JPWO2018235219A1 - Self-location estimation method, self-location estimation device, and self-location estimation program - Google Patents

Abstract

The self-position estimating device 20 is a self-position estimating device on which the imaging device 21 is mounted. The self-position estimating device 20 includes a self-position estimating device 20 based on a moving image acquired from the imaging device 21. An extraction unit 22 that extracts a feature point image candidate that is an image in which a feature point that is a region is captured by a predetermined method, and an estimation unit 23 that estimates a self-position using the extracted feature point image candidate Prepare.

Description

  The present invention relates to a self-position estimation method, a self-position estimation device, and a self-position estimation program, and more particularly to a self-position estimation method, a self-position estimation device, and a self-position estimation program that use video captured by a camera.

  The measurement of the position and orientation of an imaging device based on image information is performed by augmented reality (AR: Augmented Reality) or mixed reality (MR: Mixed Reality). It is used for position estimation, three-dimensional modeling of objects and scenes, and the like.

  Non-Patent Document 1 holds information on feature points in a scene as a three-dimensional map, and positions and poses of an imaging device based on correspondence between feature points detected in an image and feature points in the three-dimensional map. Is described.

  The method described in Non-Patent Document 1 uses the position and orientation of the imaging device measured in the previous frame to measure the position and orientation of the imaging device in the current frame.

  For example, the method described in Non-Patent Document 1 predicts the position and orientation of the imaging device in the current frame based on the position and orientation of the imaging device and the motion model measured in the previous frame. The method described in Non-Patent Document 1 uses the predicted position and orientation of the imaging device for associating a feature point in a three-dimensional map with a feature point in an image.

  Also, the method described in Non-Patent Document 1 uses the predicted position and orientation of the imaging device as initial values of calculation processing repeatedly executed to obtain the position and orientation of the imaging device in the current frame. .

  When the motion of the imaging device is large between frames or when the number of feature points detected in an image is small, measurement of the position and orientation of the imaging device by the method described in Non-Patent Document 1 may fail. .

  In order to prevent the measurement from failing, it is important to use an image suitable for self-position estimation as a feature point used in a calculation process repeatedly executed. For example, there are a method of obtaining a non-moving feature point in an image in the process of estimating a self-position, and a method of obtaining an appropriate feature point by excluding a moving feature point in an image.

  In view of the above, for example, US Pat. No. 6,064,097, discloses a special method that, when a robot encounters an undesired event (eg, accidental collision), the data algorithm before the danger is identified by a learning algorithm. A mobile robotic system is described in which a "danger" tag is inserted into the dataset.

  Patent Literature 2 discloses a self-position re-estimation method in which a result obtained by applying a posture detector to a key frame is compared with a current view in consideration of a case where self-position estimation fails. .

  In the self-position re-estimation method described in Patent Literature 2, a shape estimation process using a machine learning classifier or the like, a semantic image labeling process is applied to a current frame and a previous frame, and the like. The feature of is seen.

  In addition, Patent Document 3 discloses that when a set of feature points extracted from a detected scene is recognized in any one of previously detected and stored experiences, based on the recognized experience. A sensor position estimating method for evaluating the current position of a vehicle is described.

  An example of a device to which each of the above-described self-position estimation techniques is applied will be described below. FIG. 6 is a block diagram illustrating a configuration example of a general self-position estimation system. As shown in FIG. 6, the self-position estimation system 90 includes a camera 200 and a self-position estimation device 900.

  Further, as shown in FIG. 6, the self-position estimation device 900 includes a signal processing unit 190. Further, the camera 200 is communicably connected to the self-position estimation device 900. The connection between the camera 200 and the self-position estimation device 900 may be a wired connection or a wireless connection. Video data captured by the camera 200 is transmitted to the self-position estimation device 900.

  FIG. 7 is a block diagram showing another configuration example of a general self-position estimation system. The self-position estimation system 91 shown in FIG. 7 is used when the acquired video data does not need to be processed immediately, such as when obtaining only the trajectory of a moving object.

  As shown in FIG. 7, the self-position estimating device 900 is communicably connected to a recording medium 210 instead of the camera 200. In the example illustrated in FIG. 7, the video data stored in the recording medium 210 is transmitted to the self-position estimation device 900 instead of the video data captured by the camera 200.

  As shown in FIGS. 6 and 7, video data captured by the camera 200 or video data stored in the recording medium 210 is transmitted to the self-position estimation device 900 and processed by the signal processing unit 190.

  Next, the configuration of the signal processing unit 190 will be described with reference to FIG. FIG. 8 is a block diagram illustrating a configuration example of a general signal processing unit. The signal processing unit 190 illustrated in FIG. 8 includes an image acquisition unit 191, a feature point tracking unit 193, a position / posture estimation unit 194, a map generation unit 195, and a minimization unit 196.

  The video data transmitted to self-position estimation device 900 is input to image acquisition section 191 of signal processing section 190. All the video data input to the image acquisition unit 191 is input to the feature point tracking unit 193.

  Next, the feature point tracking unit 193 performs a process of acquiring the movement amount of the feature point on the input video data. Next, the position / posture estimation unit 194 estimates the camera position and the camera posture using the acquired movement amount of the feature point.

  Next, the map generation unit 195 estimates the three-dimensional positions of the feature points and locally generates a map of the environment. Next, the minimizing unit 196 matches the locally generated environment map with the entire image by minimizing the reprojection error in which the entire image is used. The self-position based on the video data is estimated by the matching performed by the minimizing unit 196.

  Since the map generation unit 195 locally generates a map of the environment, if the generated map of the environment is directly applied to the entire image, distortion may occur. The minimizing unit 196 matches the entire image so that no distortion occurs even when the map of the environment to be generated is overlooked.

Japanese Patent No. 5629390 Japanese Patent No. 5881743 JP-T-2015-508163A JP 2010-152787 A JP-A-2006-157197 JP 2017-021427 A

G. Klein and D. Murray, “Parallel Tracking and Mapping for Small AR Workspaces,” Proc. 6th IEEE and ACM International Symposium on Mixed and Augmented Reality (ISMAR'07), 2007.

  The mobile robot system described in Patent Literature 1 predicts danger based on acquired images with reference to information measured in the past. Further, the self-position re-estimation method described in Patent Document 2 recovers from estimation failure by referring to information measured in the past based on an acquired image. The sensor position estimating method described in Patent Document 3 performs self-position estimation evaluation and the like based on acquired images with reference to information measured in the past.

  That is, in each of the above-described self-position estimation techniques, the types and patterns of the feature points to be referred to are enormous, and the reference process requires a lot of time. Further, in each of the above-described self-position estimation techniques, a lot of cost is required for a process of extracting a feature point suitable for reference. Therefore, when the self-position is immediately estimated using each of the above-described self-position estimation techniques, there is a problem that the reference process is not executed in a short time.

  Patent Literatures 4 to 6 disclose techniques for limiting the feature points to be referred to. Patent Literature 4 discloses an environment map that estimates a current position and a position of a measurement point based on a measurement value of a measurement point measured while autonomously moving in a predetermined environment, and causes a computer to execute a process of generating a map of the environment. The generation program is described. The environment map generation program described in Patent Literature 4 causes a computer to delete position information corresponding to a measurement point determined to be a movement measurement point from the storage unit.

  Patent Document 5 discloses a self-position estimating apparatus that can generate a map of an initial environment used for estimating a self-position and accurately estimate a self-position. A self-position estimating device described in Patent Literature 5 extracts a feature point of an object from a time-series image input from an image input unit, and calculates a distance from the image input unit to the feature point. A moving object removing unit that removes an object corresponding to the feature point from the time-series image when it is determined that the feature point is a moving object.

  Patent Literature 6 discloses a self-position estimation device that improves the self-position estimation accuracy. The self-position estimating device described in Patent Literature 6 includes a feature point extracting unit that extracts an actual feature point from an image acquired by an image acquiring unit, and a moving object feature extracting unit that extracts an actual feature point extracted by the feature point extracting unit. Feature point removing means for removing dynamic feature points of a target object.

  As described above, techniques for limiting the feature points to be referenced and extracted have already been provided. However, in order to perform the self-position estimation faster, it is considered effective to reduce the amount of the video data itself used for the estimation.

[Object of the invention]
Therefore, an object of the present invention is to provide a self-position estimating method, a self-position estimating device, and a self-position estimating program that can solve the above-described problem and can estimate a self-position more quickly.

  The self-position estimating method according to the present invention is a self-position estimating method executed in a self-position estimating device equipped with an imaging device, and is used for estimating the self-position of the self-position estimating device from a moving image acquired from the imaging device. A feature point image candidate which is an image of a feature point which is an area in a moving image to be used is extracted by a predetermined method, and a self-position is estimated using the extracted feature point image candidate. Features.

  A self-position estimating device according to the present invention is a self-position estimating device equipped with an image capturing device, and a region in a moving image used for estimating a self-position of the self-position estimating device from a moving image acquired from the image capturing device. An extraction unit that extracts a feature point image candidate that is a captured image of a feature point by a predetermined method, and an estimation unit that estimates a self-position using the extracted feature point image candidate. Features.

  A self-position estimating program according to the present invention is a self-position estimating program executed on a computer equipped with an imaging device, and is used by the computer to estimate the self-position of the computer from a moving image acquired from the imaging device. An extraction process of extracting a feature point image candidate that is an image of a feature point that is an area in a moving image by a predetermined method, and an estimation process of estimating a self-position using the extracted feature point image candidate Is executed.

  According to the present invention, the self-position can be estimated more quickly.

1 is a block diagram illustrating a configuration example of a first embodiment of a self-position estimation system according to the present invention. It is a block diagram showing other examples of composition of a 1st embodiment of a self-position estimation system by the present invention. FIG. 3 is a block diagram illustrating a configuration example of a signal processing unit 110 according to the first embodiment. 5 is a flowchart illustrating an operation of a self-position estimation process performed by the self-position estimation device 100 according to the first embodiment. It is a block diagram showing the outline of the self-position estimation device by the present invention. It is a block diagram which shows the example of a structure of a general self-position estimation system. It is a block diagram which shows the other example of a structure of the general self-position estimation system. FIG. 3 is a block diagram illustrating a configuration example of a general signal processing unit.

[First Embodiment]
[Structure description]
Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a configuration example of a first embodiment of a self-position estimation system according to the present invention. FIG. 2 is a block diagram showing another configuration example of the first embodiment of the self-position estimation system according to the present invention.

  The self-position estimation system according to the present embodiment provides a video selection method used for selecting a key frame to be referred to. The self-position estimation system according to the present embodiment efficiently extracts an image in which a feature point suitable for reference is captured from video data captured by a camera, and compares the feature points in the extracted image. By extracting an image obtained by capturing a feature point suitable for reference, the self-position estimation system according to the present embodiment realizes high-speed self-position estimation.

  As shown in FIG. 1, the self-position estimation system 10 of the present embodiment includes a self-position estimation device 100 and a camera 200. The configuration of the self-position estimation system 10 of the present embodiment is the same as the configuration of the self-position estimation system 90 shown in FIG.

  Further, as shown in FIG. 2, the self-position estimation system 11 of the present embodiment includes a self-position estimation device 100 and a recording medium 210. The configuration of the self-position estimation system 11 of the present embodiment is the same as the configuration of the self-position estimation system 91 shown in FIG.

  The self-position estimation device 100 shown in FIGS. 1 and 2 includes a signal processing unit 110 instead of the signal processing unit 190, unlike the self-position estimation device 900 shown in FIGS. 6 and 7. Hereinafter, the configuration of the signal processing unit 110 will be described with reference to FIG.

  FIG. 3 is a block diagram illustrating a configuration example of the signal processing unit 110 according to the first embodiment. The signal processing unit 110 illustrated in FIG. 3 includes an image acquisition unit 111, a filtering unit 112, a feature point tracking unit 113, a position / posture estimation unit 114, a map generation unit 115, a minimization unit 116, and a learning unit. 117.

  The configuration of the signal processing unit 110 of the present embodiment is the same as the configuration of the signal processing unit 190 illustrated in FIG. 8 except for the filtering unit 112 and the learning unit 117. The functions of the image acquisition unit 111, the feature point tracking unit 113, the position / posture estimation unit 114, the map generation unit 115, and the minimization unit 116 include an image acquisition unit 191, a feature point tracking unit 193, and a position / posture. The functions are the same as those of the estimation unit 194, the map generation unit 195, and the minimization unit 196.

  The filtering unit 112 has a function of extracting, from the video data input from the image acquisition unit 111, an image in which a region that may be used as a feature point is captured. Hereinafter, an image in which a feature point is captured is referred to as a feature point image. That is, the filtering unit 112 extracts feature point image candidates.

  The feature point image candidates extracted by the filtering unit 112 are used for self-position estimation processing. However, in the present embodiment, the image candidates extracted by the filtering unit 112 are not always adopted as the feature point images. When a difference equal to or more than a predetermined threshold value occurs between a candidate for an image to be extracted and a reference point (feature point), an area appropriate as a reference point is registered as a new reference point.

  In the signal processing unit 110 of the present embodiment, the acquired images are sorted each time into an image adopted as a feature point image and an image not adopted as a feature point image. The minimizing unit 116 of the present embodiment inputs the two types of images sorted as described above to the learning unit 117.

  The learning unit 117 has a function of acquiring a feature amount of an image adopted as a feature point image and a feature amount of an image not adopted as a feature point image by executing a learning process.

  The learning unit 117 inputs the feature amount obtained in the learning process to the filtering unit 112. The filtering unit 112 extracts a feature point image candidate using the input feature amount. That is, in the signal processing unit 110 of the present embodiment, an image extracted as a candidate for a feature point image is selected in the learning process.

  The learning unit 117 learns the feature amount of the feature point image based on the image used for the feature point image in the self-position estimation processing. Further, the learning unit 117 learns a feature amount of an image that is not suitable for a feature point image based on an image that is not used as a feature point image in the self-position estimation processing. The filtering unit 112 may exclude an image that is not suitable for the feature point image from the reference image in the self-position estimation processing using the learned feature amount.

  The filtering unit 112 may be input in advance as a selection criterion with a feature model of an image in which an area useful as a reference point (feature point) is captured. The filtering unit 112 may extract feature point image candidates using the input selection criteria.

  As described above, the self-position estimating apparatus 100 of the present embodiment patterns an image in which appropriate feature points required for estimating the self-position are captured in advance by setting selection criteria or the like.

  In a general self-position estimation technique, all acquired images are referred to, and the types and patterns of extracted feature points are enormous. That is, since it takes a lot of time to execute the reference processing, there is a problem that the reference processing is not completed in a short time when the self-position is estimated at high speed.

  When a pattern of an image in which an appropriate feature point has been captured is grasped in advance, all acquired images are referred to, and compared with a case where the types and patterns of feature points extracted become enormous. Because the images are limited, calculation time and cost are reduced. That is, when the self-position estimation device 100 of the present embodiment is used, the calculation time required for the self-position estimation is shortened, so that the self-position estimation and the environment mapping are executed at higher speed.

  In addition, the self-position estimating apparatus 100 of the present embodiment selects an image in which an appropriate reference point (feature point) has been captured, so that the instability of the reference image can be eliminated. That is, when the self-position estimation device 100 of the present embodiment is used, the estimation accuracy of the self-position is further improved.

[Description of operation]
Hereinafter, the operation of the self-position estimating apparatus 100 of the present embodiment for estimating the self-position will be described with reference to FIG. FIG. 4 is a flowchart illustrating the operation of the self-position estimation process performed by the self-position estimation device 100 according to the first embodiment.

  First, the video data transmitted to the self-position estimation device 100 is input to the image acquisition unit 111 of the signal processing unit 110 (Step S101). The image acquisition unit 111 inputs the input video data to the filtering unit 112.

  Next, the filtering unit 112 extracts feature point image candidates from the input video data (step S102). The filtering unit 112 inputs the extracted feature point image candidates to the feature point tracking unit 113.

  Next, the feature point tracking unit 113 acquires the movement amount of the feature point based on the input feature point image candidates (step S103). The feature point tracking unit 113 inputs the obtained movement amount of the feature point to the position / posture estimation unit 114.

  If it is determined that the captured feature point is not appropriate for the input feature point image candidate, the feature point tracking unit 113 may acquire the movement amount of the feature point based on another image.

  Next, the position / posture estimation unit 114 estimates a camera position and a camera posture using the input movement amount of the feature point (step S104). The position / posture estimation unit 114 inputs the estimated camera position and camera posture to the map generation unit 115.

  Next, the map generation unit 115 estimates the three-dimensional position of the feature point based on the input camera position and camera posture, and locally generates a map of the environment (step S105). The map generation unit 115 inputs the generated environment map to the minimization unit 116.

  Next, the minimizing unit 116 matches the locally generated environment map with the entire image by minimizing the reprojection error in which the entire image is used (step S106). Further, the minimizing unit 116 inputs the image adopted as the feature point image and the image not adopted as the feature point image to the learning unit 117.

  Next, the learning unit 117 performs a learning process based on the input image, thereby acquiring the feature amount of the image adopted as the feature point image and the feature amount of the image not adopted as the feature point image. (Step S107).

  Next, the learning unit 117 inputs the obtained feature amount to the filtering unit 112 (Step S108). After the input, the self-position estimation device 100 ends the self-position estimation processing. Note that the self-position estimation device 100 of the present embodiment may repeatedly execute the self-position estimation process illustrated in FIG.

[Explanation of effects]
The self-position estimating apparatus 100 according to the present embodiment selects an image in which an area that may be used as a feature point is captured from a video used for self-position estimation captured by a camera, and selects the selected image. Is extracted.

  The learning unit 117 of the self-position estimation device 100 performs a learning process on a feature amount of an image in which an area that may be used as a feature point or a line feature is captured based on a video (environment) used for self-position estimation. Get in. The filtering unit 112 selects a stable feature point or line feature from a video in which natural features are captured using the feature amount of the acquired image, and sets the selected feature point or line feature as a reference point. be able to.

  In contrast to a general self-position estimating device in which the types and patterns of feature points referred to because all the acquired images are used are large, the self-position estimating device 100 of the present embodiment refers to An appropriate feature point is selected as a point, and the selected feature point can be extracted.

  When the self-position estimating apparatus 100 of the present embodiment is used, it is not necessary to consider all the feature points in the video. That is, since only an appropriate image is focused on as a feature point image, calculation time and calculation cost required for self-position estimation are reduced.

  For example, when the self-position is estimated based on a structure having high similarity over a plurality of images in an infrastructure inspection, a plant inspection, or the like, a feature of an image appropriate as a reference image and an image inappropriate as a reference image Are clearly separated.

  For example, a feature of an image suitable as a reference image is that an invariant reference target such as a pier on an expressway is photographed. Further, for example, a feature of an image that is inappropriate as a reference image is that a reference target that may move, such as a car parked on the road, is captured.

  The filtering unit 112 according to the present embodiment selects and uses video information in which an invariant reference target is photographed and is easily used as a reference image. That is, the self-position estimating apparatus 100 according to the present embodiment can reduce the calculation cost required for extracting the feature points in the image. When the self-position estimation device 100 of the present embodiment is used, the calculation time required for the self-position estimation is shortened, so that the self-position estimation and the environment mapping are executed at higher speed.

  In addition, the self-position estimating apparatus 100 of the present embodiment selects an image in which an appropriate reference point has been captured, so that the instability of the reference image can be eliminated. That is, when the self-position estimation device 100 of the present embodiment is used, the estimation accuracy of the self-position is further improved.

  The self-position estimation device 100 of the present embodiment can reduce the calculation time required for the self-position estimation as compared with a general self-position estimation method, and thus can estimate the self position more quickly. In addition, the self-position estimating apparatus 100 of the present embodiment selects an image in which an appropriate reference point has been captured, so that the instability of the reference image can be eliminated. That is, the self-position estimating apparatus 100 can further improve the self-position estimation accuracy.

  In addition, the self-position estimation device 100 of the present embodiment may be realized by, for example, a CPU (Central Processing Unit) that executes a process according to a program stored in a non-temporary storage medium. That is, the image acquisition unit 111, the filtering unit 112, the feature point tracking unit 113, the position / posture estimation unit 114, the map generation unit 115, the minimization unit 116, and the learning unit 117 execute, for example, a CPU that executes processing according to program control. It may be realized by.

  Further, each unit in the self-position estimation device 100 of the present embodiment may be realized by a hardware circuit. As an example, the image acquisition unit 111, the filtering unit 112, the feature point tracking unit 113, the position / posture estimation unit 114, the map generation unit 115, the minimization unit 116, and the learning unit 117 are each implemented by an LSI (Large Scale Integration). Is done. Further, they may be realized by one LSI.

  Next, the outline of the present invention will be described. FIG. 5 is a block diagram showing an outline of the self-position estimation device according to the present invention. The self-position estimating device 20 according to the present invention is a self-position estimating device equipped with an imaging device 21 (for example, a camera 200), and calculates the self-position of the self-position estimating device 20 from a moving image acquired from the imaging device 21. An extraction unit 22 (for example, a filtering unit 112) that extracts a feature point image candidate that is a captured image of a feature point that is a region in a moving image used for estimation by a predetermined method; An estimating unit 23 (for example, a feature point tracking unit 113, a position / posture estimating unit 114, a map generating unit 115, and a minimizing unit 116) that estimates a self-position using image candidates is provided.

  With such a configuration, the self-position estimation device can estimate the self-position at higher speed.

  In addition, the self-position estimating device 20 includes an acquisition unit (for example, a learning unit 117) that executes a learning process and acquires information on an image used for extracting a candidate for a feature point image. A feature point image candidate may be extracted using information of the obtained image.

  With such a configuration, the self-position estimating apparatus can extract an image learned in the learning processing as a candidate for a feature point image.

  Further, the acquisition unit may execute the learning process based on an image used for estimating its own position as a feature point image among feature point image candidates.

  With such a configuration, the self-position estimation device can learn the feature amount of the feature point image.

  In addition, the acquiring unit performs the learning process based on an image other than the image used for estimating the self position as the feature point image among the feature point image candidates, and the extracting unit 22 performs the image acquiring process in the learning process. May be extracted from the moving image acquired from the imaging device 21, and then a feature point image candidate may be extracted.

  With such a configuration, the self-position estimation device can estimate the self-position at higher speed.

  The extraction unit 22 may extract feature point image candidates using a model of an area that can be used as a feature point.

  With such a configuration, the self-position estimating apparatus can extract feature point image candidates reflecting the content specified in advance by the user.

  In addition, the self-position estimating device 20 includes a storage unit (for example, a recording medium 210) that stores a moving image acquired from the imaging device 21, and the extraction unit 22 extracts a feature point from the moving image stored in the storage unit. Image candidates may be extracted.

  With such a configuration, the self-position estimating device can estimate the past self-position based on a moving image in which a predetermined time has elapsed since the image was taken.

  As described above, the present invention has been described with reference to the exemplary embodiments and examples. However, the present invention is not limited to the exemplary embodiments and examples. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

Industrial applicability

  INDUSTRIAL APPLICABILITY The present invention is suitably applied to an application in which a robot represented by an unmanned flying robot grasps a space by itself and immediately detects and avoids danger. INDUSTRIAL APPLICABILITY The present invention is suitably applied when a robot autonomously executes a task in a place where it is difficult for a person to enter or in a dangerous place, and in a vast area that cannot be dealt with manually.

  In particular, the present invention is industrially effectively applied to infrastructure that deteriorates over time, autonomous inspection of plants by robots, and the like. When the present invention is applied, manpower, cost, and time required for the autonomous inspection may be reduced.

10, 11, 90, 91 Self-location estimation system 20, 100, 900 Self-location estimation device 21 Imaging device 22 Extraction unit 23 Estimation unit 110, 190 Signal processing unit 111, 191 Image acquisition unit 112 Filtering unit 113, 193 Feature point tracking Units 114, 194 position / posture estimation units 115, 195 map generation units 116, 196 minimization units 117 learning units 200 cameras 210 recording media

Claims (10)

A self-position estimation method executed in a self-position estimation device equipped with an imaging device,
A method for determining a feature point image candidate, which is an image in which a feature point which is an area in the moving image used for estimating a self-position of the self-position estimating device from a moving image acquired from the imaging device, is performed by a predetermined method. Extract with
A self-position estimating method, comprising estimating the self-position using extracted feature point image candidates. By performing a learning process to obtain information on an image used for extracting a candidate for a feature point image,
The self-position estimating method according to claim 1, wherein feature point image candidates are extracted using the acquired image information. The self-position estimation method according to claim 2, wherein a learning process is performed based on an image used for estimating the self-position as a feature point image among the candidate feature point images. A learning process is performed based on an image other than the image used for estimating its own position as a feature point image among feature point image candidates,
The self-position estimation method according to claim 2, wherein a region indicated by information of the image acquired in the learning process is excluded from the moving image acquired from the imaging device, and a feature point image candidate is extracted. The self-position estimation method according to claim 1, wherein a candidate of a feature point image is extracted using a model of an area usable as a feature point. Storing a moving image acquired from the imaging device,
The self-position estimating method according to any one of claims 1 to 5, wherein a feature point image candidate is extracted from the stored moving image. A self-position estimation device equipped with an imaging device,
A method for determining a feature point image candidate, which is an image in which a feature point which is an area in the moving image used for estimating a self-position of the self-position estimating device from a moving image acquired from the imaging device, is performed by a predetermined method. An extraction unit for extracting with
An estimating unit for estimating the self-position using the extracted feature point image candidates. An acquisition unit that executes a learning process and acquires information of an image used for extraction of a candidate for a feature point image,
The self-position estimation device according to claim 7, wherein the extraction unit extracts a feature point image candidate using the acquired image information. A self-position estimation program executed in a computer equipped with an imaging device,
On the computer,
A predetermined method is used to extract a feature point image candidate, which is an image in which a feature point, which is an area in the moving image, used for estimating the self-position of the computer from the moving image acquired from the imaging device. A self-position estimation program for executing an extraction process and an estimation process for estimating the self-position using the extracted feature point image candidates. On the computer,
Performing a learning process to perform an acquisition process of acquiring information of an image used for extracting a feature point image candidate,
The self-position estimation program according to claim 9, wherein in the extraction processing, a feature point image candidate is extracted using information of the acquired image.

Images