JP2010112836A - Self-position identification device and mobile robot provided with same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a self-position identification device capable of identifying a self-position without using a map in a more versatile environment and a mobile robot provided with the self-position identification device. <P>SOLUTION: The self-position identification device has a characteristic point extracting section 11 for extracting characteristic points of environmental information, a characteristic point relating section 12 for relating an aggregate of characteristic points between two successive frames, a self-move distance computing section 13 for computing a self-move distance out of an aggregate of related points, a position converting section 14 for making a coordinate transformation of the self-move distance to obtain the self-position, a position storing section 15 for storing the self-position, and a dead reckoning computing section 16 for computing a dead reckoning value, and further includes a characteristic segment extracting section 17 for extracting characteristic segments of the environmental information based on measurement data from an environment recognition sensor, a virtual characteristic point extracting section 19 for obtaining virtual characteristic points out of extracted characteristic segments, and a characteristic segment relating section 18 for relating an aggregate of characteristic segments between the two successive frames. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a self-position identification apparatus mounted on a mobile robot that measures environmental information in a facility such as an office or factory and moves using the result.

Even if a mobile robot uses a drive mechanism such as a wheel to determine and control its own position using an internal sensor such as an encoder, the position error from the target position is large due to the occurrence of slippage between the wheel and the floor. Become. For this reason, an external sensor such as a laser sensor or an ultrasonic sensor is used to measure the positional relationship with the surrounding environment, identify itself, and correct the deviation from the target position.
In the conventional mobile robot self-position identification, the surrounding environment is periodically measured using a laser sensor such as a laser range finder as an external sensor, and measurement data for one cycle is regarded as one frame, and two consecutive frames are obtained. A feature point set in each is obtained, the feature points are associated between the frames, and the mobile robot's self-movement amount (Δx, Δy) so as to minimize the sum of the square distances between the feature points associated between the frames. , Δθ), and the coordinate of the amount of self movement is converted to obtain the self position (see, for example, Non-Patent Document 1).
In addition, a line segment element is generated from the measurement data obtained by the laser sensor, the line segment element is extracted from the positional relationship between the inclination of the line segment element and the midpoint of the line segment element, and the line segment information is extracted. Some have identified their own positions by comparing with map information stored in advance (see, for example, Patent Document 1).

FIG. 9 is a flowchart showing a processing procedure for calculating the self-position in Non-Patent Document 1. This will be described in order with reference to this figure.
In step 101, feature points are extracted from the measurement data obtained by the laser sensor. The measurement data includes data obtained by measuring a wall surface far from the mobile robot in the surrounding environment. A threshold is provided to eliminate them, and a portion of the measurement data that protrudes from the background more than the threshold is used as a feature point.
In step 102, a dead reckoning value is fetched, coordinate conversion is performed on the feature point acquired this time, and an absolute position is obtained.
The absolute position is a position in the world coordinate system having a predetermined point in the environment where the mobile robot moves as the origin. On the other hand, the position in the robot coordinate system with the mobile robot as the origin is set as the relative position.
In step 103, the absolute position of the previous feature point is acquired.
In step 104, the absolute positions of the feature points are compared and the distances between the feature points that are closer than the threshold are associated with each other.
In step 105, the set of feature points P _j (X _j , Y _j ) (feature points of the previous frame) and P ′ _j (X ′ _j , Y ′ _j ) (current frame) associated in step 104 Are output as a corresponding point set of feature points.
In step 106, self-movement amounts (Δx, Δy, Δθ) are calculated by the following equations (1) to (3).

However, in the formulas (1) to (3), [•] represents the total sum when j is changed from 1 to N. Here, N is the number of associated feature point pairs.
In step 107, the self-movement amount calculated in step 106 is coordinate-transformed using the previous self-position to calculate the current self-position. Also, the absolute position of the feature point is calculated using the feature relative position that is the measurement position of the feature point and the current self-position.
In step 108, the self position calculated in step 107 and the absolute position of the feature point are stored.

FIG. 10 is a flowchart showing a processing procedure for calculating the self-position in Patent Document 1. This will be described in order with reference to this figure.
In step 201, line segment elements are generated from four adjacent measurement points, and line segment elements are extracted from the positional relationship between the inclination of the line segment elements and the midpoints of the line segment elements to extract line segment information. In addition, the distance and angle relationship between the two lines is determined, and a defect caused by occlusion is interpolated.
In step 202, two combinations are made for all line segments, and five feature quantities are extracted from the length, inclination, and positional relationship of the combined two line segments.
In step 203, a binary tree previously generated as map information is searched, and a search unit having a high degree of approximation to five feature quantities is selected. As a result, the comparison with the map is performed. Then, a position candidate is calculated from the selected search unit, the position candidate is assigned to the density map, the position of the small area with the highest density is selected as the maximum likelihood position candidate, and is identified as the self position.
As described above, the conventional self-position identification uses the external sensor to measure the positional relationship with the environment and obtain the self-position.
Japanese Patent Laid-Open No. 4-216487 (page 3-5, FIG. 4) Negishi, Miura, Shirai "Map Generation for Mobile Robots by Integration of Omnidirectional Stereo and Laser Range Finder" Journal of the Robotics Society of Japan Vol. 21 No. 6, pp.690-696, 2003

However, the self-position identification in Non-Patent Document 1 extracts corners from measurement data as feature points and associates them between frames, but there are flat walls on both sides of the mobile robot, such as a corridor. However, when the measurement data is a straight line, there is a problem that the correlation cannot be made and the self-position cannot be obtained.
Further, in the case of Patent Document 1, since a feature is obtained from measurement data and a self-position is obtained by comparison with a map, there is a problem that a map is necessary to obtain the self-position. In addition, the line segment is extracted from the measurement data to derive the feature, but the line segment element is obtained from the positional relationship of the midpoints of the line segment elements to obtain line segment information. There was also a problem that the line segment could not be extracted correctly because the result of whether or not it was possible to change. Further, although the distance and angle relationship between the two line segments is determined and the defect is interpolated, there is a problem that the feature amount cannot be obtained when the defect cannot be interpolated.
Furthermore, since the shape to be interpolated is limited to steps and corners, there is a possibility that an incorrect result is output compared to the map when the actual shape is different from the interpolated shape.
The present invention has been made in view of such problems, and not only a complicated environment with many corners, but also an environment having flat wall surfaces on both sides such as a corridor, or an environment in which they are combined, It is an object of the present invention to provide a self-position identification device capable of obtaining a self-position without using a map in a more general-purpose environment such as an environment where there is a defect due to occlusion, and a mobile robot equipped with the self-position identification device. To do.

In order to solve the above problem, the present invention is configured as follows.
According to the first aspect of the present invention, a feature point extraction unit that extracts feature points of environment information based on measurement data periodically obtained by an environment recognition sensor, and the measurement data for one period as one frame continuously. A feature point correspondence unit that associates a feature point set between two frames, a self-movement amount calculation unit that calculates a self-movement amount from the corresponding point set, a position conversion unit that obtains a self-position by performing coordinate transformation of the self-movement amount, A self-position identification device comprising a position storage section for storing the self-position and a dead reckoning calculation section for obtaining a dead reckoning value based on information from an internal sensor, based on measurement data from the environment recognition sensor. A feature line segment extracting unit for extracting feature line segments of environmental information, a virtual feature point extracting unit for obtaining a virtual feature point from the extracted feature line segment, and a feature line segment collection between the two consecutive frames. And wherein the segment corresponding unit associating, is characterized in that it comprises a.

  Further, in the invention according to claim 2, the feature line segment extraction unit divides measurement data from the environment recognition sensor into a plurality of measurement point groups, and the least square approximation of the measurement points in each measurement point group. The measurement point group is further divided to extract a line segment so that the distance between the straight line and each measurement point is within a predetermined threshold.

  Further, in the invention according to claim 3, the feature line segment correspondence unit is configured such that the degree of coincidence of inclinations of two line segments and the absolute value of the end points of the line segments in the association of the feature line segment sets between the two consecutive frames. It is characterized by evaluating the overlap between the end point and the line segment based on the position, and calculating the relative position of the corresponding point from the absolute position of the end point overlapped with the line segment and the end point of the other line segment. is there.

  In the invention according to claim 4, the evaluation of the degree of overlap between the end point and the line segment is performed by comparing the lengths of the two line segments and selecting the short line segment and the long line segment. This is performed depending on whether the distance from the minute is within a predetermined threshold.

  In the invention according to claim 5, the virtual feature point extraction unit obtains an intersection of straight lines obtained by extending a plurality of feature line segments extracted by the feature line segment extraction unit. The intersection point is extracted as a virtual feature point.

  In the invention according to claim 6, when the total of the number of the feature points and the number of the virtual feature points is larger than a predetermined constant, the feature line segment correspondence in the feature line segment correspondence unit It is characterized by not attaching.

  The invention according to claim 7 includes an azimuth sensor as the internal sensor, and in the self-movement amount, only the posture rotation amount is obtained from the information of the azimuth sensor, and the data of the current frame is calculated only by the posture rotation amount. A feature point, a feature line segment, and a virtual feature point are obtained from the rotated data and associated with the data of the previous frame, and the translation amount of each point from the corresponding point set is calculated by the self-movement amount calculation unit. The translational movement amount of the self is obtained by taking the average, and the self-movement amount is combined with the posture rotation amount.

  The invention according to claim 8 includes an azimuth sensor as the inner world sensor, and in the self-movement amount, only the posture rotation amount is obtained from the information of the azimuth sensor, and the data of the current frame is calculated only by the posture rotation amount. A feature point, a feature line segment, and a virtual feature point are obtained from the rotated data and associated with the data of the previous frame, and the translation amount of each point from the corresponding point set is calculated by the self-movement amount calculation unit. And the translational movement amount of the self is obtained by taking the mode value by voting, and the self-movement amount is combined with the posture rotation amount.

  According to a ninth aspect of the present invention, there is provided a mobile robot comprising the self-position identification device according to the first to eighth aspects.

According to the first to eighth aspects of the invention, not only the feature point but also the line segment including the virtual feature point is used together, and the virtual feature point obtains the shift amount of the measurement data of two frames. It is not necessary to match the actual shape because it is only determined as a corresponding point at the time, and even if there is a defect due to occlusion etc., the deviation amount of the measurement data of 2 frames can be accurately calculated without interpolating the shape Can do. For this reason, maps are used not only in complex environments with many corners, but also in more general-purpose environments such as corridors and other environments where there are flat walls on both sides, or a combination of these, as well as environments with defects. The self-position can be obtained without.
According to the ninth aspect of the present invention, it is possible to realize a mobile robot that can move within the environment by obtaining a self-position without using a map by the self-position identification device.

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

  FIG. 1 is a diagram showing an internal configuration of the self-position identification device of the present invention. The self-position identification device 2 is, for example, a computer incorporating a CPU and a memory, and includes a feature point extraction unit 11, a feature point correspondence unit 12, a self-movement amount calculation unit 13, a position conversion unit 14, a position storage unit 15, and dead reckoning. The calculation unit 16, the feature line segment extraction unit 17, the feature line segment correspondence unit 18, and the virtual feature point extraction unit 19 input information such as an environment recognition sensor to calculate and output the mobile robot's own position.

The feature point extraction unit 11 recognizes the environment shape around the mobile robot based on information from the environment recognition sensor and extracts feature point sets such as corners.
The feature point set is extracted according to this period. If the feature point information obtained periodically is one frame, the feature point correspondence unit 12 stores the feature point set for each frame, and the feature point set between two consecutive frames in time is the absolute value of the feature point. Assess and correlate position.
The self-movement amount calculation unit 13 uses the correspondence point set output from the feature point correspondence unit 12 to minimize the sum of the square distances between the feature points associated between the frames. Calculate The position conversion unit 14 performs coordinate conversion of the self-movement amount output from the self-movement amount calculation unit 13 to calculate the self position.
Also, for the feature point, virtual feature point, and feature line segment end point of the current frame, the relative position (hereinafter referred to as feature relative position) is input, and the feature point, virtual feature point, The absolute position of the feature line segment end point (hereinafter referred to as the feature position) is also calculated.

The position storage unit 15 stores the self position and the feature position output from the position conversion unit 14. The dead reckoning calculation unit 16 calculates a dead reckoning value by adding the movement amount obtained from the internal sensor information by an encoder or the like to the self position output from the position storage unit 15.
Similar to the feature point extraction unit 11, the feature line segment extraction unit 17 recognizes the environment shape based on the information of the environment recognition sensor, but extracts a feature line segment set such as a wall surface portion instead of the feature point set.
Like the feature point correspondence unit 12, the feature line segment correspondence unit 18 associates the feature line segment set between two consecutive frames by evaluating the inclination of the line segment and the absolute position of the end point of the line segment.
The virtual feature point extraction unit 19 extracts two adjacent feature line segments from the feature line segment set extracted from the feature line segment extraction unit 17, and obtains intersections of straight lines obtained by extending the two line segments. The obtained intersection is output as a virtual feature point.

2A and 2B are diagrams showing the internal configuration of a mobile robot equipped with the self-position identification device 2 of the present invention. The mobile robot 1 includes a self-position identification device 2, a mobile robot control device 3, a motor 4, an encoder 5, a drive mechanism 6, and an environment recognition sensor 7. The mobile robot control device 3 acquires the self-position calculated by the self-position identification device 2, performs command generation and servo control, and controls the drive of the motor 4. The motor 4 drives the drive mechanism 6. The encoder 5 measures the amount of rotation of the motor 4 and generates internal sensor information. The inner world sensor information may be applied by incorporating not only an encoder but also an orientation sensor such as a gyro sensor in the mobile robot 1. The drive mechanism 6 has a plurality of sets of wheels or steering shafts connected to the output shaft of the motor 4 and drives the mobile robot 1.
The environment recognition sensor 7 is a two-dimensional scanning distance sensor such as a two-dimensional laser range finder, for example, and a fan-shaped laser beam centered on the sensor mounting position in a plane that is not tilted to the left or right (scan plane). To measure the distance to surrounding objects such as walls. In such a configuration, the self-position identification device 2 takes in information from the environment recognition sensor 7 and the encoder 5, calculates the self-position, and generates information necessary for controlling the mobile robot 1. .
The present invention is different from the prior art in that the self-position identification device 2 is provided with a feature line segment extracting unit 17, a feature line segment corresponding unit 18, and a virtual feature point extracting unit 19.

FIG. 3 is a flowchart showing a processing procedure for calculating the self-position using the feature line segment in the self-position identification device 2. The processing in the present invention will be described step by step with reference to this figure.
For the sake of simplicity, it is assumed that the sensor coordinate system having the environment recognition sensor 7 as the origin is the same as the robot coordinate system.
In step 1, the feature line segment extraction unit 17 extracts a straight line by dividing a plurality of measurement points by the environment recognition sensor 7 into line segments by the following method.
First, the distance between adjacent measurement points is evaluated and divided into continuous measurement point groups. That is, when the distance between adjacent measurement points is large, it is assumed that the measurement object is discontinuous between the adjacent measurement points, and the measurement point sequence is disassembled into a measurement point group.
FIG. 4 shows how the measurement point sequence is decomposed into measurement point groups. When the object 1 and the object 2 are measured by the environment recognition sensor 7, the distance between the measurement points increases between the object 1 and the object 2, so that the measurement point sequence is decomposed into the measurement point groups 1 and 2. .
Subsequently, line segments are extracted for each measurement point group. To extract a line segment, first, a least square approximation straight line for successive measurement points is generated, and it is evaluated whether the maximum distance from the measurement point is within a predetermined threshold. If the maximum distance is within the threshold value, the same evaluation as described above is performed by merging with adjacent measurement points. If the distance between the measurement point and the least square approximation straight line exceeds the threshold value, the measurement point is divided at the measurement point before the measurement point, and the next line segment is extracted starting from the measurement point. By repeating such division, a line segment is extracted in the measurement point group.

FIGS. 5A and 5B are diagrams for explaining how line segments are extracted from a certain group of measurement points. In FIG. 5A, since the distance between the measurement point 2 and the least square approximation line exceeds the threshold value, the measurement point group is divided at the previous measurement point 1, and a part of the least square approximation line at that time is segmented. Extract as
Then, a new line segment is extracted starting from the measurement point 2. In this way, line segment 1 and line segment 2 as shown in FIG. 5B are extracted. The end point of the line segment is the intersection point when the line segments intersect with each other by extending the adjacent line segment, and the measurement point that is the start point or end point of the division when not intersecting or the start point or end point of the measurement point group.

In step 2, two adjacent feature line segments are extracted by the virtual feature point extraction unit 19 from the feature line segment set extracted by the feature line segment extraction unit 17, and the intersections of the straight lines obtained by extending the two line segments are obtained. Ask. The obtained intersection is output to the feature point correspondence unit 12 as a virtual feature point.
6 (a) and 6 (b) are diagrams for explaining how virtual feature points are extracted. In FIG. 6 (a), there is a defect between line segment 1 and line segment 2, but there is no feature point. The virtual feature point is obtained by obtaining the intersection of the straight lines obtained by extending the respective line segments. 6B, the virtual feature points are obtained in the same manner as in FIG. 6A, but the virtual feature points are not on the missing portion on the line segment 4 as shown in FIG. 6B. May be. If the sum of the number of feature points extracted from the feature point extraction unit 11 and the number of virtual feature points is larger than a predetermined constant, the processing from step 4 to step 6 is not performed.

In step 3, a dead reckoning value is taken from the dead reckoning calculation unit 16, coordinate conversion is performed on the end points and virtual feature points of the line segment acquired this time, and an absolute position is obtained.
In step 4, the previous feature position, that is, the absolute position of the end point of the previous feature line segment set is acquired from the position storage unit 15.
In step 5, the feature line segment correspondence unit 18 associates the line segment sets between the two frames based on the degree of coincidence of the line segment inclinations and the end point positions obtained in steps 3 and 4. To evaluate.

FIG. 7 is an explanatory diagram for associating line segment sets. The xy coordinate system in FIG. 7 is a world coordinate system that serves as a reference for the absolute position described above.
First, one line segment is extracted from the line segment set of each frame, and the lengths of the two extracted line segments are compared to select the longer one as the long line segment and the shorter one as the short line segment. Thereafter, the degree of coincidence of the slopes of the long line segment and the short line segment and the degree of overlap between the long line segment and the short line segment are evaluated. As shown in FIG. 7, the degree of coincidence of the slopes of the line segments is evaluated according to the following equation, assuming that the angles formed with the x axis are θ _l and θ _s respectively, and the threshold is e ₁ .

In the evaluation of the overlapping state of the line segments, it is determined by the following formula whether the distance r between the end point of the short line segment and the long line segment is smaller than the threshold value e _{2 as} shown in FIG. However, when the perpendicular line dropped from the end point of the short line segment to the long line segment does not intersect with the long line segment (when there is no intersection point on the long line segment), the determination is not performed on the end point.

In the above expression, x ₁₄ and y ₁₄ are the x coordinate and y coordinate of the end point 4, and xi and yi (i = 0, 1) are the x coordinate and y coordinate of the end point of the short line segment. The determination is performed with respect to both end points of the short line segment, and it is evaluated that the long line segment and the short line segment overlap if at least one satisfies the conditions of equations (5) and (6).
Note that the long line segment and the short line segment have a positional relationship in which the portions overlap each other exactly as shown in FIG. 8, but in FIG. 7, the explanation about each end point and the distance r of the long line segment and the short line segment is easy to understand. In order to enlarge, the long line segment and the short line segment are drawn apart.

Subsequently, in step 6, the relative position (robot coordinate system) of the corresponding point (end point 1 'in FIG. 8) on the long line segment from the absolute position of the end point of the short line segment that overlaps the long line segment and the absolute position of both end points of the long line segment. Position).
FIG. 8 is an explanatory diagram for obtaining the position of the corresponding point, which will be described based on this diagram. Since the absolute position of the end point 1 ′ corresponding to the end point 1 is the same, the end point 1 ′ is on the long line segment from the absolute position of the end point 1 overlapping the long line segment and the end points 3 and 4 of the long line segment as shown in FIG. 8. Wherein, that is, the straight line parameter t in the long line expression expressed by Expression (7) is obtained by Expression (8).

  Here, p ′ is a vector representing the absolute position of the end point 1 ′, d ′ is a position vector of the end point 3 (absolute position) viewed from the end point 4 (absolute position), a ′ is a vector representing the absolute position of the end point 4, e ′ Represents the position vector of the end point 1 (absolute position) viewed from the end point 4 (absolute position), and || represents the magnitude of the vector. After that, since the relative position of each end point is known from the measurement data of the environment recognition sensor, the relative position of the end point 1 'is obtained from the straight line parameter t obtained by the equation (8) by the following equation.

Here, p is a vector representing the relative position of the end point 1 ′, d is a position vector of the end point 3 (relative position) viewed from the end point 4 (relative position), and a is a vector representing the relative position of the end point 4.
Then, by making the previous value and the current value correspond to each other between the end point 1 and the end point 1 ′, a set of corresponding points P _j (X _j , Y _j ) (previous value) and P ′ _j (X ′) _j , Y ′ _j ) (current value) is obtained as a corresponding point set of feature line segments.
Specifically, if the end point 1 is originally data of the previous value, the end point 1 is additionally registered in P, and the end point 1 ′ is additionally registered in P ′. If the end point 1 is the current value, the end point 1 is registered in P ′ and the end point 1 ′ is registered in P.

In step 7, the corresponding point set obtained in step 6 is output as a corresponding point set in addition to the corresponding point set of the feature points and the virtual feature points. In step 8, self-movement amounts (Δx, Δy, Δθ) are calculated by equations (1) to (3).
In step 9, the position conversion unit 14 performs coordinate conversion of the self-movement amount calculated in step 8 using the previous self-position, and calculates the current self-position (absolute position). Also, the feature position is calculated using the current feature relative position and the self-position.
In step 10, the self position and feature position calculated in step 9 are stored in the position storage unit 15.

  As described above, according to the present invention, the feature line segment extraction unit 17, the feature line segment correspondence unit 18, and the virtual feature point extraction unit 19 handle not only the feature points but also the line segments including the virtual feature points. Because the virtual feature point is only obtained as a corresponding point when calculating the deviation of the measurement data of two frames, it does not need to match the actual shape, and even if there is a defect due to occlusion, the shape is interpolated Without this, the deviation amount of the measurement data of the two frames can be accurately obtained. Therefore, not only in a complicated environment with many corners, but also in a more general-purpose environment, such as an environment with flat walls on both sides such as a corridor, an environment where they are combined, or an environment with defects, without using a map Self-position can be obtained.

In the first embodiment, the self-movement amount calculation unit 13 calculates the self-movement amount (Δx, Δy, Δθ) by the equations (1) to (3) using the corresponding point set output from the feature point correspondence unit 12. However, in this embodiment, it is configured to include an orientation sensor such as a gyro, and the posture rotation amount Δθ is a value of the orientation θ acquired from the orientation sensor, that is, a value at the previous frame measurement and a value at the current frame measurement. The feature point, the feature line segment, and the virtual feature point are obtained from the data obtained by rotating the data of the current frame by Δθ, and are associated with the data of the previous frame.
Thereafter, the translation amount of each point (from the corresponding point set P _j (X _j , Y _j ) (previous value)) and P ′ _j (X ′ _j , Y ′ _j ) (current value) is calculated by the self-movement calculating unit 13. ΔX _j , ΔY _j ) = (X _j −X ′ _j , Y _j −Y ′ _j ) is obtained, and the translational movement amounts Δx and Δy of its own are obtained by taking an average as in the following equation, and combined with Δθ The amount of self movement.

However, in Expressions (10) and (11), [•] represents the total sum when j is changed from 1 to N. Here, N is the number of associated feature point pairs.
Further, instead of obtaining a translation amount of the self-taking average translational movement amount of each point (ΔX _j, ΔY _j) accuracy, and integer a change in units according to the processing speed, integer been [Delta] X _j , ΔY _j , the translation amount of the self may be obtained by taking a mode value by voting.

Device configuration diagram of self-position identification device of the present invention Device configuration diagram of a mobile robot equipped with the self-position identification device of the present invention The flowchart which shows the process sequence of the self-position identification apparatus of this invention. Explanatory drawing which decomposes | disassembles the measurement point sequence of this invention into a measurement point group Explanatory drawing which extracts a line segment from the measurement point group of this invention Explanatory drawing which extracts the virtual feature point of this invention Explanatory drawing of line segment set matching of this invention Explanatory drawing which calculates | requires the position of the corresponding end point of this invention The flowchart which shows the process sequence of the self-position identification calculation of the prior art example 1. The flowchart which shows the process sequence of the self-position identification calculation of the prior art example 2.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Mobile robot 2 Self-position identification apparatus 3 Mobile robot control apparatus 4 Motor 5 Encoder 6 Drive mechanism 7 Environment recognition sensor 11 Feature point extraction part 12 Feature point correspondence part 13 Self-movement amount calculation part 14 Position conversion part 15 Position storage part 16 Dead Reckoning calculation unit 17 Feature line segment extraction unit 18 Feature line segment correspondence unit 19 Virtual feature point extraction unit

Claims (9)

A feature point extraction unit that extracts feature points of environment information based on measurement data periodically obtained by the environment recognition sensor;
A feature point corresponding unit that associates a set of feature points between two consecutive frames, with the measurement data for one period as one frame,
A self-movement amount calculation unit for calculating a self-movement amount from the corresponding point set;
A position conversion unit that determines the self position by performing coordinate conversion of the self-movement amount;
A position storage unit for storing the self-position;
In the self-position identification device provided with a dead reckoning calculation unit for obtaining a dead reckoning value based on information from the internal sensor,
A feature line segment extraction unit for extracting feature line segments of environment information based on measurement data from the environment recognition sensor;
A virtual feature point extraction unit for obtaining a virtual feature point from the extracted feature line segment;
A feature line segment corresponding unit that associates a set of feature line segments between the two consecutive frames;
A self-position identification device comprising:   The feature line segment extraction unit divides measurement data from the environment recognition sensor into a plurality of measurement point groups, and within each measurement point group, a distance between the least square approximation straight line of the measurement points and each measurement point is a predetermined distance. 2. The self-position identification apparatus according to claim 1, wherein the measurement point group is further divided so as to be within a threshold value and a line segment is extracted. The feature line segment correspondence unit overlaps the end points and the line segments based on the degree of coincidence of the inclinations of the two line segments and the absolute position of the end points of the line segments in the association of the feature line segment sets between the two consecutive frames. Evaluate the condition,
2. The self-position identification device according to claim 1, wherein the relative position of the corresponding point is obtained from the absolute position of the end point overlapped with the line segment and the end point of the other line segment.   The evaluation of the degree of overlap between the end point and the line segment is made by comparing the lengths of the two line segments and selecting the short line segment and the long line segment, and the distance between the end point of the short line segment and the long line segment is within a predetermined threshold. 4. The self-position identification device according to claim 3, wherein the self-position identification device is performed depending on whether or not.   The virtual feature point extraction unit obtains intersections of straight lines obtained by extending a plurality of feature line segments extracted by the feature line segment extraction unit. The self-position identification apparatus according to claim 1, wherein the intersection is extracted as a virtual feature point.   The feature line segment is not associated in the feature line segment correspondence unit when the sum of the number of feature points and the number of virtual feature points is greater than a predetermined constant. The self-position identification apparatus according to 1. An azimuth sensor is provided as the internal sensor, and in the amount of self-movement, only the posture rotation amount is obtained from the information of the azimuth sensor, and feature points and feature line segments are obtained from data obtained by rotating the data of the current frame by the posture rotation amount. , And the virtual feature point is determined and associated with the data of the previous frame,
In the self-movement amount calculation unit, the translational movement amount of each point is obtained from the corresponding point set, and the translational movement amount of the self is obtained by taking an average, and the self-movement amount is combined with the posture rotation amount. The self-position identification device according to any one of claims 1 to 6. An azimuth sensor is provided as the internal sensor, and in the amount of self-movement, only the posture rotation amount is obtained from the information of the azimuth sensor, and feature points and feature line segments are obtained from data obtained by rotating the data of the current frame by the posture rotation amount. , And the virtual feature point is determined and associated with the data of the previous frame,
In the self-moving amount calculation unit, a translational movement amount of each point is obtained from the corresponding point set, and a self-translational movement amount is obtained by taking a mode value by voting. The self-position identification device according to any one of claims 1 to 6, wherein   A mobile robot comprising the self-position identification device according to claim 1.