Posture detection device and posture detection method

Abstract

To provide a posture detection device and a posture detection method capable of performing accurate posture detection.SOLUTION: A posture detection device comprises: a replacement unit that generates a third parameter by replacing a portion corresponding to a portion of a second parameter which indicates the posture of a moving body obtained by a second method different from a first method by a portion of a first parameter which indicates the posture of the moving body obtained by the first method; an evaluation unit that evaluates the reliability of the second parameter and the reliability of the third parameter; and a selection unit that selects the first parameter as the parameter indicating the posture of the moving body when the reliability of the third parameter is higher than the reliability of the second parameter, and selects the second parameter as the parameter indicating the posture of the moving body when the reliability of the third parameter is lower than the reliability of the second parameter.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present disclosure relates to a posture detection device and a posture detection method.

Background Art

[0002] As an existing automotive safety technology, there is a technology for recognizing stationary and moving objects using a monocular camera. A typical technology thereof is SFM (Structure from Motion). In applications used in ADAS (Advanced Driver-Assistance Systems), since processing is performed according to TTC (Time To Collision), calculation of an accurate distance between a detection target object and the host vehicle may be required.

[0003] Note that estimation of the posture of the host vehicle contributes to calculation of an accurate distance. Further, for posture estimation, there are a method of estimating from image information and a method of estimating using vehicle information from CAN (Controller Area Network). The former is called visual odometry, and the latter is called wheel odometry. In Patent Document 1, it is proposed to use these two methods in combination or alone.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, although visual odometry uses real image information and thus has high accuracy, it may calculate an incorrect posture suddenly. Further, wheel odometry does not suddenly generate an error, but has poor accuracy except in an ideal state. Further, Patent Document 1 does not mention how to select an existing method.

[0006] Note that there is an inlier rate as a method for measuring the reliability of the posture information. This indicates the ratio of the feature points that match the calculated posture information among all the feature points in the camera image, and can be considered as the reliability of the calculated posture information.

[0007] However, since the inlier rate of visual odometry calculated from the image information tends to be high compared to wheel odometry, it has been difficult to make a correct selection when choosing visual odometry and wheel odometry based on the inlier rate.

[0008] The non-limiting embodiments of the present disclosure contribute to providing a posture detection device and a posture detection method capable of performing accurate posture detection.

Means for Solving the Problem

[0009] For this purpose, one aspect of the posture detection device according to the present disclosure includes a replacement unit that generates a third parameter by replacing a part of the first parameter indicating the posture of the moving body obtained by the first method with a part corresponding to a part of the second parameter indicating the posture of the moving body obtained by a second method different from the first method, an evaluation unit that evaluates the reliability of the second parameter and the reliability of the third parameter, and when the reliability of the third parameter is higher than the reliability of the second parameter, selects the first parameter as the parameter indicating the posture of the moving body, and when the reliability of the third parameter is lower than the reliability of the second parameter, a selection unit that selects the second parameter as the parameter indicating the posture of the moving body.

[0010] In addition, one aspect of the posture detection method according to the present disclosure is to generate a third parameter by replacing a part corresponding to a part of the second parameter indicating the posture of the moving body obtained by a second method different from the first method, from among parts of the first parameter indicating the posture of the moving body obtained by the first method, evaluate the reliability of the second parameter and the reliability of the third parameter, and when the reliability of the third parameter is higher than the reliability of the second parameter, select the first parameter as the parameter indicating the posture of the moving body, and when the reliability of the third parameter is lower than the reliability of the second parameter, select the second parameter as the parameter indicating the posture of the moving body.

Advantages of the Invention

[0011] According to the present disclosure, accurate posture detection can be performed.

Brief Description of the Drawings

[0012]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Embodiments for Carrying Out the Invention

[0013] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. Note that each of the embodiments described below shows a specific example of the present disclosure. Therefore, each component shown in the following embodiments, the arrangement position and connection form of each component, and the order of each step and each step are examples and are not intended to limit the present disclosure. In addition, among the components in the following embodiments, the components not described in the independent claims are described as optional components.

[0014] Also, each drawing is a schematic diagram and is not necessarily drawn precisely. In each drawing, the same reference numerals are given to substantially the same configurations, and overlapping explanations are omitted or simplified.

[0015] Hereinafter, the case where the attitude detection device detects the attitude of the vehicle will be described, but the object to be detected for the attitude is not limited to the vehicle and may be other moving bodies.

[0016] In FIG. 1, SFM is a technique for restoring three-dimensional information from a two-dimensional camera image, and the stationary detection target object 1 is photographed by the camera 2 from a plurality of viewpoints. Then, based on how the feature points 3, which are parts of the detection target object 1, appear at different viewpoints, the attitude change of the camera 2 and the three-dimensional information of the detection target object 1 are restored.

[0017] For example, the detection target object 1 is a person, and the part that is the feature point 3 is the person's hand. The camera 2 photographs the feature point 3 of the person's right hand near the center of the camera image 4 at time t0. The camera 2 moves in the direction of the arrow shown in FIG. 1, and the attitude of the camera 2 changes. In the camera image 4 photographed by the camera 2 at time t1, the feature point 3 appears closer to the edge than the center.

[0018] Specifically, SFM performs (1) extraction and tracking of feature points as shown in FIG. 1, (2) estimation of camera pose, and (3) 3D reconstruction by triangulation. There are two types of camera pose estimation: visual odometry and wheel odometry, which will be described later.

[0019] In the 3D reconstruction image of FIG. 2, the person who is the above-mentioned detection target object 1 restored in 3D, the feature point 3, and many other feature points 3 are displayed. Also, in this 3D reconstruction image, the feature points of another person 11 standing facing the person who is the detection target object 1 and the feature points of the wall 12 behind those people are displayed. Note that in FIG. 2, the detection target object 1, another person 11, and the wall 12 are surrounded by dotted lines, but they are not included in the 3D reconstruction image.

[0020] In FIG. 3, epipolar geometry is a geometry that describes the relationship between multiple images when the same 3D object is photographed from different viewpoints with multiple cameras, or when the same 3D object is photographed from different viewpoints before and after movement with a single moving camera, and it is the core technology of SFM.

[0021] In the example of FIG. 3, camera 2 photographs the feature point P, which is a point in 3D space, at position C, and the feature point p corresponding to the feature point P appears in the camera image 21. Camera 2 moves from position C to position C', camera 2 photographs the feature point P at position C', and the feature point p' corresponding to the feature point P appears in the camera image 22.

[0022] In FIG. 3, the movement amount of camera 2 is represented by vector t, and the rotation amount is represented by vector R. In 3D space, since both the movement amount t and the rotation amount R have x, y, and z components, they can be expressed as t = (tx, ty, tz) and R = (Rx, Ry, Rz) (the same applies hereinafter).

[0023] As shown in FIG. 4, the component Rx of the rotation amount R represents the roll angle, the component Ry represents the pitch angle, and the component Rz represents the yaw angle.

[0024] The following equation 1 holds between the point p in the camera image 21 and the point p' in the camera image 22 shown in FIG. 3. p’ TEp = 0 ···(Equation 1)

[0025] The constraint condition represented by Equation 1 is called the epipolar constraint, and the matrix E is called the fundamental matrix in the equation representing the epipolar constraint. The matrix E is a matrix related to the translation amount t and the rotation amount R of Camera 2, and the camera pose rotation amount R and translation amount t are calculated from the matrix E.

[0026] The method of calculating R and t from the image information in this way is called visual odometry. Selecting five points from all the feature points in the camera image and solving Equation 1 is the existing technology.

[0027] When Camera 2 is fixed to the vehicle, the translation amount t and the rotation amount R of Camera 2 are equal to the translation amount and the rotation amount of the vehicle. Therefore, the translation amount and the rotation amount of the vehicle are calculated from the translation amount t and the rotation amount R of Camera 2.

[0028] Since visual odometry calculates the pose of the vehicle from image information, it has the advantage of being robust to the slope of the road surface, steering, and acceleration / deceleration.

[0029] On the other hand, in visual odometry, since five points are selected from the camera image and Equation 1 is solved, the performance may deteriorate depending on the selection pattern of the feature points. For example, when the tracking is incorrect due to noise or the like among the selected five points, or when the feature points include non-stationary objects. Also, even when the translation amount t and the rotation amount R deviate greatly from the true values, the inlier rate may be high.

[0030] In contrast, wheel odometry estimates the pose of the host vehicle from vehicle information. As vehicle information, the steering angle, wheel speed, height sensor, etc. are used.

[0031] Wheel odometry has the advantage of exhibiting high performance during constant-speed straight-line driving. On the other hand, wheel odometry may deteriorate in performance with respect to the slope of the road surface, steering, and acceleration / deceleration.

[0032] For example, in the case of an existing method that selects either visual odometry or wheel odometry, based on the vehicle's movement amount t and rotation amount R obtained by each of the visual odometry and wheel odometry methods, Equation 1 is calculated for all feature points in the camera image, and the odometry with a higher ratio of feature points satisfying Equation 1 may be selected. The ratio of the number of feature points satisfying Equation 1 to the total number of feature points in the camera image is called the inlier rate.

[0033] In visual odometry, even if the errors in the movement amount t and rotation amount R are large, the inlier rate is higher than that of the wheel. The reason is that in visual odometry, the movement amount t and rotation amount R that satisfy Equation 1 are obtained, but even if some of the parameters of the movement amount t and rotation amount R are incorrect, the other parameters are generated to satisfy Equation 1, so Equation 1, which is also used for calculating the inlier rate, is satisfied.

[0034] Also, in visual odometry, Equation 1 is solved from five feature points to calculate the movement amount t and rotation amount R. However, in practice, since the movement amount t and rotation amount R are calculated within the given processing time, a plurality of movement amounts t and rotation amounts R are generated, and the movement amount t and rotation amount R with a higher inlier rate are selected from them. Therefore, the selected movement amount t and rotation amount R are in a state with a relatively high inlier rate to some extent.

[0035] Thus, in the method according to the existing technology that selects either visual odometry or wheel odometry, there may be a case where visual odometry calculates a suddenly incorrect posture, the inlier rate in visual odometry tends to be higher than that in wheel odometry, and it may be difficult to correctly select odometry simply by calculating the inlier rate.

[0036] For this purpose, the posture detection device of the present disclosure performs the following processing. First, when the inlier rate of visual odometry is higher than the inlier rate of wheel odometry and visual odometry is selected in the existing method, the movement amount t and a part of the rotation amount R indicating the posture of the wheel odometry are replaced with a part of the movement amount t and the rotation amount R of the visual odometry.

[0037] For example, when the movement amount tv obtained by visual odometry is tv = (tvx, tvy, tvz), the rotation amount Rv is Rv = (Rvx, Rvy, Rvz), the movement amount tw obtained by wheel odometry is tw = (twx, twy, twz), the rotation amount Rw is Rw = (Rwx, Rwy, Rwz), and the y component Rwy of the rotation amount Rw obtained by wheel odometry is replaced with the y component Rvy of the rotation amount Rv obtained by visual odometry, the movement amount tw_r of the wheel odometry after replacement is tw_r = (twx, twy, twz), and the rotation amount Rw_r is Rw_r = (Rwx, Rvy, Rwz).

[0038] Note that replacing the y component Rwy of the rotation amount Rw with the y component Rvy of the rotation amount Rv is just an example. The x component Rwx of the rotation amount Rw may be replaced with the x component Rvx of the rotation amount Rv, or the z component Rwz of the rotation amount Rw may be replaced with the z component Rvz of the rotation amount Rv.

[0039] Also, the x component twx of the movement amount tw may be replaced with the x component tvx of the movement amount tv, the y component twy of the movement amount tw may be replaced with the y component tvy of the movement amount tv, or the z component twz of the movement amount tw may be replaced with the z component tvz of the movement amount tv.

[0040] Also, one or more components to be replaced in the movement amount tw and the rotation amount Rw may be provided.

[0041] Note that in the posture detection of the vehicle, since the accuracy of the z component of the rotation amount, for example, the yaw angle, contributes to the accuracy of the vehicle posture detection, it is desirable to replace at least the z component Rwz of the rotation amount Rw with the z component Rvz of the rotation amount Rv.

[0042] The pose detection device calculates the inlier rate using the pre-replacement movement amount tw and rotation amount Rw and the post-replacement movement amount tw_r and rotation amount Rw_r, respectively. When the inlier rate after replacement improves, the movement amount tv and rotation amount Rv obtained by visual odometry are selected as the final vehicle pose. When the inlier rate does not improve, the movement amount tw and rotation amount Rw obtained by wheel odometry are selected as the final vehicle pose. Thereby, the pose detection device can make a more accurate vehicle pose determination.

[0043] In FIG. 5, the pose detection device 30 includes a first reliability evaluation unit 34, a parameter replacement unit 35, a second reliability evaluation unit 36, a selection unit 37, and a three-dimensional restoration unit 38.

[0044] The first reliability evaluation unit 34 acquires the information of the first vehicle pose 31, the information of the second vehicle pose 32, and the camera image 33 including the information of the feature points. The information of the first vehicle pose 31 includes the information of the movement amount tv and rotation amount Rv of the vehicle obtained by visual odometry. The information of the second vehicle pose 32 includes the information of the movement amount tw and rotation amount Rw of the vehicle obtained by wheel odometry.

[0045] Then, the first reliability evaluation unit 34 uses the information of the first vehicle pose 31 to determine whether the movement amount tv and rotation amount Rv of the vehicle and each feature point included in the camera image 33 satisfy the epipolar constraint represented by Equation 1.

[0046] Furthermore, the first reliability evaluation unit 34 calculates the inlier rate, which is the ratio of the feature points that satisfy the epipolar constraint among all the feature points included in the camera image 33, and evaluates the reliability of the movement amount tv and rotation amount Rv of the vehicle.

[0047] Similarly, the first reliability evaluation unit 34 uses the information of the second vehicle pose 32 to determine whether the movement amount tw and rotation amount Rw of the vehicle and each feature point included in the camera image 33 satisfy the epipolar constraint represented by Equation 1.

[0048] Furthermore, the first reliability evaluation unit 34 calculates an inlier rate, which is the ratio of feature points that satisfy the epipolar constraint among all the feature points included in the camera image 33, and evaluates the reliability of the vehicle movement amount tw and the rotation amount Rw.

[0049] When it is determined that the inlier rate of the visual odometry is higher than the inlier rate of the wheel odometry, the parameter replacement unit 35 replaces a part of the movement amount tv and the rotation amount Rv, which are the parameters included in the first vehicle posture 31 obtained by the visual odometry, with a part corresponding to the above part of the parameters included in the second vehicle posture 32 obtained by the wheel odometry, and generates the movement amount tw_r and the rotation amount Rw_r of the wheel odometry, which are the parameters after replacement.

[0050] The second reliability evaluation unit 36 acquires the camera image 33 including the feature point information and the information on the movement amount tw_r and the rotation amount Rw_r in which the above part is replaced.

[0051] Then, the second reliability evaluation unit 36 determines whether the vehicle movement amount tw_r and rotation amount Rw_r and each feature point included in the camera image 33 satisfy the epipolar constraint represented by Equation 1.

[0052] Furthermore, the second reliability evaluation unit 36 calculates an inlier rate, which is the ratio of feature points that satisfy the epipolar constraint among all the feature points included in the camera image 33, and evaluates the reliability of the vehicle movement amount tw_r and the rotation amount Rw_r.

[0053] When the reliability of the movement amount tv and the rotation amount Rv in the visual odometry is lower than the reliability of the movement amount tw and the rotation amount Rw in the wheel odometry, the selection unit 37 selects the movement amount tw and the rotation amount Rw as the parameters indicating the vehicle posture.

[0054] Further, when the reliability of the movement amount tv and the rotation amount Rv in visual odometry is higher than the reliability of the movement amount tw and the rotation amount Rw in wheel odometry, the selection unit 37 determines whether the reliability of the movement amount tw_r and the rotation amount Rw_r after replacing a part of the parameters is higher than the reliability of the movement amount tw and the rotation amount Rw.

[0055] Then, when the reliability of the movement amount tw_r and the rotation amount Rw_r is higher than the reliability of the movement amount tw and the rotation amount Rw, the selection unit 37 selects the movement amount tv and the rotation amount Rv as the parameters indicating the vehicle posture. When the reliability of the movement amount tw_r and the rotation amount Rw_r is lower than the reliability of the movement amount tw and the rotation amount Rw, the selection unit 37 selects the movement amount tw and the rotation amount Rw as the parameters indicating the vehicle posture.

[0056] The three-dimensional restoration unit 38 uses the information on the movement amount and the rotation amount of the vehicle selected by the selection unit 37 to restore the three-dimensional arrangement of the objects around the vehicle based on the principle of triangulation.

[0057] In FIG. 6, first, the first reliability evaluation unit 34 calculates the inlier rate Inl1 for the first vehicle posture 31 using the information on the first vehicle posture 31 obtained from visual odometry and the camera image 33 from which feature points are extracted (step S1).

[0058] Also, the first reliability evaluation unit 34 calculates the inlier rate Inl2 for the second vehicle posture 32 using the information on the second vehicle posture 32 obtained from wheel odometry and the camera image 33 from which feature points are extracted (step S2).

[0059] Next, the selection unit 37 determines whether the inlier rate Inl1 is greater than the inlier rate Inl2 (step S3).

[0060] When the inlier rate Inl1 is not greater than the inlier rate Inl2 (step S3, NO), the selection unit 37 selects, as the final attitude of the vehicle, a second vehicle attitude 32 represented by a movement amount tw and a rotation amount Rw (step S7).

[0061] When the inlier rate Inl1 is greater than the inlier rate Inl2 (step S3, YES), the parameter replacement unit 35 replaces a part of the movement amount tv and the rotation amount Rv, which are parameters included in the first vehicle attitude 31 obtained by visual odometry, with a part corresponding to the above part of the parameters included in the second vehicle attitude 32 obtained by wheel odometry, to generate a wheel odometry movement amount tw_r and a rotation amount Rw_r.

[0062] Then, the second reliability evaluation unit 36 calculates an inlier rate Inl2_r after replacing the above part of the parameters (step S4).

[0063] Thereafter, the selection unit 37 determines whether the inlier rate Inl2_r is greater than the inlier rate Inl2 (step S5).

[0064] When the inlier rate Inl2_r is greater than the inlier rate Inl2 (step S5, YES), the selection unit 37 selects, as the final attitude of the vehicle, a first vehicle attitude 31 represented by a movement amount tv and a rotation amount Rv (step S6). When the inlier rate Inl2_r is not greater than the inlier rate Inl2 (step S5, NO), the selection unit 37 selects, as the final attitude of the vehicle, a second vehicle attitude 32 represented by a movement amount tw and a rotation amount Rw (step S7).

[0065] In FIG. 7, the host vehicle 41 steers and travels on an inclined road surface, and moves into a parking space 42 while accelerating and decelerating. In this case, the attitude detection device 30 restores the three-dimensional shapes of the wall 43 and the other vehicle 44 with feature points.

[0066] In Fig. 8, in the image 55 of Fig. 8(a), the feature points 51 obtained by visual odometry, and the wall 43’ and the vehicle 44’ whose three-dimensional shape has been restored by the feature points 51 are shown as viewed from directly above (with the XY plane from the positive direction of the Z axis).

[0067] In the image 55, since the number of feature points 51 is large and the dispersion of the feature points 51 is small, the wall 43 and the vehicle 44 are accurately restored like the wall 43’ and the vehicle 44’.

[0068] The image 56 of Fig. 8(b) shows the feature points 52 obtained by wheel odometry and the wall 43’ whose shape has been restored by the feature points 52 as viewed from directly above (with the XY plane from the positive direction of the Z axis).

[0069] In the image 56, since the number of feature points 52 is small and the dispersion of the feature points 52 is large, the positional accuracy of the wall 43’ is poor and the restoration of the three-dimensional shape of the vehicle 44 has failed.

[0070] As described above, the posture detection device 30 in the present embodiment replaces a part of the movement amount t and the rotation amount R indicating the posture obtained by wheel odometry with a part of the movement amount t and the rotation amount R obtained by visual odometry, and calculates the inlier rate using the movement amount tw and the rotation amount Rw before replacement and the movement amount tw_r and the rotation amount Rw_r after replacement, respectively.

[0071] Then, when the inlier rate corresponding to the movement amount tw_r and the rotation amount Rw_r is larger than the inlier rate corresponding to the movement amount tw and the rotation amount Rw, the posture detection device 30 selects the movement amount tv and the rotation amount Rv as the parameters indicating the posture of the vehicle.

[0072] In the image 57 of Fig. 8(c), the feature points 53 obtained using the movement amount tw_r and the rotation amount Rw_r after replacement, and the wall 43’ and the vehicle 44’ whose three-dimensional shape has been restored by the feature points 53 are shown as viewed from directly above (with the XY plane from the positive direction of the Z axis).

[0073] Compared with the image 56 before replacement, in the image 57 after replacement, the positional accuracy of the wall 43' has been improved, and the three-dimensional shape of the vehicle 44 has been successfully restored.

[0074] In such a case, since the inlier rate corresponding to the movement amount tw_r and the rotation amount Rw_r is larger than the inlier rate corresponding to the movement amount tw and the rotation amount Rw, the attitude detection device 30 selects the movement amount tv and the rotation amount Rv obtained by visual odometry as the parameters indicating the final attitude of the vehicle.

[0075] In FIG. 9, the host vehicle 41 is steered and travels on an inclined road surface, and moves into the parking space 42 while accelerating and decelerating. In this case, the attitude detection device 30 wants to three-dimensionally restore the feature points between the wall 43 and the vehicle 44.

[0076] In FIG. 10, in the image 65 of FIG. 10(a), the feature points 61 obtained by visual odometry and the wall 43' and the vehicle 44' whose three-dimensional shape has been restored by the feature points 61 are shown as viewed from directly above (with the XY plane from the positive direction of the Z axis). In the image 65, the positional accuracy of the wall 43' and the vehicle 44' is poor.

[0077] In the image 66 of FIG. 10(b), the feature points 62 obtained by wheel odometry and the wall 43' and the vehicle 44' whose shape has been restored by the feature points 62 are shown as viewed from directly above (with the XY plane from the positive direction of the Z axis). In the image 66, the three-dimensional shapes of both the wall 43' and the vehicle 44' have been restored.

[0078] In the image 67 of FIG. 10(c), similar to the image 57 in FIG. 8, the feature points 63 obtained using the movement amount tw_r and the rotation amount Rw_r after replacing a part of the movement amount t and the rotation amount R indicating the attitude obtained by wheel odometry, and the wall 43' and the vehicle 44' whose three-dimensional shape has been restored by the feature points 63 are shown as viewed from directly above (with the XY plane from the positive direction of the Z axis).

[0079] Compared with the image 66 before replacement, in the image 67 after replacement, the positional accuracy of the wall 43' and the vehicle 44' has decreased.

[0080] In such a case, since the inlier rate corresponding to the movement amount tw_r and the rotation amount Rw_r is smaller than the inlier rate corresponding to the movement amount tw and the rotation amount Rw, the attitude detection device 30 selects the movement amount tw and the rotation amount Rw obtained by wheel odometry as parameters indicating the final attitude of the vehicle.

[0081] The example of FIG. 10, like the example of FIG. 8, has no significant difference between visual odometry and wheel odometry. In wheel odometry, the same kind of error occurs for all feature points, while in visual odometry, the magnitude of the error differs for each feature point.

[0082] When the detected object includes feature points with a large error, since the error in the position of the detected object increases, wheel odometry with a smaller position error is used.

[0083] As described above, the embodiments have been explained, but the present disclosure is not limited to the above embodiments.

[0084] For example, in the above embodiment, the attitude detection device replaces a part of the first parameter indicating the attitude of the vehicle obtained by visual odometry with a part corresponding to the above part of the second parameter indicating the attitude of the vehicle obtained by wheel odometry to generate a third parameter, and evaluates the reliability of the second parameter and the reliability of the third parameter.

[0085] Then, when the reliability of the third parameter is higher than the reliability of the second parameter, the attitude detection device selects the first parameter as the parameter indicating the attitude of the vehicle, and when the reliability of the third parameter is lower than the reliability of the second parameter, the attitude detection device selects the second parameter as the parameter indicating the attitude of the vehicle.

[0086] However, the method for obtaining the first parameter is not limited to visual odometry, and the method for obtaining the second parameter is not limited to wheel odometry. Any method different from the method for obtaining the first parameter and the method for obtaining the second parameter may be used.

[0087] In the above embodiment, the reliability of the movement amount and the rotation amount of the vehicle is evaluated based on the inlier rate, which is the ratio of the feature points satisfying the epipolar constraint among all the feature points included in the camera image. However, the criterion used for evaluating the reliability is not limited to the inlier rate.

[0088] For example, it may be the number of feature points satisfying the epipolar constraint among all the feature points included in the camera image, or other criteria may be used.

[0089] The attitude detection device may be a device mounted on the vehicle or a device installed outside the vehicle.

[0090] In the above embodiment, each component may be realized by executing a software program suitable for each component. Each component may be realized by a program execution unit such as a CPU or a processor reading and executing a software program recorded on a recording medium such as a hard disk or a semiconductor memory.

[0091] Furthermore, the general or specific aspects of the present disclosure may be realized by an apparatus, a method, an integrated circuit, a computer program, or a recording medium such as a computer-readable CD-ROM. Also, they may be realized by any combination of an apparatus, a method, an integrated circuit, a computer program, and a recording medium.

[0092] In addition, forms obtained by applying various modifications conceivable by those skilled in the art to each embodiment, or forms realized by arbitrarily combining the components and functions in each embodiment without departing from the spirit of the present disclosure are also included in the present disclosure.

Industrial Applicability

[0093] The present disclosure can be used in a posture detection device and a posture detection method.

Description of Signs

[0094] 1 Object to be detected 2 Camera 3 Feature point 4 Camera image 31 First vehicle posture 32 Second vehicle posture 33 Input image 34 Inlier rate calculation unit 35 Parameter replacement unit 36 Reliability evaluation unit 37 Selection unit 38 3D restoration unit 41 Own vehicle 42 Parking space 43 Wall 44 Vehicle 51 Feature point 55 Image

Claims

1. A replacement unit that generates a third parameter by replacing a part of a first parameter indicating the attitude of a moving body obtained by a first method with a part corresponding to a second parameter indicating the attitude of the moving body obtained by a second method different from the first method; An evaluation unit that evaluates the reliability of the second parameter and the reliability of the third parameter; A selection unit that selects the first parameter as the parameter indicating the attitude of the moving body when the reliability of the third parameter is higher than the reliability of the second parameter, and selects the second parameter as the parameter indicating the attitude of the moving body when the reliability of the third parameter is lower than the reliability of the second parameter; An attitude detection device comprising the above.

2. The first method is visual odometry, and the second method is wheel odometry. The attitude detection device according to Claim 1.

3. Each of the first parameter and the second parameter includes a parameter indicating at least one of the movement amount of the moving body and the rotation amount of the moving body. The attitude detection device according to Claim 1.

4. Each of the first parameter and the second parameter includes at least the parameters of the roll angle, pitch angle, and yaw angle of the moving body, and a part of the first parameter includes the yaw angle. The attitude detection device according to Claim 1.

5. The evaluation unit evaluates the reliability of the second parameter and the reliability of the third parameter based on the number or ratio of feature points that satisfy the epipolar constraint among a plurality of feature points included in an image captured by an imaging device mounted on the moving body. The attitude detection device according to Claim 1.

6. The selection unit selects the second parameter as the parameter indicating the attitude of the moving body when the reliability of the first parameter is lower than the reliability of the second parameter, and the replacement unit generates the third parameter when the reliability of the first parameter is higher than the reliability of the second parameter. The attitude detection device according to Claim 1.

7. Generate a third parameter by replacing a part of a first parameter indicating the attitude of a moving body obtained by a first method with a part corresponding to a second parameter indicating the attitude of the moving body obtained by a second method different from the first method. Evaluate the reliability of the second parameter and the reliability of the third parameter, When the reliability of the third parameter is higher than the reliability of the second parameter, select the first parameter as the parameter indicating the attitude of the moving body. When the reliability of the third parameter is lower than the reliability of the second parameter, select the second parameter as the parameter indicating the attitude of the moving body. Attitude detection method.

8. The first method is visual odometry, and the second method is wheel odometry. The attitude detection method according to claim 7.

9. The first parameter and the second parameter each include a parameter indicating at least one of the movement amount of the moving body and the rotation amount of the moving body. The attitude detection method according to claim 7.

10. The first parameter and the second parameter each include at least the parameters of the roll angle, pitch angle, and yaw angle of the moving body, and a part of the first parameter includes the yaw angle. The attitude detection method according to claim 7.

11. Furthermore, based on the number or ratio of feature points satisfying the epipolar constraint among a plurality of feature points included in the image captured by the imaging device mounted on the moving body, evaluate the reliability of the second parameter and the reliability of the third parameter. The attitude detection method according to claim 7.

12. Furthermore, when the reliability of the first parameter is lower than the reliability of the second parameter, select the second parameter as the parameter indicating the attitude of the moving body. When the reliability of the first parameter is higher than the reliability of the second parameter, generate the third parameter. The attitude detection method according to claim 7.

Images