. Also, in the following, when referring to a process in which image data is generated or a process stored in the storage device 20, an image is simply generated or stored in the storage device 20 for simplicity. There is.

  FIG. 3A is a diagram schematically showing the captured image data 21. As shown in the figure, the captured image data 21 is rectangular (square or rectangular), the origin O is defined at the upper left corner of the captured image, the x axis in the vertical downward direction and the horizontal right direction in the figure, Assume that the y-axis is defined. The captured image data 21 is assumed to be image data representing a black and white grayscale image. Further, it is assumed that the captured image data 21 represents an image in which the density becomes higher (whiter) in the area where the automobile is present.

  In the figure, reference numeral 210 schematically represents one road on the ground. Reference numeral 220 schematically shows a spot image representing one of a plurality of automobiles traveling on the road 210. FIG. 2B shows an example of an actual geographic image obtained by actually photographing the ground from an artificial satellite. The size is 313 × 145 pixels, and the density is a value from 0 to 255. The resolution of this type of geographic image is generally low, for example, about 50 to 60 cm. As shown in the figure, the automobile is represented by a very small white island-like spot image, and the outline of the automobile is not clear. The number of pixels constituting a spot image representing an automobile is, for example, 3 × 6 pixels to 6 × 20 pixels.

  In addition, when applying this invention to the detection of the spot image with respect to a rectangular detection target like a motor vehicle, it is convenient to align the direction of a motor vehicle before application. For example, it is assumed that the road direction is the horizontal direction of the captured image as shown in FIG. For this purpose, it is effective to extract an edge from the captured image, generate a histogram for the direction of the edge, determine the direction of the road from the peak of the histogram, and rotate the captured image so that it is in the horizontal direction. . See, for example, the following literature for specific processing for implementing this method. T. Zhao and R. Nevatia, “Car Detection in Low Resolution Aerial Image,” [online], [searched September 1, 2004], Internet <URL: http://iris.usc.edu/~taozhao/ See papers / ICCV01 / ICCV01_ZhaoNevatia_CarDetection.pdf>. In the following, it is assumed that the captured image has already been rotated and the direction of the road, and thus the direction of the automobile, is directed horizontally.

  FIG. 4 schematically shows region attribute data 22 corresponding to the captured image data 21. In the present embodiment, the region attribute data 22 has a field that stores region attribute values corresponding to all the pixels of the captured image data 21. Since the region attribute data 22 includes the region attribute of the pixel corresponding to each pixel of the captured image data 21, the region attribute data 22 also represents an image having the value of the region attribute as an image value, like the captured image data 21. Since it can be considered as image data, as shown in FIG. 4, the region attribute data 22 is considered as image data having a rectangular region, and the origin 0 is located at the upper left corner of the figure as with the captured image data 21. Assume that the x-axis and the y-axis are defined in the vertical downward direction and the horizontal right direction in the figure. As will be described later, the region attribute value for each pixel included in the region attribute data 22 is a value indicating that the pixel does not belong to the extreme value region (eg, “0”), and the candidate that the pixel belongs to the extreme value region Or a value (for example, “1”) indicating that the pixel belongs to the extreme value region.

  Even if the region attribute data 22 has a field that holds region attribute values for all the pixels of the captured image data 21 as in the present embodiment, in practice, any one of the road regions 210 in the captured image data 21 is present. It suffices to have a field for holding a region attribute value for each of a plurality of pixels belonging to (FIG. 3A), but here, which pixel in the captured image data 21 belongs to the road region Since the region attribute data 22 varies depending on the region represented by the image 21, it is assumed that the region attribute data 22 has a field that holds a region attribute value corresponding to each of all the pixels of the captured image data 21.

  Returning to FIG. 1, when the spot image detection program 40 is executed, the initial value setting unit 100 is first executed. The initial value setting unit 100 sets the region attribute of the region attribute data 22 (FIG. 4) for each of all the pixels of the captured image data 21 to an initial value. Here, the initial value is a value “0” indicating that the corresponding pixel does not belong to the extreme value region.

  The spot image detection program 40 executes the extreme point pixel search unit 200 after the execution of the initial value setting unit 100. FIG. 5 is a schematic flowchart of an example of processing of the extreme point pixel search unit 200. The extreme point pixel search unit 200 sequentially selects pixels of the captured image 21 in a certain order, and determines whether the selected pixel is a pixel having an extreme value in density (referred to as an extreme point pixel). The extreme point pixel is detected as a pixel that may belong to the spot image representing the detection target in the captured image 21. As in the current assumption, when detecting a spot image of an automobile that is whiter (higher density) than the background, a maximum point pixel having a maximum density is detected as an extreme point pixel. Conversely, when detecting a spot image of a car that is blacker than the background (smaller in density), a minimum point pixel having a minimum value is detected as an extreme point pixel. An extreme point pixel, for example, a maximum point pixel, means that the density of the maximum point pixel is a maximum point compared to the density value of eight neighboring pixels surrounding the pixel. Therefore, in order to determine whether a certain pixel is an extreme point pixel, the density of the target pixel is compared with the density of eight neighboring pixels of the pixel.

  Specifically, the extreme point pixel search unit 200 is executed as follows. First, it is determined whether or not the selection of pixels from the captured image 21 is completed (step S201). If the pixel selection is not completed, the pixel to be processed next is selected (step S202). The spot image detection program 40 does not need to process all the pixels of the captured image 21 in order to detect a spot image representing a car that is a search target, and belongs to a road region in which the road in the captured image 21 exists. Only a plurality of pixels need be processed. Therefore, in step S202, a plurality of pixels belonging to the road region may be sequentially selected in a certain order, and pixels that do not belong to the road region may not be selected. Selection of a plurality of pixels belonging to a road area in the captured image 21 can be performed based on data indicating a road area (not shown) stored in advance in the storage device 20.

  More specifically, the extreme point pixel search unit 200 compares the density of the pixel selected in step S202 with the density of adjacent pixels of the pixel in order to search for extreme point pixels. In step S203, it is determined whether the pixel is an extreme point pixel. Here, the adjacent pixels of the selected pixel are eight adjacent pixels (eight neighboring pixels) surrounding the selected pixel. Therefore, in order to determine whether or not the selected pixel is an extreme point pixel, the density of pixels around that pixel is also necessary. Therefore, in the extreme point pixel search unit 200, the width of one pixel around the road region is determined. It is determined whether or not the pixels belonging to the internal area excluding the area are extreme point pixels. Therefore, in step S202, all the pixels in the region one pixel inside the road region in the captured image 21 may be sequentially selected in a certain order. The order of selection may be, for example, the x coordinate order and the y coordinate order. That is, a plurality of pixels having the same y coordinate may be selected in ascending order of the x coordinate, and a plurality of pixels having different y coordinates may be selected in the ascending order of the y coordinate.

  Hereinafter, the pixel selected in step S202 is referred to as a pixel p. After the pixel p is selected, it is determined whether or not the selected pixel p is an extreme point pixel (step S203). In the present example, the extreme point pixel is a maximum point pixel. Here, it is assumed that the selected pixel p is a maximal point pixel having a maximal density when the density of the selected pixel p is not smaller than any of the density values of the eight neighboring pixels surrounding the pixel. In this definition, the density of the selected pixel p has the same density as any of the density of the eight neighboring pixels, and any density of the other pixels in the eight neighboring pixels is equal to or smaller than the density of the selected pixel p. For example, the density of the selected pixel p is regarded as a maximum value. Extremely, even when the density of the selected pixel p and the density of all the 8 neighboring pixels are the same, the selected pixel p is regarded as the maximum point pixel when searching for the maximum point pixel. However, when the density of the selected pixel p is lower than any one of the densities of the eight neighboring pixels, the density of the selected pixel p is not a maximum value. It should be noted that the density of the selected pixel p is not compared with the density of pixels farther than the eight neighboring pixels of the selected pixel p.

  If it is determined in step S203 that the selected pixel p is not an extreme point pixel (maximum point pixel), the process returns to step S201, and it is determined whether the pixel selection is completed. That is, it is determined whether or not a pixel that has not been determined whether or not it is an extreme point pixel is in an area one pixel inside the road area. If there is still an unselected pixel, the new pixel is selected in step S202, and it is determined whether the new selected pixel p is an extreme point pixel in step S203. Such pixel selection end determination step S201, pixel selection step S202, and extreme point pixel determination step S203 are repeated until an extreme point pixel is detected.

  As a result of the above repetition, when it is determined that any selected pixel p is an extreme point pixel, it represents that the region attribute value in the region attribute data 22 relating to the pixel p belongs to the extreme value region. The attribute value is changed to “1” (step S204). When the extreme point pixel is selected, the extreme point pixel search unit 200 ends the process, returns to “success”, and returns to the main routine of the spot image detection program 40.

  Returning to FIG. 1, in the spot image detection program 40, after the extreme point pixel search unit 200 is executed, an extreme point pixel search in the extreme point pixel search unit 200 is performed by an extreme point pixel search success / failure determination step 250. Is determined based on a return value from the extreme point pixel search unit 200. As assumed now, when the return value from the extreme point pixel search unit 200 is “success”, the extreme value region growing unit 300 is executed. The extreme value region growing unit 300 detects an extreme value region including the extreme value point using the detected extreme value point as a starting point. Here, the extreme value region is a region made up of the extreme point pixel (maximum point pixel in the present example) and one or a plurality of peripheral pixels that are in a monotonous change relationship with the pixel. Here, a pixel that is in a monotonic change relationship with an extreme point pixel indicates a monotone decreasing relationship when the extreme point pixel is a maximum point pixel, as assumed now, and the extreme point pixel is a minimum. In the case of a point pixel, it indicates a monotonically increasing relationship.

  A peripheral pixel having a monotonic change relationship with an extreme point pixel means that there is a path from the extreme point pixel to the surrounding pixel via a pixel whose density is monotonously changing. . When the extreme point pixel is a local maximum pixel, the surrounding pixels that are monotonically decreasing with respect to the local maximum pixel are referred to as the peripheral pixels via the pixels whose density is monotonously decreasing from the local maximum pixel. This means that there is a path to the pixel. Here, the monotonic decrease means that it is smaller or equal. Similarly, when the extreme point pixel is a minimum point pixel, the surrounding pixels that are in a monotonically increasing relationship with the minimum point pixel are connected via a pixel whose density is monotonically increasing from the minimum point pixel. This means that there is a route to the surrounding pixels. Here, the monotonous increase means that it is larger or equal.

  Here, the above-described path with respect to the maximal point pixel means that, for example, the maximal point pixel (first pixel), the second pixel adjacent thereto, and the third pixel adjacent to the second pixel are included. In some cases, the density of the second pixel is equal to or less than the density of the maximum point pixel from the definition of the maximum point pixel, and the density of the second pixel inevitably has a monotonically decreasing relationship with the density of the maximum point pixel. Even when the density of the second pixel is the same as the density of the maximum point pixel, it is still considered that the density of the second pixel is monotonously decreased with respect to the density of the maximum point pixel. Furthermore, when the density of the third pixel is equal to or lower than the density of the second pixel, the density of the third pixel is in a monotonically decreasing relationship with the density of the second pixel. Even when the density of the third pixel is the same as the density of the second pixel, it is assumed that the density of the third pixel is in a monotonically decreasing relationship with the density of the second pixel. Is no different. After all, the density of the pixel on the path from the maximum point pixel through the second pixel to the third pixel changes in the order of the maximum point pixel density, the second pixel density, and the third pixel density. However, the density of these three pixels is monotonously decreasing.

  Alternatively, there is a fourth pixel adjacent to both the second and third pixels, the density of the fourth pixel is equal to or lower than the density of the second pixel, and the density of the third pixel is the fourth pixel. If the pixel density is equal to or lower than the pixel density, the density of the third pixel is monotonically decreased with respect to the density of the fourth pixel, and the density of the fourth pixel is monotonous with respect to the density of the second pixel. It is in a decreasing relationship. In the end, on the path from the maximum point pixel to the third pixel via the second and fourth pixels, the pixel density is the maximum pixel density, the second pixel density, and the fourth pixel density. The density changes in the order of the density of the third pixel, and the density of these four pixels decreases monotonously.

  FIG. 6 is a schematic flowchart of an example of processing of the extreme value region growing unit 300. The extreme value region growing unit 300 uses the search condition data 25 (the search condition data 25 ( Detection is performed from pixels within the maximum occupation range specified in FIG. As described below, the extreme value region growing unit 300 sequentially detects pixels in a monotonous change relationship with respect to the detected extreme value point pixel within the maximum occupation range, and the process is referred to as region growth. The processing unit 300 is referred to as a region growth unit, and the maximum occupation range is sometimes referred to as a region growth range.

  In the present embodiment, as illustrated in FIG. 2A, a rectangular area that is slightly larger than the vehicle occupation range S1 is assumed as the maximum occupation range R1 of the spot image. In this example, since the center of the maximum occupation range R1 coincides with the center of the roof existence range S2, there is no positional shift between the centers. Therefore, the maximum occupied range R1 for performing the region growth is a rectangular region including the detected extreme point pixel p at the center thereof. Here, it is assumed that the maximum occupation range R1 designated for the automobile of the detection target is a rectangular area having the center at the origin and the width and height of 2Si + 1 and 2Sj + 1, respectively. In this case, the maximum occupied range R1 is a rectangular region including the detected extreme point pixel p at the center and having a width and a height of 2Si + 1, 2Sj + 1, respectively.

  FIG. 7A is a diagram illustrating the maximum occupation range used in the present embodiment. Reference numeral 26 shows an example of the maximum occupation range R1 when the coordinates of the detected extreme point pixel exemplified in FIG. 2A are p (x, y). The coordinates of the pixels Q, R, S, and T at the vertex of the maximum occupied range 26 are (x-Si, y-Sj), (x + Si, y-Sj), (x + Si, y + Sj), and (x-Si, y + Sj), respectively. And the size of the maximum occupation range 26 is (2Si + 1) × (2Sj + 1).

  FIG. 7B shows the relationship between the pixels in the maximum occupied range 26 in FIG. 7A and the coordinate change amounts i and j from the x coordinate and y coordinate of the extreme point pixel p (x, y). FIG. In this figure, qm (m = 1 to 8, 21 to 36, or 41 to 50) shown in each square is a pixel number used for explanation. In the following description, FIG. 7B is mainly used instead of FIG. 7A. FIG. 7C is a diagram illustrating an example of the density of pixels within the maximum occupation range 26. The density values shown here are numerical values for explanation, and do not represent the actual image density.

  Returning to FIG. 6, the extreme value region growing unit 300 first sets the region growth success flag flag to the initial value NO in step S <b> 301. This value NO represents unsuccessful region growth. Next, the extreme value region adjacent pixel search unit 350 is executed. Each time the extremum area neighboring pixel search unit 350 is activated by the extremum area growing unit 300, an unprocessed pixel adjacent to the detected extremum point pixel p or an adjacent extremum point pixel p. When a pixel determined to be present in the value region has already been detected, an unprocessed pixel that is adjacent to the pixel determined to be in the extreme value region and is within the maximum occupation range is detected.

  In the present embodiment, the extreme value region adjacent pixel search unit 350 starts from the pixel q41 in the upper left corner of FIG. 7B and starts to change the pixels in the maximum occupied range 26 from left to right when activated for the first time. Search sequentially from top to bottom. That is, if the coordinates of the pixel to be searched are (x + i, y + j), i is changed from -Si to + Si in the state of j = -Sj, j is then increased by 1, and i is increased within the same range. Change with. The same is repeated until j exceeds Sj.

  FIG. 8 is a schematic flowchart of an example of processing performed by the extreme value region adjacent pixel search unit 350. Specifically, when activated, the extreme value region neighboring pixel search unit 350 determines whether or not the value of the region growth success flag flag is YES in step S351. When the extreme value region adjacent pixel search unit 350 is first activated by the extreme value region growth unit 300, the value of the region growth success flag flag is set to NO in step S301 of FIG. In this case, since the value of the region growth success flag flag is not YES, the search pixel is initially set in step S352. In the present embodiment, the pixel q41 at the upper left corner of the maximum occupation range 26 shown in FIG. 7B is set as the initial value of the search pixel. In step S353, the region attribute value of the search pixel is read from the region attribute data 22, and it is determined whether or not the region attribute value is “0”.

  FIG. 9 is a diagram illustrating a change example of the region attribute data 22 for the pixels in the maximum occupation range 26. In the figure, instead of the coordinates (x, y) of the corresponding pixel, the coordinates of the region attribute data 22 are derived from the x and y coordinates of the extreme point pixel p (x, y) as in FIG. 7B. The coordinate change amount (i, j) is shown. FIG. 11A shows a state in which the initial value setting unit 100 (FIG. 1) sets a value “0” indicating that the extreme region does not belong as the initial value of the region attribute data 22 for all the pixels of the captured image 21. Indicates. In FIG. 5B, when the extreme point pixel p is detected by the extreme point pixel search unit 200, the extreme value is set as the region attribute value for the extreme point pixel p in step S204 (FIG. 5). It is a figure which shows the state in which the value "1" which shows area | region affiliation was set. The region of the extreme point pixel p is the 0th-order maximum region M0 (p) referred to in Equation 1.

  As now assumed, when the extreme value region adjacent pixel search unit 350 is first activated and the pixel q41 (FIG. 7B) is a search pixel, the region attribute of the pixel is “0”. Returning to FIG. 8, when it is determined in step S353 that the region attribute of the search pixel is “0”, in step S354, the region attribute value “adjacent to the search pixel (that is, one of 8 neighboring pixels)” It is determined whether or not there is a pixel “1”. When the extreme region neighboring pixel search unit 350 is first activated as assumed, the pixel having the region attribute value “1” is only the detected extreme point pixel p. Therefore, since the search pixel q41 is not a pixel adjacent to the extreme point pixel p, it is determined in step S354 that there is no adjacent pixel having the region attribute value “1”.

  In this case, in step S355, the next pixel, that is, the right adjacent pixel q21 is set as a search pixel. Thereafter, in step S356, it is determined whether or not the increment j of the y coordinate of the new search pixel q21 has exceeded the limit value Sj. If not, the processes in and after step S353 are repeated. As a result, in the case of FIG. 7B, when the pixel q1 becomes the search pixel, the region attribute value that is adjacent to the pixel p becomes the pixel having “1” for the first time. Therefore, the determination is Yes in both steps S353 and S354. In that case, the extreme value region adjacent pixel search unit 350 succeeds in searching for an adjacent pixel, and returns to the extreme value region growth unit 300 with a return value of “success” in step S357.

  Returning to FIG. 6, in the extreme value region growing unit 300, in step S <b> 302, it is determined whether or not the search for the adjacent pixel is successful. The density of all the pixels (extreme region belonging pixels) r having the region attribute value “1” adjacent to (that is, 8 neighboring pixels) is obtained. In the present assumption, when the pixel q1 is a search success pixel, the adjacent extreme value region belonging pixel detected in step S303 is only the extreme value point pixel p. In step S304, it is determined whether or not the condition that at least one pixel equal to or higher than the density of the search pixel q1 is present among the obtained densities of all the extreme value region belonging pixels r is satisfied. When this condition is satisfied, the search success pixel q1 has a monotonously decreasing relationship with the adjacent extreme value region belonging pixel.

  As described above, in the present assumption, the extreme value region belonging pixel adjacent to the pixel q1 is only the extreme value pixel p, and the density thereof is the density of the pixel q1 in view of the definition of the extreme value pixel p. Not lower. Accordingly, in step S304, it is determined that the pixel q1 satisfies the monotonous change relationship. When the search pixel satisfies the monotonous change relationship, in step S305, the region attribute value of the pixel is changed to a value “0.5” representing the extreme region belonging candidate. In step S306, the region growth success flag flag is set to YES, and the process returns to the extreme value region adjacent pixel search unit 350.

  Referring to FIG. 8, in the case of the current assumption, the extreme value region adjacent pixel search unit 350 determines that the region growth success flag flag is YES in step S <b> 351. In this case, the process proceeds to step S355, and the next pixel q2 of the immediately preceding search pixel (q1 in FIG. 7B under the current assumption) is set as the search pixel. Thereafter, in step S356, it is determined that the difference between the y coordinate of the new search pixel q2 and the y coordinate of the extreme point pixel p is not larger than the limit value Sj of the distance, and the processing after step S353 is executed for the search pixel q2. The Since the search pixel q2 is also a pixel that does not belong to the extreme value region and is adjacent to the extreme value point pixel p, the extreme value region adjacent pixel search unit 350 succeeds in searching for an adjacent pixel, and the extreme value region growth unit 300 Return to.

  The extreme value region growing unit 300 (FIG. 6) executes the processing subsequent to step S303 in the same manner as the processing for the pixel q1 for the pixel q2. The pixel belonging to the adjacent region with respect to the pixel q2 is only the extreme point pixel p, and the pixel q2 is in a monotonous change relationship with the pixel p, which is the same as the pixel q1. Therefore, as in the case of the pixel q1, in step S305, the area attribute value of the searched adjacent pixel q2 is changed to “0.5”, and in step S306, the area growth success flag flag is set to YES again. Since the region growth success flag flag has already been set to YES, it is needless to say that the second and subsequent executions of step S306 may be omitted.

  The extreme value region growing unit 300 then executes the extreme value region adjacent pixel searching unit 350 again. In 350, each of the pixels q3 to q8 in FIG. The extreme value region growing unit 300 similarly performs the processing from step S303 onward for each of the pixels q3 to q8. In step S304, it is determined that these pixels are monotonically changing pixels for the same reason as described for the pixel q1, and therefore, the area attribute of the pixel is set to “0.5” as in the pixels q1 and q2. Be changed. FIG. 9C shows an example of region attribute data for pixels in the maximum occupation range 26 at this stage. In this way, the pixels q1 to q8 adjacent to the maximum point pixel p and in a monotonically decreasing relationship are detected. The region occupied by these pixels q1 to q8 is the first-order incremental region B1 (p) described in Equation 1.

  Returning to FIG. 6, after the pixel q8 is processed as described above, the processing returns to the extreme value region adjacent pixel search unit 350 again. In FIG. 8, the extreme value area adjacent pixel search unit 350 repeats the process from step S355 onward. In this case, as can be seen from FIG. 7B, in step S354, the adjacent extreme point pixel p is detected. It is determined that no other pixel exists. Therefore, when other pixels are sequentially determined as search pixels in step S355, and finally a pixel outside the search range (pixel (x−Si, y + Sj + 1) in FIG. 7A) is set as the search pixel, In step S356, it is determined that the increment j of the y coordinate of the search pixel has become larger than the limit value Sj of the distance. The process returns to the extreme value region growth unit 300 as “unsuccessful”.

  Returning to FIG. 8, when the extreme value region growing unit 300 determines that the search for the adjacent pixel is unsuccessful in step S <b> 302, in step S <b> 307, it is determined whether or not the region growth success flag flag is YES. If so, in step S308, the pixel having the region attribute value “0.5”, in this example, all the region attribute values of the pixels q1 to q8 are changed to the value “1” representing the extreme region belonging. change. FIG. 9D shows an example of region attribute values for pixels in the maximum occupation range 26 at this stage. Thus, at this stage, the region including the maximum point pixel p and the adjacent pixels q1 to q8 that are monotonously decreasing is detected as the extreme value region (maximum region in this example). The region occupied by these pixels p, q1 to q8 is the first maximum region M1 (p) described in Expression 1.

  Returning to FIG. 6, the extreme value region growing unit 300 thereafter repeats the processing from step S301 onward. That is, in step S301, the region growth success flag flag is changed to NO again, and the extreme value region adjacent pixel search unit 350 is activated. Referring to FIG. 9, when the extreme value region neighboring pixel search unit 350 is called from the extreme value region growth unit 300 and determines that the value of the region growth success flag flag is NO in step S <b> 351, it has already been described. In the same manner as described above, adjacent pixels are searched for all the pixels within the maximum occupation range 26. However, in this case, as shown in FIG. 9D, the region attribute data for the pixels in the maximum occupied range 26 is the extreme point pixel p and the eight pixels q1 to q8 surrounding it. The value is “1”. As a result, the search for adjacent pixels succeeds for the pixels adjacent to these eight pixels having the region attribute value “1”. That is, in the example of FIG. 7B, the search for adjacent pixels succeeds for each of the pixels q21 to q36.

  Returning to FIG. 6, the extreme value region growing unit 300 performs the processing from step S <b> 303 on and after each search pixel in the same manner as when the adjacent pixel search is successful for each of the pixels q <b> 2 to q <b> 8. However, as shown in FIG. 9D, when each of the pixels q21 to q36 is a search pixel, the adjacent pixel is not the extreme point pixel p but the pixels q2 to q8. It differs from the case of pixels q2 to q8 in that it is one, two or three. Accordingly, the determination result of whether or not the search pixel is in a monotonous change relationship in steps S303 and S304 differs depending on each of the pixels q21 to q36. For example, when the pixel q21 is an adjacent pixel, only the pixel q1 is a pixel adjacent to the pixel and having a region attribute value “1”. In the example of FIG. 7C, since the densities of the pixels q1 and q21 are both 1, it can be seen that the pixel q21 is in a monotonically decreasing relationship with respect to the pixel q1, and the determination result in step S304 is YES, step S305 The region attribute value for the pixel q21 is set to a value “0.5” indicating the extreme region affiliation candidate. Thereafter, the region growth success flag flag is set to YES in step S306, the process returns to the extreme value region adjacent pixel search unit 350, and the extreme value region adjacent pixel search unit 350 is executed.

  Referring to FIG. 8, the extreme region neighboring pixel search unit 350 is executed for the pixel q22 subsequent to the pixel q21 processed last because the region growth success flag flag is set to YES. . The pixel subsequent to the pixel q22 is processed in the same manner. However, if it is determined in step S304 in the extreme value region growing unit 300 that the pixel searched for as an adjacent pixel by the extreme value region adjacent pixel search unit 350 is determined not to satisfy the monotonic decrease relationship, Nothing is changed in the region attribute data corresponding to the pixel, and the process moves again to the extreme value region adjacent pixel search unit 350. For example, when the pixel q25 is determined as a search pixel by the extreme value region adjacent pixel search unit 350, in the extreme value region growth unit 300, the pixel belonging to the extreme value region adjacent to the pixel q25 is only the pixel q3. In the example of 7 (c), since the density of the pixel q25 is higher than the density of the pixel q3, it is determined that the pixel q25 is not in a monotonously decreasing relationship. Therefore, the region attribute value corresponding to the pixel q25 remains “0”. For pixels located at the boundary of the maximum occupation range 26, such as the pixel q25, other adjacent pixels whose density is compared with respect to whether the pixel is monotonously decreasing or monotonically increasing are included in the maximum occupation range 26. Only pixels (q3 only in the above example) are used.

  On the other hand, in the density example of FIG. 7C, all the pixels other than the pixel q25 are in a monotonously decreasing relationship. Therefore. After each of the pixels q21 to q36 is searched as an adjacent pixel and steps S304 to S306 are performed on the pixel, the region attribute values for the pixels q21 to q36 are “0.5” except for the region attribute value of the pixel q25. Changed to FIG. 9E shows region attribute data at this stage. That is, the region attribute values for the pixels q21 to q24 and q26 to q36 are changed to “0.5”. The area occupied by these pixels is the second incremental area B2 (p) described in Equation 1.

  In the extreme value region growing unit 300, after executing steps S305 and S306 for the pixel q36, the extreme value region adjacent pixel searching unit 350 is executed again. Referring to FIG. 8, in the extreme value region adjacent pixel search unit 350, after processing the pixel q <b> 36, the pixel adjacent to the pixel whose region attribute value is “1” is behind the pixel q <b> 36. 26, it is determined in step S356 that the increase j in the y coordinate of the search pixel is larger than the distance limit value Sj, and the extreme value region adjacent pixel search unit 350 searches for the adjacent pixel. Is unsuccessful, the return value is set to “unsuccessful”, and the process returns to the extreme value region growing unit 300. Since the region growth success flag flag is YES in this state, the region attribute value “0.5” for the pixels q21 to q24 and q26 to q38 is changed to the value “1” in step S308 as already described. FIG. 9F shows the value of the region attribute data 22 at this stage. Within the maximum occupied range 26, the region attribute value is 1 for pixels other than the pixel q25. The area occupied by the pixel having the attribute value “1” is an example of the second maximum area M2 (p) in Expression 1.

  Thereafter, in the extreme value region growing unit 300, the region growth success flag flag is set to NO in step S301, and the extreme value region adjacent pixel searching unit 350 is executed again. Similarly to the above-described processing related to the pixel q21, the search for the neighboring pixels having the region attribute “1” for each of the pixels q41, q42, q43, q44, q45, q46, q47, q48, q49, and q50 has succeeded, and Since these pixels are monotonously decreasing with respect to the pixels adjacent to the pixels, the region attribute value for each pixel is changed to “0.5” in step S305, and the region growth success flag flag is set to YES. . FIG. 9G shows the value of the region attribute data 22 at this stage. The area occupied by these pixels is the third-order incremental area B3 (p) described in Equation 1.

  After the above pixel processing, the extreme value region adjacent pixel search unit 350 is executed again, but the adjacent pixel search is unsuccessful and the return value is “unsuccessful”, and the processing is performed by the extreme value region growth unit 300. Return. Since the region growth success flag flag is YES in this state, the region attribute value “0.5” for the pixels q41 to q50 is changed to the value “1” in step S308 as described above. FIG. 9H shows the value of the region attribute data 22 at this stage. Within the maximum occupied range 26, the region attribute value is 1 for pixels other than the pixel q25. The area occupied by the pixels having the attribute value “1” is an example of the third maximum area M3 (p) in Expression 1.

  Thereafter, the extreme value region adjacent pixel search unit 350 is executed again, but since the processing for all the pixels within the maximum occupied range 26 has been completed, in step S356 (FIG. 8), the y coordinate of the search pixel is incremented. In step S358, the return value is set to “unsuccessful” and the process returns to the extreme value region growing unit 300. This means that the fourth-order incremental region B4 (p) in Equation 1 is an empty region. Therefore, as shown in Expression 2, it means that the pixel region (M3 (p)) of the value “1” shown in FIG. 9F becomes the final maximum region M (p). In the extreme value region growing unit 300, it is determined in step S302 that the search for adjacent pixels is unsuccessful, and if it is determined in step S307 that the region growth success flag flag is not YES, it is determined that the region growth has ended and spot image detection is performed. Return to program 40. Thus, the processing of the extreme value region growing unit 300 for the detected extreme value point pixel p is completed.

  FIG. 10A is a diagram schematically illustrating another example of the captured image data 21. FIG. 4B is a diagram showing an example of region attribute data for the captured image. The state of execution of region growth processing in the extreme value region growth unit 300 will be described more specifically with reference to these drawings. FIG. 6A shows the pixels in the maximum occupied range 26 with respect to the maximum point pixel (pixel surrounded by a double circle) having the maximum value “5” in the center, and FIG. The attribute values for the pixels in H.26 are shown.

  In the captured image of FIG. 9A, a plurality of pixels having a pixel value “5” are arranged. In the following, a pixel is designated by its coordinates (i, j). The local maximum pixel is the pixel (0, 0), and it is determined in step S203 (FIG. 5) of the local extreme point pixel search unit 200 that this pixel is the local maximum pixel. As in (b), it is assumed that the value “1” is set in step S204 (FIG. 5) of the extreme point pixel search unit 200. When the extreme value region growing unit 300 (FIG. 6) is executed for this pixel, pixels having the region attribute value “0” in the maximum occupied range 26 are sequentially selected from the pixels (−2, −3) in the upper left corner. It is determined whether the pixel is adjacent to the pixel having the region attribute value “1” (only the pixel (0, 0) at this stage) and the density is monotonously decreasing. As a result, the pixel (0, Each of the 8 pixels around 0) is determined to have a monotonic change relationship with respect to the maximum point pixel (0, 0), and the region attribute of each pixel is the extreme region affiliation candidate as in FIG. 9C. Is changed to the area attribute value “0.5”, and thereafter, the area attribute value “0.5” is changed to the value “1” as in FIG. 9D. At this stage, the region attribute value for the pixel having the value “5” in the pixel (0, 1) is also “1”.

  After that, when the processing of the extreme value region growing unit 300 is repeated again, pixels whose region attribute value within the maximum occupied range 26 is “0” are sequentially selected from the pixels (−2, −3) in the upper left corner, It is determined whether or not the pixel having the region attribute value “1” (at this stage, the maximum point pixel (0, 0) and the surrounding eight pixels) is monotonously decreasing. At this stage, the above determination is made for each of the 16 pixels around the 8 pixels, and as a result, the pixels (−2, −2) to (−2, 2), (−1, −2), ( -1, 2), (0, -2), (0, 2), (1, -2), (1, 2), (2, -2) are determined as extreme value region affiliation candidates, and the region The attribute value is changed to “0.5”. For example, the pixels (0, 2), (1, 2) having a density value of 5 are both adjacent to the pixel (0, 1) having a region attribute value “1”, and the density is the pixel. It is judged that there is a monotonically decreasing relationship with the concentration of. This is because the density values of the pixels (0, 2) and (1, 2) are the same as the density value “5” of the pixel (1, 2).

  On the other hand, the leftmost pixel (2, -1) among the four pixels having the value 5 surrounded by a square is not in a monotonously decreasing relationship with the adjacent pixel having the region attribute value "1". To be judged. The same applies to the two pixels (2, 0) and (2, 1) on the right side. Pixel (2, 2) is also not determined to be in a monotonous decreasing relationship. This is because the density of this pixel is higher than the density of any of the adjacent inner pixels whose area attribute value is “1”. It should be noted that this pixel (2, 2) is adjacent to the pixel (1, 2) one level above it, and the density is equal to the density, but at this time, the region attribute of the pixel (1, 2) The value is “0.5”, and even if it is adjacent to the pixel (2, 2), the extreme value region growing unit 300 does not make it a target for comparing the density with the pixel (2, 2). Thus, whether or not the pixel (1, 2) belongs to the extreme value region even if the pixel (1, 2) is an extreme value region affiliation candidate and the region attribute value is set to “0.5”. When determining, it is not determined whether or not there is a simple decrease relationship with the density of the pixel.

  As described above, when the extreme value region growing unit 300 determines whether or not the pixels around the maximum point pixel belong to the extreme value region, the density of the adjacent pixel whose region attribute value is “0.5” is determined. Whether or not there is a monotonic change relationship is determined by not determining whether the region attribute value is adjacent to a pixel having a value of “0.5”, but the pixel density on the path is a monotonous change relationship as a path from the maximum point p. This is because it can be surely prevented that a pixel having no path exists in the extreme value region by mistake. Therefore, in the present embodiment, a pixel (for example, (1,2,1) whose density is monotonously changed with respect to the pixel adjacent to the pixel (for example, the pixel (0,1)) whose region attribute value is “1”. ) Is detected, the region attribute value of the detected pixel (1, 2) is temporarily set to “0.5”, and the region growth is successful (step S306 (FIG. 6)). When the process of 350 is repeated and the search for adjacent pixels for the extreme value region is completed, the region attribute value of the pixel (1, 2) or the like whose region attribute value is “0.5” is changed to “1”. (Step S308 (FIG. 6)), and then the process of the extreme value region growing unit 300 is repeated again.

  In this example, at this stage, four pixels (2, -1) to (2, 2) to (2, 2) to pixels (-2, -3) to (2, -3) and ( The area attribute values of -2, 3) to (2, 3) are "0", and the area attribute values for all other pixels are "1". In this state, when the processing of the extreme value region growing unit 300 is repeated, the region attribute values of the pixels (−2, −3) to (2, −3) and (−2, 3) to (2, 3) are changed. The rightmost pixel (2, 2) of the above four pixels is adjacent to the pixel (1, 2) whose area attribute value is “1” immediately above it. In addition, since the densities of these two pixels are both equal to 5, it is determined that the density of the pixel (2, 2) is monotonically decreasing with respect to the density of the pixel (1, 2). The region attribute value of the pixel (2, 2) is set to a value “0.5” indicating the extreme region belonging candidate. The region attribute values of the other three pixels (2, −1), (2, 0), and (2, 1) remain “0”. Thereafter, the process of the extreme value region growing unit 300 is repeated, and the region attribute values are “−0.5” (−2, −3) to (2, −3) and (−2, 3) to (2, 3). The region attribute value of the pixel (2, 2) is set to “1”.

  Thereafter, the processing of the extreme value region growing unit 300 is repeated, and this time, the pixel (2, 1) is adjacent to the pixel (2, 2) having the region attribute value “1”, and the density is monotonously decreasing. And the region attribute value of the pixel (2, 1) is set to “0.5”. The region attribute values of the other two pixels (2, −1) and (2, 0) remain “0”. Thereafter, in the extreme value region growing unit 300, the region attribute value of the pixel (2, 1) having the region attribute value “0.5” is set to “1”. Thereafter, the processing of the extreme value region growing unit 300 is repeated. As a result, the pixel (2, 0) is now adjacent to the pixel (2, 1) having the region attribute value “1”, and the density is monotonous. It is determined that there is a decreasing relationship, and the region attribute value of the pixel (2, 0) is set to “0.5”. The area attribute values of the other pixels (2, −1) remain “0”.

  Thereafter, the processing of the extreme value region growing unit 300 is repeated, the region attribute value of the pixel (2, 0) having the region attribute value “0.5” is set to “1”, and the processing of the extreme value region growing unit 300 is repeated. It is. Thereafter, similarly, the region attribute value of the pixel (2, −1) is set to “1”, and then the processing of the extreme value region growing unit 300 is finished. As a result, for the captured image shown in FIG. 10A, the region attribute is “1” for all pixels in the maximum occupation range 26 as shown in FIG. As can be seen from the above processing, every time the processing of the extreme value region growing unit 300 is repeated, the extreme value region growing unit 300 searches for the pixel again from the initial value pixel in the maximum occupied range 26 to the entire maximum occupied range 26. It repeats with respect to the pixel of area | region attribute value "0" in. Thus, after the pixel search is repeatedly performed on the pixels having the region attribute value “0” in the entire maximum occupied range 26, the region attribute value of the pixel at the special position becomes “1”. However, it is possible to prevent a pixel having a region attribute value “0” adjacent to the pixel from being a search target.

As described above, when the processing of the extreme value region growing unit 300 ends, the processing returns to the main routine of the spot image detection program 40. Referring to FIG. 1, in the spot image detection program 40, an extreme value area determination unit 400 is executed. FIG. 11 is a schematic flowchart of an example of processing of the extreme value region determination unit 400. Here, it is determined whether or not the detected extreme value region is a region included in the spot image representing the automobile that is the detection target. First,
In step S401, it is determined whether or not all the pixels in the designated center minimum occupation range R2 are included in the detected extreme value region. If it is determined by this determination that any pixel in the designated center minimum occupation range R2 is not included in the detected extreme value region, the detected extreme value region is a spot for the detection target. It is determined that it does not represent an image, and in step S407, the return value is set to “fail” and the process returns to the main routine of the spot image detection program 40. If it is determined in step S401 that all the pixels in the specified center minimum occupation range R2 are included in the detected extreme value region, the detected extreme value region is determined in step S402. Is a maximum region, the maximum density M of the pixels in the peripheral portion in the maximum region may be obtained. Here, the peripheral pixels in the maximum region are pixels adjacent to pixels that are not the maximum region among the pixels in the maximum region. In other words, it is a pixel located at the outermost periphery among the pixels in the maximum region. In the case of the density distribution in FIG. 7C, the extreme value region is composed of pixels other than the pixel (−2, 2) among the pixels in the maximum occupied range 26, as shown in FIG. 9H. Therefore, the maximum density value M of the peripheral pixels in the maximum region is the density “2” of the pixels (−2, −1), (−1, 1), and (−1, 2).

  In step S403, it is determined whether or not the absolute value of the difference between the density of the extreme point pixel p and the detected maximum density value M is equal to or greater than the allowable minimum density difference T1 specified in the search condition data 25. When the result of the determination in step S403 is YES, in step S404, the minimum value m of the density within the central minimum occupation range R2 of the maximum area is detected. In step S405, it is determined whether or not the difference between the maximum pixel density and the minimum value m is equal to or smaller than the allowable maximum density difference T2 specified in the search condition data 25. When the extreme value region is the minimal region, whether or not the difference between the maximum value of the concentration within the central minimum occupation range R2 of the minimal region and the detected concentration of the minimal point is equal to or smaller than the allowable maximum concentration difference T2. Can be determined. When the determination result in step S405 is YES, it is determined that the detected extreme value region is a region representing a spot image representing the detection target, and in step S406, the extreme value region determination unit 400 returns. Return to the main routine of the spot image detection program 40 with the value “pass”.

  When the result of determination in either step S403 or S405 is NO, it is determined that the detected extreme value region is not a region representing a spot image representing the detection target, and in step S407, the extreme value region determination unit 400 is determined. Returns to the main routine of the spot image detection program 40 with the return value as “fail”.

  Returning to FIG. 1, in the spot image detection program 40, after the execution of the extreme value region determination unit 400, the detected extreme value region is detected in the extreme value region pass / fail determination step 450 based on the return value therefrom. It is determined whether or not it has passed as an extreme value region representing a spot image for use. If it is acceptable, the spot image related data generation unit 500 is executed. The spot image related data generation unit 500 generates spot image related data for the extreme value area determined to be acceptable and stores it in the storage device 20. The spot image related data generated by the spot image related data generation unit 500 includes extreme value region data 23 and image data 24 with a spot image marker.

  FIG. 12A is a diagram illustrating an example of the extreme value region data 23. As will be described later, for the extreme value region determined to be acceptable, the identification information of the extreme value region is determined, and among the detected extreme value regions, as the presence region of the spot image represented by the extreme value region, The minimum rectangular area ABCD surrounding all the pixels whose area attribute data is “1” is determined as a spot image existence area rectangle, and the determined extreme value area identification information 231 and the extreme value point are included in the extreme value area data 23. The coordinates (x, y) (232) of the pixel p and the coordinates (x, y) (233) of the vertices A, B, C, and D of the rectangle ABCD are stored.

  FIG. 6B schematically shows the image data 24 with spot image markers. The example of region attribute data obtained for the case of FIG. 7C has a value “1” for pixels other than the pixel q25 shown in FIG. 7B as shown in FIG. However, the spot image existence area rectangle ABCD is determined so as to include the pixel q25 as indicated by the reference numeral 242, although it has the value “0” for the pixel q25. The marker graphic 243 is drawn in addition to the outer periphery of the passed extreme value region 241 on the copy image of the captured image 21 to indicate that the passed extreme value region is a region representing the spot image representing the detection target. It is a figure to be made. In the example of FIG. 12B, a white band-like figure 243 having a one-pixel width indicating the spot image existence area rectangle ABCD (242) including the extreme value area 241 is used as a marker figure.

  FIG. 12C shows an example of the image data 24 with spot image markers for the spot image detected from the captured image 21 of FIG. A spot image surrounded by a white ellipse is shown. The figure shows that spot images of four large vehicles have been detected. This detection result is a detection result when the maximum occupation range R1 is designated as 25 × 17 pixels and the central minimum occupation range R2 is designated as 13 × 3 pixels. The allowable minimum density difference T1 is 130, and the allowable maximum density difference T2 is 90. In this search, there is a significant difference between the maximum occupation range R1 and the central minimum occupation range R2. As a result, the condition that the maximum occupation range is smaller than 25 × 17 pixels and the central minimum occupation range of the roof is larger than 13 × 3 pixels, rather than detecting a vehicle having a specific vehicle occupation range and a roof occupation range. Multiple types of automobiles are detected as detection objects.

  FIG. 13 is a schematic flowchart of an example of processing of the spot image related data generation unit 500. First, as already described, the spot image existence area rectangle ABCD including the passed extreme value area is determined in step S501. In the present embodiment, a maximum region may be detected at a different position with respect to the same vehicle. In the present embodiment, when such overlapping extreme value areas are detected, one of them is invalidated. For this reason, when the newly determined rectangle and the data of the other extremum area already detected are stored in the extremum area data 23 in step S502, the already determined extremum area is already determined. It is determined whether or not the rectangles overlapped. In the present embodiment, in determining the overlap of the extreme value areas determined to be acceptable, it is determined whether or not the minimum rectangle including the extreme value area determined to be acceptable overlaps. It is easier than accurately determining overlapping areas. Of course, it goes without saying that the overlap of the extreme value areas may be directly determined without determining the overlap of the extreme value area inclusion rectangles. Hereinafter, even when the extreme value region inclusion rectangles overlap, they may be simply referred to as extreme value region overlaps.

  If no rectangular overlap is detected, identification information is determined for the extreme value area determined to be acceptable in step S503 and stored in the field 231 of the extreme value area data 23 shown in FIG. In step S504, the x and y coordinates of the detected extreme point pixel p are stored in the field 232 of the extreme value region data 23. In step S 505, the x and y coordinates of the newly determined rectangular vertices A, B, C, and D are stored in the field 233 of the extreme value region data 23. In step S506, the position and size of the detected marker pattern for the extreme value region are determined, and the resultant is superimposed and drawn on the image data 24 with the spot image marker. By executing the spot image-related data generation unit 500 in this manner for each of the plurality of extreme value regions detected and determined to be acceptable, the image data with spot image markers as illustrated in FIG. 24 is generated.

  However, as a result of the determination in step S502, if it is determined that the newly detected passed extreme value region is to be superimposed on another extreme value region that has already been registered, in step S507, of the two overlapping rectangles Then, a more desirable rectangle is selected as a region representing the spot image based on a predetermined criterion. For example, when the densities of the maximum points in two overlapping extreme value regions are different, a rectangle including the higher density can be selected. Alternatively, a rectangle including the larger number of pixels included in each of the two extreme value regions can be selected. Alternatively, it is possible to select the one having the larger difference between the density of the maximum point pixel and the maximum value of the density of the peripheral pixels in the extreme value region. Furthermore, by combining several selection criteria as described above, a criterion for selecting two extreme value regions where the extreme value region inclusion rectangles overlap may be determined. For example, when the densities of the extreme points in the two extreme value regions are different, a rectangle including the extreme value region with the higher density is selected. You may choose a rectangle to contain.

  It is determined in step S508 whether another previously generated rectangle has been selected. If the determination result is No, in step S509, the image is attached to the other extreme value region from the spot image marker-added image 24. The value of the pixel constituting the marker graphic is rewritten to the value of the corresponding pixel in the original captured image 21. Thus, the marker graphic attached to the other extreme value area is deleted. In step S510, data related to the other extreme value region is deleted from the extreme value region data 23.

  By these processes, the other extreme value areas detected earlier are invalidated and are regarded as not detected. Thereafter, the process proceeds to step S503, and as described above, steps S503 to S506 are executed for the extreme value region related to the newly determined rectangle, and the process returns to the main routine of the spot image detection program 40.

  When another extreme value region is selected in step S507, and as a result, the determination result in step S508 is Yes, the process returns to the main routine of the spot image detection program 40. As a result, the passed extreme value region detected this time is regarded as not detected. Thus, the process of the spot image related data generation unit 500 is completed. Returning to FIG. 1, in the spot image detection program 40, the process proceeds to the initial value setting unit 100. When the initial value setting unit 100 is repeated after the second time, the maximum value centered on the extreme point pixel processed by the extreme value region growing unit 300 in the captured image data 21 is different from the first execution. Only the region attribute value in the region attribute data 22 is set to the initial value 0 for all pixels in the occupied range. This is because, as already described, when the extreme value region growing unit 300 is executed first, the region attribute values for all the pixels within the maximum occupied range may be changed. Thereafter, the search for new extreme pixels is continued by the extreme point pixel search unit 200. When a new extreme value pixel is found, the processing after the extreme value region determination unit 400 is executed as described above. As a result of repeating the above processing, when it is determined in the search success / failure determination step 250 that a new extreme pixel has not been detected, the spot image detection program 40 ends the processing. As described above, as illustrated in FIG. 12C, spot image marker-added image data 24 in which a marker graphic is added to a plurality of spot images representing a detection target is generated.

  As is clear from the above, in this embodiment, an extreme value region including a maximum point pixel, which is considered to be included in a spot image of a detection object such as an automobile, can be easily detected. Specify the maximum occupation range of the spot image that reflects the vehicle occupation area of the automobile you want to detect as an object and the center minimum occupation area of the spot image that reflects the roof presence area of the automobile and corresponds to the detection object It is possible to detect a car that performs.

  Needless to say, the present invention is not limited to the following embodiments. For example, in the above embodiment, the example in which the field attribute data 22 has a field that stores the area attribute value corresponding to all the pixels of the captured image data has been described. May have a field for holding a region attribute value. In this case, the region attribute data 22 is used in common for the maximum occupation range of all extreme point pixels. Therefore, the area attribute data having such an aspect requires less data. In order to detect a car having a darker density than the background, the minimum area may be detected in the same manner as the maximum area described above. Further, when the captured image is a color image, lightness data generated from the color image may be used as the density data.

1 is a schematic block diagram of an embodiment of a spot image detection device according to the present invention. It is a figure for demonstrating the search condition data designated at the time of the detection of the spot image by this invention. (A) is a figure which shows captured image data typically. (B) is a figure which shows the example of the actual geographic image obtained by actually image | photographing the ground from an artificial satellite. It is a figure which shows typically the area | region attribute data corresponding to captured image data. It is a schematic flowchart of an example of a process of an extreme point pixel search part. It is a schematic flowchart of an example of the process of an extreme value area | region growth part. (A) is a figure which illustrates the maximum occupation range. (B) is a figure which shows the relationship between the coordinate change amount i and j from the x coordinate of the pixel in the maximum occupation range, and the extreme value point pixel p (x, y), y coordinate. (C) is a figure which shows the example of the density | concentration of the pixel in the largest occupation range. It is a schematic flowchart of an example of the process of the adjacent pixel search part for extreme value areas. It is a figure which shows an example of the change of the area | region attribute data in the largest occupation range. It is a figure which shows the other example of a captured image, and the example of the area | region attribute data with respect to it. It is a schematic flowchart of an example of a process of an extreme value area | region determination part. (A) is a figure which shows the example of extreme value area | region data. (B) is a figure which shows typically the image data with a spot image marker. (C) is a figure which shows the example of image data with a spot image marker. 5 is a schematic flowchart of an example of processing of a spot image related data generation unit 500.

Explanation of symbols

  P: Maximum region, p: Maximum point, f (p): Concentration distribution, R1: Maximum occupied area of spot image, R2: Minimum occupied area of spot image, S1: Vehicle occupied area of vehicle, S2: Roof of vehicle Existence range, C1... Center of vehicle occupation range of car, C2. Center of automobile roof existence range, T1... Allowable minimum density difference, T2... Allowable maximum density difference, 241. 243 ... Spot image marker graphics.

Claims (4)

It is obtained by imaging a space where a plurality of objects including the detection target are located, and from a captured image including a plurality of spot images each representing one of the plurality of objects, the density is relative to the density of surrounding pixels. An extreme point pixel detection step of detecting at least one extreme point pixel having a density that is a predetermined extreme value of a maximum value or a minimum value;
As a range on the captured image that can be occupied by the spot image representing the detection target object, from the pixel within the maximum occupation range specified in advance for the detection target object, to the density of the detected extreme point pixel On the other hand, a monotonous change pixel group detection step for detecting a group of monotone change pixels whose density is in a monotone change relationship;
Depending on whether or not an extreme value region including the detected extreme value pixel and the monotonically changing pixel group detected for the extreme value pixel satisfies a predetermined condition, the extreme value region is An extreme value region determination step for determining whether or not the region represents a spot image for the detection target;
Including
The predetermined condition is: (a) a center specified in advance for the detection target as a range that should be occupied by a central portion of a spot image of the detection target in the detected monotonically changing pixel group; A plurality of pixels located within the minimum occupation range all belong to the monotonically changing pixel group, and (b) a density of the detected extreme point pixel and a density of the plurality of pixels within the central minimum occupation range Is equal to or smaller than an allowable maximum density difference specified in advance for the central minimum occupation range with respect to the detection target for any pixel, and (c) the detected extreme point pixel The density difference between the density and the density of a plurality of pixels located in the peripheral part of the detected monotonically changing pixel group is designated in advance for the maximum occupied range of the detection target for any pixel. Was is that tolerance is the minimum density difference or more,
The spot image detection method characterized by the above-mentioned. The object to be detected is a desired range of a specific part having a density different from that of other parts when picked up among a plurality of cars on the road, and the vehicle occupation range is the center of the specific part. A vehicle having a desired area and a desired area,
The maximum occupation range is designated as a range wider than the desired vehicle occupation range of the detection target vehicle, and the central minimum occupation range is designated as a value narrower than the existence range of the specific portion of the vehicle.
The spot image detection method according to claim 1. It is obtained by imaging a space where a plurality of objects including the detection target are located, and from a captured image including a plurality of spot images each representing one of the plurality of objects, the density is relative to the density of surrounding pixels. Extreme point pixel detection means for detecting at least one extreme point pixel having a density which is a predetermined one of the maximum value or the minimum value;
As a range on the captured image that can be occupied by the spot image representing the detection target object, from the pixel within the maximum occupation range specified in advance for the detection target object, to the density of the detected extreme point pixel A monotonous change pixel group detecting means for detecting a group of monotone change pixels having a monotonous change in density;
Depending on whether or not an extreme value region including the detected extreme value pixel and the monotonically changing pixel group detected for the extreme value pixel satisfies a predetermined condition, the extreme value region is Extreme value region determination means for determining whether or not the region represents a spot image with respect to the detection object;
Function as a computer
The predetermined condition is: (a) a center specified in advance for the detection target as a range that should be occupied by a central portion of a spot image of the detection target in the detected monotonically changing pixel group; A plurality of pixels located within the minimum occupation range all belong to the monotonically changing pixel group, and (b) a density of the detected extreme point pixel and a density of the plurality of pixels within the central minimum occupation range Is equal to or smaller than an allowable maximum density difference specified in advance for the central minimum occupation range with respect to the detection target for any pixel, and (c) the detected extreme point pixel The density difference between the density and the density of a plurality of pixels located in the peripheral part of the detected monotonically changing pixel group is designated in advance for the maximum occupied range of the detection target for any pixel. And is the minimum allowable density difference above is that,
A spot image detection program characterized by that. The object to be detected is a desired range of a specific part having a density different from that of other parts when picked up among a plurality of cars on the road, and the vehicle occupation range is the center of the specific part. A vehicle having a desired area and a desired area,
The maximum occupation range is designated as a range wider than the desired vehicle occupation range of the detection target automobile, and the central minimum occupation range is designated as a value narrower than the existence range of the specific portion of the vehicle.
The spot image detection program according to claim 3.