JP5614118B2 - Landmark detection method, robot and program - Google Patents

Description

  The present invention relates to a landmark detection method, a robot having a landmark detection function, and a program for causing a computer to execute a landmark detection procedure. The present invention also relates to a computer-readable storage medium storing such a program.

Some autonomously traveling robots use landmarks. First, in the state where the position of the robot is known, an object to be a mark on the traveling route of the robot is registered as a landmark. FIG. 1 is a plan view showing an example of landmarks registered with respect to the position of the robot. In FIG. 1 and FIG. 2 described later, thick parallel vertical lines indicate paths through which the robot can move, the position of the robot is indicated by a circle, and the position of the landmark is indicated by an x mark. In the example of FIG. 1, the xy coordinates of the landmark mk ₁ registered with respect to the robot position p ₁ are (x ₁₁ , y ₁₁ ), and the xy of the landmark mk ₂ registered with respect to the robot position p ₂ . coordinate is the _(x l2, _yl2).

When the landmark is registered, the positions of the landmarks mk ₁ and mk ₂ are measured by a distance sensor mounted on the robot. Specifically, by adding the relative coordinates of the landmarks mk ₁ and mk ₂ viewed from the robot measured by the distance sensor to the xy coordinates of the known robot position, each of the landmarks mk ₁ and mk ₂ Find the coordinates. Also, landmarks _mk 1, mk ₂ corresponding the respective positions p1, p2 a camera mounted on the robot by imaging, obtains a feature vector representing a feature of the landmark _mk 1, mk ₂ of the captured image. As the landmarks mk ₁ and mk ₂ , for example, characteristic points in the captured image such as door corners, that is, characteristic points are used. The feature point extraction method itself is well known, and various extraction methods have been proposed. The positions (xy coordinates) and feature vectors of the landmarks mk ₁ and mk ₂ obtained in this way are registered, and the current robot position (that is, the current position of the robot) when the landmark where the robot autonomously travels is detected. ) Is used to calculate.

At the time of detecting the landmark where the robot autonomously travels, the landmark mk ₁ , mk ₂ is picked up from the captured images captured by the camera when the robot travels using the feature vectors of the registered landmarks mk ₁ , mk _2. Is detected. FIG. 2 is a diagram illustrating a case where the landmark mk ₁ is detected from the captured image. The distance from the robot to the detected landmark mk ₁ is obtained by a distance sensor, the relative position from the current position p1 ′ of the robot to the detected landmark mk ₁ is obtained, and this relative position and the registered landmark mk ₁ The current position of the robot can be calculated from the position (xy coordinates). In FIG. 2, the xy coordinates (x _r , r _y ) of the current position p1 ′ of the robot are unknown, and the relative position to the landmark mk ₁ detected from the robot is (x ′ _l1 , y ′ _l1 ). And Strictly speaking, the equation does not hold, but if the error is negligible, the xy coordinates (x ₁ , y ₁ ) of the current position p1 ′ of the robot are (x ₁ , y ₁ ) = (x _r , r _y ) + (x ₁₁ , y ₁₁ ).

  In the conventional landmark detection method, if the position of the robot when the landmark is registered is close to the current position of the robot when the landmark is detected when the robot is running, the landmark detection is likely to succeed. . However, if the position of the robot when the landmark is registered and the current position of the robot when detecting the landmark when the robot is traveling are far apart, the landmark detection and landmark detection The viewpoint of the captured image is shifted from time to time, and the captured image changes greatly. For this reason, the feature vector also changes greatly with a large change in the captured image, and the landmark may not be detected when the robot travels. Therefore, for example, when registering landmarks, it is possible to register feature vectors at many positions of the robot, and when detecting landmarks, it is possible to match using all of the feature vectors, but the amount of calculation becomes very large, Considering the performance of the computer installed in the robot, it is not realistic. It is also possible to find the average of feature vectors and use it for matching. However, if the amount of calculation is reduced using simple statistics, there is a possibility of erroneous detection of landmarks. It cannot be expected to improve accuracy.

JP 2006-172016 A JP 2008-102856 A

  In the conventional landmark detection method, there is a problem in that it is difficult to improve the detection accuracy of the landmark because erroneous detection of the landmark may occur.

  Therefore, an object of the present invention is to provide a landmark detection method, a robot, and a program that can improve the detection accuracy of landmarks.

According to one aspect of the present invention, there is provided a landmark detection method for detecting a landmark from an autonomously traveling robot, wherein a feature vector generating device uses a landmark vector to generate a landmark feature point based on a captured image output by a camera within the robot. And a step of calculating a Mahalanobis distance in the robot based on the feature vector and the feature vector of the feature point at the time of landmark registration registered in the storage device by the Mahalanobis distance calculation device. When, in the robot, and outputting a detection result of the landmark the Mahalanobis distance is selected by the minimum of the feature vector landmark detection result generating device, within the robot,-out based on the output of the drive device , measurements distributed according to the movement amount and the normal distribution of the robot by the current position candidate generation unit A step of determining a candidate of the current position of the robot based on the error, in the robot, the candidate of the current position of the robot, consistency evaluate the consistency with the position of the landmark which is registered in the storage device A step of evaluating by the apparatus and outputting an evaluation result; and within the robot, calculating a weighted average of coordinate values of the robot position candidates by a weighted average calculating apparatus, and calculating a weighted average corresponding to the evaluation result landmark detection method comprising the step of obtaining an estimated position of said robot is provided asking.

According to an aspect of the present invention, there is provided an autonomous traveling robot that detects a landmark, the feature of a landmark feature point based on a camera, a storage device, a traveling device, and a captured image output by the camera. A feature vector generating device for generating a vector; a Mahalanobis distance calculating device for calculating a Mahalanobis distance based on the feature vector and a feature vector of a feature point registered in the storage device; and the Mahalanobis distance a landmark detection result generating unit for outputting a detection result of the landmarks but by selecting the minimum of the feature vector,-out based on the output of the traveling device, the error of measurement distributed in accordance with the movement amount and the normal distribution of the robot A current position candidate generating device for obtaining a current position candidate of the robot based on the current position candidate of the robot; For it, the integrity evaluation device for outputting an evaluation result by evaluating the consistency of the position of the landmark which is registered in the storage device, a weighted average of the coordinate values of the candidate of the robot position, the evaluation result There is provided a robot provided with a weighted average calculating device that obtains an estimated position of the robot by calculating a corresponding weighted average .

According to one aspect of the present invention, the computer, a program which Ru is detected landmark from autonomous robots, the feature vector of the feature point of the landmark based on the captured image by the camera of the robot output A step of generating, a step of calculating a Mahalanobis distance based on the feature vector and a feature vector of a feature point at the time of landmark registration registered in the storage device of the robot, and a feature vector having the smallest Mahalanobis distance. a step of outputting a detection result of the landmark selected, the-out based on the output of the traveling device of the robot, the current position of the robot based on the error of measurement distributed in accordance with the movement amount and the normal distribution of the robot candidate And a landmarker registered in the storage device for the current position candidate of the robot. A step of outputting to the evaluation result evaluated consistent with the position of the weighted average of the coordinate values of the candidate robot position, estimation of the robot determined by calculating the weighted average in accordance with the evaluation result A program for causing the computer to execute a procedure for obtaining a position is provided.

  According to the disclosed landmark detection method, robot, and program, landmark detection accuracy can be improved.

It is a top view which shows an example of the landmark registered with respect to the position of a robot. It is a figure which shows the case where the landmark is detected from the captured image. It is a figure explaining an example of the extraction method of the feature point at the time of landmark registration. It is a figure explaining an example of the extraction method of the feature point at the time of the detection of a landmark. It is a figure explaining a mode that the distance from the position of a robot to a landmark is measured, and a landmark is imaged. It is a figure which represents on the graph the feature vector calculated | required by the 1st comparative example when a captured image changes a lot. It is a figure showing the feature vector calculated | required by the 2nd comparative example on a graph. It is a figure explaining evaluation based on Mahalanobis distance of a feature vector. It is a figure explaining a mode that the distance from the position of a robot to a landmark is measured, and a landmark is imaged. It is a figure explaining the approximation of a feature vector. It is a figure explaining the jth component of a feature vector. It is a figure explaining a three-dimensional coordinate. It is a figure explaining the case where the similar thing is contained in the feature vector produced with respect to a different landmark. It is a figure explaining creation of a vector which combined each component of a feature vector, and a coordinate value of a position. It is a figure explaining the registration of the landmark which calculates | requires the average vector and covariance matrix of the vector which combined each component of the feature vector, and the coordinate value of a position. It is a figure explaining the production | generation of the predicted position of the present robot. It is a figure explaining creation of the candidate of the position of the present robot using a particle filter. It is a block diagram which shows an example of a structure of a robot. It is a block diagram explaining the distance calculation apparatus between vectors. It is a flowchart explaining the operation | movement at the time of landmark registration of a robot. It is a flowchart explaining the operation | movement at the time of the landmark detection which a robot drive | works. It is a flowchart explaining operation | movement of the feature vector approximation apparatus. It is a flowchart explaining operation | movement of the feature vector and robot coordinate value combination apparatus. It is a flowchart explaining operation | movement of an average and a covariance matrix calculation apparatus.

  In the disclosed landmark detection method, robot, and program, a landmark is detected using a discriminator created by learning a landmark feature vector in advance.

  The classifier may be extended so that the landmark is detected using a vector that combines the feature vector and the coordinate value of the robot.

  Hereinafter, embodiments of the disclosed landmark detection method, robot, and program will be described with reference to the drawings.

  In one embodiment of the present invention, the landmark detection method is applied to an autonomous traveling robot (not shown).

  At the time of registration of landmarks, the landmarks are imaged at a plurality of positions of the robot by a camera mounted on the robot, and feature vectors are learned from the captured images of the landmarks obtained at the respective positions to identify the classifier, that is, the Mahalanobis distance. Create (Mahalanobis Distance). On the other hand, when a landmark where the robot autonomously travels is detected, the landmark is detected using an identifier.

  The classifier is expanded to create a vector of combinations of the landmark image feature vector and coordinate values as components for the captured image of the landmark at each position, and a classifier is created from this combination vector. It may be used for landmark detection.

As the landmark, for example, a characteristic point in the captured image of the camera such as a corner of the door, that is, a characteristic point is used. The feature point extraction method itself is well known, and various extraction methods using a Harris Operator, SIFT (Scale Invariant Feature Transform), or the like have been proposed. FIG. 3 is a diagram for explaining an example of a feature point extraction method at the time of landmark registration. As shown in FIG. 3, for example, 16 (= 4 × 4) areas having a size of, for example, 20 pixels × 20 pixels around the feature point (ie, landmark) 102 in the captured image 101 are cut out and cut out. A vector in which the gray values ν _{i, M} of the pixels in each region are arranged is created to obtain an M-dimensional feature vector. The following equation represents the feature vector ν _i of the i-th landmark obtained in this example, and the superscript t represents transpose.

  When a landmark is detected, feature points are extracted from the captured image of the camera. FIG. 4 is a diagram for explaining an example of a feature point extraction method when a landmark is detected. As shown in FIG. 4A, one feature point 102-1A of the captured image 101A is selected, and the feature vector of the feature point 102-1A and each of the registered landmarks 102-1 to 102-i are selected. The Euclidean Metric with the feature vector, that is, the sum of squares of the difference of each component of the feature vector is obtained. The landmark 102-1 having the feature vector having the smallest Euclidean distance is selected as the landmark corresponding to the feature point 102-1A. In this case, the detection result of the feature point 102-1A indicates that it is the landmark 102-1. Similarly, as shown in FIG. 4B, another feature point 102-2A of the captured image 101A is selected, and each landmark 102-1 registered with the feature vector of the feature point 102-2A is selected. The Euclidean distance between the feature vectors of ˜102 and i is calculated. The landmark 102-2 having the feature vector having the smallest Euclidean distance is selected as the landmark corresponding to the feature point 102-2A. In this case, the detection result of the feature point 102-2A indicates that it is the landmark 102-2.

  Next, the effects of using the Mahalanobis distance will be described below. When the average vector is described in detail, the following equation is obtained.

  The covariance matrix is as follows.

  Further, the Mahalanobis distance is as follows.

  The feature vector is an M-dimensional vector (that is, the number of values held by the vector is M). In the case of FIG. 3 described above, M = 16 for convenience of explanation. The value of M is several tens to several hundreds. If the dimension of the feature vector is large, it is difficult to illustrate the feature vector. In this example, for the sake of convenience, description will be made assuming that the feature vector is one-dimensional (M = 1). Even if the dimension of the feature vector is large, the basic concept is the same as the one-dimensional case described below.

  When the feature vector is one-dimensional, the Mahalanobis distance is as follows:

  As described above, in the first comparative example, since erroneous detection of the landmark may occur, it is difficult to improve the detection accuracy of the landmark. On the other hand, in the second comparative example, the amount of calculation is very large in order to prevent erroneous detection of the landmark, and it is difficult to improve the detection accuracy of the landmark in consideration of the performance of the computer mounted on the robot. That is, in the first and second comparative examples in which the feature vector is evaluated based on the Euclidean distance, it is difficult to improve the landmark detection accuracy.

  For learning, it is necessary to obtain a large number of feature vectors. However, if the robot is moved to the landmark imaging position and an image is taken to obtain a feature vector, the amount of work is large. Therefore, by moving the robot to each landmark imaging position by approximating and interpolating the feature vector component at a certain position using the feature vector component at the nearby position in the above procedure, the image is obtained. The amount of work is greatly reduced as compared with the case of obtaining a feature by imaging.

  Needless to say, the approximation and interpolation of the feature vector components in this modification may be omitted.

  By the way, if similar features are included in feature vectors created for different landmarks, the detection performance may be degraded. FIG. 13 is a diagram for explaining a case in which similar features are included in feature vectors created for different landmarks k ′ and k ′ ″. FIG. 13 (b) shows a landmark covariance matrix to explain the evaluation of the feature vector, and the distribution of the two landmarks is similar as shown in FIG. The feature vectors overlap as shown by being surrounded by a broken line. The reason why similar feature vectors are included in the feature vectors created for different landmarks k ', k' " It is possible that similar features may be included in the feature vectors created for the different landmarks k ′ and k ′ ″.

Therefore, in the second modification of the present embodiment, in order to distinguish similar feature vectors from feature vectors created for different landmarks k ′, k ′ ″, landmarks k ′, k ′ ″ are used. The robot position at the time of registration is used. At each of the landmark imaging position and the position approximating the feature vector, as shown in FIG. 14, each component ν _{i, 1} , ..., ν _{i, M} of the feature vector ν _i and the position coordinate value x _ri , y _ri is combined to create a vector V _i = (ν _{i, 1} , ..., ν _{i, M} , x _ri , y _ri ). FIG. 14 is a diagram illustrating the creation of a vector that combines each component of the feature vector and the coordinate value of the position.

  Needless to say, the generation of the vector combining the feature vector and the coordinate value of the robot position in this modification may be omitted, and the average vector and covariance matrix of the feature vector may be calculated.

  Needless to say, the generation of a vector combining the feature vector and the coordinate value of the predicted robot position in this modification may be omitted, and the average vector of the feature vectors and the covariance matrix may be calculated.

  Next, an example of the configuration of the robot in the present embodiment will be described with reference to FIG. FIG. 18 is a block diagram illustrating an example of the configuration of the robot.

  18 includes a camera 1, an image input device 2, a feature point extraction device 3, a feature vector generation device 4, a feature vector approximation device 5, a feature vector and robot coordinate value combination device 6, and an average covariance matrix calculation device. 7, distance sensor 8, landmark storage device 9, traveling device 10, encoder 11, predicted position generation device 12, Mahalanobis distance calculation device 13, landmark detection result generation device 14, current position candidate generation device 15, consistency evaluation device 16 and a weighted average calculation device 17.

  The camera 1 captures an image including a landmark and outputs a video signal representing the captured image. The image input device 2 receives the video signal output from the camera 1, and outputs image data indicating the gray value of each pixel to the feature point extraction device 3. The feature point extraction device 3 performs feature point extraction processing on the image data and outputs the feature point data to the feature vector generation device 4. The feature vector generation device 4 generates and outputs a feature vector from the gray value of each pixel around the feature point based on the feature point data.

  At the time of landmark registration, the feature vector generation device 4 outputs the feature vector to the feature vector approximation device 5 and the feature vector / robot coordinate value combination device 6. The feature vector approximating device 5 approximates the feature vector with the feature vector of the nearby landmark imaging position, and outputs the feature vector at the approximated position to the feature vector and robot coordinate combining device 6. The feature vector and robot coordinate combination device 6 creates a vector by combining each component of the feature vector from the feature vector generation device 4 or the feature vector approximation device 5 and the coordinate value of the position of the robot 20, and calculates the mean and covariance matrix. Output to the device 7. At the time of landmark registration, the robot 20 is moved to a position where the landmark should be imaged, for example, by a human hand, so the coordinate value of the position of the robot 20 is known. However, when the robot 20 is moved to the position where the landmark should be imaged by using the traveling device 10 described later, the coordinate value of the predicted position of the robot 20 generated by the predicted position generation device 12 described later is used as the robot. You may use as a coordinate value of 20 positions. The average and covariance matrix calculation device 7 calculates the average and covariance matrix of a vector (that is, learning data) obtained by combining each component of the feature vector and the coordinate value of the position of the robot 20 and outputs the average and covariance matrix to the landmark storage device 9. Remember.

  When the landmark is registered, the distance sensor 8 measures the distance from the robot 20 to the landmark, and outputs the landmark coordinates based on the current position of the robot 20 to the landmark storage device 9. Remember. As a result, the landmark storage device 9 has the average value and the covariance matrix of the coordinate value of the landmark position, and the vector (or learning data) obtained by combining each component of the feature vector and the coordinate value of the position of the robot 20. Stored (ie, registered).

  The traveling device 10 has a known configuration in which the robot 20 is caused to travel by rotating wheels (not shown) by a motor (not shown). The encoder 11 is attached to a wheel, for example, and obtains and outputs wheel rotation number data, that is, a movement amount of the robot 20 based on wheel rotation information output from the traveling device 10. The predicted position generation device 12 obtains the current predicted position of the robot 20 (that is, the predicted robot position) from the past position of the robot 20 and the movement amount of the robot 20 obtained from the encoder 11, and combines the feature vector and the robot coordinate system. 6 and the current position candidate generation device 15.

  When the landmark is detected, the feature vector generation device 4 uses the feature vector generated based on the video signal (that is, the captured image) output from the camera 1 at the current position of the robot 20 as the feature vector and robot coordinate combination device 6. Output to. The feature vector / robot coordinate combination device 6 creates a vector by combining each component of the feature vector from the feature vector generation device 4 and the coordinate value of the robot predicted position from the prediction position generation device 12, and sends it to the Mahalanobis distance calculation device 13. Output. In addition, the coordinate value of the landmark stored in the landmark storage device 9 is read and output to the Mahalanobis distance calculation device 13.

  The Mahalanobis distance calculation device 13 reads the mean and covariance matrix at the position of the robot 20 at the time of landmark registration read out from the landmark storage device 9, and the current robot predicted position from the feature vector and robot coordinate combination device 6. The Mahalanobis distance is calculated based on the mean and covariance matrix at, and is output to the landmark detection result generation device 14. The landmark detection result generation device 14 selects a landmark having the shortest Mahalanobis distance for each feature point of the captured image and outputs it to the consistency evaluation device 16 as a landmark detection result. On the other hand, the current position candidate generation device 15 determines the current position candidate of the robot 20 (that is, the robot) based on the movement amount of the robot 20 output from the encoder 11 and the normal distribution error based on the current predicted robot position. Position candidates). The robot position candidates generated by the current position candidate generation device 15 are output to the consistency evaluation device 16 and the weighted average calculation device 17.

  Therefore, at the time of landmark detection, the consistency evaluation device 16 uses the landmark coordinate values measured by the distance sensor 8 at the current position of the robot 20 and the robot position candidates from the current position candidate generation device 15 as references. The landmark detection result (or the position of the landmark) from the landmark detection result generation device 14 is read from the landmark storage device 9 and the landmark coordinate value (or the landmark information) when the landmark is registered. Position). The evaluation result of the consistency evaluation device 16 is output to the weighted average calculation device 17. The weighted average calculation device 17 obtains an estimated position of the robot 20 by calculating a weighted average for the coordinates of the robot position candidate according to the evaluation result output from the consistency evaluation device 16.

  In FIG. 18, the functions of the components (that is, the device) of the robot 20 other than the camera 1, the distance sensor 8, the landmark storage device 9, the traveling device 10, and the encoder 11 are a processor such as a CPU (Central Processing Unit), and the like. This can be realized by a known general-purpose computer (not shown) provided with a storage device. In this case, the storage device of the computer may function as the landmark storage device 9. The storage device of the computer stores various data including a program executed by the CPU and intermediate results of operations executed by the CPU. When the CPU executes the program, each function of the robot 20 is realized, and each procedure of processing executed by the robot 20 is executed. The program can be stored in various computer-readable storage media including a computer storage device.

  FIG. 19 is a block diagram for explaining an inter-vector distance calculation apparatus. In FIG. 19, the same parts as those in FIG. 19, the inter-vector distance calculation device 40 is formed by the feature vector and robot coordinate value combination device 6 and the Mahalanobis distance calculation device 13 shown in FIG. The functions of the inter-vector distance calculation device 40 can also be realized by a known general-purpose computer including a processor such as a CPU and a storage device.

  FIG. 20 is a flowchart for explaining the operation at the time of landmark registration of the robot 20 shown in FIG. In FIG. 20, in step S1, an arbitrary position for photographing a landmark is selected. In step S2, photographing with the camera 1 and distance measurement by the distance sensor 8 are performed on the landmark from the position selected in step S1. In step S3, the image input device 2, the feature point extraction device 3, and the feature vector generation device 4 extract (or generate) a landmark feature vector based on the output video signal of the camera 1. In step S4, it is determined whether or not a sufficient number of landmarks have been photographed. If the determination result is NO, the process returns to step S1. On the other hand, if the determination result in step S4 is YES, in step S5, the feature vector approximation device 5 selects the coordinates to be approximated, and obtains the shooting positions of three nearby landmarks. In step S6, the feature vector approximating device 5 performs approximation from the feature vectors of the three nearby landmarks selected in step S5, and generates a feature vector at the approximated position. In step S7, the robot coordinate value combination device 6 and the average and covariance matrix calculation device 7 calculate the average of the vector obtained by combining the feature vector from the feature vector generation device 4 and the feature vector approximation device 5 and the known coordinate value of the robot 20, and The covariance matrix is calculated and stored in the landmark storage device 9, and the process ends.

  FIG. 21 is a flowchart for explaining the operation at the time of detecting a landmark traveled by the robot 20 shown in FIG. In FIG. 21, in step S <b> 11, the image input device 2, the feature point extraction device 3, and the feature vector generation device 4 obtain (or generate) a landmark feature vector based on the output video signal of the camera 1. In step S <b> 12, the feature vector and robot coordinate value combination device 6 calculates a vector that combines the feature vector from the feature vector generation device 4 and the coordinate value of the predicted position of the robot from the predicted position generation device 12. In step S13, the Mahalanobis distance calculation device 13 reads the mean and covariance matrix at the position of the robot 20 at the time of landmark registration read from the landmark storage device 9, and the current vector from the feature vector and robot coordinate combination device 6. The Mahalanobis distance is calculated based on the mean and covariance matrix at the predicted robot position. In step S14, the landmark detection result generation device 14 selects a landmark having the shortest Mahalanobis distance for each feature point of the captured image and obtains a landmark detection result. In step S <b> 15, the encoder 11, the predicted position generation device 12, and the current position candidate generation device 15 based on the wheel rotation information output from the traveling device 10 cause the current robot 20 to move based on the movement amount of the robot 20 and the normal distribution error. Find position candidates. In step S16, the consistency evaluation device 16 is obtained in step S14 based on the landmark coordinate value measured by the distance sensor 8 at the current position of the robot 20 and the robot position candidate obtained in step S15. The consistency of the detected landmark detection result (or landmark position) with the landmark coordinate value (or landmark position) when the landmark is registered in the landmark storage device 9 is evaluated. To do. In step S17, the weighted average calculation device 17 calculates the weighted average for the coordinates of the candidate robot position according to the consistency evaluation result obtained in step S16 to obtain the estimated position of the robot 20, and the process ends. To do.

  FIG. 22 is a flowchart for explaining the operation of the feature vector approximating apparatus 5 shown in FIG. FIG. 22 shows processing for one landmark, and such processing is performed for each landmark.

  FIG. 23 is a flowchart for explaining the operation of the feature vector and robot coordinate value combination device 6 shown in FIG. FIG. 23 shows processing for one landmark, and such processing is performed for each landmark.

In FIG. 23, in step S41, a variable i representing a learning data number is set to 1. In step S42, a vector V _i is generated by combining the component of the feature vector ν _i and the coordinate values x _ri and y _ri of the robot 20. In step S43, it is determined whether i ≧ N. If the determination result is NO, i is incremented by 1 in step S44, and then the process returns to step S42. On the other hand, the process ends if the decision result in the step S43 is YES.

  In the above-described embodiments and modifications, the landmark detection method is applied to an autonomous traveling robot. However, a well-known GPS (Global Positioning System) is mounted on the robot, and detection is performed based on the landmark detection method. Based on the position of the detected robot and the position of the robot detected by GPS, the position detection accuracy of the robot can be further improved. In this case, the position of the robot can be detected based on the landmark detection method even when the GPS does not operate because the radio wave from the artificial satellite does not reach the robot.

The following additional notes are further disclosed with respect to the embodiment including the above examples.
(Appendix 1)
A landmark detection method for detecting a landmark from an autonomous traveling robot,
Generating a feature vector of a landmark feature point by a feature vector generation device based on a captured image output by a camera in the robot;
In the robot, a Mahalanobis distance is calculated based on the feature vector registered in the storage device by the Mahalanobis distance calculation device based on the feature vector at the time of landmark registration;
In the robot, a feature vector having the smallest Mahalanobis distance is selected by a landmark detection result generation device and is output as a landmark detection result;
In the robot, obtaining a candidate for the current position of the robot based on a movement amount and a normal distribution error of the robot by a current position candidate generation device based on the output of the traveling device;
In the robot, for the candidate of the current position of the robot, a process of evaluating the consistency with the position of the landmark registered in the storage device and outputting an evaluation result;
A landmark detection method comprising: calculating a weighted average according to the evaluation result on the coordinates of the robot position candidate by a weighted average calculation device in the robot to obtain an estimated position of the robot.
(Appendix 2)
Within the robot, the method further includes a step of obtaining a predicted position of the current robot from a past position of the robot and a movement amount of the robot by a predicted position generation device,
The step of generating the feature vector creates a vector that combines each component of the feature vector and the coordinate value of the predicted position of the robot by a feature vector and robot coordinate combination device,
The step of calculating the Mahalanobis distance is performed by combining the vector of the combination by the Mahalanobis distance calculation device, each component of the feature vector registered in the storage device and the coordinate value of the position of the robot. The landmark detection method according to appendix 1, wherein the Mahalanobis distance is calculated based on:
(Appendix 3)
The step of calculating the Mahalanobis distance is performed by combining the vector of the combination by the Mahalanobis distance calculation device, each component of the feature vector registered in the storage device and the coordinate value of the position of the robot. The landmark detection method according to appendix 2, wherein the Mahalanobis distance is calculated based on the mean and the covariance matrix.
(Appendix 4)
The step of outputting the evaluation result includes the landmark position registered in the storage device with reference to the coordinate value of the landmark measured by the distance sensor at the current robot position and the current position candidate of the robot. The landmark detection method according to any one of appendices 1 to 3, wherein the consistency evaluation apparatus evaluates consistency with the consistency evaluation apparatus and outputs an evaluation result.
(Appendix 5)
The step of outputting the evaluation result evaluates the consistency of the current position candidate of the robot with the positions of a plurality of landmarks registered in the storage device, and outputs the evaluation result. The landmark detection method according to any one of appendices 1 to 4.
(Appendix 6)
An autonomous traveling robot that detects landmarks,
A camera,
A storage device;
A traveling device;
A feature vector generation device for generating a feature vector of a landmark feature point based on a captured image output by the camera;
A Mahalanobis distance calculation device for calculating a Mahalanobis distance based on the feature vector and the feature vector of the feature point at the time of landmark registration registered in the storage device;
A landmark detection result generating device for selecting the feature vector having the smallest Mahalanobis distance and outputting it as a landmark detection result;
A current position candidate generation device that obtains a candidate for the current position of the robot based on a movement amount of the robot and a normal distribution error based on the output of the traveling device;
Consistency evaluation device that evaluates the consistency with the position of the landmark registered in the storage device and outputs an evaluation result for the current position candidate of the robot;
A robot comprising a weighted average calculation device that calculates a weighted average of the coordinates of the robot position candidate according to the evaluation result to obtain an estimated position of the robot.
(Appendix 7)
A predicted position generation device for obtaining a predicted position of the current robot from a past position of the robot and a movement amount of the robot;
Further comprising a feature vector and robot coordinate combination device for creating a vector combining each component of the feature vector and the coordinate value of the predicted position of the robot;
The Mahalanobis distance calculation device calculates the Mahalanobis distance based on the vector of the combination, each vector of the feature vector registered in the storage device at the time of landmark registration, and a vector obtained by combining the coordinate values of the position of the robot. The robot according to appendix 5.
(Appendix 8)
The Mahalanobis distance calculation device uses the combination vector, a vector vector and a covariance matrix obtained by combining each component of the feature vector registered in the storage device and the coordinate value of the position of the robot. The robot according to appendix 7, which calculates a Mahalanobis distance based on it.
(Appendix 9)
A distance sensor;
The consistency evaluation device uses the landmark coordinate value measured by the distance sensor at the current robot position and the current position candidate of the robot as a reference, and the landmark position registered in the storage device. The robot according to any one of appendices 6 to 8, wherein the consistency is evaluated and an evaluation result is output.
(Appendix 10)
Any one of appendices 6 to 9, wherein the consistency evaluation device evaluates the consistency of the current position candidate of the robot with the positions of a plurality of landmarks registered in the storage device, and outputs an evaluation result. The robot according to claim 1.
(Appendix 11)
A computer program that detects landmarks from an autonomous robot.
Generating a feature vector of landmark feature points based on a captured image output by the camera of the robot;
Calculating Mahalanobis distance based on the feature vector and the feature vector of the feature point at the time of landmark registration registered in the storage device of the robot;
A procedure for selecting the feature vector having the smallest Mahalanobis distance and outputting it as a landmark detection result;
A procedure for obtaining a candidate for the current position of the robot based on a movement amount and a normal distribution error of the robot based on an output of the traveling device of the robot;
For the current position candidate of the robot, a procedure for evaluating the consistency with the position of the landmark registered in the storage device and outputting an evaluation result;
A program that causes the computer to execute a procedure of calculating a weighted average of coordinates of the robot position candidates according to the evaluation result to obtain an estimated position of the robot.
(Appendix 12)
Causing the computer to further execute a procedure for obtaining a predicted position of the current robot from a past position of the robot and a movement amount of the robot;
The procedure for generating the feature vector is to create a vector combining each component of the feature vector and the coordinate value of the predicted position of the robot,
The Mahalanobis distance is calculated based on the Mahalanobis distance based on the combination vector, a vector obtained by combining each component of the feature vector registered in the storage device and the coordinate value of the position of the robot. The program according to appendix 11, which calculates
(Appendix 13)
The procedure for calculating the Mahalanobis distance includes the vector and the covariance matrix of the combination vector, each component of the feature vector registered in the storage device and the coordinate value of the position of the robot. The program according to appendix 12, wherein the Mahalanobis distance is calculated based on:
(Appendix 14)
The procedure for outputting the evaluation result is based on the landmark coordinate value measured by the distance sensor of the robot at the current robot position and the candidate for the current position of the robot as a reference, and the land registered in the storage device. 14. The program according to any one of appendices 11 to 13, which evaluates consistency with a mark position and outputs an evaluation result.
(Appendix 15)
The procedure for outputting the evaluation result includes evaluating the consistency of the current position candidate of the robot with the positions of a plurality of landmarks registered in the storage device, and outputting the evaluation result. The program according to any one of the above.
(Appendix 16)
A computer-readable storage medium storing the program according to any one of appendices 11 to 15.

  As mentioned above, although the disclosed landmark detection method, robot, and program have been described by the embodiments, the present invention is not limited to the above embodiments, and various modifications and improvements can be made within the scope of the present invention. Needless to say.

DESCRIPTION OF SYMBOLS 1 Camera 2 Image input device 3 Feature point extraction device 4 Feature vector generation device 5 Feature vector approximation device 6 Feature vector and robot coordinate value combination device 7 Average covariance matrix calculation device 8 Distance sensor 9 Landmark storage device 10 Traveling device 11 Encoder 12 Predicted position generation device 13 Mahalanobis distance calculation device 14 Landmark detection result generation device 15 Current position candidate generation device 16 Consistency evaluation device 17 Weighted average calculation device 20 Robot

Claims (6)

An autonomous traveling robot that detects landmarks,
A camera,
A storage device;
A traveling device;
A feature vector generation device for generating a feature vector of a landmark feature point based on a captured image output by the camera;
A Mahalanobis distance calculation device for calculating a Mahalanobis distance based on the feature vector and the feature vector of the feature point at the time of landmark registration registered in the storage device;
A landmark detection result generating device for selecting the feature vector having the smallest Mahalanobis distance and outputting it as a landmark detection result;
-Out based on the output of the traveling device, the current position candidate generating apparatus for determining the candidate of the current position of the robot based on the error of measurement distributed in accordance with the movement amount and the normal distribution of the robot,
Consistency evaluation device that evaluates the consistency with the position of the landmark registered in the storage device and outputs an evaluation result for the current position candidate of the robot;
A robot comprising a weighted average calculation device that obtains an estimated position of the robot by calculating a weighted average of coordinate values of the robot position candidates by calculating a weighted average corresponding to the evaluation result. A predicted position generation device for obtaining a predicted position of the current robot from a past position of the robot and a movement amount of the robot;
Further comprising a feature vector and robot coordinate combination device for creating a vector combining each component of the feature vector and the coordinate value of the predicted position of the robot;
The Mahalanobis distance calculation device calculates the Mahalanobis distance based on the vector of the combination, each vector of the feature vector registered in the storage device at the time of landmark registration, and a vector obtained by combining the coordinate values of the position of the robot. The robot according to claim 1.   The Mahalanobis distance calculation device uses the combination vector, a vector vector and a covariance matrix obtained by combining each component of the feature vector registered in the storage device and the coordinate value of the position of the robot. The robot according to claim 2, wherein the Mahalanobis distance is calculated based on the distance. A distance sensor;
The consistency evaluation device uses the landmark coordinate value measured by the distance sensor at the current robot position and the current position candidate of the robot as a reference, and the landmark position registered in the storage device. The robot according to claim 1, wherein the consistency is evaluated and an evaluation result is output. To the computer, a program that Ru to detect the landmark from the autonomous robot,
Generating a feature vector of landmark feature points based on a captured image output by the camera of the robot;
Calculating Mahalanobis distance based on the feature vector and the feature vector of the feature point at the time of landmark registration registered in the storage device of the robot;
A procedure for selecting the feature vector having the smallest Mahalanobis distance and outputting it as a landmark detection result;
A step of determining a candidate of the current position of the robot the-out based on the output of the traveling device of the robot, based on the error of measurement distributed in accordance with the movement amount and the normal distribution of the robot,
For the current position candidate of the robot, a procedure for evaluating the consistency with the position of the landmark registered in the storage device and outputting an evaluation result;
A program for causing the computer to execute a procedure for obtaining a weighted average of coordinate values of the robot position candidate by calculating a weighted average corresponding to the evaluation result to obtain an estimated position of the robot. A landmark detection method for detecting a landmark from an autonomous traveling robot,
Generating a feature vector of a landmark feature point by a feature vector generation device based on a captured image output by a camera in the robot;
In the robot, a Mahalanobis distance is calculated based on the feature vector registered in the storage device by the Mahalanobis distance calculation device based on the feature vector at the time of landmark registration;
In the robot, a feature vector having the smallest Mahalanobis distance is selected by a landmark detection result generation device and is output as a landmark detection result;
Within the robot,-out based on the output of the drive apparatus, a step of determining a candidate of the current position of the robot based on the error of measurement distributed in accordance with the movement amount and the normal distribution of the robot by the current position candidate generating unit,
In the robot, for the candidate of the current position of the robot, a process of evaluating the consistency with the position of the landmark registered in the storage device and outputting an evaluation result;
In the robot, a weighted average of coordinate values of the robot position candidates is obtained by calculating a weighted average according to the evaluation result by a weighted average calculation device, and the estimated position of the robot is obtained. Landmark detection method.

Images