JP2005061931A - Three-dimensional shape recovery method, its apparatus, and three-dimensional shape recovery program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To substantially reduce time and labor required for shape recovery work from time-series images. <P>SOLUTION: A sequence determining mechanism 11 creates information required for dividing the time-series images into sequences. A profile creating mechanism 12 creates profiles for executing a point group acquisition program 17, and all the created profiles are stored in a profile storage device 13. The created profiles are selected by a profile selecting mechanism 14. The profiles are selected again on the basis of the results of point group acquisition to alter parameters to be provided for the point acquisition program 17 and execute processing. In the case of the absence of profiles to be executed again, processing is shifted to a point group integrating mechanism 15. The point group integrating mechanism 15 combines point groups acquired from each sequence into spatial information and inputs it to a file server 16. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a method, apparatus, and program for restoring a three-dimensional shape in a wide area from a time-series image obtained by photographing a whole area by performing partial photographing of a photographing target area a plurality of times.

  A technique (a method, an apparatus, or the like) that acquires a three-dimensional point group from an image and restores the shape from the point group is known (see Non-Patent Document 1). (Refer nonpatent literature 2.).

  With such a technique, for example, as shown in FIG. 16, three-dimensional shape data (city space model) of an urban area is constructed from an aerial image of a helicopter, and numerical simulations such as city planning, car navigation, virtual mall, disaster, etc. It is used for basic data of various applications such as (see Non-Patent Document 3).

  As shown in FIG. 17, the image (video) is composed of an image sequence called a frame taken every unit time (photographing times 001 to 005) by a video camera installed in the helicopter. Is called a time-series image.

  FIG. 18 shows an example of time-series images (shooting times 001 to 005) obtained by aerial imaging of the target building on the ground. As can be seen from FIG. 18, since the number of frames on which the target building on the ground is projected is limited, when restoring a wide-area three-dimensional shape from a long time-series image, each partial image sequence (sequence) It is necessary to acquire a point cloud.

  Furthermore, since a large structure may span a plurality of sequences, it is necessary to restore the shape after integrating point clouds acquired from a plurality of consecutive sequences into one point cloud. In order to distinguish the point group after integration from the point group for each sequence, it is hereinafter referred to as spatial information.

  FIG. 19 shows an example in which a sequence is generated every N frames from a time-series image. When restoring a wide-area 3D shape from a long time-series image, a point cloud with 3D coordinate values is acquired for each sequence, the point cloud is integrated to generate spatial information, and the shape is restored from the spatial information. Do.

  In the following description, it is assumed that a program for acquiring a point cloud from a sequence (point cloud acquisition program) and a program for restoring a solid shape from spatial information (shape restoration program) are prepared as conventional techniques. An example of executing the point cloud acquisition program is shown in FIG. In this example, the threshold value (Threshold) of the program, the shooting time of the first frame (Initial Frame), and the number of frames in one sequence (Frames) are specified, and the point cloud acquisition program is executed.

  The output includes the name of the image file displaying the point cloud acquisition result (Initial Image), the number of points acquired (Feature Points), the name of the file describing the 3D coordinate values of each point (Shape3D), and the acquired position It is an error (Score).

  An example of executing the shape restoration program is shown in FIG. In this example, information such as the file name of the spatial information (World Point File) and the number of points included in the spatial information (World Points) are entered, and the shape information generated after execution is saved in the file (Model File) To do.

  In addition, as an existing technique for acquiring three-dimensional information from the above time-series images, a plurality of cameras are used so that an object can always be captured in a frame even if it is not projected on the frame by some cameras. There is a technique (see Patent Document 1).

Furthermore, there is a technique that enables acquisition of three-dimensional information from a plurality of frames that are not temporally continuous from a time-series image (see Patent Document 2).
Yuji Ishikawa, Isao Miyagawa, Kaori Wakabayashi, Tomohiko Arikawa, “Reconstruction of shape model from 3D point set based on MDL criteria”, Proceedings of Symposium on Image Recognition and Understanding (MIRU2002), vol. 165--170, Information Processing Society of Japan, 2002 Isao Miyagawa, Shigeru Nagai, Tomohiko Arikawa, “Reconstruction of Spatial Information by Factorization with Camera Motion Constrained”, IEICE Transactions, vol.J85-D-II, no.5, pp. 898-966, 2002 Proc. Of the 10th Functional Graphic Information System Symposium April 1999 Disaster prevention information provision system using urban 3D maps Naotori Otani, Yuji Ishikawa, Yoshitaka Kuwata, Shige Inoue JP 2001-101419 A JP 2002-032741 A

  In the above-described conventional technique, a three-dimensional point group can be acquired for each sequence, but since the actual input is given as a time-series image, the spatial information cannot be restored as it is.

  In a time-series image obtained by photographing a wide area, the number of images is enormous. Therefore, it is necessary to automatically divide a time-series image into sequences and acquire a three-dimensional point group.

  When the point cloud acquisition program for each sequence is executed manually, the operations of input setting, program execution, and result confirmation must be performed by the number of sequences, which takes a lot of time and effort. Also, setting mistakes during program execution are likely to occur.

  Furthermore, if the task of confirming the point cloud acquisition result, changing the setting, and re-executing is performed, it takes much more time and effort. Furthermore, after acquiring point clouds from all sequences, there is a problem that it is difficult to select which point group should be selected and combined into one spatial information.

  As a technique for acquiring three-dimensional information from a time-series image and a technique for enabling acquisition of three-dimensional information from a plurality of frames that are not temporally continuous from a time-series image, There is technology.

  However, each technique restores the three-dimensional shape on the assumption that the same object is projected between frames, and can be used as a technique for acquiring three-dimensional information from a single sequence. . However, in these techniques, there are a plurality of sequences, and the above-mentioned problems that occur when three-dimensional information is acquired for each sequence and shape restoration is performed after the integration of the three-dimensional information is taken into consideration. There is no current situation.

  The present invention has been made in view of the above circumstances. A point cloud is automatically acquired for each sequence only by providing a time-series image, and finally a set of three-dimensional coordinates in a form integrated as spatial information. 3D shape restoration method and apparatus, and 3D shape restoration, which can greatly save labor in shape restoration work from time-series images and reduce work errors. The challenge is to provide a program.

  In order to achieve the above object, the present invention provides a method for restoring a three-dimensional shape of a wide area from a time-series image obtained by photographing a whole area by repeating partial photographing of a photographing target area. A step of dividing the time-series image into a plurality of image sequences, a step of acquiring a three-dimensional point group from each divided image sequence a plurality of times, and acquiring a plurality of three-dimensional point groups; And a step of collecting a three-dimensional point group into one to acquire a wide-area three-dimensional point group (hereinafter referred to as spatial information) and a step of restoring a wide-area three-dimensional shape from the spatial information. As a means for solving the three-dimensional shape restoration method.

  In the second invention, the step of dividing the time-series image into a plurality of image sequences is a step of determining the number of images to be included in the image sequence based on the number of feature points in the image projected in the time-series image. And a step of determining a head image of the next image sequence based on a change amount of the image coordinates of the feature points in the image.

  In the third invention, a step of combining partial image sequence designation and parameters necessary for obtaining a three-dimensional point cloud into one file, and generating the above-mentioned file for all image sequences. Selecting a sequence of images for acquiring a point cloud by selecting the file, adding a point cloud acquisition result to the file, and selecting the file for spatial information And a step of selecting a group of point groups to be included in the three-dimensional shape restoration method.

  The fourth invention solves the three-dimensional shape restoration method comprising: collecting only point groups that satisfy the condition given the evaluation value of the three-dimensional point group and generating spatial information. Means.

  A fifth aspect of the invention is an apparatus for restoring a three-dimensional shape in a wide area from a time-series image obtained by repeatedly capturing a partial image of a region to be imaged, and dividing the time-series image into a plurality of image sequences. One means, a second means for obtaining a plurality of three-dimensional point groups by performing a process of obtaining a three-dimensional point group from each image sequence divided by the first means a plurality of times, and a second means for obtaining A third means for collecting a plurality of acquired three-dimensional point groups into one to acquire a wide-area three-dimensional point group (hereinafter referred to as spatial information), and a wide area 3 from the spatial information acquired by the third means. A fourth means for restoring a three-dimensional shape; and means for solving the three-dimensional shape restoration device.

  In a sixth aspect of the present invention, the first means for dividing the time-series image into a plurality of image sequences determines the number of images included in the image sequence based on the number of feature points in the image projected in the time-series image. A three-dimensional shape restoration comprising: a first determination unit configured to perform determination; and a second determination unit configured to determine a leading image of a next image sequence based on a change amount of an image coordinate of a feature point in the image. Means to solve the apparatus.

  The seventh aspect of the invention is an integration means for combining partial image sequence designation and parameters necessary for obtaining a three-dimensional point cloud into one file, and generating the above file for all image sequences. Selecting means, selecting means for selecting an image sequence for acquiring a point cloud by selecting the file, adding means for adding a point cloud acquisition result to the file, and selecting the file And a group selecting unit that selects a group of point groups to be included in the spatial information.

  According to an eighth aspect of the invention, the third means collects only point groups that satisfy the condition given the evaluation value of the three-dimensional point group, and generates spatial information, and restores the three-dimensional shape Means to solve the apparatus.

  According to a ninth aspect of the present invention, there is provided a means for solving a three-dimensional shape restoration program, characterized in that the steps in the three-dimensional shape restoration method are programs for causing a computer to execute the steps.

  Further, the present invention uses the configuration shown in FIG. 1 in order to solve the above respective inventions. That is, the present invention includes a sequence determination mechanism 11, a profile generation mechanism 12, a profile storage device 13, a profile selection mechanism 14, and a point group integration mechanism 15 in addition to the components of the existing technology (point group acquisition program 17, shape restoration program 18). And file server 16 (image file, point cloud file, spatial information file, shape file, etc.). Hereinafter, each mechanism will be described in order.

  The sequence determination mechanism 11 generates information necessary for dividing the time series image into sequences. Specifically, the number of frames in the sequence is determined based on the number of feature points in the frame projected on the time-series image, and further, based on the amount of movement of the image coordinates of the feature points, The start frame of the sequence is determined.

  The profile generation mechanism 12 generates a setting file (profile) for executing the point cloud acquisition program 17 for each sequence. Specifically, information for specifying a sequence from the time-series image and parameters given to the point cloud acquisition program 17 are described in the profile. All generated profiles are stored in the profile storage device 13.

  The profile selection mechanism 14 selects the generated profiles and executes the point cloud acquisition program 17 for those profiles. Based on the point cloud acquisition result, a profile is selected again, and the parameters given to the point cloud acquisition program 17 are changed to execute processing. If there is no profile to be executed again, the process moves to the point group integration mechanism 15.

  In the processing of the profile selection mechanism 14, the point cloud acquisition result may be added to the profile, a list of the profiles may be displayed together with the contents, and input relating to parameter change or profile selection may be received.

  The point group integration mechanism 15 collects the point group acquired from each sequence into one or a plurality of spatial information. Based on the point cloud acquisition result, it is determined which point cloud is to be combined into one spatial information, and the spatial information generated by the determination is input to the file server 16.

  In the processing of the point group integration mechanism 15, the point cloud acquisition result is added to the profile, a list of the profiles is displayed together with the contents, and an input related to the selection of the profile is received to select the point group to be included in the spatial information. Also good.

  With the above contrivance, the point cloud acquisition program 17 and the shape restoration program 18 are executed with the time series image as an input, and wide-area three-dimensional shape data can be obtained fully automatically.

  As described above, according to the present invention, a point cloud is automatically acquired for each sequence only by providing a time-series image, and finally a set of three-dimensional coordinates is obtained in a form integrated as spatial information. Can do. Therefore, there is an advantage that the troubles such as a to d described below can be saved.

a. Create a configuration file for each sequence and execute the program b. Confirmation of point cloud acquisition result, resetting of parameters and re-execution of program c. Selection of point cloud for grouping into spatial information d. Create a configuration file required for the shape restoration program.

  For example, in the example using the profiles I0001.prf to I0004.prf as described above, the parameter setting and the point group acquisition program are executed seven times. In a longer time-series image, an increase in the number of sequences is expected due to an increase in the number of sequences and an increase in the number of sequences that need to be re-executed. Therefore, a great labor saving can be achieved in the shape restoration work from the time series image, and work mistakes can be reduced.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  Normally, 30 images are recorded per second for video such as NTSC and Hi-Vision. From there, images are taken out at appropriate intervals and stored in a recording device. The recording device manages time-series images by storing a set of the shooting time and the file name of the image data. An example of time-series image management in the file server is shown in FIG. In FIG. 2, the number sequence on the left side in the figure indicates the shooting time, and the character string on the right side in the figure indicates an image file.

  Next, the sequence determination mechanism 11 will be described with reference to the flowchart shown in FIG. In this example, it is assumed that a time-series image is captured by linearly moving the camera upward in the captured image. In this case, the position of the photographed object in the frame moves from the top to the bottom of the frame as in the image example of each frame shown in FIG. Therefore, the initial feature points are assumed to be generated in the upper region of the initial frame (upper stage in FIG. 4).

Here, if the vertical width of the generation region is L, the value of L can be determined as 1/3 of the height of the frame F, for example. The generation of feature points is possible by automatically detecting a portion where the pixel value changes greatly. In order to know the position where the feature points are distributed, the center of gravity is obtained, and the height H ₀ from the lower end of the frame F is stored.

In FIG. 3, first, in step S1, the frame number counter N is set to "1" (initialized). In step S2, feature points are generated in the upper part of the initial frame of the time-series image (the vertical width of the feature point generation region is L). In step S3, the height H ₀ of the center of gravity of the feature point set in the initial frame is obtained.

After that, in step S4, the position of the feature point in the next frame F is obtained by searching for the image pattern around the feature point in the next frame (tracking the feature point). That is, feature point tracking is performed for only one frame, the frame number counter N is incremented by “1” in step S5, and the center-of-gravity height H _f of the feature point in frame F is obtained in step S6.

  Since the object photographed as described above moves from the top to the bottom in the frame, the feature point distribution range gradually decreases as the feature point tracking is performed sequentially.

  When it is determined that the feature point distribution area has moved by L, a new feature point generation area having an upper and lower width L is formed in the upper part of the frame F, and the next sequence starts from that frame. (Middle of FIG. 4).

This determination is made based on the height H _f of the barycentric position of the feature point set. When H _f becomes lower than H ₀ by L or more, the frame number counter N is stored as N ′ in step S7, and the frame Is the start frame of the next sequence.

  Hereinafter, it is determined in step S8 whether or not an arbitrary point of the feature points has reached the lower end of the frame, and feature point tracking is performed until the frame is out (Yes in step S8) (lower part of FIG. 4). ).

  Since the frame immediately before the occurrence of frame-out becomes the last frame in the sequence, the number (N−1) obtained by subtracting “1” from the counter N of the number of frames used up to this step is the sequence. (Step S9). In addition, the number of frames (overlap number) overlapping the next sequence is obtained as (N−N′−1) (step S10).

  In this example, it is assumed that the camera moves at a constant speed, and all the sequences have the same number. Therefore, if the number of sequences and the number of overlaps are determined, the start frames of all sequences can be specified. If the camera motion is not constant, the procedure of the flowchart shown in FIG. 3 may be repeated over the entire time series image to determine the number of frames and the start frame for each sequence. Since the feature point tracking process is also required when acquiring the point cloud, the result of the tracking process may be stored and used when acquiring the point cloud.

  Next, the profile generation mechanism 12 will be described. Here, taking as an example the case of generating a profile having the setting values shown in FIG. 20, the contents of a file (initial setting file) describing data to be input to the profile generation mechanism 12 are shown in FIG. Hereinafter, each line will be described.

  Threshold: Default value of the threshold value Threshold of the point cloud acquisition program 17 set for each profile. Depending on the sequence, the acquisition of the point cloud may not be successful. In this case, it is necessary to change the threshold value and re-execute, but the default value is used in the first point cloud acquisition process.

  Initial Frame: Designates the first frame of a time-series image by the shooting time of that frame.

  End Frame: Designates the last frame of the time-series image by the shooting time of that frame.

Frames: Number of frames in one sequence when time-series images are divided into sequences
Overlap Frames: The number of overlaps between sequences (see FIG. 19).

  Next, the profile generation mechanism 12 will be described based on the flowchart shown in FIG. In FIG. 6, first, the initial setting file is read in step S11. In the following description, for example, the setting value of the Initial Frame is expressed by enclosing it with “[“, “]” as [Initial Frame] or the like.

Next, in step S12, [Initial Frame] is set to the variable T, and in step S13, the shooting time of the [Frames] frames before the frame indicated by [End Frame] is set to the variable T _end .

When a sequence is started from a frame shot at a time later than T _end , the sequence range extends beyond the range specified in the initialization file ([Initial Frame] to [End Frame]). The start time T is checked.

  In step S14, the profile is generated with the setting value blank, and the setting value of the initial setting file is set as it is in the Frames and Threshold columns. T is set in the Initial Frame field (step S15), and the profile is stored in the profile storage device 13 (step S16).

  The first frame of the next sequence is the frame [Frames]-[Overlap Frame] ahead of the frame shot at time T. Therefore, the shooting time of the frame is set to T (step S17).

Here, if T is larger than T _end (Yes) in step S18 for the reason described above, the process is terminated. Otherwise (no), a new profile is generated for the sequence starting with the new T.

  FIG. 7 shows an example in which a profile is generated based on the contents of the initial setting file shown in FIG. A time-series image from the photographing time 001 to 200 is divided into four sequences, and a profile of each sequence is automatically generated. Assume that each profile is stored in the profile storage device 13 with file names I0001.prf to I0004.prf.

  In the above example, the time-series image to be restored is specified by Initial Frame and End Frame, but is specified by Initial Frame (or End Frame) and the number of frames.

  In the description of the profile selection mechanism 14, the setting value of the profile is expressed by enclosing it with “[“, “]”. [Score] is an average error of point positions, and is a value set after the point group acquisition program 17 is executed. The point group acquisition program 17 can reduce points with large errors by lowering [Threshold] and reduce the average error.

  The upper diagram in FIG. 8 is a diagram of the point cloud that has acquired the building shape as viewed from the side. In this diagram, by lowering [Threshold], points with large positional errors are removed, as shown in the lower diagram in FIG. Become.

  The profile selection mechanism 14 described here aims to set [Score] to be equal to or lower than a predetermined threshold value Vs in all profiles. However, in order to leave as many points as possible, the [Threshold] is gradually reduced.

  FIG. 9 is a flowchart for explaining the profile selection mechanism 14. In FIG. 9, all profiles generated by the profile generation mechanism 12 first are read from the storage device 13 (step S21), and all profiles are added to the profile list (step S22).

  Next, while repeating [Threshold] repeatedly, the value is reduced to half as long as [Score] becomes less than or equal to the threshold value Vs in step S23, the profile is removed from the profile list, and steps S24 and S25 are followed. The point cloud acquisition program is executed one by one for all the profiles included in the profile list.

  However, in order to reduce the processing time, it is determined in step S24 whether the profile list is empty or the number of repetitions is greater than the threshold N times, the threshold is limited to N times, and the profiles included in the profile list in step S25 Halve the [Threshold]. Through the above processing, a profile is automatically selected, and the point cloud acquisition program 17 is executed and input to the file server 16.

  As a method of decreasing [Threshold], there are a method of subtracting each time, a method of accepting manual input, and the like. In addition, when restoring a shape from a point cloud, not only [Score] but also the number of points contained in the point cloud (Feature Points) is an important factor. If the threshold is below a certain threshold, the target profile is deleted from the profile list.

  The selection of the profile to be executed by the point cloud acquisition program 17 includes an automatic method using the threshold value and the evaluation function as described above, and a method of visually confirming the three-dimensional arrangement of the point cloud and receiving manual input. Either of these may be selected.

  As a method based on manual selection, as shown in FIG. 10, a list of profile selection screens is displayed together with [Score] and the like, and a field for directly inputting a Threshold value or an execution target of the point cloud acquisition program 17 A check box for selecting a profile to be used is provided. In this way, by accepting inputs related to all profiles at once, it is possible to reduce labor and mistakes associated with profile selection.

  A case where the processing of the profile selection mechanism 14 is performed on the profile of FIG. 7 will be described with reference to FIG. The contents of each profile before the execution of processing are set to Threshold values, but Score (average error) and Feature Points (number of points included in the point cloud) are not set.

  Since all profiles are initially added to the profile list, the point cloud acquisition program 17 is executed for each profile of I0001.prf to I0004.prf. The result after execution is shown in FIG. 11 as the end of the first process.

  Here, if the threshold Vs of [Score] is 4.0, since [Score] of I0001.prf and I0003.prf is smaller than the threshold Vs, the two profiles are deleted from the profile list.

  With respect to the remaining I0002.prf and I0004.prf, [Threshold] is set to a half value (= 10) as shown at the end of the second processing in FIG. 11, and the point cloud acquisition program 17 is executed again. However, I0004.prf is re-executed with [Threshold] set to “5” as shown in FIG. 11 when all the processes are completed because [Score] is still larger than the threshold Vs even if [Threshold] is set to “10”. .

  Assuming that the number of repetitions is limited to 2 (N = 2), the final result is obtained as in the end of all processes shown in the lower part of FIG. As a result of reducing [Threshold], the number of points is reduced in I0002.prf and I0004.prf.

  In particular, in I0004.prf, [Score] could not be made lower than the threshold value Vs, and most of the points were deleted. This indicates that it is difficult to accurately obtain the positions of a large number of points from the sequence of I0004.prf. The handling of the point cloud of I0004.prf will be described in the next point cloud integration mechanism 15.

  The point group integration mechanism 15 generates spatial information by collecting the point groups acquired from the profiles into one set. The process is shown in the flowchart of FIG. First, in FIG. 12, all profiles are read in step S31 (hereinafter, each setting value of the image profile is expressed as [Threshold]).

  As described above, there is a sequence in which a point cloud with high accuracy cannot be obtained. Therefore, in step S32, [Shape3D] having a profile whose [Score] is equal to or lower than the threshold value Vs is read and the content is added to the spatial information file. That is, only the point group whose [Score] is equal to or less than the threshold value Vs is added to the spatial information file.

  After generating the spatial information, a setting file necessary for executing the shape restoration program 18 is generated in step S33. The point group integration mechanism 15 may also accept a manual method in addition to an automatic method using a threshold for profile selection.

  Specifically, as shown in the profile selection screen of FIG. 13, the profiles are listed together with [Score], [Feature Points], and the like so that selection of profiles to be integrated into the spatial intelligence can be input in a lump.

  Next, a case where the processing of the point group integration mechanism 15 is performed on the profiles I0001.prf to I0004.prf generated by the processing example of FIG. 11 will be described.

  First, all profiles from I0001.prf to I0004.prf are read. Among these, since [Score] (= 4.2) of I0004.prf is larger than the threshold value Vs, it is determined that the position accuracy of the point cloud is low, and the point cloud from I0001.prf to I0003.prf is stored in the spatial information file M0001.XYZ. Added.

  Specifically, the file I0001.xyz describing the three-dimensional coordinate values of each point registered as [Shape3D] in I0001.prf is read and written to M0001.XYZ. Similarly, I0002.xyz and I0003.xyz are read and added to M0001.XYZ.

  Next, a setting file M0001.prf for the shape restoration program 18 is generated. “M0001.XYZ” is registered in the World Point Set as the name of the file describing the spatial information. Further, the sum of [Feature Points] of the three profiles is registered in the World Points column of M0001.prf as the number of three-dimensional coordinate values described in M0001.XYZ.

  In addition, Model File is added as a column describing the file name of the shape file generated after execution of the shape restoration program, and the setting file shown in FIG. 14 is automatically generated.

  FIG. 15 shows an example in which spatial information is restored from a time-series image acquired by taking an aerial image of an urban area and a three-dimensional model of the urban area is constructed. In a highly accurate point cloud automatic acquisition process, a profile generation mechanism and a profile selection mechanism are used, and in a spatial information restoration process, a point cloud integration mechanism is used.

  In the present invention, a point cloud acquisition program and a shape restoration program are executed using a time-series image as input, and wide-area three-dimensional shape data (city space model) can be obtained fully automatically. Can be used for basic data.

The block diagram which shows the structure of this invention. Explanatory drawing which shows the example of the data storage format of a time series image. The flowchart of a sequence determination mechanism. Explanatory drawing explaining the procedure which determines the number of frames of a sequence. Explanatory drawing which shows the example of a description of the initial setting file of a profile production | generation mechanism. The flowchart of a profile production | generation mechanism. Explanatory drawing which shows the example of a production | generation of a profile. Explanatory drawing which shows the result of having removed the point with a big position error. The flowchart of a profile selection mechanism. Explanatory drawing which shows the selection screen of the profile used as a point cloud acquisition program execution object. Explanatory drawing which shows the example of a process of a profile selection mechanism. The flowchart of a point cloud integration mechanism. Explanatory drawing which shows the selection screen of the profile used as the object of point cloud integration. Explanatory drawing which shows the example of the setting file of a shape restoration program. Explanatory drawing which shows the example which built the urban area space model by embodiment. Explanatory drawing which shows the usage example of the three-dimensional shape acquisition technique from an image | video. Explanatory drawing which shows the aerial photography example of a building. Explanatory drawing which shows the example of the image of each flame | frame. The figure explaining the division | segmentation into the sequence of a fixed number of sheets from a time series image. Explanatory drawing which shows the example of the execution method of a point cloud acquisition program. Explanatory drawing which shows the example of the execution method of a shape restoration program.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 ... Sequence determination mechanism 12 ... Profile generation mechanism 13 ... Profile storage apparatus 14 ... Profile selection mechanism 15 ... Point cloud integration mechanism 16 ... File server 17 ... Point cloud acquisition program 18 ... Shape restoration program

Claims (9)

A method for reconstructing a three-dimensional shape of a wide area from a time-series image obtained by photographing a whole area by repeating partial photographing of a photographing target area,
Dividing the time-series image into a plurality of image sequences;
Performing a process of acquiring a three-dimensional point group from each divided image sequence a plurality of times to acquire a plurality of three-dimensional point groups;
Combining a plurality of three-dimensional point groups into one to obtain a wide-area three-dimensional point group (hereinafter referred to as spatial information);
And a step of restoring a wide area three-dimensional shape from the spatial information. The three-dimensional shape restoration method according to claim 1,
The step of dividing the time-series image into a plurality of image sequences is as follows:
Determining the number of images to be included in the image sequence based on the number of feature points in the image projected in the time-series image;
Determining the top image of the next image sequence based on the amount of change in the image coordinates of the feature points in the image. The three-dimensional shape restoration method according to claim 1,
Combining a partial image sequence specification and parameters necessary for obtaining a three-dimensional point cloud into one file;
Generating the above file for all image sequences;
Selecting an image sequence to acquire a point cloud by selecting the file;
Adding a point cloud acquisition result to the above file;
Selecting a group of point groups to be included in the spatial information by selecting the file, and a method of restoring a three-dimensional shape. The three-dimensional shape restoration method according to claim 1,
Collecting only point groups that satisfy the condition for which the evaluation value of the three-dimensional point group is given, and generating spatial information. A method for restoring a three-dimensional shape, comprising: An apparatus that restores a wide-area three-dimensional shape from a time-series image obtained by repeatedly performing partial imaging of an imaging target area,
First means for dividing a time-series image into a plurality of image sequences;
A second means for obtaining a plurality of three-dimensional point groups by performing a process of obtaining a three-dimensional point group from each image sequence after being divided by the first means a plurality of times;
A plurality of three-dimensional point groups acquired by the second means, and a third means for acquiring a wide-area three-dimensional point group (hereinafter referred to as spatial information);
And a fourth means for restoring a wide-area three-dimensional shape from the spatial information acquired by the third means. The three-dimensional shape restoration apparatus according to claim 5,
The first means for dividing the time-series image into a plurality of image sequences includes:
A first determination unit that determines the number of images to be included in the image sequence based on the number of feature points in the image displayed in the time-series image;
A three-dimensional shape restoration apparatus comprising: a second determination unit that determines a head image of a next image sequence based on a change amount of an image coordinate of a feature point in an image. The three-dimensional shape restoration apparatus according to claim 5,
An integration means for combining partial image sequence specification and parameters necessary for acquiring a three-dimensional point cloud into one file;
Generating means for generating the above file for all image sequences;
A selection means for selecting an image sequence for acquiring a point cloud by selecting the above file;
Additional means for adding point cloud acquisition results to the above file;
A three-dimensional shape restoration apparatus comprising: a group selection unit that selects a group of point groups to be included in the spatial information by selecting the file. The three-dimensional shape restoration apparatus according to claim 5,
The third means collects only the point group that satisfies the condition for which the evaluation value of the three-dimensional point group is given, and generates spatial information.
A three-dimensional shape restoration apparatus. The step in the three-dimensional shape restoration method according to any one of claims 1 to 5 is a program for causing a computer to execute.
A program for restoring a three-dimensional shape.

Images