JP2009080572A - Apparatus and method for detecting moving body - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a moving body detection apparatus, capable of detecting a moving body with good accuracy from an image of a wide view angle obtained from a mobile body. <P>SOLUTION: The moving body detection apparatus includes an imaging device 10 for obtaining wide view angle images from the mobile body at first and second time points; an edge acquisition section 101 for obtaining edges of radial direction from the wide view angle images at the first and the second time points, respectively; a frequency distribution generation section 102 for generating edge frequency distribution by defining a frequency to be the magnitude of the edge in the radial direction at each of the first and the second time points; a clustering section 104 for clustering a consecutive plurality of histogram bins included in the edge frequency distribution at each of the first and the second time points, as an identical class; a moving amount estimation section 105 for estimating the moving amount of the class between the first and the second time points; a candidate detection section 106 for detecting moving image candidates based on the moving amount; and a moving body detection section 107 for detecting the moving body from the moving body candidates. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a moving object detection device and a moving object detection method for detecting a moving object from a wide viewing angle image.

  It is important for an autonomous moving body such as a robot to detect surrounding animals and perform an avoidance operation in order to perform safe traveling. Therefore, a camera capable of acquiring a wide viewing angle image has been developed. If a wide viewing angle image is used, the surrounding situation can be grasped even with a single camera element, and the downsizing of the apparatus and the trouble of synchronization processing can be saved. In addition, the moving body can detect a wide range of moving objects (obstacles) and can run independently. However, since a wide viewing angle image has a large distortion, there are few examples of performing moving object detection using a moving wide viewing angle image.

  As a technique using a wide viewing angle image during movement, there is a technique of creating an edge histogram from the wide viewing angle image and detecting a moving object (see, for example, Patent Documents 1 and 2). However, since edges are detected for the entire wide viewing angle image, the edge histogram is flattened in indoor facilities where there are many objects with circumferential edges on the wide viewing angle image, such as shelves. This makes it difficult to detect moving objects. In addition, since only a large edge is used for threshold processing, and a single similarity measure is used for the entire edge histogram, background information having a small edge cannot be used, and only moving objects that move in a specific direction can be detected. .

For example, in Patent Documents 1 and 2, small edge information for knowing the overall background flow cannot be used by applying threshold processing when creating an edge histogram. In these examples, using the fact that the car is moving forward, a large edge flowing from the front to the back is a stationary object, and a large edge flowing from the back to the front is a moving object. Therefore, a moving object cannot be detected in a robot that moves in various directions.
JP 2001-301552 A JP 2001-8196 A

  An object of the present invention is to provide a moving object detection apparatus and a moving object detection method capable of accurately detecting a moving object from a wide viewing angle image from a moving object.

  According to one aspect of the present invention, (b) an imaging device that acquires a wide viewing angle image from a moving body at first and second times, and (b) a wide field of view at each of first and second times. An edge acquisition unit that acquires each radial edge from a corner image; and (c) a frequency distribution generation unit that generates an edge frequency distribution using the size of each radial edge at the first and second times as a frequency, and (D) a clustering unit that clusters a plurality of continuous histogram bins included in each edge frequency distribution at each of the first and second times as the same class; and (e) between the first and second times. A movement amount estimation unit that estimates a movement amount of a class, (f) a candidate detection unit that detects a moving object candidate based on the movement amount, and (g) a moving object detection unit that detects a moving object from the moving object candidate. animal Detecting apparatus is provided.

  According to another aspect of the present invention, (b) a step of acquiring a wide viewing angle image from a moving body, and (b) a radial edge from each of the wide viewing angle images at the first and second times. (C) a step of creating an edge frequency distribution using the size of the radial edge at each of the first and second times as a frequency, and (d) each of the first and second times. Clustering a plurality of consecutive histogram bins included in the edge frequency distribution as the same class, (e) estimating the amount of movement of the class between the first and second times, and (f) moving An object detection method is provided that includes a step of detecting a moving object candidate based on the amount, and (g) a step of detecting the moving object from the moving object candidate.

  ADVANTAGE OF THE INVENTION According to this invention, the moving body detection apparatus and moving body detection method which can detect a moving body accurately from the wide viewing angle image from a moving body can be provided.

  Next, embodiments of the present invention will be described with reference to the drawings. In the following description of the drawings, the same or similar parts are denoted by the same or similar reference numerals. However, it should be noted that the drawings are schematic, and the relationship between the thickness and the planar dimensions, the ratio of the thickness of each layer, and the like are different from the actual ones. Therefore, specific thicknesses and dimensions should be determined in consideration of the following description. Moreover, it is a matter of course that portions having different dimensional relationships and ratios are included between the drawings.

(Animal body detection device)
As shown in FIG. 1, the moving object detection device according to the embodiment of the present invention includes an imaging device 10 that acquires wide viewing angle images from a moving body at first and second times, and first and second. An edge acquisition unit 101 that acquires radial edges from each wide viewing angle image at each time, and a frequency that creates an edge frequency distribution with the size of each radial edge at the first and second times as frequencies. A distribution creation unit 102; a clustering unit 104 that clusters a plurality of continuous histogram bins included in each edge frequency distribution at each of the first and second times as the same class; and between the first and second times A moving amount estimating unit 105 for estimating the moving amount of the class, a candidate detecting unit 106 for detecting a moving object candidate based on the moving amount, and detecting a moving object from the moving object candidate And a object detection unit 107.

  The edge acquisition unit 101, the frequency distribution creation unit 102, the clustering unit 104, the movement amount estimation unit 105, the candidate detection unit 106, and the moving object detection unit 107 are a central processing unit (module) as a hardware resource module (logic circuit). CPU) 100 is logically provided. The dividing unit 103 is a module (logic circuit) as a hardware resource that divides the edge frequency distribution in the moving direction of the moving object. A left frequency distribution storage device 108, a right frequency distribution storage device 109, a left class storage device 110, a right class storage device 111, a data storage device 112, and the like are connected to the CPU 100. Here, the right side and the left side are defined by defining an arbitrary angle of the wide viewing angle image as an origin and defining one as a right side and the other as a left side with respect to the origin.

  The imaging device 10 is installed on a moving body such as a robot. As the imaging device 10, a camera equipped with an optical device for securing a wide field of view such as a curved mirror or a fisheye lens can be used. The imaging device 10 captures an annular wide viewing angle image as shown in FIG. 2 at the first and second times. Moving bodies (persons) A1 and A2 are shown at positions P1 and P2 in the wide viewing angle image shown in FIG.

  The edge acquisition unit 101 performs panoramic development about the annular wide viewing angle images acquired by the imaging device 10 at the first and second times with the moving direction of the moving body as the center, as shown in FIG. Get panoramic images. In the panoramic image shown in FIG. 3, the vertical axis direction represents the radial direction, and the horizontal axis direction represents the circumferential direction.

  Further, the edge acquisition unit 101 acquires a radial edge having a strong intensity in the radial direction as shown in white in FIG. 4 by differentiating the density of each pixel of the panoramic image in the vertical axis direction (circumferential direction). To do.

  The frequency distribution creating unit 102 creates an edge frequency distribution with the size of the radial edge of each azimuth angle acquired by the edge acquiring unit 101 as the frequency. Here, the size of the edge in the radial direction refers to the cumulative frequency of the edge in the radial direction or the cumulative luminance of the edge. FIG. 5 shows an edge frequency distribution diagram (edge histogram) expressed by an edge frequency distribution. The plurality of histogram bins constituting the edge frequency distribution are set to one-degree intervals here, but are not particularly limited.

  The dividing unit 103 divides the edge frequency distribution created by the frequency distribution creating unit 102 into the left side and the right side at the origin (for example, the traveling direction of the moving body). For example, the 180 degree direction indicated by the dotted line in FIG. 5 is the traveling direction, the left edge frequency distribution is 0 to 180 degrees, and the right edge frequency distribution is 180 to 360 degrees. The left edge frequency distribution and the right edge frequency distribution are stored in the left frequency distribution storage device 108 and the right frequency distribution storage device 109, respectively.

  The clustering unit 104 reads the threshold value stored in the data storage device 112, and the frequency is greater than 0 in each of the right edge frequency distributions at the first and second times stored in the right frequency distribution storage device 109. Among histogram bins, a plurality of continuous histogram bins whose distances on the azimuth angle in the circumferential direction are within a threshold value are clustered as the same class. This threshold value can be set arbitrarily. Class information such as the class number assigned to each class, the number of histogram bins (number of members), the edge frequency distribution constituting the class, the azimuth angle of the histogram bin as the start point of the class, and the azimuth angle of the histogram bin as the end point The right class storage device 111 stores each class.

  FIG. 6 shows an example of clustering the right edge frequency distribution of 180 degrees to 360 degrees. In FIG. 6, the histogram bin is indicated by a circle mark. At the first time t−Δt, class numbers 1 to 10 are assigned for each class. At the second time t, class numbers 1 to 9 are assigned for each class.

  Similarly, the clustering unit 104 reads the threshold value stored in the data storage device 112, and the left edge of each of the first time t−Δt and the second time t stored in the left frequency distribution storage device 108. In the frequency distribution, among histogram bins whose frequency is greater than 0, a plurality of consecutive histogram bins whose distances on the circumferential direction azimuth are within a threshold are clustered as the same class, and class information is stored in the left class. Store in device 110.

  The movement amount estimation unit 105 stores the right edge frequency distribution at the first time t−Δt and the second time t stored in the right frequency distribution storage device 109 and the right class storage device 111. The respective class information at the first time t−Δt and the second time t is read out. For the class number n (n = 1, 2, 3..., The total number of classes at the first time t−Δt) at the first time t−Δt, the edge frequency distribution at the second time t using the similarity measure Correspondingly, the movement amount of the class between the first time t−Δt and the second time t is estimated.

Here, a plurality of types of similarity measures are defined so that both small edges and large edges can be handled, and the similarity measures are selected and used according to the characteristics of the edge frequency distribution constituting each class. In the embodiment of the present invention, “bin matching degree” and “frequency difference degree” are calculated as a plurality of types of similarity measures. When the frequency on the azimuth angle for shifting the class of the class number n at the first time t−Δt is set to shift, the bin coincidence degree Cn (shift) and the frequency difference degree Rn (shift) are, for example, formulas (1), It is defined as (2).

Here, Ht (k) represents the histogram bin frequency of the azimuth angle k at the second time t, sn represents the azimuth angle that is the starting point of the class number n, and fn represents the azimuth angle that is the end point of the class number n. . The function Z (•) returns 1 if the argument is positive, and returns 0 if the argument is 0 or less.

  When the frequency difference degree is used between the first time t−Δt class and the second time t class as shown in FIG. 7, the frequency difference of the histogram bins is reduced as shown in FIG. 8. Is associated with. On the other hand, the class at the first time t−Δt and the class at the second time t as shown in FIG. 9 correspond so that the histogram bins overlap as shown in FIG. Attached.

  Here, as shown in FIG. 11, when noise or the like is applied to a class having a small edge as a whole and having a relatively large number of members, the frequency difference is used as shown in FIG. There is a case where an error occurs in the movement amount estimation due to a large histogram bin. On the other hand, when the bin coincidence is used when the number of members is small, an error may occur in the movement amount estimation due to a background edge that is generated and disappears like noise. Thus, when only one similarity measure is used, it may not be possible to cope with noise or the like. On the other hand, in the embodiment of the present invention, it is possible to estimate a more appropriate movement amount by using a plurality of types of similarity scales.

  Here, an example of the movement amount estimation method will be described with reference to the flowchart of FIG.

  (A) In step S101, the initial value of “Shift” in degrees on the azimuth angle for shifting the class of class number n at the first time t−Δt is set to “−S_thresh−1”. In step S102, “Shift” is increased once. In step S103, it is determined whether “Shift” has reached the azimuth angle “S_thresh” of the end point of the class. If “Shift” has not reached “S_thresh”, the bin coincidence is calculated in step S104, and the frequency difference is calculated in step S105. Returning to step S102, “Shift” is increased once. As a result, for the class number n at the first time t−Δt, the bin coincidence and the frequency difference are shifted one by one from the azimuth angle “−S_thresh” of the class start point to the azimuth angle “S_thresh” of the class end point. Each of the degrees is compared with the edge frequency distribution at the second time t.

  (B) In step S106, the number of members of the class number n is compared with the threshold value read from the data storage device 112. When the number of members of the class number n is equal to or smaller than the threshold, in step S107, the movement amount “shift_Rn” when the frequency difference is the minimum is calculated and set as the class movement amount “shift_n”. In step S108, “shift_Rn” is compared with the threshold value read from the data storage device 112. When “shift_Rn” is smaller than the threshold value, “shift_Rn” is estimated as the movement amount and output. On the other hand, if “shift_Rn” is larger than the threshold value, “shift_Rn” is determined as an abnormal value, and the shift amount “shift_Cn” having the largest bin matching degree is set to “shift_n” in step S109. In step S110, “shift_Cn” is compared with a threshold value. When “shift_Cn” is larger than the threshold value, “shift_Cn” is estimated as a movement amount and output. On the other hand, if “shift_Cn” is smaller than the threshold value, “shift_Cn” is also determined to be an abnormal value, and the value of “shift_n” is set to “NULL” in step S115 and is not used for the subsequent processing. As described above, it is possible to determine the class movement abnormality after the association when the bin coincidence degree and the frequency difference degree are respectively used, and to appropriately estimate the class movement amount.

  (C) If the number of members of class number n is larger than the threshold value in step S106, “shift_Cn” is set to “shift_n” in step S111. In step S112, “shift_Cn” is compared with the threshold value read from the data storage device 112. When “shift_Cn” is larger than the threshold value, “shift_Cn” is estimated as a movement amount and output. On the other hand, if “shift_Cn” is smaller than the threshold value, “shift_Cn” is determined to be an abnormal value, and “shift_Rn” is calculated in step S113 to be “shift_n”. In step S113, “shift_Rn” is compared with a threshold value. When “shift_Rn” is smaller than the threshold value, “shift_Rn” is estimated as the movement amount and output. On the other hand, if “shift_Rn” is larger than the threshold value, “shift_Rn” is also determined to be an abnormal value, and the value of “shift_n” is set to “NULL” in step S115 and is not used for subsequent processing. “Shift_n” obtained in this way is “Shift_n”, and the circle mark between the class at the first time t−Δt and the class at the second time t in FIG. 6 is the first time t−. The class at Δt is moved by the movement amount “Shift_n”.

  Similarly, the movement amount estimation unit 105 receives the left edge frequency distribution from the left frequency distribution storage device 108 at the first time t−Δt and the second time t, and the first time t− from the left class storage device 110. The class information of the left edge frequency distribution at each of Δt and the second time t is read out, and the movement amount between the first time t−Δt and the second time t of the class created from the left edge frequency distribution is calculated. presume.

  The candidate detection unit 106 uses the representative value of the movement amount of each class estimated by the movement amount estimation unit 105 as a background flow, and detects a class having a movement amount significantly different from the background flow as a moving object candidate. That is, the candidate detection unit 106 determines the variance of “shift_n” except for the movement amount “shift_n” of each class of the left edge frequency distribution estimated by the movement amount estimation unit 105, except for the value “NULL”. An intermediate value is calculated, and the intermediate value is obtained as a representative value. Then, the threshold value is read from the data storage device 112 and the variance is compared with the threshold value. On the other hand, if the variance is smaller than the threshold value, the absolute value of the difference between the movement amount of each class and the threshold value are compared, and a class having a movement amount whose absolute value is equal to or greater than the threshold value is detected as a moving object candidate. . This threshold value may be varied according to the variance or may be fixed. Further, an average value or a mode value may be used for the intermediate value.

  Similarly, the candidate detection unit 106 detects a specific class as a moving object candidate based on the movement amount “shift_n” of each class of the left edge frequency distribution estimated by the movement amount estimation unit 105.

  The moving object detecting unit 107 distributes the movement amount of all classes, the frequency of the edge frequency distribution of the class that becomes a moving object candidate, the number of members, and the histogram bins independently for the class that is determined as the moving object candidate by the candidate detecting unit 106. Based on the gender, etc., it is determined whether or not the class is a moving object. In other words, the determination is based on an index representing the character of an animal such as “the number of members is equal to or greater than the threshold” or “there are few histogram bins isolated in the class”. These indicators may be stored in the data storage device 112 in advance. When these indices are satisfied, the azimuth angle where the corresponding class exists is detected and output as the position of the moving object. As a result, in FIG. 6, the class of class number 8 at the second time t is detected as the moving object, and the position P2 on the azimuth angle is detected.

  Similarly, the moving object detection unit 107 determines whether the class of the left edge frequency distribution determined as the moving object candidate by the candidate detection unit 106 is a moving object, and detects the moving object.

  Examples of the left frequency distribution storage device 108, the right frequency distribution storage device 109, the left class storage device 110, the right class storage device 111, and the data storage device 112 include a semiconductor memory, a magnetic disk, an optical disk, a magneto-optical disk, and a magnetic tape. It can be adopted. As the semiconductor memory, for example, a ROM and a RAM can be adopted. The ROM can function as a program storage device that stores a program executed by the CPU 100 (details of the program will be described later). The RAM can temporarily store data and the like used during program execution processing in the CPU 100, and can function as a temporary data memory or the like used as a work area.

(Animal body detection method)
Next, the moving object detection method according to the embodiment of the present invention will be described with reference to the flowchart of FIG.

  (A) In step S10, as shown in FIG. 2, the imaging device 10 captures and acquires wide viewing angle images from the moving body at the first time t-Δt and the second time t.

  (B) In step S11, the edge acquisition unit 101 panoramicly develops the wide viewing angle image acquired by the imaging device 10, as shown in FIG. Further, as shown in FIG. 4, the edge acquisition unit 101 differentiates in the circumferential direction of the wide viewing angle images at the first time t−Δt and the second time t, so that the wide viewing angle image is displayed. Obtain radial edges from each.

  (C) In step S12, the frequency distribution creating unit 102 generates an edge frequency distribution with the size of each radial edge at the first time t−Δt and the second time t as the frequency, as shown in FIG. Create each one.

  (D) In step S13, the dividing unit 103 divides each edge frequency distribution at the first time t−Δt and the second time t created by the frequency distribution creating unit 102 in the traveling direction.

  (E) In step S14, the clustering unit 104 clusters a plurality of continuous histogram bins included in the edge frequency distribution as the same class as shown in FIG. The class information of the left edge frequency distribution is stored in the left class storage device 110 in time series, and the class information of the right edge frequency distribution is stored in the right class storage device 111.

  (F) In step S15, the movement amount estimation unit 105 reads out the classes of the first time t-Δt and the second time t from the left class storage device 110 and the right class storage device 111, and obtains a plurality of similar measures. Using this, the movement amount of the class between the first time t−Δt and the second time t is estimated. For example, the movement amount is estimated using the frequency difference degree and the bin coincidence degree in the procedure of steps S101 to S115 shown in FIG.

  (G) In step S <b> 16, the candidate detection unit 106 detects a specific class as a moving object candidate from a plurality of classes based on the movement amount estimated by the movement amount estimation unit 105.

  (H) In step S17, the moving object detection unit 107 determines whether the class detected as the moving object candidate by the candidate detection unit 106 is a moving object, and when determining that the moving object is an moving object, the azimuth angle of the moving object Detect the upper position.

  According to the embodiment of the present invention, by acquiring and using a radial edge with relatively small distortion in a wide viewing angle image, background information having a small edge can be used, and threshold processing is not performed. An available edge frequency distribution can be created. Therefore, it is possible to detect a moving object with high accuracy using a wide viewing angle image from a moving object such as a robot.

  Further, by using a plurality of types of similarity scales, it is possible to remove an influence such as noise and estimate an appropriate movement amount.

(program)
Further, a series of procedures shown in FIG. 14, that is, the step in which the imaging device 10 captures and acquires a wide viewing angle image from the moving moving body; the edge acquisition unit 101 acquires the first image acquired by the imaging device 10. A step of acquiring radial edges from the wide viewing angle images at time t-Δt of time 1 and time t of second time; a frequency distribution creation unit 102 creates an edge frequency distribution with the size of the radial edge as a frequency A step in which the dividing unit 103 divides the edge frequency distribution in the traveling direction; a clustering unit 104 clusters a plurality of continuous histogram bins included in the edge frequency distribution as the same class, and classifies the classes in time series. Step of storing in the left class storage device 110 and the right class storage device 111; The class of the first time t−Δt and the second time t is read from the right class storage device 111, and a class between the first time t−Δt and the second time t is used using a plurality of types of similarity measures. A step in which the candidate detection unit 106 detects a specific class from among a plurality of classes as a moving object candidate based on the movement amount; The step of detecting the moving object based on the detected body class and the class can be executed by controlling the moving object detecting apparatus shown in FIG. 1 by the algorithm program equivalent to FIG.

  This program may be stored in the data storage device 112 of the computer system constituting the moving object detection device of the present invention. The program can be stored in a computer-readable recording medium, and the recording medium can be read into the data storage device 112 of the moving object detection apparatus, thereby executing a series of procedures according to the embodiment of the present invention. it can.

  Here, the “computer-readable recording medium” means a medium capable of recording a program such as an external memory device of a computer, a semiconductor memory, a magnetic disk, an optical disk, a magneto-optical disk, and a magnetic tape. To do. Specifically, a flexible disk, CD-ROM, MO disk, etc. are included in the “computer-readable recording medium”. For example, the main body of the moving object detection apparatus can be configured to incorporate or externally connect a flexible disk device (flexible disk drive) and an optical disk device (optical disk drive).

  A flexible disk is inserted into the flexible disk drive, and a CD-ROM is inserted into the optical disk drive from the insertion slot, and the program stored in these recording media is stored as data by performing a predetermined read operation. It can be installed on the device 112. Further, for example, a ROM or a magnetic tape device can be used by connecting a predetermined drive device. Furthermore, this program can be stored in the data storage device 112 via an information processing network such as the Internet.

(Other embodiments)
As described above, the present invention has been described according to the embodiment. However, it should not be understood that the description and drawings constituting a part of this disclosure limit the present invention. From this disclosure, various alternative embodiments, examples, and operational techniques will be apparent to those skilled in the art.

  In the embodiment of the present invention, the bin coincidence degree and the frequency difference degree are used as the similarity measure. However, the similarity measure defined here is an example. Any similar scale may be used as long as it is focused on size. For example, a similarity measure such as an absolute difference (SAD) or a histogram intersection can be used instead of the frequency difference. In the embodiment of the present invention, two similarity measures are used, but more similarity measures may be used.

  In the embodiment of the present invention, an example in which the moving object detection is performed on the wide viewing angle images at the first time t−Δt and the second time t has been described. Imaging may be performed, and the moving object detection may be performed from the amount of movement of the class between a plurality of times.

  As described above, the present invention naturally includes various embodiments not described herein. Accordingly, the technical scope of the present invention is defined only by the invention specifying matters according to the scope of claims reasonable from the above description.

It is a block diagram which shows an example of a logical structure of the moving object detection apparatus which concerns on embodiment of this invention. It is an image which shows an example of a wide viewing angle image. It is an image which shows an example of the image which panoramic-expanded the wide viewing angle image. It is an image which shows an example of a radial direction edge. It is an image which shows an example of the edge histogram expressed by edge frequency distribution. It is an image which shows the clustering result of edge frequency distribution of 1st time and 2nd time. It is the schematic which shows an example of the edge histogram expressed by the edge frequency distribution of the class of 1st time and 2nd time. It is the schematic for demonstrating the class movement amount estimation method by a frequency difference degree. It is the schematic which shows another example of the edge histogram expressed by the edge frequency distribution of the class of 1st time and 2nd time. It is the schematic for demonstrating the class movement amount estimation method by bin matching degree. It is the schematic which shows the edge histogram represented by the edge frequency distribution of the class with noise. It is the schematic for demonstrating the problem of the movement amount estimation using a frequency difference degree. It is a flowchart for demonstrating an example of the class movement amount estimation process which concerns on embodiment of this invention. It is a flowchart for demonstrating an example of the moving body detection method which concerns on embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Imaging device 100 ... Central processing unit (CPU)
DESCRIPTION OF SYMBOLS 101 ... Edge acquisition part 102 ... Frequency distribution creation part 103 ... Dividing part 104 ... Clustering part 105 ... Movement amount estimation part 106 ... Candidate detection part 107 ... Moving object detection part 108 ... Left frequency distribution storage device 109 ... Right frequency distribution storage device 110: Left class storage device 111 ... Right class storage device 112 ... Data storage device

Claims (6)

An imaging device that acquires a wide viewing angle image from a moving body at first and second times;
An edge acquisition unit for acquiring a radial edge from each of the wide viewing angle images at each of the first and second times;
A frequency distribution creating unit that creates an edge frequency distribution using the size of the radial edge at each of the first and second times as a frequency; and
A clustering unit that respectively clusters a plurality of continuous histogram bins included in the edge frequency distribution at each of the first and second times as the same class;
A movement amount estimation unit for estimating a movement amount of the class between the first and second times;
A candidate detection unit for detecting moving object candidates based on the amount of movement;
A moving object detection device comprising: a moving object detection unit that detects a moving object from the moving object candidates. The edge acquisition unit
Panoramic development of the wide viewing angle image,
The moving object detection apparatus according to claim 1, wherein the edge in the radial direction is acquired by taking a differential in a circumferential direction of the panoramic wide-angle image. The movement amount estimation unit
The movement amount of the class between the first and second times is estimated by associating the classes of the first and second times with each other using a plurality of types of similarity measures. Item 3. The moving object detection device according to Item 1 or 2. The multiple types of similarity measures are:
A first similarity measure that takes into account the frequency difference of the histogram bins;
The moving object detection apparatus according to claim 3, further comprising: a second similarity measure that takes into account the degree of overlap of the histogram bins.   The moving object detection apparatus according to claim 1, further comprising a dividing unit that divides the edge frequency distribution in a traveling direction of the moving body. Obtaining a wide viewing angle image from a moving object;
Obtaining a radial edge from each of the wide viewing angle images at each of the first and second times; and
Creating an edge frequency distribution using the size of the radial edge at each of the first and second times as a frequency; and
Clustering a plurality of successive histogram bins included in each edge frequency distribution at each of the first and second times as the same class,
Estimating the amount of movement of the class between the first and second times;
Detecting a moving object candidate based on the amount of movement;
A moving body detecting method comprising: detecting a moving body from the moving body candidates.