Information processing method and correction method

By using a correction coefficient to align the self-position of a mobile robot with a reference point through visual odometry and external sensors, the method addresses installation errors and enhances self-position estimation accuracy outdoors.

JP2024166806APending Publication Date: 2024-11-29NIPPON TELEGRAPH & TELEPHONE CORP +1
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
JP2023083157
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-19
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

Existing methods for determining the self-position of mobile robots outdoors, such as using cameras installed on cranes or WiFi radio waves, are impractical, costly, or prone to errors due to electromagnetic interference, while visual odometry methods suffer from installation errors that require time-consuming and precise alignment.

Method used

A method involving a visual odometry unit and external sensors to calculate a correction coefficient that aligns the estimated self-position of a mobile robot with a reference point, using a sensor unit to determine the movement trajectory of the robot and an external sensor to determine the reference point, minimizing the distance between their data to correct the self-position.

Benefits of technology

This approach allows for highly accurate self-position estimation by correcting installation errors, enabling precise alignment of the camera's mounting position and reducing the need for costly and time-consuming alignment processes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024166806000001_ABST
    Figure 2024166806000001_ABST
Patent Text Reader

Abstract

To enable high-precision self-position estimation by grasping an attachment position of a camera more accurately.SOLUTION: A correction coefficient for correcting a self-position estimated by a mobile robot 1, which can move two-dimensionally, is obtained. The mobile robot 1 includes a VO part 20 for estimating the self-position using the visual odometry method. Movement locus data of the mobile robot 1, which the VO part 20 has obtained by the visual odometry method, and movement locus data of a reference point 31 of the mobile robot 1 obtained by an external sensor are obtained. A correction coefficient which minimizes a total value of a distance between the movement locus data obtained by the visual odometry method and the movement locus data obtained by the external sensor is obtained.SELECTED DRAWING: Figure 6
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present disclosure relates to an information processing method and a correction method. [Background technology]

[0002] Inspection of infrastructure structures installed outdoors uses environmental measurement devices using a mobile cart operated by a human operator or mobile robots. For example, inspection of underground buried objects uses a device equipped with a ground penetrating radar and capable of moving in two dimensions. In order to accurately grasp the location of underground buried objects in underground buried object exploration, it is necessary to determine the relative position between the location of the underground buried object and the reference position of the robot with high accuracy.

[0003] From the viewpoint of improving the efficiency of the exploration work, it is desirable for the underground radar to be able to move freely in all directions. Also, there are environments where the inspection location is narrow and it is not possible to make a large turn on the spot. In such cases, a robot with a holonomic omnidirectional mobile mechanism that can move in any direction, as described in Non-Patent Document 1, can be used.

[0004] In a mobile measurement device using an omnidirectional movement mechanism, measurement data is analyzed and managed by linking it to the position information of the measurement point. Therefore, the accuracy of the position information is closely related to the quality of the measurement data, so it is necessary to improve the position accuracy on a two-dimensional plane. Regarding the position, it is sufficient to be able to estimate the amount of relative movement from the movement origin with high accuracy.

[0005] For indoor use, Non-Patent Document 2 proposes a method for calculating the position of a robot moving indoors using images from a camera installed near the ceiling. Non-Patent Document 3 proposes a method for calculating the position of a mobile robot by estimating the distance from a transmitter using WiFi radio waves. Non-Patent Document 4 proposes a method for estimating the self-position using a ground image. [Prior art documents] [Non-patent literature]

[0006] [Non-Patent Document 1] K. Tadakuma, R. Tadakuma and J. Berengeres, “Development of Holonomic Omnidirectional Vehicle with 'Omini-Ball':Spherical Wheels,” Proceedings of the 2007 IEEE / RSJ International Conference on Intelligent Robots and Systems, San Diego, CA, USA 2007 [Non-Patent Document 2] Jae Hong Shim and Young Im Cho, “A Mobile Robot Localization via Indoor Fixed Remote Surveillance Cameras,” Sensors. 16. 195. 2016. [Non-Patent Document 3] Joydeep Biswas and Manuela Veloso,“WiFi localization and navigation for autonomous indoor mobile robots,”Proc. of 2010 IEEE International Conference on Robotics and Automation [Non-Patent Document 4] Linguang Zhang, Adam Finkelstein and Szymon Rusinkiewicz, “High-Precision Localization Using Ground Texture,” ICRA2019 [Non-Patent Document 5] T. Kosakai, Y. Kataoka, Y. Ebina, M. Mizukami, and S. Mochizuki, “Two-dimensional self-localization using sensor-fusion of mobile robot for inspecting underground infrastructure facilities,” Proc. of ICPE2022 (2022) Summary of the Invention [Problem to be solved by the invention]

[0007] When targeting outdoor mobile measurements, the method of Non-Patent Document 2 requires a crane or similar device to be used to install a camera above the vehicle, which is not practical in terms of high costs and cumbersome work. The method of Non-Patent Document 3, which uses WiFi radio waves, is concerned that errors may increase if electromagnetic obstructions are present outdoors. Non-Patent Document 4 requires that ground images be acquired in advance, and in the case of a larger-scale system, the map database becomes larger in capacity, which results in longer search times.

[0008] A known method for estimating the robot's self-position is the visual odometry method, which uses a camera to capture video of the road surface while the robot is moving, and calculates the amount of movement of the mobile robot by measuring the amount of image displacement between frames in the captured video. The visual odometry method is considered to be effective in terms of accuracy, since in principle it can obtain self-position estimation accuracy at the image resolution.

[0009] However, it is difficult to perfectly match the mounting position of the camera mounted on a mobile robot with the reference position set on the robot. Even if the self-position can be estimated with high accuracy by the visual odometry method, an error will always remain between the self-position estimated by the visual odometry method and the reference position of the robot due to mounting error. To precisely match the reference position of the robot with the mounting position of the camera, it is necessary to take measurements and use a highly accurate laser displacement meter.

[0010] The present disclosure has been made in consideration of the above, and aims to more accurately grasp the mounting position of a camera and enable highly accurate self-position estimation. [Means for solving the problem]

[0011] An information processing method of one embodiment of the present disclosure is an information processing method for determining a correction coefficient for correcting a self-position estimated by a moving body capable of moving two-dimensionally, wherein the moving body is equipped with a sensor unit for estimating its self-position using a visual odometry method, and a computer obtains first movement trajectory data indicating a movement trajectory of the moving body obtained by the sensor unit and second movement trajectory data indicating a movement trajectory of a reference point of the moving body obtained by an external sensor, and determines a correction coefficient that minimizes the sum of the distances between the first movement trajectory data and the second movement trajectory data.

[0012] A correction method of one embodiment of the present disclosure is a correction method for correcting a self-position estimated by a moving body capable of moving two-dimensionally, in which a sensor unit estimates the self-position, and a computer corrects the self-position using a correction coefficient that includes a relative positional relationship between the sensor unit and a reference point of the moving body and a relationship between the orientation of the sensor unit and the orientation of the moving body. Effect of the Invention

[0013] According to the present disclosure, it is possible to grasp the mounting position of the camera more accurately, enabling highly accurate self-position estimation. [Brief description of the drawings]

[0014] [Figure 1] FIG. 1 is a side view showing an example of the configuration of a mobile robot. [Diagram 2] FIG. 2 is a front view showing an example of the configuration of a mobile robot. [Diagram 3] FIG. 3 is a top view showing an example of the configuration of a mobile robot. [Figure 4] FIG. 4 is a flow chart showing an example of a process for obtaining movement trajectory data of a mobile robot. [Diagram 5] FIG. 5 is a diagram showing an example of the coordinate system of the mobile robot and the coordinate system of the camera. [Figure 6] FIG. 6 is a flowchart showing an example of the flow of a process for deriving a correction coefficient. [Figure 7] FIG. 7 is a flowchart showing an example of a process flow for deriving correction coefficients by the simulated annealing method. [Figure 8] FIG. 8 is a diagram illustrating an example of movement trajectory data of an external sensor. [Figure 9] FIG. 9 is a diagram illustrating an example of movement trajectory data of an internal sensor. [Figure 10] FIG. 10 is a flowchart illustrating an example of a process for correcting movement trajectory data. [Figure 11] FIG. 11 is a diagram showing an example of a movement trajectory of a mobile robot. [Figure 12] FIG. 12 is a diagram showing an example of movement trajectory data before correction. [Figure 13] FIG. 13 is a diagram showing an example of the movement trajectory data after correction. [Figure 14] FIG. 14 is a diagram showing an example of a trajectory of a mobile robot. [Figure 15] FIG. 15 is a diagram showing an example of a movement trajectory that combines a linear movement trajectory and a turning movement trajectory. [Figure 16] FIG. 16 is a top view showing an example of the configuration of a mobile robot on which two markers are installed. [Figure 17] FIG. 17 is a diagram showing the positional relationship between the two markers and the camera. [Figure 18] FIG. 18 is a diagram showing an example of a trajectory acquired by an external sensor and a trajectory acquired by an internal sensor. [Figure 19] FIG. 18 is a diagram showing an example of a trajectory acquired by an external sensor and a trajectory acquired by an internal sensor. [Figure 20] FIG. 20 is a diagram illustrating an example of a hardware configuration of an apparatus that executes the processing of this embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] [Mobile robot configuration] An example of the configuration of a mobile robot 1 for which a correction coefficient is to be calculated will be described with reference to Figures 1 to 3. Figure 1 is a side view of the mobile robot 1, Figure 2 is a front view of the mobile robot 1, and Figure 3 is a top view of the mobile robot 1.

[0016] The mobile robot 1 is equipped with wheels 12 on a housing 11. The mobile robot 1 can move on a two-dimensional plane (the XY plane in the figure) by using the wheels 12. It is preferable to use omni-wheels or Mecanum wheels for the wheels 12 to create a holonomic omnidirectional mobile mechanism that can move in any direction on the two-dimensional plane.

[0017] The mobile robot 1 includes a visual odometry (VO) unit 20 in the housing 11. The VO unit 20 is disposed in the lower part of the housing 11 and captures a road surface video, and analyzes the road surface video to obtain movement trajectory data of the mobile robot 1. The VO unit 20 includes a camera 21 that captures the road surface video, a distance sensor that measures the shooting distance from the road surface, a box that blocks external light, lighting that illuminates the road surface with a stable amount of light, and a control unit that controls each unit to obtain movement trajectory data. The VO unit 20 is used as a sensor unit for obtaining the self-position (movement trajectory data) of the mobile robot 1.

[0018] The VO unit 20 captures a video of the road surface with the camera 21 and records the shooting distance with a distance sensor. The VO unit 20 derives the physical length per pixel from the relationship between the road surface video and the shooting distance, and derives the number of pixels moving between image frames of the road surface video by image processing. The VO unit 20 calculates the amount of movement of the mobile robot 1 based on the physical length per pixel and the number of pixels moving between image frames, and estimates the self-position of the mobile robot 1.

[0019] The VO unit 20 corrects the estimated self-position to the position of the reference point of the mobile robot 1 using a correction coefficient indicating the relative positional relationship between the camera 21 and the reference point of the mobile robot 1. The correction coefficient is calculated in advance based on the movement trajectory data obtained by the VO unit 20 and the movement trajectory data of the reference point of the mobile robot 1 obtained by the external sensor. The correction coefficient includes not only the positional relationship between the camera 21 and the reference point on the XY plane, but also the relationship between the orientation of the mobile robot 1 and the orientation of the camera 21. The derivation of the correction coefficient will be described later.

[0020] When deriving the correction coefficient, the mobile robot 1 is provided with a marker 30 on the upper surface of the housing 11. The marker 30 is a mark used by an external sensor to estimate the position and orientation of the mobile robot 1. For example, a point within the marker 30 is set as the reference point 31 of the mobile robot 1. In this embodiment, using the technology of Non-Patent Document 5, a video is taken with an overhead camera that covers the mobile robot 1 and the moving range of the mobile robot 1, the marker 30 is detected from the video, and the amount of movement of the marker 30 is calculated with high accuracy to obtain time-series movement trajectory data.

[0021] In this embodiment, an overhead camera that captures video and a device that detects the markers 30 from the video to obtain movement trajectory data are used as the external sensor. The external sensor is not limited to this, and any sensor that can obtain movement trajectory data of the mobile robot 1 from outside the mobile robot 1 may be used.

[0022] The marker 30 and the external sensor are components used to calculate the correction coefficient, and are no longer necessary after the correction coefficient is calculated.

[0023] [Operation] Next, an example of a process for obtaining movement trajectory data of the mobile robot 1 will be described with reference to the flowchart of FIG.

[0024] In step S1, a correction coefficient is calculated from the movement trajectory data of the mobile robot 1 measured by the VO unit 20 and the movement trajectory data of the mobile robot 1 measured by an external sensor, to minimize an evaluation function using the distance between the movement trajectory data.

[0025] In step S2, the mobile robot 1 is moved outdoors (or indoors), and the movement trajectory data measured by the VO unit 20 is corrected with the correction coefficient to obtain movement trajectory data of the reference point 31 of the mobile robot 1.

[0026] The derivation of the correction coefficient in step S1 is performed only once at the time of mounting on the device in order to calibrate the influence of the mounting error of the device. After the correction coefficient is derived in step S1, the marker 30 of the mobile robot 1 may be removed.

[0027] [Derivation of correction coefficient] 5 shows the coordinate system of the mobile robot 1 and the coordinate system of the camera 21. The coordinate system of the mobile robot 1 is an orthogonal coordinate system of a rectangular area set by an external sensor (the XY coordinate system in the figure), and the coordinate system of the camera 21 is an orthogonal coordinate system (the PQ coordinate system in the figure) with the camera 21 as the origin and the orientation of the camera 21 as one of the axes. Since the camera 21 and the marker 30 are fixed to the mobile robot 1, the relative positional relationship (amount of deviation) between the camera 21 and the reference point 31 does not change even if the mobile robot 1 moves. The amount of deviation D (d x , d y , d θ ) is known, the movement trajectory data obtained in the VO unit 20 can be corrected to obtain the movement trajectory data of the reference point 31.

[0028] When the amount of deviation D is known, the position coordinates (v' xk ,v' yk ,v' θk ) can be obtained by the following procedure. Using equation (1), the position coordinates (v pk ,v qk ,v φk ) to the initial position coordinates of the reference point 31 (a x0 ,a y0 ) and translate it based on the initial angle a θ0 The camera 21 is rotated based on the equation (2), taking into account the amount of deviation D, and the position coordinates (internal sensor coordinate system) of the camera 21 are aligned with the position coordinates (external sensor coordinate system) of the reference point 31. The orientation v of the camera 21 is φkand the direction of mobile robot 1 a θk and the difference is constant, and equation (3) holds.

[0029]

number

[0030] When formula (2) is used as the correction formula, the direction a of the mobile robot 1 is θk The deviation of the direction d θ and the orientation of camera 21 v φk As a result, the position coordinates of the camera 21 obtained by the visual odometry method (v pk ,v qk ,v φk ) and the amount of deviation D(d x ,d y ,d θ ) the position coordinates of the camera 21 can be corrected to the position coordinates of the reference point 31.

[0031] Next, an example of a process for deriving a correction coefficient (amount of deviation) will be described with reference to the flowchart in FIG.

[0032] In step S11, while moving the mobile robot 1, movement trajectory data of the mobile robot 1 by the VO unit 20 (internal sensor data) and movement trajectory data of the mobile robot 1 by the external sensor (external sensor data) are acquired.

[0033] In step S12, the initial position coordinates (a x0 ,a y0 ) and the initial angle a θ0 Extract.

[0034] In step S13, the initial position coordinate (a x0 ,a y0 ) and the initial angle a θ0 The internal sensor data is offset corrected.

[0035] In step S14, the correction coefficient is derived by searching for the minimum value.

[0036] Here, the derivation of the correction coefficient by minimum value search will be described.

[0037] The corrected position coordinates (v') of the camera 21 obtained by the formulas (1) to (3) x ,v' y ,v' θ ) and the position coordinates of the reference point 31 obtained by the external sensor (a x ,a y ,a θ ), the evaluation function S using the sum of Euclidean distances is determined by the following equation (4).

[0038]

number

[0039] Here, i represents each time (frame number), and n represents the number of frames (number of data) used to derive the correction coefficient.

[0040] When the evaluation function S is minimized, it may not necessarily be the shortest distance at each time point i, but it will be an optimal fit for the entire trajectory. In other words, when the evaluation function S is minimized, it means that the difference between the camera 21 and the reference point 31 is minimized in all n frames. Therefore, d' that minimizes the evaluation function S is x ,d' y ,d' θ The amount of deviation D(d x ,d y ,d θ ) The minimum derivation of the evaluation function S is a minimum value search problem, and can be found using the steepest descent method, Newton's method, etc.

[0041] d' where the evaluation function S is minimized x ,d' y ,d' θIt is difficult to analytically obtain an exact solution. In this embodiment, an approximate solution is obtained using a minimum value search by simulated annealing, which is a randomized algorithm. Simulated annealing is a method of considering a neighborhood u' of the current state u and probabilistically determining whether the next state will be u or u'. The transition probability p is given by the following equation.

[0042]

number

[0043] Here, T is a parameter equivalent to temperature. If the new state S(u') is better than the current state S(u) (S(u)>S(u')), the state is set to S(u'). Even if the new state S(u') is worse than the current state S(u) (S(u)≦S(u')), the state is set to S(u') with the probability of equation (5). The higher the temperature T, the higher the probability that the state will transition to S(u'), and the lower the temperature T, the lower the probability that the state will transition to S(u'). The temperature T decreases over time. At first, a wide area is searched in search of the optimal solution, and the search range gradually narrows, heading towards an area with a low value of the evaluation function S, and finally converging to a state where the evaluation function S is at a minimum.

[0044] An example of a process for deriving correction coefficients by simulated annealing will be described with reference to the flowchart in FIG.

[0045] In step S151, the initial state u=(d x ,d y ,d θ ) and the initial temperature T0.

[0046] In step S152, the temperature T is lowered to generate the next state u' (near the current state). Specifically, the next state u' is selected from the three states of the following equation (6).

[0047]

number

[0048] The probability of choosing one of these three states is equal to one in three. δ x , δ y , δ θ is set by the following formula (7) so that it is determined randomly within a certain range, but the range is set so that it decreases with temperature.

[0049]

number

[0050] where j and h are constants and T is the current temperature.

[0051] In step S153, the score (evaluation value S(u')) of the new state u' is calculated.

[0052] In step S154, the evaluation value S(u) of the current state is compared with the evaluation value S(u') of the new state.

[0053] If S(u)>S(u'), that is, if the evaluation value S(u') of the new state is good, then in step S155, the new state u' is adopted, and the process returns to step S152.

[0054] Even if S(u)≦S(u′), in step S156, it is determined, based on the probability of equation (5), whether to proceed to step S155 and adopt the new state u′, or to keep the current state u and return to step S152.

[0055] The above processes from step S152 to step S156 are repeated for a certain period of time, and the state u'=(d' x ,d' y ,d' θ ) is the correction coefficient.

[0056] An example will be shown in which movement trajectory data acquired by an external sensor and movement trajectory data acquired by an internal sensor are used as input data to find the minimum value of the evaluation function S using the simulated annealing method.

[0057] An example of movement trajectory data is shown in Fig. 8 and Fig. 9. Fig. 8 shows movement trajectory data acquired by an external sensor. Fig. 9 shows movement trajectory data acquired by an internal sensor. The mobile robot 1 was moved around the periphery of a rectangular area by combining linear movement and 90-degree rotations. The movement trajectory data of the internal sensor in Fig. 9 is the initial position coordinates (a x0 ,a y0 ,a θ0 ) is taken into consideration and offset correction is applied to the data. At the points marked "Trajectory Deviation" in the figure, there is a difference in position between the movement trajectory data of the external sensor and the movement trajectory data of the internal sensor. This is known to be due to the effect of a relative positional deviation between the camera 21 and the reference point 31.

[0058] The initial state is d x =0, d y =0, d θ = 0, initial temperature T0 = 20°C, and the constants j and h in equation (7) were set to j = 1000 and h = 0.087, respectively, and the deviation D was calculated using the simulated annealing method. The deviation D is d x =2.6721[mm],d y =-13.8073[mm],d θ = 0.0652 [rad]. This deviation amount D indicates the deviation amount of the camera 21 as viewed from the reference point 31.

[0059] [Correction of movement trajectory data] Next, an example of a process for correcting movement trajectory data acquired by the visual odometry method will be described with reference to the flowchart of FIG.

[0060] In step S21, the VO unit 20 captures an image of the road surface with the camera 21 and acquires VO data (captured video). At the same time, the mobile robot 1 may perform measurements with the on-board measuring device.

[0061] In step S22, the VO unit 20 extracts feature points from frames of the captured video, and in step S23, estimates its own position by calculating the amount of movement of the feature points pair between frames.

[0062] In the processes from steps S21 to S23, the VO unit 20 acquires movement trajectory data by the visual odometry method.

[0063] In step S24, the VO unit 20 uses the formula (2) and the correction coefficient to obtain the position coordinates of the reference point 31 from the self-position obtained by the visual odometry method. After the measurement by the mobile robot 1, another device may correct the movement trajectory data obtained by the VO unit 20 to obtain the position coordinates of the reference point 31.

[0064] As a result, movement trajectory data of the coordinate system with the initial position of the mobile robot 1 as the origin is obtained. x0 ,a y0 ,a θ0 ) may be corrected.

[0065] Next, an example of the result of correcting the movement trajectory data obtained by the visual odometry method will be described with reference to FIGS.

[0066] As shown in Fig. 11, the mobile robot 1 was moved in a pattern of moving back and forth in a 45-degree diagonal direction with a constant posture, and movement trajectory data was obtained by the visual odometry method. For comparison, movement trajectory data was also obtained by an external sensor.

[0067] The initial position coordinates of the external sensor (a x0 ,a y0 ,a θ0 ) is applied to the offset correction factor d x =2.6721[mm],d y =-13.8073[mm],d θ = 0.0652 [rad] and equation (2) were used to obtain the corrected movement trajectory data. The corrected movement trajectory data is data defined in the XY coordinate system of the external sensor.

[0068] Figure 12 shows the movement trajectory data before correction. For comparison, movement trajectory data acquired by an external sensor is also shown. The movement trajectory data before correction shown in Figure 12 was created by applying offset correction to the movement trajectory data acquired by the visual odometry method to align the movement trajectory data acquired by the external sensor with the initial position. Figure 13 shows the movement trajectory data after correction using the correction coefficient.

[0069] As shown in FIG. 13, it can be seen that the corrected movement trajectory data and the movement trajectory data obtained by the external sensor are in good agreement, confirming the usefulness of the correction method of this embodiment.

[0070] [Trajectory example] Next, an example of the trajectory of the mobile robot 1 when the correction coefficient is derived will be described.

[0071] The correction coefficient is derived using trajectory data in the camera coordinate system measured by an internal sensor and trajectory data in the external sensor coordinate system measured by an external sensor. The greater the number of internal sensor data and external sensor data measured at the same time, the greater the impact on the accuracy of the minimum value search. There are two types of trajectories for acquiring trajectory data: linear trajectories and turning trajectories.

[0072] As shown in Fig. 14(a), the linear movement trajectory includes the forward, backward, sideways, and diagonal directions of the mobile robot 1. By fixing the orientation of the mobile robot 1 and moving it linearly in translation, the deviation between the orientation of the camera 21 and the orientation of the mobile robot 1 can be obtained. If the orientation of the camera 21 and the orientation of the mobile robot 1 are misaligned, for example, even if the mobile robot 1 moves along the axes of the XY coordinate system of the external sensor, the movement trajectory obtained by the camera 21 will not be aligned with the axes of the PQ coordinate system. By obtaining the deviation in the movement direction, the deviation between the orientation of the camera 21 and the orientation of the mobile robot 1 can be obtained.

[0073] The movement trajectory of the rotation system includes a circular movement as shown in FIG. 14(b) and a four-way movement as shown in FIG. 14(c). A circular movement is a movement that moves forward on a circumference of a circle with a radius R. When the radius is 0, it is a rotation on the spot. A four-way movement is a movement along the periphery of a rectangular area. There are four-way movements that are composed only of linear movements and those that are composed of a combination of rotational movements or turning movements. By rotating or turning the mobile robot 1, the positional deviation between the camera 21 and the reference point 31 can be obtained. When the mobile robot 1 rotates or turns, the movement trajectory of the camera 21 and the reference point 31 becomes a circle. The center position of the rotational movement or turning movement can be specified from the circular trajectory. Since the center positions of the circular trajectories of the camera 21 and the reference point 31 coincide with each other, the positional deviation between the camera 21 and the reference point 31 can be obtained from the positional relationship between the center position and the start position or end position of the circular trajectory. The misalignment between the axes of the XY coordinate system of the external sensor and the PQ coordinate system of the camera 21 can be corrected using the misalignment determined from the movement trajectory of the straight line system.

[0074] By acquiring the linear movement trajectory and the turning movement trajectory, it is possible to obtain the positional deviation and the orientation deviation between the camera 21 and the reference point 31. An example of a movement trajectory that combines the linear movement trajectory and the turning movement trajectory is shown in Fig. 15. The linear movement trajectory and the turning movement trajectory may be acquired in separate processes.

[0075] However, it is difficult to continue moving in translation while maintaining the posture, or to rotate or pivot around one point. x ,d y and the misalignment d between the XY coordinate system and the PQ coordinate system θ and construct an equation (2) for converting the PQ coordinate system into the XY coordinate system using the above equation, and calculate the displacement amount D(d x ,d y ,d θ ) and derives the optimal value to obtain the relative positional relationship between the camera 21 and the reference point 31.

[0076] [Correction to any point] Next, correction of movement trajectory data to an arbitrary point will be described. When carrying out measurements by mounting a measuring device on the mobile robot 1, it is desirable to acquire movement trajectory data for the position of the measuring device (target position). By calculating a correction coefficient in advance, movement trajectory data acquired by the visual odometry method using the VO unit 20 can be corrected to movement trajectory data for the reference point 31. If the relative positional relationship between the reference point 31 and the target position is known, positional information of the target position can be obtained from the positional information acquired by the VO unit 20. The relative positional relationship between the reference point 31 and the target position can be grasped from design values, etc.

[0077] As shown in FIG. 16, two markers 30A and 30B were placed on the mobile robot 1. The VO unit 20 acquired the movement trajectory data of the mobile robot 1, and the external sensor acquired the movement trajectory data of points 31A and 31B. The amount of deviation between each of the points 31A and 31B and the camera 21 was derived by the simulated annealing method. The movement trajectory data used to derive the amount of deviation was a trajectory of movement in all four directions, rotated 90 degrees at each vertex. The amount of deviation of the camera 21 as seen from point 31A was d x =2.6721[mm],d y =-13.8073[mm],d θ =0.0652 [rad]. The deviation of the camera 21 from the point 31B is d x =0.7058[mm],d y =541.5261[mm],d θ 17 shows the positional relationship between points 31A, 31B and camera 21 based on the derived deviation amount. It was confirmed that the derived result was almost the same positional relationship as the design value, even taking into account the inclusion of assembly errors and the like.

[0078] FIG. 18 shows the trajectory of point 31A acquired by the external sensor and the trajectory acquired by the internal sensor (VO unit 20) corrected by the correction coefficient. FIG. 19 shows the trajectory of point 31B acquired by the external sensor and the trajectory acquired by the internal sensor corrected by the correction coefficient. The correction coefficient used to correct the trajectory of point 31B is the correction coefficient obtained for point 31A with offset correction of the relative positional relationship between points 31A and 31B. For example, in the example of FIG. 17, the amount of deviation dx=0.7121[mm], dy=-541.4927[mm] obtained by offset correcting the relative positional relationship (1.96,-555.3) from point 31A to the amount of deviation dx=2.6721[mm], dy=-13.8073[mm] of the camera 21 as seen from point 31A is used as the correction coefficient. The movement trajectories are significantly different due to the different positions of points 31A and 31B. In both Figures 18 and 19, the trajectory of the internal sensor is corrected to obtain a trajectory similar to the trajectory obtained by the external sensor.

[0079] It can be seen that if a correction coefficient for the reference point 31 is obtained, then by correcting the movement trajectory data acquired by the VO unit 20, movement trajectory data for any target position can be acquired.

[0080] As described above, the method of calculating the correction coefficient in this embodiment involves obtaining the movement trajectory data of the mobile robot 1 obtained by the VO unit 20 using the visual odometry method and the movement trajectory data of the reference point 31 of the mobile robot 1 obtained by the external sensor, and calculating a correction coefficient that minimizes the total value of the distance between the movement trajectory data obtained by the visual odometry method and the movement trajectory data obtained by the external sensor. This makes it possible to calculate the deviation (correction coefficient) between the mounting position of the camera 21 and the reference point 31 set on the mobile robot 1 simply by moving the mobile robot 1 and obtaining the movement trajectory data by the VO unit 20 and the external sensor. In addition, since the accuracy required for the mounting position of the VO unit 20 can be reduced, the system can be configured at low cost. Even if the VO unit 20 is installed in an invisible position, the mounting position of the camera 21 can be grasped.

[0081] In addition, if the relative position of any point on the mobile robot 1 from the reference point 31 is known, the movement trajectory of the any point can be obtained by correcting the correction coefficient with the relative position. As a result, it becomes possible to collectively obtain the movement trajectories of multiple sensors at different positions.

[0082] An apparatus for carrying out each process of the present embodiment described above may be, for example, a general-purpose computer system including a central processing unit (CPU) 901, a memory 902, a storage 903, a communication device 904, an input device 905, and an output device 906, as shown in Fig. 20. In this computer system, the apparatus is realized by the CPU 901 executing a predetermined program loaded onto the memory 902. This program can be recorded on a non-transitory computer-readable recording medium such as a magnetic disk, an optical disk, or a semiconductor memory, or can be distributed via a network. [Explanation of symbols]

[0083] 1. Mobile robot 11. Cabinet 12 wheels 20 VO Section 21 Camera 30 Marker 31 Reference point

Claims

1. 1. An information processing method for determining a correction coefficient for correcting a self-position estimated by a moving body capable of moving two-dimensionally, comprising: the moving body includes a sensor unit for estimating its own position by a visual odometry method; The computer Acquire first movement trajectory data indicating a movement trajectory of the moving object acquired by the sensor unit, and acquire second movement trajectory data indicating a movement trajectory of a reference point of the moving object acquired by an external sensor; A correction coefficient that minimizes the total distance between the first movement trajectory data and the second movement trajectory data is obtained. Information processing methods.

2. 2. The information processing method according to claim 1, the correction coefficient includes a relative positional relationship between the sensor unit and a reference point of the moving body and a relationship between an orientation of the sensor unit and an orientation of the moving body, The first movement trajectory data and the second movement trajectory data include a movement trajectory in which the moving body moves linearly and a movement trajectory in which the moving body moves rotationally or circularly. Information processing methods.

3. 2. The information processing method according to claim 1, A correction factor that minimizes the sum of the distances is obtained using the simulated annealing method. Information processing methods.

4. A method for correcting a self-position estimated by a moving body capable of moving two-dimensionally, comprising the steps of: The sensor unit estimates its own position, The computer The self-position is corrected using a correction coefficient including a relative positional relationship between the sensor unit and a reference point of the moving body and a relationship between an orientation of the sensor unit and an orientation of the moving body. Correction method.

5. The correction method according to claim 4, The correction coefficient is corrected based on a relative position from a reference point of the moving body to an arbitrary point on the moving body. Correction method.

Citation Information

Patent Citations

  • Method and system for localizing mobile robot using external surveillance cameras

    US20170252925A1