JP2014085721A - Moving celestial body detection apparatus and control method thereof - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect a moving celestial body more properly in images of multiple frames captured by changing a direction of an imaging apparatus according to diurnal motion, in consideration of noise characteristics or influence of weather.SOLUTION: A moving celestial body detection apparatus for detecting a moving celestial body imaged in a line segment-shape in image data captured by changing a direction of an imaging apparatus according to diurnal motion includes: acquisition means for acquiring image data; calculation means for calculating a sky level and noise standard deviation in the image data for each partial region obtained by dividing the image data; normalization means for normalizing the image data for each partial region, on the basis of the sky level and the noise standard deviation; and detection means for detecting a moving celestial body imaged in a line segment-shape in the image data normalized by the normalization means.

Description

  The present invention relates to a technique for detecting a moving celestial body from an image obtained by photographing a sky.

  In space from about 200 km to 36000 km from the ground, the existence of artificial satellites and used rockets that have passed their lifetimes, and broken pieces that collided with them has become a problem. These artificial moving objects are called space debris or space debris. Space debris is flying at high speed in outer space. There is a serious problem that the space debris collides with an operating satellite or space station, and the necessary satellite or space station is destroyed. In order to avoid such problems, it is desired to observe space from the ground with a telescope to detect space debris and monitor them.

  As a method for observing a moving celestial object such as space debris, a method of taking an image by changing the direction of the imaging device in accordance with the diurnal motion of a stationary celestial object such as a star is known. At this time, the moving celestial body is observed as a linear image in the image. Patent Document 1 discloses a method for specifying a target object by calculating a motion trajectory of an object to be detected over a plurality of frames and adding image data on the trajectory.

Japanese Patent Laid-Open No. 2002-220098

  However, image data obtained by observing a celestial object may not be able to detect a moving celestial object appropriately due to the influence of fog, clouds, etc. or noise from an image sensor. In the method disclosed in Patent Document 1, since the motion trajectory is calculated without considering the undesired components generated in the image data as described above, the moving celestial object may not be detected correctly. Therefore, the present invention has an object to more appropriately detect a moving celestial object in consideration of noise characteristics and the influence of weather in a plurality of frames of images taken by changing the direction of the imaging device in accordance with the diurnal motion. .

  In order to solve the above-described problems, the present invention provides a moving celestial body detection device that detects a moving celestial body that is captured in a line segment in image data that is taken by changing the direction of the imaging device in accordance with the diurnal motion, Obtaining means for obtaining, for each partial region obtained by dividing the image data, calculating means for calculating a sky level and a noise standard deviation in the image data, the sky level and the noise standard deviation, A normalization processing means for normalizing the image data for each of the partial areas, and a moving celestial object that is imaged in a line segment with respect to the image data that has been normalized by the normalization processing means. And detecting means for detecting.

  According to the present invention, in a multi-frame image captured by changing the direction of the imaging device in accordance with the diurnal motion, it is possible to more appropriately detect a moving celestial object in consideration of noise characteristics and the influence of weather.

Block diagram of moving object detection device Block diagram showing the detailed configuration of the image data correction processing unit Block diagram showing the detailed configuration of the data generation unit for moving celestial object detection processing Flowchart showing generation processing of moving object detection processing data Example of detection filter A photograph of the sky when you can observe celestial bodies when clouds are appearing Diagram explaining the principle of dimming processing Example of arrangement of end points of moving objects Block diagram showing the detailed configuration of the moving object detection unit Flow chart of processing for generating detection data for each pixel Block diagram showing the detailed configuration of the normalization processing unit Flow chart showing the flow of normalization processing Block diagram showing the detailed configuration of the intra-frame integration processing unit A flowchart of processing for generating detection data in a frame by integrating detection data for each pixel Block diagram showing details of interframe integration processor A flowchart of processing for generating detection data of moving objects by integrating in-frame detection data Block diagram showing details of identity determination unit Flowchart of processing for determining the same celestial nature of the in-frame detection data and the moving celestial detection data

  Preferred embodiments of the present invention will be described below with reference to the accompanying drawings. In addition, the structure shown in the following Examples is only an example, and this invention is not limited to the structure shown in figure.

<First Embodiment>
● Outline of Moving Object Detection Process First, the configuration of the entire moving object detection process for detecting moving objects such as meteors and space debris will be described. The moving celestial object detection process includes pre-processing for generating various data necessary for detecting the moving celestial object and detection processing for detecting the moving celestial object.

  Data generated in the preprocessing includes correction processing data for correcting an image obtained by imaging according to the characteristics of the sensor and the optical system, and detection processing data for detecting a moving celestial body. . Data for correction processing includes flat correction data obtained by photographing a surface with constant luminance, bias correction data obtained by closing the shutter and no light, dark correction data obtained by exposure time 0, and aberrations such as distortion aberration. It consists of representing aberration data. The correction processing data is acquired in advance before observing the moving object. However, if the observation conditions change, the observation is interrupted as necessary, and preprocessing is performed again to obtain correction processing data. The detection processing data includes a filter used for the detection processing, mask data for masking non-moving objects, a threshold value for detection determination, and the like. Details will be described later.

  In the present embodiment, when photographing the sky, the exposure time is adjusted so that the moving amount of the moving celestial body is sufficiently longer than one pixel and shorter than the length of one side of the image, and the orientation of the imaging device is adjusted according to the diurnal motion. Shoot while changing. Such a photographing method is called a star tracking mode or the like, and a moving celestial body is captured as a line segment in one image data, and the line segment is captured over a plurality of frames.

  The detection process for detecting a moving celestial object reflected as a line segment in this way mainly consists of three stages. In the first stage, a moving celestial object is detected for each pixel in a certain frame. Specifically, pixels constituting a line segment that is considered to be a moving celestial object are detected in a processing target frame (image data). Hereinafter, data obtained by the first stage processing is referred to as pixel-by-pixel detection data.

  In the second stage, among the detection data for each pixel detected for a certain frame, the data considered to be the trajectory of the same moving object is integrated. That is, one line segment is detected by integrating detection data for each pixel. Data obtained in the second stage is called intra-frame detection data. When a plurality of moving objects appear in one frame, a plurality of in-frame detection data is calculated for each frame.

  Finally, in the third stage, among the plurality of in-frame detection data calculated for each frame, data that is considered to be of the same moving object is integrated. In other words, line segments that exist across a plurality of frames are integrated, and a moving celestial object is detected as a line segment that moves with time. Hereinafter, data obtained in the third stage is referred to as moving object detection data.

  There is a possibility that a plurality of moving celestial bodies are detected in a frame constituting a series of moving images. When a plurality of moving celestial objects are observed in a series of moving images, there are a plurality of moving celestial body detection data correspondingly. By analyzing the moving object detection data and the captured image obtained in this way, the position, velocity, luminous intensity, etc. of the moving object in outer space can be analyzed. As a result of the analysis, the trajectory of the moving object can be estimated from the position and speed of the moving object.

  As described above, in the present embodiment, first, preprocessing is performed to generate various data, and then processing is performed on frames obtained one after another by photographing or moving image input to detect moving celestial bodies.

Apparatus Configuration FIG. 1 is a block diagram showing a detection apparatus applicable to the first embodiment.

  In FIG. 1, an operation unit 108 receives an instruction such as data input, analysis, and output from a user, and sends a signal corresponding to the instruction to the CPU 102. The CPU 102 uses the RAM 103 as a work memory and executes an OS and various programs stored in the ROM 104. Data and signals are transmitted and received between the processing units through the bus 105.

  The image data input unit 101 is an interface for inputting a frame (image data) of a moving image shot using a telescope or a digital camera. Further, in addition to the image data obtained by photographing the sky to be subjected to the moving object detection process, flat correction data, bias correction data, and dark correction data for correcting the image data are also input.

  The image data correction processing unit 109 corrects various aberrations of the optical system such as fixed pattern noise reduction processing and distortion aberration on the image data input from the image data input unit. Hereinafter, data processed by the image data correction processing unit 109 is referred to as corrected image data. That is, the image data correction processing unit 109 corrects the moving object detection target image data in order to detect the moving object more accurately.

  The moving celestial object detection processing data generation unit 110 reads the corrected image data and generates detection processing data necessary for the moving celestial object detection process. Here, a detection filter, mask data, and average image data are generated as detection processing data.

  The moving celestial body detection unit 111 uses the detection processing data to detect pixels constituting a line segment that is highly likely to indicate a moving celestial body in the corrected image data. The moving celestial body detection unit 111 generates detection data for each pixel. This corresponds to the first stage in the moving object detection process described above.

  The intra-frame integration processing unit 112 generates intra-frame detection data from the detection data for each pixel generated by the moving celestial body detection unit 111. In a frame to be processed, line segments that are considered to represent the same moving object are integrated. If no moving object is observed, no intra-frame detection data is generated. If one or more moving objects are detected in the processing target frame, one or more in-frame detection data is generated for each moving object. This corresponds to the second stage in the aforementioned moving celestial body detection process.

  The inter-frame integration processing unit 113 integrates the intra-frame detection data of each frame considered to indicate the same moving celestial object between the frames to generate moving celestial object detection data. This corresponds to the third stage in the aforementioned moving object detection process.

  The output unit 106 is an interface that outputs various data to a hard disk, an external memory, a network, and the like. Specifically, it outputs moving object detection data, image data, corrected image data, detection processing data, pixel-by-pixel detection data, in-frame detection data, and the like. Whether or not to store these data may be set in advance by the user via the operation unit 108.

  Hereinafter, each process will be described in detail.

Details of Image Data Correction Processing FIG. 2 is a block diagram showing a detailed configuration of the image data correction processing unit 109. The sensor noise reduction processing unit 201 performs processing for reducing a noise pattern specific to the sensor in the image data. For the sensor noise reduction process, flat correction data, bias correction data, and dark correction data acquired in advance from the correction processing data storage unit 202 in the RAM 103 are used. After that, the aberration correction unit 203 reads the aberration characteristics of the device that captured the image data from the aberration characteristic storage unit 204 for the image data with reduced sensor noise, and performs aberration correction processing such as distortion aberration. The aberration characteristic storage unit 204 outputs corrected image data. When the correction image data is set to be stored, the aberration correction unit 203 stores the correction image data in the correction image data storage unit 205. The corrected image data is sent to the moving celestial body detection processing data generation unit 110 and the moving celestial body detection unit 111.

Details of Detection Processing Data Generation Processing FIG. 3 is a block diagram showing a detailed configuration of the moving object detection processing data generation unit 110. The moving celestial body detection processing data generation unit 110 generates average image data, mask data, and a detection filter used for the moving celestial body detection processing as detection processing data. Each of the average image generation unit 301, the mask data generation unit 302, and the detection filter generation unit 303 receives necessary detection parameters from the detection parameter storage unit 305, and generates average image data, mask data, and detection filters.

  The average image data averages a plurality of corrected image data to be processed, and generates average image data. This average image data is data in which random noise contained in each of the corrected image data to be processed is reduced.

  The mask data is data indicating a region corresponding to a position where a bright celestial body that does not move, such as a star, a nebula, or a galaxy, or a region in which image data is defective. The mask data is used in order not to detect a moving celestial object in a specific area. Here, the pixel value of a pixel in a bright celestial or defective area is set to 0, and the pixel values of other pixels are set to 1. If the pixel value is 0, it is masked out of celestial object detection.

  The detection filter is used for a filter process performed on the corrected image data in order to detect a moving celestial body. The generation of the detection filter requires a limit grade calculated by the limit grade calculation unit 304 and a detection parameter. A threshold for determining the significance of the filter processing result using the detection filter generated here is also generated together with the detection filter. Details of the detection filter will be described later.

  The moving celestial body detection processing data generation unit 110 reads and processes one or more corrected image data.

  FIG. 4 is a flowchart of the moving celestial object detection processing data generation process in the moving celestial object detection process data generation unit 110. First, in step S <b> 401, the average image generation unit 301 acquires a processing target frame from the detection parameter storage unit 305 as a detection parameter.

  In step S402, the average image generating unit 301 reads each corrected image data obtained from a plurality of frames, and averages the plurality of corrected image data for each pixel to generate one average image data. The “average” referred to here may be a simple arithmetic average for each pixel, but it is preferable to employ a statistic that is robust with respect to the median or the like. This is because a phenomenon in which the brightness value around the collided pixel increases is rarely caused by collision of cosmic rays, which are high-energy radiation flying in outer space, with the sensor. The arithmetic average is easily affected by outliers, and the arithmetic average when this phenomenon occurs tends to be larger than the actual average value. On the other hand, when the median is used, it is difficult to be influenced by such outliers, and better average image data can be obtained.

  In step S <b> 403, the average image generation unit 301 stores the generated average image data in the average image storage unit 306 in the RAM 103.

  Next, in step S404, the mask data generation unit 302 generates mask data for masking a celestial body or the like that is to be excluded from the detection target in advance. First, the mask data generation unit 302 acquires parameters used for mask data generation from the detection parameter storage unit 305. As a specific method for generating mask data, a mask data may be generated by detecting a star that is not a detection target using a highly accurate astronomical object detection method conventionally used in astronomy. Alternatively, as a simple method, in the average image data, a pixel having a pixel value exceeding a predetermined threshold may be masked as a processing target pixel region. This is because a pixel having a protruding pixel value in the average image data is likely to be a stationary object such as a star. If it is possible to visually confirm that there is no obvious outlier, it is possible to generate mask data by performing threshold processing on the corrected image data. Parameters necessary for these processes are stored in the detection parameter storage unit.

  Further, the position of a region that is assumed to be defective in the image data is also stored in the detection parameter storage unit 305 in advance. In step S405, the mask data generation unit 302 generates mask data from the acquired detection parameters and average image data.

  The mask data generated in step S406 is stored in the mask data storage unit 307 in the RAM 103.

  In step S <b> 407, the limit grade calculation unit 304 acquires any one corrected image data from the image data correction processing unit 109. The limit grade calculation unit 304 identifies a standard star whose brightness is known in the corrected image data using the mask data and the shooting direction information of the telescope, and calculates the limit grade.

  In step S <b> 408, the filter generation unit 303 acquires the detection parameters for generating the limit grade and the detection filter from the limit grade calculation unit 304 from the detection parameter storage unit 305. Examples of the detection parameters used for generating the detection filter include a target moving celestial body class and a target value of a false detection rate.

  In step S409, the filter generation unit 303 generates a detection filter suitable for the situation at the time of observation from the acquired detection parameter and a threshold value for statistically determining the significance of detection in the result of the filter processing.

  In step S410, the detection filter is stored in the detection filter storage unit 308, and the threshold is stored in the threshold storage unit 309. Details of the detection filter and its generation method are described below together with the principle of the entire detection process.

● Principle of moving celestial object detection The principle of detecting a moving celestial object from each corrected image data will be explained. First, a case where the correction error of the corrected image data is sufficiently small will be described as an example. In the corrected image data, the horizontal and vertical pixel positions are x and y, respectively. Also, t is the shooting time when the image data that is the basis of the corrected image data is shot. In the corrected image data obtained by correcting the image data photographed at the photographing time t, the pixel values at the pixel positions x and y are represented as I (x, y, t). The pixel value I (x, y, t) in the corrected image data is mainly a component Os (x, y, t) from a stationary object that does not move, a component Om (x, y, t) from a moving object, night light, etc. Of the sky itself (called sky level) component S (x, y, t) and random noise component N (x, y, t). Shooting is performed while changing the shooting direction of the telescope or digital camera according to the diurnal motion. At this time, except for a rare celestial body such as a pulsar, the component Os (x, y, t) from the stationary celestial body does not vary with respect to the photographing time t. Also, the component S (x, y, t) from the sky hardly changes during the photographing time of about several minutes, except in the case where fog or clouds are present. Therefore, the fluctuation of the component S (x, y, t) from the sky with respect to the photographing time t can be ignored. Therefore, the pixel value of the correction data can be expressed as follows.

I (x, y, t)
= Os (x, y) + S (x, y) + Om (x, y, t) + N (x, y, t) (1)
The average image data is obtained by averaging a plurality of corrected image data during the photographing time with respect to time t. In the average image data, the component from the stationary object and the component from the sky are equal to Os (x, y) and S (x, y), respectively, because they do not depend on the photographing time t. On the other hand, since the moving celestial component and the noise component in the average image data change with time t, the moving celestial component Om (x, y, t) and the noise component N (x, y, t) of the correction data. Are averaged.

  First, the moving object om (x, y, t) will be described. Since the appearance of a moving celestial object is rare, Om (x, y, t) is 0 at most imaging times t. That is, when attention is paid to certain pixel positions x and y, there are many cases in which no moving object exists in any frame. When there is no moving object, the average value of the components Om (x, y, t) of the moving object is 0 regardless of whether it is an arithmetic average or a median value. In addition, when only one of the corrected image data Om (x, y, t) has a value K, the average is the value obtained by dividing K by the number of frames Nf in the case of an arithmetic average. In the case of the median value, if the number of frames Nf ≧ 2, the value is 0. When the average is calculated using a sufficient number of corrected image data, the component Om (x, y, t) of the moving celestial body in the average image data can be substantially zero at all pixel positions.

  Next, the noise component N (x, y, t) will be described. The variance of the noise component N (x, y, t) in certain corrected image data is assumed to be σ2 (x, y). When the arithmetic average of the number of frames Nf is calculated, the average value of the noise components N (x, y, t) after averaging is 0, and the variance is σ2 (x, y) / Nf. If the median is used instead of the arithmetic mean, the average value of the noise component N (x, y, t) after averaging is 0, and the variance is about 1.2 × σ 2 (x, y) / Nf. It has been known. Therefore, the noise component N (x, y, t) in the average image data can be set to 0 at most pixel positions by using a sufficient number of frames. As a result, the average image data J (x, y) can be expressed as follows.

J (x, y) = Os (x, y) + S (x, y) + N ′ (x, y) (2)
Here, N ′ (x, y) is a noise component whose variance is σ ′ 2 (x, y) smaller than σ 2 (x, y) by averaging.

In order to efficiently detect a moving celestial body, it is preferable to remove the component Os (x, y) from the stationary celestial body. Several methods can be considered for removal. Here, an example is shown in which average image data is subtracted from corrected image data to be detected. If the image data after subtraction is A (x, y, t),
A (x, y, t) = I (x, y, t)-J (x, y)
= Om (x, y, t) + N (x, y, t) -N '(x, y)
= Om (x, y, t) + N '' (x, y, t) (3)
It becomes. Here, N ″ (x, y, t) is a noise component obtained by synthesizing N (x, y, t) and N ′ (x, y), and its variance is based on the additiveity of the variance. σ2 (x, y) + σ′2 (x, y). As described above, in the image data after subtracting the average image data from the corrected image data, a random noise component N ″ (x, y, t) is added to the moving celestial component Om (x, y, t). It turns out that it becomes a thing. Therefore, if the moving image is not included in the corrected image data, since Om (x, y, t) is 0, the image has only a noise component.

  As another method for removing the component Os (x, y) from the stationary celestial body, mask processing using mask data can be cited. The mask data has a value of 1 at positions x and y where no static object exists. In other words, since the value is 1 at the point x, y where Os (x, y) is 0, only the pixel whose mask data is 1 is analyzed, so that Os (x, y) is effectively removed. Can do. In this case, the sky level S (x, y) still remains, but it is possible to analyze it by deleting the sky level S (x, y) separately. For this reason, it should be noted that it is not an essential requirement to generate average image data in the present invention.

  Here, consider image data C (x, y, t) obtained by dividing image data A (x, y, t) for each pixel by a standard deviation σ ″ (x, y, t), which is the square root of the variance. When there is no moving object, the image data C (x, y, t) is converted into image data in which the average pixel value is 0 and the variance is normalized to 1. When there was a moving celestial body, it had a value corresponding to the luminance value of the moving celestial body. Hereinafter, the image data C (x, y, t) is referred to as normalized image data.

  Hereinafter, an example in which a moving celestial object is detected from the normalized image data C (x, y, t) will be described. Since moving objects such as meteors and space debris move during the exposure time, they appear in the image data as line segments instead of stars. The moving object detection detects this line segment.

  In order to detect a line segment in the vicinity of the pixel (x0, y0) on the normalized image data C (x, y, t), a filter for line segment detection on the pixel (x0, y0) and its neighboring pixels Process. FIG. 5 shows an example of a detection filter for detecting a line segment. Pixels painted in black are target pixels, and regions including target pixels drawn with diagonal lines are line segment regions to be referred to. In the filter processing, the pixel of interest (x0, y0) and its neighboring pixels are referenced in a line segment shape, and the sum (addition value) of the pixel values of the pixels included in the line segment region is calculated. As shown in FIG. 5, a plurality of types of directions of the line segment area to be referred to are set, that is, a plurality of detection filters are used. In this example, the width W of the reference area is 2 pixels, the length L is 8 pixels, and there are 8 types of directions D from (a) to (h). In all the filters shown in FIG. 5, the filter coefficient included in the shaded portion indicating the reference area is 1, and the filter coefficient in the other areas is 0. Note that the filter coefficient of the pixel corresponding to the reference area may be weighted instead of 1, and a value weighted to the sum of the pixel values of the pixels included in the reference area is calculated as an addition value. As a result of calculating the sum of the pixel values of the eight reference areas, eight totals (added values) are obtained. The maximum value among the eight added values is selected as the selection value H of the pixel (x0, y0), and if the selection value H exceeds a certain threshold Th set statistically, the pixel (x0, y0) is near. Can be considered that a moving celestial body exists along the direction of the line segment region where the maximum value is obtained as a result of the filtering process.

  Note that the width of the line-shaped reference region is preferably set to the width of the moving celestial body to be detected, that is, the size of the point spread function of the optical system (also referred to as Point Spread Function, hereinafter referred to as PSF). . If the width W is wider than the width of the moving celestial body, more noise is picked up and the detection accuracy decreases. Also, the reference area should be small in order to increase the speed of the detection process. Therefore, it is desirable to design the pixel pitch of the optical system and the sensor in advance so that the PSF size of the optical system is 2 pixels or 3 pixels. The length L of the reference area is determined according to a limit grade that changes depending on weather conditions such as weather and humidity at the time of observation, and the brightness of the moving celestial body to be detected. For example, consider a case where the limit grade at the time of observation is 15 and it is desired to detect a moving object of 17.5 grade. There is a relationship of I∝10 ^ (−0.4 × m) between a celestial body of magnitude m and a celestial body of luminance I, and this celestial body is 10 times darker when the magnitude is 2.5 mag. Approximately, in order to detect a celestial object that is n times darker with the same significance, a reference area consisting of n ^ 2 times as many pixels is required. Accordingly, the length L when the width W is set to the width of the PSF is preferably set to a value about 100 times larger than the width W. A darker moving object can be detected by increasing the length L of the region. However, when the reference area becomes large, the detection process is delayed and the detection accuracy for a line segment shorter than the length L is lowered. Therefore, it is better to keep the length L of the region long enough to detect a celestial body having a target grade. Therefore, in the present invention, the detection filter is generated according to the limit grade at the time of observation. The direction D should be set so that there is no jump in the direction to be detected, and there should be a type of about 3L / 2W or more.

  The threshold value Th for determining that a moving celestial body exists is determined from the following viewpoints. Let P be the number of pixels in the referenced line segment region. In the example of FIG. 5, P = 16. When there is no moving celestial object, the sum T of 16 pixel values is statistically the sum of the pixel values obtained by randomly selecting P pixels from a noise distribution having an average value of 0 and a variance of 1. Behave in much the same way. Therefore, the average value of the sum T of pixel values is 0, and the variance can be regarded as P. The selection value H is obtained by calculating the sum T of the pixel values of the pixels in different D types of line segment regions, and calculating the maximum value among them. Therefore, the statistical behavior of the selected value H is almost the same as the behavior that the maximum value follows when D samples are acquired from the noise distribution with the average value 0 and the variance P described above. Specifically, when the noise component N (x, y, t) follows a Gaussian distribution, the probability distribution function PL (H) of the selection value H is as follows.

  Here, π is a circular ratio, and erf is an error function. In view of the probability distribution function of the selection value H, it is determined whether or not the selection value H indicates that a moving celestial object exists (exactly, a hypothesis that no moving celestial object exists) is rejected. The threshold value Th for this is set. Here, following a statistical test method, a certain rejection probability Pd is set, and if the probability that the selected value H exceeds the threshold Th is smaller than Pd, the hypothesis that there is no moving object is rejected, and the moving object is I conclude that there is. In a general statistical test method, the rejection probability Pd is often set to about 5%, but in the case of moving celestial object detection, a smaller one is desirable. This is because false detection occurs theoretically with a rejection probability Pd. In the case of image data consisting of several million pixels, if the rejection probability Pd is set to about 5%, about 100,000 erroneous detections occur for each corrected image data, and subsequent processing becomes difficult. The rejection probability Pd is preferably set so that the number of erroneous detections is within a few or a few hundred per corrected image data. For example, if the corrected image data is data composed of several million pixels, the rejection probability Pd may be set to about 10 minus 4 to about 10 minus 6 to the reciprocal of the number of pixels. And let K corresponding to this rejection probability be threshold value Th. Specifically, K satisfying the following expression is set as the threshold Th.

  The expected value of the selection value H according to the probability distribution function PL (H) is about two to three times the square root of the number of pixels P, although it depends on the number of patterns D in the line segment region. The threshold value Th determined from the above equation should be at least larger than the expected value of the selection value H. Normally, the threshold Th is about five times the square root of the number of pixels P. If the threshold Th is smaller than this, it may be considered that the rejection probability Pd is too large or the calculation is incorrect. In other words, by setting the threshold Th to a large value that is five times or more the square root of the number of pixels P, it is possible to detect a moving celestial object while sufficiently suppressing erroneous detection.

  By the above analysis, the presence / absence of a moving celestial object passing through the pixel (x0, y0) on the normalized image data C (x, y, t) and the direction of the moving celestial object are obtained. By performing this process at a plurality of pixel positions, a detection result for each pixel can be obtained.

  The above is an example where the correction error of the corrected image data is sufficiently small. Next, a process for dealing with a case where the correction error cannot be ignored in practice or a case where the number of false detections becomes very large due to the presence of a thin cloud will be described. Further, this process does not perform the process of subtracting the average image data from the corrected image data, and enables the above-described detection process even when the sky level S (x, y) exists in the corrected image data. As examples where the correction error cannot be ignored, there are cases where the sky is not uniform due to the influence of the weather, or noise is generated in the sensor due to heat or power. Furthermore, since the influence of the weather and the noise characteristics of the sensor differ for each area of the obtained image data, the average value and the noise variance also differ depending on the area of the image data. In this case, as described above, the normalized image data is not obtained simply by performing the normalization process of dividing the image data A (x, y, t) by a single standard deviation. That is, as a result of the normalization process, sufficiently normalized image data is not always obtained. Therefore, in this embodiment, the entire image data is divided to define partial areas, and an average value and a standard deviation are obtained and normalized for each area. The partial area may be divided according to, for example, a sensor reading channel. FIG. 6 shows an example in which a thin cloud appears in the captured frame. In the case where a thin cloud is present and the sky level S (x, y) exists, the detection principle is almost equivalent.

  However, if the partial area obtained by dividing is too small, the signal of the moving celestial body is weakened during the normalization process, and the detection sensitivity is lowered. In order to avoid this decrease in detection sensitivity, the normalization process is preferably divided into three stages or two stages. In the case of the three stages, first, the first normalization process is performed on the entire image. For the image data normalized by the first normalization process, check whether each partial area divided according to the sensor characteristics is fully normalized, and only if the partial area is not normalized The second normalization process is performed every time. Then, for the image data that has been normalized by the second normalization process, check whether it is sufficiently normalized for each small area obtained by further dividing the partial area. A third normalization process is performed for each small area. Note that the size of the small area in the third normalization process is set so that the length of one side is about twice or more of the line segment to be detected. When normalization processing is performed in two stages, the entire normalization processing may be skipped, and normalization for each partial region arbitrarily set from the beginning may be performed. Through these series of operations, even if there is a correction error or a thin cloud that cannot be ignored in the image data, it is possible to obtain normalized image data C (x, y, t) appropriately normalized in each region. Note that when the mask data M (x, y, t) is used, the normalization process is performed in consideration of this. That is, when calculating the average value and the standard deviation for each region, the calculation is performed by ignoring the pixels indicating that the mask data M (x, y, t) is not subject to processing.

  After this normalization process, a dimming process is performed. The dimming process is a process of replacing a pixel value exceeding a predetermined threshold Th2 with a predetermined value in the normalized image data C (x, y, t). The predetermined value is smaller than the pixel value of the pixel to be replaced and is a value of 0 or more. For example, when the threshold value Th2 and the predetermined value are both 3, a pixel value exceeding 3 is replaced with 3. When the sky level in the image data is not 0, the predetermined value is desirably equal to or higher than the sky level. Such light reduction processing can reduce false detection of bright spots due to cosmic rays or the like.

  FIG. 7 is a conceptual diagram showing the effect of the dimming process. As described above, the filter processing for detecting the line segment indicating the moving celestial object calculates the sum of the pixel values in the line segment-like reference region. Then, a moving object is detected by determining a selection value based on the sum. When the light reduction process is not performed, if even one very bright pixel exists in the line-shaped reference area, the sum of the calculated pixel values becomes large and is detected as a line segment indicating a moving object. May occur. However, a bright spot having no bright pixels having only one or two pixels is almost always a value due to some noise such as cosmic rays, and is not a line segment, so it is a false detection as a moving object. Therefore, by performing the above-described dimming process, a pixel value representing a pixel that is a bright luminescent spot is replaced with a small pixel value, and no line segment is detected even if the filtering process is performed.

  On the other hand, if a pixel having a very bright pixel value is replaced with 0, a very bright moving object cannot be detected. As shown in FIG. 7, the bright line segment is detected as a line segment by the filter process, although the pixel values of many pixels are reduced by the dimming process.

  By performing the dimming process in this way, it is possible to prevent erroneous detection of an abnormally bright luminescent spot with a single determination. When the normalized image data is dimmed and the width of the detection filter used for the filter process performed by the determination unit 904 after the dimming process is equal to the PSF, the predetermined value is the threshold Th. A value obtained by dividing by the number of pixels P of the detection filter may be set to a value that is approximately 10% larger than that. By setting in this way, the selection value of the moving celestial body darkened by the dimming process is likely to exceed the threshold Th, and detection is possible.

  For example, when the threshold Th is 6 times the square root of the number of pixels P, the predetermined value is a value obtained by dividing 6 by the square root of P. When the number of pixels P is 16, as shown in FIG. 5, the predetermined value is 1.5 or 1.65 which is about 10% larger than that. Alternatively, if it is desired to detect a bright moving celestial object that is captured only at half the length of the detection filter, it may be about 3.3. Since the threshold value Th is 24, it is detected if the moving celestial object is captured with 2 × 4 = 8 pixels.

  If the width of the detection filter is larger than the PSF, the predetermined value may be equal to or greater than a value obtained by dividing the detection filter length W by the width of the PSF. This is because the width of the moving celestial body is approximately the width of the PSF, and the added value when there is a moving celestial body darkened by the dimming process is approximately the predetermined value × the width of the PSF × the detection filter length L. is there. As described above, the predetermined value may be determined according to the number of pixels P of the detection filter and the threshold value Th.

  On the other hand, the predetermined value may be determined from the standard deviation of noise. Usually, the limit grade is defined as the brightness of a celestial body having a brightness 5 to 10 times the standard deviation of noise. When the predetermined value is larger than five times the standard deviation of noise, the value is such a value that it can be considered that a significant celestial body exists with only one pixel. If the luminance of such a magnitude is added and the added value is calculated, the probability that the added value exceeds the threshold value Th even if there is no moving celestial object and false detection is increased. In order to avoid erroneous detection of moving celestial bodies, one pixel cannot be regarded as detection because the light is too weak. However, if the weak light is lined up in a line, it can be considered that the celestial body exists statistically. It is necessary to set a predetermined value to such a value. That is, it must be at least smaller than 5 times the standard deviation of noise, and more than the standard deviation of noise. When the predetermined value is smaller than the standard deviation, the added value is difficult to exceed the threshold and cannot be detected. Preferably, the predetermined value is set to a value not more than four times the standard deviation of noise.

  Through the above processing, the moving celestial object can be detected with high accuracy along with the direction information for each pixel. If the sky level (average value) and the standard deviation of noise are known for the entire image, or for each section or partial area, mathematically equivalent processing is possible without generating a normalized image that normalizes the entire image. It is possible to Therefore, although it is necessary to derive the average value and the standard deviation (or variance), whether or not to generate a normalized image does not limit the present invention.

  Next, the detection results for each pixel are integrated. Since the moving celestial object is shown as a line segment, when the moving celestial object is detected at a certain pixel, in most cases, the same moving celestial object is detected at a plurality of pixels along the direction. These are collected as one moving celestial body, and the data format is converted into the coordinates of the two end points of the line segment from a plurality of pixel position information. This is the integration of detection results for each pixel.

  The processing will be specifically described below. The pixel in which the moving celestial object is detected includes direction data of the moving celestial object. Therefore, along with the direction in which the moving celestial object is detected, a search is made for other moving celestial objects having substantially the same direction, and if they are found, they are determined to be the same moving celestial object. Then, the same operation is recursively repeated for newly found moving objects. Thereby, the detection information (pixel position and direction) for each pixel belonging to the same moving object can be linked. When a plurality of pixel positions are combined as being the same moving celestial body, a point group of pixel positions aligned in a line segment shape is obtained. From this point group, the end point of the line segment is calculated. There are several methods for calculating the end points. For example, when a straight line that approximates the obtained point group is obtained and each point is projected onto the approximated straight line to find an end point, a stable result can be obtained. In addition, when the speed of the moving celestial body on the celestial sphere is slow and the length of the line segment is short, there are only about 1 to 3 pixels to be connected, and an approximate straight line is accurately obtained from the point group. difficult. In this case, the direction of the straight line is determined from the detection result for each pixel, and only the intercept may be obtained. When it is necessary to obtain the end points of the line segment at higher speed, the horizontal and vertical maximum and minimum values may be obtained from the point group, and the corresponding vertical or horizontal pixel positions may be extracted.

  Finally, the detection results of moving objects are integrated between frames. Normally, moving objects are shown across multiple frames until they disappear or go out of sight. The moving celestial bodies detected as line segments for each frame by the above-described processing are integrated by linking the same moving celestial bodies by taking correspondence between the line segments between the frames. The detection result integrated in this way is the inter-frame detection result. By the integration process between the frames, a frame number and a trajectory of the moving celestial object until a certain moving celestial object appears and disappears are obtained. Note that there are cases where the moving celestial bodies to be integrated cannot be found in several frames before and after the moving celestial bodies detected within the frame. Since this is highly likely to be erroneously detected and subsequent analysis is difficult, it is better to discard the detection results that have no moving celestial bodies to be integrated between frames.

  The identity of a moving celestial object detected in one frame and a moving celestial object detected in another frame is determined by whether or not a plurality of conditions as shown below are satisfied. The first condition is whether the two line segments have the same direction. If the directions are completely different, it is determined that they are separate moving objects. The next condition is whether or not two moving objects are aligned on a straight line. Here, if the two line segments are not aligned on a straight line, it is determined that they are separate moving objects. Conversely, if two line segments are arranged on a straight line and have not yet been associated with other in-frame detection data, it may be determined that they are the same moving object at this point. Strictly speaking, this is because the possibility that different moving celestial objects occur simultaneously on a straight line is very low, and the possibility that different moving celestial objects occur simultaneously on a straight line can be ignored. If the frequency of detection of moving celestial objects, including false detections, is high and the possibility that two moving celestial objects are generated in a line segment cannot be ignored, additional conditions are added to the interval between the two moving celestial objects. You can determine by imposing. For example, when the interval between two line segments is more than three times larger than the length of the line segment in a certain frame, it is determined that they are separate celestial bodies. This is a possible decision because a moving object cannot physically accelerate on the celestial sphere. However, if the time between the exposure of one frame and the exposure of the next frame is long, the threshold for the interval between line segments needs to be increased accordingly.

  If any one of the intra-frame detections is already associated, the non-retrograde condition is further imposed to determine the identity in order to increase the accuracy. Non-retrograde means the property that a moving celestial body does not travel backwards on the celestial sphere. Once the correspondence of the moving celestial object is established between the frames, the traveling direction of the moving celestial object that cannot be understood from the detection result of one frame is known. Specifically, the start point and end point of the moving object can be determined by comparing the coordinates of the end points of two line segments. For example, in FIG. 8A, when the moving object of the frame f and the moving object of the frame f + 1 are integrated, the end point A is the start point and the end point D is the end point. On the other hand, in FIG. 8B, when the moving object of the frame f and the moving object of the frame f + 1 are integrated, the end point B is the start point and the end point C is the end point. A similar method can be used when a moving object that has already been handled in two frames is assigned a correspondence in the third and subsequent frames. By using the traveling direction of the moving celestial body thus obtained, it is possible to determine non-retroversion. Specifically, this is performed using the start and end points of a moving celestial body that has already been dealt with and whose traveling direction is known, and two end points of the in-frame detection data to be matched. Whether the coordinates of the start point and the end point are (xi, yi) and (xf, yf), respectively, and the coordinates of the end points of the detection data in the frame are (x1, y1) and (x2, y2) Judge by how.

{(Xi <xf <max (x1, x2)) or (xi>xf> min (x1, x2))}
And {(yi <yf <max (y1, y2)) or or (yi>yf> min (y1, y2))} (6)
When this equation is satisfied, it can be said that the x coordinate and y coordinate of the moving celestial body are monotonously increasing or decreasing and are not retrograde. In addition, when the moving celestial object to be detected is limited to a moving celestial object such as an artificial satellite or space debris, the following more severe conditional expression can be imposed.

{(Xi <xf ≦ min (x1, x2)) or (xi> xf ≧ max (x1, x2))} and {(yi <yf ≦ min (y1, y2)) or (yi> yf ≧ y (y , Y2))}} (7)
By adding this non-retrograde determination, it is possible to reduce erroneous determination that associates different moving celestial bodies as the same object, and to increase the reliability of the in-frame detection data.

  Based on the above principle, it is possible to detect a moving celestial object from image data consisting of a plurality of frames, and further obtain a frame in which the moving celestial object is observed, a frame in which the moving celestial object has been observed, a start point and an end point of the moving celestial object it can.

  Hereinafter, details of processing in the moving object detection unit 111, the intra-frame integration processing unit 112, and the inter-frame integration processing unit 113 will be sequentially shown.

Details of Moving Object Detection Unit 111 FIG. 9 is a block diagram showing a detailed configuration of the moving object detection unit 111. The moving celestial body detection unit 111 includes a difference image generation unit 901, a normalization processing unit 902, a light reduction processing unit 903, and a determination unit 904. FIG. 10 is a flowchart showing a flow of operations in the moving celestial body detection unit 111. FIG. 11 is a block diagram showing a more detailed configuration of the normalization processing unit 902 in the moving celestial body detection unit 111. FIG. 12 is a flowchart showing the flow of normalization processing in the normalization processing unit 902. Hereinafter, the operation of the moving object detection unit 111 will be described with reference to FIGS. 9 to 12.

  First, in step S <b> 801, the difference image generation unit 901 reads the corrected image data processed by the image data correction processing unit 109 and the average image data stored in the average image storage unit 305. Next, in step S802, the difference image generation unit 901 generates difference image data by subtracting the average image data from the corrected image data.

  In step S <b> 803, the normalization processing unit 902 reads mask data from the mask data storage unit 306. Further, in step S804, the normalization processing unit 902 performs mask processing on the difference image data, and then normalizes the difference image data.

  Here, an example of normalizing the difference image data in two stages will be described with reference to FIGS. First, in step S <b> 1201, the average variance calculation unit 1101 acquires from the division method storage unit 906 a parameter indicating a predetermined segment based on sensor characteristics. Then, the average variance calculation unit 1101 calculates an average value and a variance value for each section of the difference image data. In step S1202, the first normalization processing unit 1102 normalizes the difference image data for each section based on the average value and the variance value calculated by the average variance calculation unit 1101.

  Next, it is determined whether or not the normalization process is hindered by the presence of a cloud or the like. First, in step S1203, the average variance recalculation unit 1103 divides the difference image data into partial areas that are finer than the section, and calculates an average value and a variance value for each partial area. In step S1204, based on the average value and the variance value, it is determined whether the normalization process in the first normalization result determination unit 1104 is successful. Specifically, the average value is not significantly different from the sky level (0 by the first normalization process), and this variance is more significant than the original partition variance (1 by the first normalization process). Judgment is made from the viewpoint of whether it is larger. For example, the deviation of the average value can be judged by the t test, and the deviation of the variance can be judged by the F test. If it is determined by this determination that normalization has failed, the process proceeds to step S1205, and the renormalization processing unit 1105 uses the average value and the standard deviation calculated by the average variance recalculation unit 1103. Normalize the region and generate normalized image data. If normalization is successful, renormalization processing is not performed and this flow is terminated.

  Hereinafter, the description will be continued returning to FIG. 9 and FIG. In step S805, the light reduction processing unit 903 obtains the threshold value Th2 and a predetermined value from the detection parameter storage unit 305 and performs light reduction processing in order not to erroneously detect a bright spot such as a cosmic ray as a line segment.

  In step S <b> 806, the determination unit 904 acquires a threshold Th necessary for determining whether or not the detected line segment indicates a moving object from the detection parameter storage unit, and further reads a detection filter group from the output filter storage unit 308.

  In step S807, the determination unit 904 performs processing for detecting a line segment for each pixel. First, filter processing is performed by applying the detection filter group to each pixel of interest. In the filtering process performed here, the sum of the pixel values in the line-segment reference region including the target pixel is calculated. The maximum line segment direction is searched from among the plurality of sums calculated by the filter processing, and the maximum sum is selected as a selection value.

  In step S808, the calculated selection value is compared with the threshold Th to determine the presence or absence of a moving object. If it is determined that there is a moving celestial object, the process proceeds to step S809. If it is determined that there is no moving celestial object, the process proceeds to step S810. In step S809, the selection value determined as the moving object and the direction of the line segment corresponding to the selection value are recorded.

  In step S810, it is determined whether processing has been performed for all pixels in the corrected image data to be processed. If there are remaining pixels to be determined, the process returns to step S807. If there is no pixel to be determined, the process proceeds to step S811.

  In this way, the processing from step S807 to step S809 is repeated for all the pixels in the corrected image data to be processed to obtain detection data for each pixel. In step S 811, it is determined whether or not the pixel-by-pixel detection data is set to be stored in the pixel-by-pixel detection data storage unit 905. If it is determined in step S811 that the setting is made to save, the process advances to step S812 to save the pixel-by-pixel detection data in the pixel-by-pixel detection data storage unit 905. The above is the process in the moving celestial body detection unit 111.

Details of Intraframe Integration Processing Unit 112 FIG. 13 is a block diagram showing a detailed configuration of the intraframe integration processing unit 112. The intra-frame integration processing unit 112 includes a detection pixel integration unit 1301, an end point calculation unit 1302, and a luminance value calculation unit 1303. FIG. 14 is a flowchart showing the flow of the intra-frame integration processing in the intra-frame integration processing unit 112.

  First, in step S1401, the detection pixel integration unit 1301 acquires each parameter necessary for integration from the detection parameter storage unit 305. Specifically, a threshold for determining whether the direction is the same as the search range of the same moving object corresponding to the direction information added by the moving object detection unit 110 is acquired as a detection parameter. Various examples of the search range corresponding to the direction information are conceivable, but the shape of the detection filter group used for detecting each pixel of the moving celestial body is suitable.

  After obtaining the detection parameters, in step S1402, the detection pixel integration unit 1301 examines the pixels in order and determines whether a moving celestial object has been detected. When a moving celestial object is detected, a search is made for other pixels in which the line segment is detected along the direction of the line segment indicating the moving celestial object.

  In step S1403, it is checked whether there is a line segment other than the search target pixel. If a moving celestial object that has not yet been integrated is found in step S1403, the process proceeds to step S1404. In step S1404, it is determined from the identity of the direction whether or not it is the same moving celestial object.

  If they are the same moving object in step S1405, the information is integrated, and the search is further performed recursively. If it is not the same moving object, do nothing and proceed.

  In step S1406, it is determined whether all the surrounding areas have been searched. If there are unsearched pixels, the process returns to step S1403. If all the surrounding areas have been searched, the process ends. Note that the recursive search in step S1405 is to search the moving celestial bodies determined to be the same in step S1404 by repeating the processing from step S1403 to step S1406. By this recursive search, detection data for each pixel belonging to the same moving object in the processing target frame is integrated.

  Here, in step S1407, it is checked whether an integrated moving celestial body is detected by a plurality of pixels. If the moving celestial object is not detected by a plurality of pixels, the process proceeds to step S1411 and it is determined that the pixel is erroneously detected. If a moving celestial object is detected by a plurality of pixels, the process advances to step S1408, and the result of integration is sent to the endpoint calculation unit 1302. The end point calculation unit 1302 calculates the end point of the moving celestial object from the integrated detection result for each pixel. In step S1409, the luminance value calculation unit 1303 acquires the corrected image data from the image data preprocessing unit 109, and calculates the luminance value for each moving object.

  In step S1410, information in which the number of detected pixels and the frame number are added together with the calculated end point and luminance value is additionally recorded in the intra-frame detection data storage unit 1304 as intra-frame detection data.

  In step S1412, it is determined whether all pixels in the processing target frame have been processed. If there are unprocessed pixels, the process returns to step S1402, and if all pixels have been processed, pixel-by-pixel detection is performed. The data integration process is complete.

  Through the intra-frame integration process as described above, the intra-frame detection data is generated for the number of moving objects existing in the frame.

Details of Interframe Integration Processing Unit 113 FIG. 15 is a block diagram showing a detailed configuration of the interframe integration processing unit 113. The inter-frame integration processing unit 113 includes an identity determination unit 1501, a moving celestial body detection data buffer 1502, and a detection information calculation unit 1503.

  FIG. 16 is a flowchart showing a flow of interframe integration processing performed by the interframe integration processing unit 113.

  The identity determination unit 1501 acquires two pieces of data indicating moving celestial bodies from the in-frame detection data storage unit 1304 or a moving celestial body detection data buffer described later, and determines whether the moving celestial bodies indicated by each are the same. FIG. 17 is a block diagram showing further details of the identity determination unit 1501. The identity determination unit 1501 includes a direction coincidence determination unit 1701, an identical line determination unit 1702, a non-retrograde determination unit 1703, and a moving celestial body detection data update unit 1704.

  The direction coincidence determination unit 1701 determines whether the directions of the two moving objects coincide. The same straight line determination unit 1702 determines whether the two moving objects determined by the direction coincidence determination unit 1701 exist on the same straight line. The non-retrograde determination unit 1703 further determines whether the two moving objects to be determined are not retrograde. When it is determined that the two moving objects are the same, the moving object detection data update unit 1704 updates the moving object detection data. The moving celestial body detection data is data obtained by inter-frame integration processing. That is, it is data obtained by integrating the line segments indicating the moving celestial bodies from the in-frame detection data of different frames. In the present embodiment, first, when the moving celestial object indicated by the intra-frame detection data and the moving celestial object indicated by the different intra-frame detection data are the same, they are integrated into the moving celestial object detection data. Thereafter, it is determined at any time whether there is any in-frame detection data to be integrated in the moving celestial body detection data, and the moving celestial body detection data is updated. The moving object detection data buffer 1502 is a buffer for temporarily storing a plurality of moving object detection data. The detection information calculation unit 1503 adds information such as the speed of the moving object to the moving object detection data.

  The interframe integration process proceeds as follows. First, in step S1601, the identity determination unit 1501 acquires two pieces of data indicating a moving celestial object from the in-frame detection data storage unit 1304 or the moving celestial object detection data buffer 1502. The identity determination unit 1501 acquires parameters for determining whether or not the processing target moving celestial bodies are the same from the detection parameter storage unit 305, and the direction match determination unit 1701, collinear line determination unit 1702, non-retrograde property The data is stored in the determination unit 1703. Specifically, this parameter is a threshold value for considering that two moving celestial bodies are in the same direction, a threshold value for considering that two moving objects are arranged in a straight line, and the like.

  In step S1602, the identity determination unit 1501 obtains one intra-frame detection data from the intra-frame detection data storage unit 1304. It is assumed that the in-frame detection data is acquired in the order of the frames.

  In step S1603, the identity determination unit 1501 checks whether there is moving object detection data stored in the moving object detection data buffer 1502, and if not, the process proceeds to step S1610. If there is moving object detection data, the process proceeds to step S1604 to acquire one. In step 1605, it is determined whether the in-frame detection data and the moving object detection data are the same moving object.

  The details of determining whether or not they are the same moving object are as shown in the flowchart of FIG. First, in step S1801, the direction coincidence determination unit 1701 determines whether or not the traveling directions of the two moving objects coincide within a predetermined range.

  If it is determined that the directions match, in step S1802, the same straight line determination unit 1702 determines whether the two moving celestial bodies are on the same straight line within a predetermined range. If step S1801 or step S1802 is negative, it is determined that the two moving objects are not the same, and the process proceeds to step S1803.

  If it is determined in step S1802 that they are on the same straight line, in step S1804, the non-retrograde determination unit 1703 determines whether the moving celestial body detection data extends over a plurality of frames. Specifically, it may be confirmed that the start frame number and the end frame number included in the moving object detection data do not match.

  If it is across multiple frames, the process proceeds to step S1805, and it is considered as the same moving object based on the end point of the moving object detection data in the frame to be determined or the coordinates of the start point and end point of the moving object detection data. Judge whether the moving object will go backwards.

  If it is retrograde, the process proceeds to step S1803, where it is determined that the two moving celestial bodies are not the same. If it is not retrograde, it is determined in step S1806 that they are the same moving celestial object. If the moving celestial body detection data does not extend over a plurality of frames in step S1804, the start point and the end point are unknown, and it is determined that they are the same celestial body.

  Hereinafter, returning to FIG. 16, the detailed description of the inter-frame integration processing will be continued.

  If it is determined in step S1605 that the in-frame detection data and the moving celestial body detection data are the same celestial body, in step S1606, the moving celestial body detection data is integrated to update the values of the start and end points of the moving celestial body and the end frame number. To do.

  In step S1607, it is determined whether there is undecided intra-frame detection data in the intra-frame detection data storage unit 1304. If undetected intra-frame detection data exists, the process returns to step S1602. If it is determined in step S1605 that the in-frame detection data and the moving object detection data are not the same object, the process proceeds to step S1608, and the end frame number of the moving object detection data is more constant than the frame of the in-frame detection data. Find out if it is more than a few years old. If it is not more than a certain number in the past, nothing is done and the process returns to step S1603, but if it is in the past for a certain number or more, it is determined that the moving object has already disappeared or has gone out of the field of view, and the moving object detection data is determined. Process.

  In step S1609, the corresponding data in the moving object detection data buffer 1502 corresponding to the moving object detection data to be confirmed is deleted, and the moving object detection data to be confirmed is sent to the detection information calculation unit 1603. The detection information calculation unit 1603 calculates additional detection information such as the velocity on the celestial sphere in addition to the start point, end point, start frame, and end frame of the moving celestial object, and additionally records them in the moving celestial object detection data storage unit 1504. However, at this time, if the moving celestial object is not detected across a plurality of frames, it is preferably discarded because there is a possibility of erroneous detection. After step S1609, the process returns to step S1603, and the comparison with the next moving object detection data stored in the moving object detection data buffer 1502 is continued.

  If there is no undetermined moving object detection data in the moving object detection data buffer 1502 in step S1603, the in-frame detection data read in step S1602 is converted into moving object detection data in step S1610, and the moving object detection data buffer is obtained. Add to 1502. At this point in time, since the start point and end point of the moving celestial body are unknown, the end points of the in-frame detection data are assigned to the start point and end point, respectively, and recorded. The start frame number and the end frame number are assigned with the frame number of the detection data in the frame.

  When the process proceeds in this manner and there is no undetermined moving object detection data in the frame in step S1607, all unconfirmed moving object detection data remaining in the moving object detection data buffer 1502 is sent to the detection information calculation unit 1503. . The detection information calculation unit 1503 discards data that does not extend over a plurality of frames with respect to each of the transmitted moving object detection data, and calculates additional detection information for those that extend over and stores the result as the moving object detection data storage. Recorded in section 1504.

  Through the above series of processing, normalization processing was performed on the image data obtained by imaging the sky taking into account the case where the average value and noise variance differ for each region in the image data. Thereby, the image data can be normalized more appropriately. In the present embodiment, the region that is the processing unit for performing the normalization processing is determined in stages. Thereby, since the normalization process is performed for each region, it is possible to prevent the detection sensitivity from being lowered as much as possible.

<Other embodiments>
The present invention can also be realized by supplying a storage medium storing software program codes for realizing the functions of the above-described embodiments to a system or apparatus. In this case, the computer (or CPU or MPU) of the system or apparatus reads out and executes the computer program code stored in the storage medium so that the computer can read the functions of the above-described embodiments.

Claims (9)

A moving celestial body detection device that detects a moving celestial body that appears in a line segment in image data taken by changing the direction of the imaging device according to the diurnal motion,
Acquisition means for acquiring image data;
For each partial region obtained by dividing the image data, calculating means for calculating the sky level and noise standard deviation in the image data;
Based on the sky level and the standard deviation of the noise, normalization processing means for normalizing the image data for each of the partial regions, and image data subjected to normalization processing by the normalization processing means And a detecting means for detecting a moving celestial object reflected in a line segment. Further, the image data has a determination unit that determines whether each of the small areas smaller than the partial area is sufficiently normalized,
A second normalization processing unit that normalizes the image data for each small region when the determination unit determines that the image data is not sufficiently normalized;
The moving celestial body detection apparatus according to claim 1, wherein the detection unit detects a moving celestial body from the image data subjected to the normalization processing by the second normalization processing unit.   The moving object detection apparatus according to claim 1, wherein either the partial area or the small area is set according to a characteristic of a sensor that has captured the image data.   The mobile celestial body detection device according to claim 1, wherein either the partial region or the small region is set according to the weather when the image data is captured.   The moving celestial body according to any one of claims 2 to 4, wherein the small area is a rectangle having sides longer than a length of a line portion indicating the moving celestial body detected by the detecting means. Detection device. Further, the image data has a mask process for masking a region not to be processed,
The moving object detection apparatus according to claim 1, wherein the normalization processing unit performs normalization processing on the masked image data. The acquisition means acquires a plurality of image data obtained by photographing the sky,
Average image generation means for generating average image data by averaging pixel values representing respective pixels at corresponding pixel positions of the plurality of image data;
Among the plurality of image data, a difference image generation unit that generates difference image data by subtracting the average image data for each pixel from the image data to be processed,
The mobile celestial body detection apparatus according to claim 1, wherein the normalization processing unit performs processing on the difference image data.   A computer program for causing a computer to function as the moving celestial body detection device according to any one of claims 1 to 7 by being read and executed by a computer. A moving celestial body detection method for detecting a moving celestial body imaged in a line segment in image data taken by changing the direction of the imaging device according to the diurnal motion,
Get image data,
For each partial region obtained by dividing the image data, calculate the sky level and noise standard deviation in the image data,
Based on the sky level and the standard deviation of the noise, the image data is normalized for each partial region,
A moving celestial object detection method, comprising: detecting a moving celestial object appearing in a line segment from the image data subjected to normalization processing by the normalization processing unit.