JP2012088135A - Calibration apparatus, calibration program and calibration method for distance sensor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a calibration apparatus, a calibration program and a calibration method for a distance sensor that can calibrate positions and directions of a plurality of distance sensors arranged in a certain environment without being provided with an initial value.SOLUTION: A measurement apparatus 10 includes a computer 12, and a plurality of distance sensors 14 are connected to the computer 12. The computer 12 detects a person observation point in each sensor 14 based on an output from each sensor 14, calculates a human movement locus from a temporal change of the person observation points, and then makes the movement loci calculated based on the output from the respective sensors 14 coincide with each other among the sensors 14. The computer extracts two person observation points on the coincident movement locus in accordance with a predetermined rule, and calculates restrictions between the sensors 14 about a distance and a relative angle between the sensors 14 generating the movement locus for each group of sensors 14 whose movement loci are made coincident, by using the extracted person observation points. Then, positions and directions of all the sensors 14 are estimated by using the restrictions between the sensors 14, and the estimated positions are adjusted.

Description

  The present invention relates to a distance sensor calibration apparatus, a calibration program, and a calibration method, and in particular, a distance sensor calibration apparatus that calibrates the positions and orientations of two or more distance sensors arranged in a certain environment, for example. The present invention relates to a calibration program and a calibration method.

  An example of a conventional distance sensor calibration device of this type is disclosed in Patent Document 1. In the moving body estimated position system disclosed in Patent Document 1, the distances to the moving bodies are measured by three or more distance sensors that are arranged in a measurement region so as to be distributed to each other. The sensor measurement value at an arbitrary time is acquired and accumulated from each distance sensor. A distance reliability indicating the degree of reliability corresponding to the distance is given to the sensor measurement value of each acquired and accumulated distance sensor, and a highly reliable measurement value among the acquired and accumulated sensor measurement values is adopted. The positions of the distance sensor and the moving body are estimated. The estimated position processing is performed by calibrating the position of each distance sensor and estimating the moving position of the moving body using sensor measurement values obtained from two or more distance sensors at positions before and after the moving body moves. .

JP 2010-127650 [G01S 5/14]

  However, in the moving body estimated position system of Patent Document 1, when calculating each sensor position, an initial value of each sensor position is required. For example, the initial value is measured every time the measurement environment changes. It is necessary and troublesome.

  Therefore, a main object of the present invention is to provide a novel distance sensor calibration apparatus, calibration program, and calibration method.

  Another object of the present invention is to provide a distance sensor calibration apparatus and calibration program capable of estimating the positions and orientations of a plurality of distance sensors arbitrarily arranged in a certain environment without giving an initial value. And providing a calibration method.

  The present invention employs the following configuration in order to solve the above problems. The reference numerals in parentheses, supplementary explanations, and the like indicate correspondence relationships with embodiments described later to help understanding of the present invention, and do not limit the present invention in any way.

  A first invention is a distance sensor calibration device that is arranged in a certain environment and calibrates the positions and orientations of a plurality of distance sensors that detect distances at a certain angle. Based on the output, for each distance sensor, the human position detecting means for detecting the position of the human in a certain environment, and based on the human position detected by the human position detecting means, for each distance sensor, A trajectory generating means for generating a moving trajectory, a trajectory identifying means for identifying a moving trajectory generated for each distance sensor by the trajectory generating means between two different distance sensors based on the characteristics of the moving trajectory, and a trajectory identifying means. In the identified movement trajectory, based on two arbitrarily selected human positions, the distance between the two distance sensors and the relative distance between the two distance sensors. A sensor constraint calculation unit that calculates a sensor constraint for a direction for each set of two distance sensors that generated the identified movement trajectory, and a sensor constraint calculated by the sensor constraint calculation unit. It is a distance sensor calibration apparatus provided with the estimation means which estimates the position and direction of several distance sensors.

  In the first invention, the distance sensor calibration device (12) is arranged in a certain environment, and calibrates the positions and orientations of the plurality of distance sensors (14) that detect distances at fixed angles. The calibration apparatus includes a human position detecting means (12, S3), a locus generating means (12, S5), a locus identifying means (12, S7), a sensor constraint calculating means (12, S11), and an estimating means (12, S15). The human position detecting means detects the position of a human in a certain environment for each distance sensor based on the outputs of the plurality of distance sensors. The trajectory generating means generates a human trajectory for each distance sensor based on the human position detected by the human position detecting means. That is, the time change of the human position is generated as the movement trajectory. The trajectory identification means identifies the movement trajectory generated for each distance sensor by the trajectory generation means between two different distance sensors based on the characteristics of the movement trajectory. For example, with respect to two different movement loci, it is determined that the movement loci whose speed changes approximate to each other match. The sensor constraint calculation unit is configured to determine a distance between the two distance sensors and a relative direction of the two distance sensors based on two arbitrarily selected human positions in the movement locus identified by the locus identification unit. Is calculated for each set of two distance sensors that have generated the identified movement trajectory. The estimation unit estimates the positions and orientations of a plurality of distance sensors in a certain environment using the sensor constraint calculated by the sensor constraint calculation unit. For example, a plurality of distance sensors are sequentially arranged on a virtual map so as to maintain the distance between the distance sensors and the relative orientation.

  According to the first invention, since the positions and orientations of the plurality of distance sensors are estimated based on the outputs of the plurality of distance sensors, the position of the distance sensor arranged in a certain environment without giving an initial value. And the orientation can be calibrated.

  A second invention is dependent on the first invention, and the sensor constraint calculation means changes at least one of two arbitrarily selected human positions in the movement locus identified by the locus identification means. Thus, sensor constraint averaging means for calculating a plurality of sensor constraints for each set of two distance sensors and averaging the plurality of sensor constraints is included.

  In the second invention, the sensor constraint calculation means includes sensor constraint averaging means (12, S165). The sensor constraint averaging means changes a plurality of sensors for each set of two distance sensors by changing at least one of two arbitrarily selected human positions in the movement locus identified by the locus identification means. Calculate constraints and average multiple sensor constraints. That is, a plurality of sensor constraints of two distance sensors are calculated, and an average value thereof is obtained. It is calculated for each set of two distance sensors.

  According to the second invention, since the average of sensor constraints is obtained, the position of a plurality of distance sensors can be estimated in consideration of the error in the distance between the distance sensors.

  A third invention is according to the second invention, further comprising a variance calculating unit that calculates variances of a plurality of sensor constraints, and the estimating unit includes two distance sensors that minimize the variance calculated by the variance calculating unit Are placed on the virtual map in accordance with the average of the sensor constraints of the two distance sensors, and the two distance sensors arranged by the reference sensor placement means are arranged in the order of small variance. Sequential arrangement means for sequentially arranging distance sensors having an average of sensor constraints with the distance sensor on a virtual map according to the average of the sensor constraints.

  In the third invention, the calibration apparatus further includes a variance calculating means (12, S165). The estimation means includes reference sensor arrangement means (12, S183) and sequential arrangement means (12, S189, S191, S193, S195). The variance calculation means calculates the variance of the plurality of sensor constraints. That is, the distance error between the distance sensors is calculated. The reference sensor arrangement unit arranges the two distance sensors having the smallest variance calculated by the variance calculation unit on the virtual map according to the average of the sensor constraints of the two distance sensors. That is, two distance sensors having the most reliable sensor constraint are arranged. The sequential arrangement means, after the two distance sensors arranged by the reference sensor arrangement means, in order of decreasing variance, the distance sensors having the average of sensor constraints with the already arranged distance sensors are virtually calculated according to the average of the sensor constraints. Place them sequentially on the map. That is, the distance sensors having high reliability and having sensor restrictions with the already arranged distance sensors are sequentially arranged.

  According to the third aspect, since the distance sensors are arranged on the virtual map in descending order of the average reliability of the sensor constraints, the accuracy of the estimated position of the distance sensor can be increased.

  A fourth invention is dependent on any one of the first to third inventions, and is based on each of the distances of each set of distance sensors indicated by the sensor constraint and the positions of a plurality of distance sensors estimated by the estimating means. Difference calculating means for calculating the difference between each distance of each pair of distance sensors, minimum value determining means for determining whether the total value of the differences calculated by the difference calculating means is the minimum value, and minimum value determining means Position adjusting means for adjusting the position of each of the plurality of distance sensors based on the position of each of the other plurality of distance sensors and the sensor constraint estimated by the estimating means until the sum of the differences becomes the minimum value by Further prepare.

  In the fourth invention, the calibration apparatus further includes a difference calculating means (12, S255), a minimum value determining means (12, S225), and a position adjusting means (12, S215, S217). The difference calculation means calculates a difference between each distance of each pair of distance sensors indicated by the sensor constraint and each distance of each pair of distance sensors based on the positions of the plurality of distance sensors estimated by the estimation means. . The minimum value determining means determines whether or not the sum of the differences calculated by the difference calculating means is the minimum value. The position adjusting means determines the position of each of the plurality of distance sensors, the position of each of the other plurality of distance sensors estimated by the estimating means, and the sensor constraints until the total value of the differences becomes the minimum value by the minimum value determining means. Adjust based on.

  According to the fourth invention, the position of the distance sensor is adjusted so that an error between the distance between the distance sensors indicated by the sensor constraint and the distance between the distance sensors based on the estimated position of the distance sensor is minimized. Therefore, the position of the distance sensor can be estimated as accurately as possible.

  A fifth invention is dependent on the third invention, and based on the variance calculated by the variance calculation means, a position error calculation means for calculating an error of each position of the plurality of distance sensors estimated by the estimation means, An error average value calculating means for calculating an average value of errors at positions of each of the plurality of distance sensors calculated by the position error calculating means, and whether or not the average value calculated by the error average value calculating means is equal to or less than a predetermined value. The human position detection means, the trajectory generation means, the trajectory identification means, the sensor constraint calculation means, and the estimation means are repeated until the position error determination means and the position error determination means determine that the average value is less than or equal to the predetermined value. It further includes a repeat control means for functioning.

  In the fifth invention, the calibration apparatus further includes position error calculation means (12, S185, S197), error average value calculation means (12), position error determination means (12, S19), and repetition control means (12). Prepare. The position error calculation unit calculates an error in the position of each of the plurality of distance sensors estimated by the estimation unit based on the variance calculated by the variance calculation unit. That is, the position of the distance sensor includes an error (position error) due to the distance error between the sensors, and the distance error between the sensors is accumulated every time the distance sensor is arranged. . The error average value calculation means calculates the average value of the error of each position of the plurality of distance sensors calculated by the position error calculation means. The position error determination means determines whether the average value calculated by the error average value calculation means is equal to or less than a predetermined value. The iterative control means causes the human position detection means, the trajectory generation means, the trajectory identification means, the sensor constraint calculation means, and the estimation means to repeatedly function until the position error judgment means determines that the average value is equal to or less than a predetermined value.

  According to the fifth aspect, since the position of the distance sensor is estimated and adjusted until the average value of the position errors of the distance sensor is equal to or less than the predetermined value, the position of the distance sensor can be accurately calibrated. Can do.

  A sixth invention is a calibration program for a calibration apparatus for a distance sensor, which is arranged in a certain environment and calibrates the positions and orientations of a plurality of distance sensors that detect distances at a certain angle, A human position detecting step for detecting a human position in a certain environment based on outputs of a plurality of distance sensors based on the outputs of the plurality of distance sensors; and a human position detected by the human position detecting step. Based on the trajectory generation step for generating the human trajectory for each distance sensor, the movement trajectory generated for each distance sensor by the trajectory generation step is determined based on the characteristics of the travel trajectory. Select any of the trajectory identification step identified in step 1 and the movement trajectory identified in the trajectory identification step. Based on the two positions of the two human beings, the sensor constraints on the distance between the two distance sensors and the relative orientation of the two distance sensors, and the set of two distance sensors that generated the identified movement trajectory A calibration program that executes a sensor constraint calculation step that is calculated every time, and an estimation step that estimates the positions and orientations of a plurality of distance sensors in a certain environment using the sensor constraint calculated by the sensor constraint calculation step. .

  A seventh aspect of the invention is a distance sensor calibration method for calibrating the positions and orientations of a plurality of distance sensors that are arranged in a certain environment and detect a distance at a certain angle. Based on the output of the distance sensor, the position of a person in a certain environment is detected for each distance sensor, and (b) based on the position of the person detected in step (a), (C) identifying a movement trajectory generated for each distance sensor in step (b) between two different distance sensors based on the characteristics of the movement trajectory; In the movement trajectory identified by (c), the distance between the two distance sensors and the relative direction of the two distance sensors are determined based on the positions of two arbitrarily selected humans. A constraint is calculated for each set of two distance sensors that generated the identified movement trajectory, and (e) a plurality of distance sensors in a certain environment using the sensor constraint calculated in step (d) This is a distance sensor calibration method comprising estimation means for estimating the position and orientation of the distance sensor.

  In the sixth and seventh inventions, similarly to the first invention, the position and orientation of the distance sensor arranged in a certain environment can be calibrated without giving an initial value.

  According to the present invention, movement trajectories detected by different distance sensors are identified, and sensor restrictions on the distance and relative angle between the two distance sensors using two or more different human positions on the identified movement trajectories. Since the position and orientation of a plurality of distance sensors are estimated using sensor constraints for each set of distance sensors, a plurality of distance sensors arbitrarily arranged in a certain environment without giving an initial value Can be calibrated for position and orientation.

  The above object, other objects, features and advantages of the present invention will become more apparent from the following detailed description of embodiments with reference to the drawings.

FIG. 1 is an illustrative view showing one example of a measuring apparatus of the present invention. FIG. 2 is an illustrative view showing a detection range of the distance sensor 1 shown in FIG. FIG. 3 is an illustrative view showing an arrangement state of a plurality of distance sensors shown in FIG. 1 and their entire detection ranges. FIG. 4 is an illustrative view showing an example of sensor information detected by each distance sensor shown in FIG. 1 and a graph showing the distribution of the number of distances detected for a certain angle obtained from the result of the background scanning process. FIG. 5 is an illustrative view for explaining the outline of the background scanning process and an illustrative view for explaining a method of detecting a human observation point. FIG. 6 is an illustrative view showing a human observation point list for a person detected by each distance sensor shown in FIG. 1 and a local locus list for a movement locus detected by each distance sensor. FIG. 7 is an illustrative view showing an example of a movement locus detected by two distance sensors. FIG. 8 is a graph showing the speed change of one moving track of interest among the moving tracks detected by a certain distance sensor and the speed changes of all the moving tracks detected by other distance sensors, and their speeds. It is a graph which shows a difference. FIG. 9 is an illustrative view showing an association table in which movement loci whose speed differences are equal to or less than a threshold among movement loci detected by different distance sensors, and sharing of shared human observation points detected by different distance sensors It is an illustration figure which shows a human observation point list | wrist. FIG. 10 is an illustrative view showing parameters for a shared human observation point and two different distance sensors that detect the shared observation point, and an illustrative view showing an example in which a restriction between sensors cannot be calculated. FIG. 11 is an illustrative view showing an example of a case where the constraint between distance sensors can be calculated, and an illustrative view for explaining the calculation method. FIG. 12 is an illustrative view showing a constraint list indicating constraints between distance sensors and dispersion of estimated values. FIG. 13 is an illustrative view for explaining an arrangement method in a case where all sensors are arranged using the restriction between distance sensors. FIG. 14 is an illustrative view showing a sensor position list indicating an estimated position, an angle, and an error between the distance sensors in the global coordinate system for the arranged distance sensors. FIG. 2 is an illustrative view showing a memory map of a memory built in the computer shown in FIG. 1. FIG. 16 is an illustrative view showing specific contents of the data storage area shown in FIG. FIG. 17 is a flowchart showing calibration processing of the computer (CPU) shown in FIG. FIG. 18 is a flowchart showing background scanning processing of the computer (CPU) shown in FIG. FIG. 19 is a flowchart showing human position detection processing of the computer (CPU) shown in FIG. FIG. 20 is a flowchart showing local locus generation processing of the computer (CPU) shown in FIG. FIG. 21 is a flowchart showing a part of the locus matching process of the computer (CPU) shown in FIG. FIG. 22 is another part of the locus matching process of the computer (CPU) shown in FIG. 1, and is a flowchart subsequent to FIG. FIG. 23 is a flowchart showing a part of the process of creating a shared human observation point list of the computer (CPU) shown in FIG. FIG. 24 is another part of the process of creating the shared human observation point list of the computer (CPU) shown in FIG. 1, and is a flowchart subsequent to FIG. FIG. 25 is a flowchart showing a calculation process of constraints between sensors of the computer (CPU) shown in FIG. FIG. 26 is a flowchart showing a part of the arrangement processing of all the sensors of the computer (CPU) shown in FIG. FIG. 27 is another part of the arrangement process of all the sensors of the computer (CPU) shown in FIG. 1, and is a flowchart subsequent to FIG. FIG. 28 is a flowchart showing sensor position adjustment processing of the computer (CPU) shown in FIG. FIG. 29 is a flowchart showing a part of an overall error calculation process of the computer (CPU) shown in FIG. FIG. 30 is another part of the calculation process of the total error of the computer (CPU) shown in FIG. 1, and is a flowchart subsequent to FIG. FIG. 31 is an illustrative view for explaining another method for identifying a movement trajectory detected by two different sensors. FIG. 32 is an illustrative view for explaining another method for identifying a movement locus detected by two different sensors.

  With reference to FIG. 1, a measuring apparatus 10 of this embodiment includes a computer 12. The computer 12 is a general-purpose computer that functions as a device that tracks (detects) a human and also functions as a calibration device for a distance sensor (hereinafter simply referred to as “sensor”) 14. As can be seen from FIG. 1, a plurality of (six in this embodiment) sensors 14 a, 14 b, 14 c, 14 d, 14 e, and 14 f are connected to the computer 12.

  Hereinafter, in this embodiment, when it is not necessary to identify the sensors 14a to 14f, they are simply referred to as “sensor 14”.

  In the measurement apparatus 10 of this embodiment, six sensors 14 are provided to track the subject in a certain region (see FIG. 3), but this is merely an example. Therefore, the number of sensors 14 is appropriately changed according to the size and shape of the region and the position and number of obstacles. Although detailed description is omitted, since the measuring apparatus 10 of this embodiment measures the moving direction of the subject, at least two sensors 14 are necessary.

  The computer 12 is serially connected to the sensor 14 via an interface such as RS232C. In this embodiment, all the sensors 14 are connected to one computer 12, but may be shared by two or more computers connected by a network such as a LAN. In such a case, any one computer may function as a central computer.

  The sensor 14 is, for example, a laser range finder, and measures the distance to the object from the time it takes to irradiate the laser and reflect it back to the object. For example, the laser range finder (sensor 14) reflects a laser beam emitted from a transmitter (not shown) by a rotating mirror (not shown) and scans the front in a fan shape at a certain angle (for example, 0.5 degrees). To do.

  Here, a distance sensor (model LMS 200) manufactured by SICK can be used as the sensor 14. When this distance sensor is used, a distance of 8000 mm can be measured with an error of about ± 15 mm. Specifically, as shown in FIG. 2, the detection range of the sensor 14 is indicated by a semicircular shape (fan shape) having a radius R (R≈8000 mm). That is, the sensor 14 can measure a 90 ° right and left direction within a predetermined distance (R) when the front direction is the center. However, assuming that the front direction is 0 °, it is a positive direction of the angle at which the leftward direction is detected from the front direction, and is a negative direction of the angle at which the rightward direction is detected from the front direction.

  The laser used is a class 1 laser in the Japanese Industrial Standard JIS C 6802 “Safety Standards for Laser Products”, which is a safe level that does not affect human eyes. For example, the sampling rate of the sensor 14 is set to 37 Hz. This is for tracking (detecting) a person who moves by walking or the like.

  Further, as will be described later, the sensor 14 is disposed at an arbitrary position where a certain area can be detected. Specifically, each of the sensors 14 is arranged so that detection areas of at least two or more sensors 14 overlap each other, and although not illustrated, the sensors 14 are fixed to a height of about 900 mm from the floor surface. This height is for enabling detection of a human torso, and is calculated from, for example, the average height of a Japanese adult. Therefore, the height at which the sensor 14 is fixed may be appropriately changed according to the place (region or country) where the measurement device 10 is provided and the age or age of the subject (human) (for example, children or adults).

  As described above, the sensor 14 is disposed at an arbitrary position where a certain area can be detected. Usually, in such a measuring apparatus 10, the entire detection range (total detection range) is set by actually measuring the position and orientation of each sensor 14, or each sensor 14 is arranged at a predetermined position and orientation. In either case, it is troublesome. Furthermore, when the environment for tracking humans (where the measuring device 10 is applied) is frequently changed, or the position and orientation of the sensor 14 are frequently changed even if the environment for tracking humans is the same. Is extremely troublesome.

  Therefore, in this embodiment, the position and orientation of each sensor 14 are estimated based on the detection result of each sensor 14, and they are adjusted (calibrated).

  Briefly, based on the output from each sensor 14, a human position (human observation point) is detected for each sensor 14, and the movement trajectory of the human is calculated (generated) from the time change of the human observation point. To do. Next, the movement trajectory generated based on the output of each sensor 14 is identified (matched) between the sensors 14. Two or more human observation points on the matched movement trajectory are extracted according to a predetermined rule, and using the extracted two or more human observation points, a restriction (sensor restriction) between the sensors 14 that generate the movement trajectory is obtained. ) Average and its variance. However, the sensor constraint is a distance between the sensors 14 and a relative angle (orientation) between the sensors 14. Further, the variance is an error in the distance between the sensors 14. Such sensor constraints are calculated for each set of sensors 14 for which the movement trajectory has been identified. Then, the positions and orientations of all the sensors 14 are estimated using the calculated average of sensor constraints and the variance thereof. The estimated positions of all sensors 14 are adjusted and optimized. Details will be described below.

  FIG. 3A is an illustrative view showing a map in which the environment in which the sensors 14a to 14f are installed is viewed from directly above. In this embodiment, the environment shown in FIG. 3A is a shopping mall. As can be seen from FIG. 3A, the sensors 14a, 14c, and 14d are installed on the upper side of the map, and the sensors 14b, 14e, and 14f are installed on the lower side of the map. Moreover, the arrow extended from each of sensor 14a-14f has shown the front direction of each sensor 14a-14f. As shown in FIG. 3B, a range (total detection range) where the detection ranges of two or more sensors 14 overlap is indicated by a hatched range (region).

  However, as described above, since the initial values for the position and orientation (positional relationship) of the sensor 14 are not given, the computer 12 does not have the positional relationship as shown in FIGS. The detection range cannot be known in advance.

  Therefore, the computer 12 extracts a background from information (sensor information) detected by each sensor 14. That is, the entire detection range is defined by obtaining the boundary between the hatched range and the hatched range in FIG. This is because the measuring device 10 detects or tracks a person. Accordingly, a range where a fixed object such as a range blocked by a door or a wall or a column is arranged is extracted as a background. Thus, by extracting the background in advance, what is detected in front of the background by the sensor 14 can be easily determined as an observation point of a person or the like.

  In this embodiment, the computer 12 executes a background scanning process (background scanning process), and extracts the background from the result. Specifically, the background is extracted from the sensor information detected by each sensor 14. An example of sensor information is shown in FIG. As shown in FIG. 4 (A), the sensor information is information about a distance (mm) detected at a certain angle (0.5 °) in a range of −90 ° to 90 ° for each frame. .

  For simplicity, FIG. 4A shows that the distance is detected in units of 5 °, but actually, the distance is detected in units of 0.5 ° as described above. .

  Further, in this embodiment, the frame is a unit time for the sensor 14 to detect the distance for all the angles of the detection range, and is set to 27 ms, for example. Accordingly, each sensor 14 detects a distance of 180 ° in units of 0.5 ° and transmits the detected distance to the computer 12 during one frame (27 ms).

  Although detailed description is omitted, since the sensor information of each sensor 14 is generated by the computer 12 based on the output of each sensor 14, the number of frames of the sensor information of each sensor 14 is synchronized.

  If such a background scanning process is performed for a certain length of time (for example, 10 minutes), the maximum detectable distance for each angle is known. FIG. 4B shows a graph representing the distribution of the number of distances detected for a certain angle, which is obtained from the result of executing the background scanning process. As can be seen from this graph, the most detected distance is set as the maximum detectable distance, that is, the boundary, and the distance exceeding the boundary is regarded as the background.

  Therefore, as shown in FIG. 5 (A), by executing the background scanning process, the detection range of the sensor 14 (see FIG. 2) excluding the portion overlapping the background, that is, actually detecting a human being. A range (a human detectable range) is defined. In FIG. 5A, the human detectable range is shown by adding a shaded pattern. This human detectable range constitutes the entire detection range described above.

  As shown in FIG. 5A, human H or other objects may be detected or measurement errors may occur during the execution of the background scanning process. As can be seen, a distance different from the boundary may be detected.

  As shown in FIG. 5A, when the distance between the sensor 14 and a background such as a wall exceeds the detection range (maximum distance R) of the sensor 14, the sensor information includes the maximum distance R. (8000 mm) is described.

  Further, here, the background scanning process is executed for a certain length of time to determine the most detected distance as a boundary, but in reality, a predetermined number (for example, 100) of the same distance is detected. In such a case, the distance is determined as a boundary. Therefore, the background scanning process is terminated when the boundaries for all the angles of all the sensors 14 are obtained.

  Although detailed description is omitted, as the background data (see FIG. 16), data of the distance to the boundary or the maximum distance R for each angle is stored for each sensor 14.

  When the background scanning process is completed, information about the distance to the observation point (human observation point) that seems to be a human and the angle thereof is detected from the sensor information of each sensor 14. However, the distance and angle detected here are the distance and angle of each sensor 14 in the local coordinate system. For example, as for the local coordinates of each sensor 14, the center of its own is set to the origin O, the axis passing through the origin O and overlapping the front direction of the sensor 14 is set as the Y axis, and the axis passing through the origin O and orthogonal to the Y axis is set. Set to X-axis. The front direction of the sensor 14 is the positive direction of the Y axis, and the direction rotated 90 ° to the left from the front direction is the positive direction of the X axis.

  In this embodiment, the length (width) between observation points continuously detected within the human detectable range is within a predetermined range, and the background about the angle at which the observation point is detected When the length (depth) between them is equal to or greater than a predetermined value, it is considered that a person has been detected, and the distance and angle at that time are recorded (added) to a list (person observation point list). The human observation point list is generated based on sensor information detected after the background is extracted by executing the background scanning process.

  Here, a case where a human H (human observation point) is detected by a certain sensor 14 will be described with reference to FIG. As shown in FIG. 5B, it is assumed that the distance from the points P1, P2, P3, P4, and P5 is continuously measured by the sensor 14 for a certain five angles in a certain frame. However, for simplicity, distances measured at other angles are omitted.

  As can be seen from FIG. 5B, the point P1 and the point P5 are points on the boundary, are determined to be the background, and are excluded from the observation points. Since the point P2, the point P3, and the point P4 are more troublesome than the boundary, they are extracted as observation points. Next, using the extracted observation points, the length (width) between the observation points at both ends and the length (depth) between each observation point and the boundary at the angle at which each observation point was detected are calculated. The

As described above, since the local coordinates of the sensor 14 are set with the center as the origin O, in the example shown in FIG. 5B, the distance to the point P2 detected by the sensor 14 (first distance L1). ) And the angle at that time (first angle φ ₁ ), the distance to the point P 4 detected by the sensor 14 (second distance L 2), and the angle at that time (second angle φ ₂ ), A distance (width) W between P2 and point P4 is calculated. Specifically, it is calculated according to Equation 1.

[Equation 1]
W = √ {(x1-x2) ² + (y1-y2) ² }
However, x1 = L1 × sinφ ₁ , x2 = L2 × sinφ ₂ , y1 = L1 × cosφ ₁ , y2 = L2 × cosφ ₂ .

  For each observation point, the distance (depth) dep between the observation point and the point on the boundary at the angle at which the observation point is detected is detected. That is, the length (depth) dep of the dotted line shown in FIG. This is calculated by subtracting the distance to the observation point from the distance to the background or the maximum distance R obtained by the background scanning process.

  Then, it is determined whether or not the width W is within a predetermined range and the depth dep is greater than or equal to a predetermined distance. Specifically, it is determined whether or not Expression 2 is satisfied. The condition expressed by Equation 2 is set to discriminate one person. That is, since the height of the sensor 14 is set in the vicinity of the human torso, the numerical value shown in Equation 2 is set so that it can be determined whether the body is a torso. Further, the width W is set in a relatively wide range because a human may face the front or the side.

[Equation 2]
300 (mm) ≤ W ≤ 800 (mm)
300 (mm) ≤ dep
However, the numerical value of Formula 2 may be appropriately changed according to the age and physique of the person to be measured. However, in summer and winter, clothes are completely different and their thickness is significantly different, so it is necessary to set different values depending on the season.

  As described above, the width W is set within the predetermined range in order to detect one person. Therefore, for example, when two or more humans are lined up or overlapped when viewed from the sensor 14, it is necessary to further introduce a method for distinguishing individual humans.

  The reason why the depth dep is set to be equal to or greater than the predetermined distance is that it should not be determined as a human if there is no certain thickness.

  FIG. 5B shows a case where one human H is detected by one sensor 14, but two or more people present at a distance apart from each other when viewed from one sensor 14. Human H may be detected.

  When the condition shown in Equation 2 is satisfied, it is determined that the person H has been detected (observed), and the observation point (human observation point) is stored in polar coordinates (distance and angle) in the local coordinate system of each sensor 14. However, as can be seen from FIG. 5B, since a plurality of human observation points are detected for the same person H in one frame, the representative value (average value of distance and angle, or distance Maximum value or minimum value and angle at that time) are stored.

  Specifically, a list of human observation points (human observation point list) as shown in FIG. 6A is created, and the human observation points detected this time (in the current frame) are stored in this human observation point list ( Added). As shown in FIG. 6A, the human observation point list describes the number of frames, the distance, the angle, and the speed corresponding to the index. The index is identification information attached to the human observation point, and is represented by an alphabet and a number in this embodiment. However, it is not necessary to be limited to this, and the index may be represented by only alphabets or numbers, or may be represented by information that cannot be deciphered by humans. The number of frames is the number of frames at the time when the person observation point is detected (frame number). The distance (mm) is a distance to the human observation point detected by the sensor 14 and indicated by the corresponding index. The angle (°) is an angle with respect to the front direction of the sensor 14 when the sensor 14 detects the human observation point indicated by the corresponding index. The speed (mm / s) is the moving speed of the human observation point indicated by the corresponding index. However, the speed (mm / s) is a movement speed of the same person, and when a movement trajectory (local trajectory list) described later is generated, a distance per unit time (movement speed) between consecutive human observation points. Is calculated and listed in the list of human observation points. Therefore, when creating the human observation point list, null is described as the velocity (mm / s).

  By referring to such a human observation point list, the movement trajectory of the human H detected by each sensor 14 is generated. In this embodiment, when the distance between human observation points in a predetermined time (one frame) is a predetermined value (for example, 500 mm) or less, it is determined that the points are on the same movement locus. That is, in such a case, it is determined that the human observation points for the same person H are changing with time. On the other hand, when the distance between the human observation points exceeds a predetermined value, it is determined that the point is on a different movement locus. In such a case, a movement trajectory of a different person H is also generated. Although a detailed description is omitted, the distance between the human observation points is calculated according to the above-described formula 1.

  In this way, it is determined whether or not they are human observation points on the same moving locus, and a local locus list as shown in FIG. 6B is generated based on the determination result. That is, a movement trajectory is generated (detected). A local trajectory list is created for each sensor 14. That is, the movement trajectory of the person H is generated based on the sensor information of the local coordinate system of each sensor 14, and each local trajectory list is created.

  As shown in FIG. 6B, in the local trajectory list, the index of the human observation point list is described corresponding to the local ID. The local ID is identification information for identifying the movement locus detected by the sensor 14, and is represented by alphabets and numbers. However, the local ID may be represented by only alphabets or numbers, or may be represented by information that cannot be deciphered by humans. The index of the human observation point list is an index for the human observation point on the movement locus indicated by the corresponding local ID, and two or more indexes are described in time series.

  When the local trajectory list is created for all the sensors 14, two movement trajectories that are considered to be the same among the movement trajectories detected based on the sensor information of the two different sensors 14 are associated with each other. . That is, the movement trajectory generated based on the outputs of the two different sensors 14 is identified as being considered to be the movement trajectory of the same person H.

  For example, in the case shown in FIG. 7, two movement trajectories are detected in each of the sensor 14a and the sensor 14b. As described above, each sensor 14 detects a human observation point in the local coordinate system and cannot recognize the presence of another sensor 14 with each other. For this reason, it is impossible to immediately know which movement locus generated by each sensor 14 matches which movement locus generated by other sensors 14. Therefore, in this embodiment, a common (identical) movement locus among the movement loci detected by the two different sensors 14 based on the time change (human movement speed) of the human observation point that can be seen from the movement locus. It is designed to detect.

Specifically, of the movement trajectories detected by one sensor 14, the speed change of one movement trajectory of interest (referred to as “movement trajectory T _m ” for convenience of description) and the other sensor 14 detect the movement. Compared with the respective speed changes of all the movement trajectories (for convenience of explanation, referred to as “movement trajectory T _n ”), the movement trajectory T _n with the smallest speed change error (speed difference) is the movement trajectory T _m . It is determined that they are the same. However, the speed difference (RMS error) is calculated according to Equation 3.

[Equation 3]

However, in this embodiment, when another moving track T _n having a speed difference of the same extent as the smallest movement track T _n speed difference is present, the speed difference is even the smallest movement track T _n , not determined to be the same as the movement track T _m. In this embodiment, even if there is only a movement trajectory T _n having the smallest speed difference, if this speed difference is a certain value (100 mm / s in this embodiment) or more, the movement not determined to be the same as the trajectory T _m. In these cases, even the speed difference is a minimum, be determined and the movement track T _m and movement track T _n are coincident it is considered not appropriate.

The FIG. 8 (A), the a focused to one speed change of the movement locus T _m of a of the movement trajectory detected by the one sensor 14 (solid line), but difficult to understand in the drawing, is detected by the other sensors 14 The speed change (broken line etc.) of 7 movement locus | trajectories _Tn (here n = 1, 2, ..., 7) is described.

Figure 8 (B) is a graph showing one and movement track T _m shown in FIG. 8 (A), the speed difference between each of the seven movement track T _n. For example, the speed difference is obtained by accumulating the absolute values of the speed differences at the sample points every fixed time (5 frames in this embodiment) in the speed change. As can be seen from FIG. 8B, the speed difference from the movement trajectory T _n (n = 5) is the smallest in the one movement trajectory T _{m of} interest. Further, the speed difference of the movement trajectory T _n (n = 5) is not comparable to the speed difference of the other movement trajectories T _n (n = 1-4, 6, 7). Further, the speed difference of the movement trajectory T _n (n = 5) is less than a certain value (100 mm / s in this embodiment). Therefore, it is determined that the movement trajectory T _m and the movement trajectory T _n (n = 5) are the same, and the association is performed.

If such determination processing is performed, then between the movement track T _m and all movement track T _n of interest, the speed difference is determined whether the same movement locus there are determined. When the determination process for all the movement trajectories T _{m and} all the movement trajectories T _n are executed, the determination process for the next set of sensors 14 is executed. In this embodiment, since six sensors 14 are used, the determination process is executed for a total of 15 ( ₆ C ₂ ) combinations of sensors 14. Hereinafter, in this embodiment, when referring to the set of sensors 14, (A, B), (A, C), (A, D), (A, E), (A, F), (B, C) ), (B, D), (B, E), (B, F), (C, D), (C, E), (C, F), (D, E), (D, F), It means each set of (E, F), and when processing is executed for each set of sensors 14, the processing is executed in this order. However, alphabets in parentheses are identification information of the sensor 14. In this embodiment, the identification information of the sensor 14a is “A”, the identification information of the sensor 14b is “B”, the identification information of the sensor 14c is “C”, and the identification information of the sensor 14d is “D”. The identification information of the sensor 14e is “E”, and the identification information of the sensor 14f is “F”. However, the identification information of the sensor 14 does not need to be indicated by alphabets, and may be expressed by numbers, a combination of alphabets and numbers, or may be expressed by information that cannot be deciphered by humans.

  FIG. 9A shows an example of a table (association table) for movement loci that are determined to be the same movement locus and associated with each other. In this association table, a first sensor ID, a first local ID, a second sensor ID, and a second local ID are described corresponding to the global ID.

  As described above, the global ID is identification information about a movement locus that is determined to be the same, that is, one movement locus in the global coordinate system, and is represented by a number. However, the global ID may be represented by alphabets, may be represented by numbers and alphabets, or may be represented by information that cannot be deciphered by humans.

The first sensor ID is identification information of the sensor 14 which detects the movement track T _m of a case where it is determined that the same movement locus. First local ID is a local ID of the movement track T _m of a case where it is determined that the same movement locus. The second sensor ID is identification information of the sensor 14 that has detected the movement trajectory T _n when it is determined that they are the same movement trajectory. The second local ID is a local ID for the movement trajectory T _n when determined as the same movement trajectory.

For example, as shown in FIG. 9A, the movement trajectory with the global ID “35” has the first local ID “T ₁ ” detected by the sensor 14 a with the first sensor ID “A”. This means that the movement locus T _n is the movement locus T _n whose second local ID detected by the sensor 14b whose movement locus T _m and the second sensor ID is “B” is “T ₁ ”. Although the description is omitted, the same applies to other cases.

  The first local ID and the second local ID may be assigned the same ID, but are distinguished by the first sensor ID and the second sensor ID, so that the movement locus indicated by them can be identified. .

When the same movement locus is associated, that is, when the association table is created, a human observation point on the movement track T _m and on the movement track T _n which are determined to be the same, the same timing (same frame) The human observation points (shared human observation points) detected in (1) are extracted from the respective movement trajectories T _m and T _n .

  FIG. 9B shows an example of the extracted list of shared human observation points (shared human observation point list). As shown in FIG. 9B, in the shared human observation point list, the first distance, the first angle, the second distance, and the second angle are described corresponding to the first sensor ID and the second sensor ID. The

As described above, the first sensor ID is identification information of the sensor 14 which detects the movement track T _m which is determined to be the same movement locus. Similarly, the second sensor ID is identification information of the sensor 14 which detects the movement track T _n it is determined that the same movement locus.

  The first distance is a distance (mm) between the sensor 14 indicated by the first sensor ID and the shared human observation point OBS. The first angle is an angle (°) of the shared human observation point OBS viewed from the sensor 14 with respect to the front direction of the sensor 14 indicated by the first sensor ID.

  The second distance is a distance (mm) between the sensor 14 indicated by the second sensor ID and the shared human observation point OBS. The second angle is an angle (°) of the shared human observation point OBS viewed from the sensor 14 with respect to the front direction of the sensor 14 indicated by the second sensor ID.

  For example, as shown in FIG. 9B, the shared human observation point OBS between the sensor 14a whose first sensor ID is “A” and the sensor 14b whose second sensor ID is “B” is the sensor 14a. The first distance to the front direction of the sensor 14a is 45 °, the second distance to the sensor 14b is 2000mm, and the second angle to the front direction of the sensor 14b is 30mm. °. Although the description is omitted, the same applies to other cases.

  However, the human observation point OBS is a human observation point (H1, H2, H3, H4, H5,...) Registered in the human observation point list for each sensor 14 as described above.

FIG. 10A shows a movement locus detected by the sensor 14 (sensor S ₁ ) indicated by the first sensor ID and the sensor 14 (sensor S ₂ ) indicated by the second sensor ID and determined to be the same. FIG. 6 is an illustrative view showing a positional relationship (distance and angle) between the shared human observation point OBS and the sensors S ₁ and S ₂ . However, in FIG. 10A (the same applies to FIG. 10B described later), the first distance is “d ₁ ”, the first angle is “θ ₁ ”, the second distance is “d ₂ ”, and Two angles are indicated by “θ ₂ ”.

  In FIG. 10A, one shared person observation point OBS is shown, but actually, there are a plurality of shared person observation points OBS on the movement trajectory determined to be the same.

  Next, a method for calculating the constraint (positional relationship) between the sensors 14 will be described. However, the restriction between the sensors 14 means the distance D between the two sensors 14 and the relative orientation (α, β) between the two sensors 14 (see FIGS. 11A and 11B). ).

As described above, because the human observation point OBS represented by polar coordinates in the local coordinate system of each sensor 14, for example, as shown in FIG. 10 (B), the sensor S ₁ is a distance d ₁ and the angle theta ₁ When the rotation is performed by an arbitrary angle γ while maintaining, the restrictions on the sensor S ₁ and the sensor S ₂ are completely different from the case where the rotation is not performed by the angle γ. That is, when a shared human observation point OBS is one, and the person observation point OBS, from the positional relationship between the sensor S ₁ and the sensor S ₂ (polar), the constraint between the sensors (positional relationship) correctly It cannot be calculated.

Although illustration is omitted, the case where the sensor S ₂ rotates by an arbitrary angle while maintaining the distance d ₂ and the angle θ ₂ is the same as that shown in FIG.

Thus, in this embodiment, calculates human observation points of the two shared _OBS p, and OBS _q, from the positional relationship between the sensor _{S 1} and the sensor _{S 2,} constraint between the sensor _{S 1} and the sensor _{S 2} (positional relationship) I have to do it.

FIG. 11A is an illustrative view showing a positional relationship between two shared human observation points OBS _p and OBS _q and the sensors S ₁ and S ₂ . Although illustration is omitted, the human observation point OBS _p and the human observation point OBS _q are points on the movement trajectory determined to be the same.

Here, a method for calculating the constraint between sensors will be described with reference to FIGS. 11 (A) and 11 (B). However, in FIG. 11B, the positions of the _two sensors S ₁ and S ₂ shown in FIG. 11A are indicated by points A and B, and two shared human observation points OBS _p and OBS _q Are indicated by points P and Q. Further, the distance D between the two sensors described above and the relative angles α and β between the two sensors are shown. Specifically, shown in FIG. 11 (B), the length of the line segment AB is the distance D, the line segment AB and the front direction angle between the sensor S ₁ is the angle alpha, and the segment AB sensor S ₂ front direction angle formed is an angle beta.

As described above, the two shared human observation points OBS _p and OBS _q shown in FIG. 11A are extracted (selected) according to a predetermined rule. For example, the shared human observation point OBS _p is the newest (last detected) human observation point in time, and the shared human observation point OBS _q is older than the shared human observation point OBS _p (temporal). (Observed previously) and selected arbitrarily (randomly). However, two shared human observation points OBS _p and OBS _q may be selected at random.

(1) First, the length d _PQ of the line segment _PQ is calculated using the cosine theorem. Here, since the first angle θ ₁ and the third angle θ ₃ shown in FIG. 11A are included in the shared human observation point list, they are known values. Therefore, the angle ∠PAQ shown in FIG. 11B can be calculated using the magnitude of the first angle θ _{1 and} the magnitude of the third angle θ ₃ . However, since the signs (+, −) of the first angle θ ₁ and the third angle θ ₃ change depending on the orientation of the sensor S ₁ , another person observation point OBS from the first angle θ1 with respect to the reference person observation point OBS _p . It is calculated by subtracting the third angle θ ₃ with respect to _q . Specifically, the angle ∠PAQ is calculated according to Equation 4. Similarly, the angle ∠PBQ is calculated according to Equation 5 by using the second angle θ ₂ and the fourth angle θ ₄ . However, in Equations 4 and 5, | · | means an absolute value. Hereinafter, this is the same in this embodiment. Further, the second angle θ ₂ and the fourth angle θ ₄ shown in FIG. 11A are known values because they are included in the sensor information detected by the sensor S ₂ . The same applies hereinafter.

[Equation 4]
∠PAQ = | θ ₁ −θ ₃ |
[Equation 5]
∠PBQ = | θ ₂ −θ ₄ |
(2) Next, the length of the line segment PQ is calculated by using the cosine theorem for the triangle ΔAPQ. Here, the length d _AP of the line segment _AP and the length d _AQ of the line segment AQ correspond to the first distance d ₁ and the third distance d ₃ shown in FIG. 11A, respectively. It is known because it is included in the list of human observation points. Therefore, as shown in Equation 6, the length d _PQ of the line segment _PQ is obtained using the cosine theorem.

[Equation 6]
d _PQ = √ ((d _AP ) ² + (d _AQ ) ² −2 × d _AP × d _AQ × cos∠PAQ)
Although detailed description is omitted, since the angle ∠PBQ is calculated according to Equation 5, the length of the line segment PQ can be calculated by using the cosine theorem for the triangle ΔBPQ.

(3) Next, the angle ∠APQ and the angle ＱBPQ are calculated, and further, the angle ∠APB is calculated. The angle ∠APQ is calculated according to Equation 7 by using the sine theorem for the triangle ΔAPQ. Further, the angle ∠BPQ is calculated according to Equation 8 by using the sine theorem for the triangle ΔBPQ. However, since the angle ∠APB is the sum of the angle ∠APQ and the angle ∠BPQ, the mathematical formula is omitted. The length d _BQ line segment BQ, since corresponds to the fourth distance d ₄ shown in FIG. 11 (A), included in the shared human observation point list, is known.

[Equation 7]
∠APQ = sin ⁻¹ (d _AQ / d _PQ × sin ∠PAQ)
[Equation 8]
∠BPQ = sin ⁻¹ (d _BQ / d _PQ × sin∠PBQ)
(4) Subsequently, the length dAB of the line segment AB is calculated according to Equation 9 by using the cosine theorem for the triangle ΔPAB. However, the length d _BP line segment BP, since corresponds to a third distance d ₃ shown in FIG. 11 (A), included in the shared human observation point list, is known.

[Equation 9]
d _AB = √ ((d _AP ) ² + (d _PB ) ² −2 × d _AP × d _PB × cos∠APB)
(5) Further, the angles ∠PAB and ∠PBA are calculated according to Equations 10 and 11 by using the sine theorem for the triangle ΔPAB, respectively.

[Equation 10]
∠PAB = sin ⁻¹ (d _BP / d _AB × sin ∠APB)
[Equation 11]
_{^{∠PBA = sin -1 (d AP /}} d AB × sin∠APB)
The angle ∠PAB and angle ∠PBA are calculated, for the sensor _{S 1} and the angle theta ₁ and sensor _{S 2} and the human observation point angle theta ₂ between the OBS _p of the human observation point OBS _p is known, the sensor S the angle formed between the _first front direction and the line segment AB α (∠PAB- | θ1 |) and the sensor angle formed between the front direction and the line segment AB of _{S 2 β (∠PBA- | θ2 |} ) is easily determined that Can do.

In this way, the constraints (D, α, β) between the two sensors are calculated using Equation 9 to Equation 11. However, since the observation point OBS _q is selected at random as described above, a plurality (three in this embodiment) of observation points OBS _q are selected at random, and in each case, between the two sensors. The constraint is calculated. When the constraints between the two sensors are calculated for the observation point OBS _p and the plurality of observation points OBS _q , the average and the variance (error) of the estimated values are calculated. In this embodiment, the average of the constraints between the sensors and the variance of the estimated values are calculated by a Kalman filter. Here, the calculation method will be briefly described.

First, a new constraint (d _ij , θ _ij ) on two sensors (here, sensor i and sensor j) generated from a shared human observation point OBS is a relative coordinate between the sensors as shown in _Equation 12. In the system, the coordinate Cartesian difference _Zij is converted.

[Equation 12]

[Equation 13]

[Equation 14]

[Equation 15]

  The restrictions between sensors and the variance of estimated values for each set of sensors 14 thus calculated are registered (added) in a list (constraint list) as shown in FIG. As shown in FIG. 12, the restriction list describes the first sensor ID, the second sensor ID, the distance between the sensors, the angle, and the variance of the estimated value.

In FIGS. 11A and 11B, the calculation method is described as the distance D between the sensors S ₁ and S ₂ and the relative angles α and β of the constraints between the sensors. However, the relative angle α is the front direction of the orientation of the sensor S ₁ in the case where the sensor S ₁ of interest viewed sensor S _2, the relative angle beta, when viewed sensor S ₁ from the sensor S ₂ of interest which is the front direction of the orientation of the sensor S _2. Therefore, in the constraint list, each case is described as another sensor constraint.

  The first sensor ID is identification information of the sensor 14 of interest, and the second sensor ID is identification information of the sensor 14 viewed from the sensor 14 indicated by the first sensor ID. The inter-sensor distance is the distance (mm) between the sensor 14 indicated by the first sensor ID and the sensor 14 indicated by the second sensor ID, calculated as described above. The angle is an angle (°) calculated as described above. Specifically, when the sensor 14 indicated by the second sensor ID is viewed from the sensor 14 indicated by the first sensor ID, the first sensor ID is It is the direction of the front direction of the sensor 14 shown. The variance of the estimated value is the variance σ of the estimated value calculated as described above, and is shown here with the corresponding first sensor ID and second sensor ID.

  When the constraint list is created, the positions of all the sensors 14 are estimated using the constraint list. For example, all the sensors 14 are arranged on a virtual map. Then, by repeating the relaxation calculation for the constraints between the sensors, those estimates are refined.

  First, the constraint list is rearranged (sorted) in the order of decreasing variance σ (in order of high reliability). Then, the two sensors 14 are arranged on the virtual map according to the restriction between the sensors in the case of the highest reliability. At this time, one sensor 14 is set as a reference. For example, of the two sensors 14, the sensor 14 described on the left side (or the right side) in the group of sensors 14 in parentheses may be set as a reference. That is, the position and angle of the sensor 14 set as a reference are set as reference values. In this embodiment, (X coordinate, Y coordinate, angle) = (0, 0, 0) is set as the reference value. That is, the position of the reference sensor 14 is set to the origin on the virtual map. However, the front direction of the sensor 14 is set to the positive direction of the Y axis on the virtual map. The other sensor 14 is arranged on the virtual map according to the average of the sensor constraints calculated as described above.

For example, assuming that the most reliable set of sensors 14 is (A, C), as shown in FIG. 13A, first, the sensors 14a and 14c are set to the global coordinates with reference to the sensor 14a. It is placed on the virtual map of the system. As can be seen from FIG. 13A, the sensor 14a is disposed at the origin O (0, 0), and the direction thereof is set in the positive direction of the Y axis. Next, the sensor 14c is arranged using the constraints between the sensors (here, the distance D ₁ , the constraint angles α ₁ and β ₁ ) with the sensor 14a. Therefore, the position of the sensor 14c is represented by (D ₁ × sin | α ₁ |, D ₁ × cos | α ₁ |). However, as described above, the constraint between sensors includes a deviation, that is, an error (an error between sensors). Therefore, it can be said that the position where the sensor 14c is arranged is an estimated position. Further, the direction of the sensor 14c is determined to be − (180 ° − | α | − | β |).

  In this embodiment, the variance (error) is not calculated for the angle.

  In the global coordinate system, when the Y-axis direction is used as a reference, the right side (first quadrant, fourth quadrant side) is the plus direction of the direction (angle) of the sensor 14, and the left side (second quadrant, second quadrant). (3 quadrant side) is the minus direction of the angle of the sensor 14.

  When the two sensors 14 (here, A and C) for the set having the highest reliability are arranged, any one of the arranged sensors 14 (A and C) and another sensor having a restriction between the sensors 14 is arranged according to the constraints between the sensors. In the example shown in FIG. 13A, the sensor 14 having a restriction between the sensor 14a or the sensor 14c is next placed on a virtual map. However, when determining the sensor 14 to be placed next, by referring to the restriction list, the restriction between the sensors is read out in descending order of reliability.

  FIG. 13B shows an example of a case where the next sensor 14 (here, B) is arranged from the state shown in FIG. Here, when the sorted restriction list is searched from the first column, the sensor 14b having the restriction between the sensor 14a and the sensor 14a is detected from the sensors 14a or 14c, and this sensor 14b is arranged. is there.

As shown in FIG. 13B, the sensor 14b is arranged on the virtual map in accordance with the restriction between the sensor 14a and the already arranged sensor 14a. In the global coordinate system, the estimated position of the sensor 14b is represented by (−D ₂ × sin | α ₂ |, D2 × cos | α ₂ |). The angle of the sensor 14b is represented by 180 ° − | α ₂ | − | β ₂ |.

  In this manner, the sensors 14 having the constraints between the sensors already arranged are arranged in order on the virtual map. Further, when the sensor 14 is arranged on the virtual map, the arranged information is registered (added) in the sensor position list as shown in FIG. As illustrated in FIG. 14, the sensor position list describes a sensor ID, an estimated position, an angle, and a constraint error between sensors, corresponding to the arrangement sensor ID. The placement sensor ID is identification information about the sensor 14 placed this time. The pair sensor ID indicates identification information of the other (paired) sensor 14 indicated by the restriction between the sensors used when the sensor 14 indicated by the placement sensor ID is placed.

  Since the position of the reference sensor 14 is set to the origin O, (x1, y1) = (0, 0), and the front direction of the reference sensor 14 is set to the Y-axis direction. Is 0 °.

The estimated position is a two-dimensional coordinate indicating the position of the sensor 14 arranged this time in the global coordinate system. The angle is an angle with respect to the plus direction of the Y axis in the front direction in the global coordinate system of the sensor 14 arranged this time. The error between the sensors is the variance (error) σ _ij described in the restriction list between the sensors, and the numerical value of the error σ _ij for the sensor 14 (sensor i) paired with the sensor 14 (sensor j) arranged this time. Is described.

  If identification information (here, A-F) of all the sensors 14 is described in the placement sensor ID column of this sensor position list, all the sensors 14 are placed on a virtual map. When all the sensors 14 are arranged, the position of each sensor 14 is adjusted by the relaxation calculation so that the error (overall error) of the arrangement position for the whole is equal to or less than a predetermined threshold value. That is, the estimated position of the sensor is refined by the relaxation calculation.

  Relaxation calculation methods and proof of convergence (refining) are described in the literature (T. Duckett, S. Marsland, and J. Shapiro, Learning globally consistent maps by relaxaTmon, in Proc. IEEE Interna Tmonal Conference on RoboTmcs & AutomaTmon (ICRA), San Francisco, CA, April 2000.).

Briefly, in the relaxation calculation, the estimated position is generated according to the equation 16 for each sensor i and the sensor j arranged at the neighboring position (x _j , y _j ) related thereto. However, the sensor i and the sensor j are the two sensors 14 that have detected the common observation point OBS. In these calculations, in order to calculate the angle constraint θ _ji in the absolute coordinate system, the estimated angle constraint is used as shown in _Equation 17.

[Equation 16]

[Equation 17]

Next, the error [nu _ji of each estimated position, as the sum of the error u _ji for estimation of constraints between the error [nu _j and two sensors of the estimated position of the sensor j in the neighborhood, is calculated according to Equation 18.

[Equation 18]

By combining these estimated positions, a new estimated position (x _i , y _i ) of the sensor 14 and its error ν _j are calculated according to Equation 19. In Equation 19, N _i represents a set of all sensors j other than the sensor i. However, i ≠ j.

[Equation 19]

  Then, the angles of all the sensors 14 are adjusted using Equation 20 so that the average error between the angle constraint and the estimated position of the sensor 14 is minimized.

[Equation 20]

  Such a procedure is repeated up to an arbitrary stop point. In the measuring apparatus 10 of this embodiment, when the above-described procedure is repeated a predetermined number of times (for example, three times), the measuring apparatus 10 tends to converge accurately. However, in this embodiment, the above-described procedure is repeated until the minimum value of the overall error is obtained. Here, the total error is the sum of the difference between the distance between the sensors obtained when all the sensors 14 are arranged on the virtual map according to the restriction between the sensors and the distance between the sensors in the calculated restriction between the sensors. It is. However, the distance between the sensors 14 arranged on the virtual map is calculated only between the sensors 14 having the restriction between the sensors. In this way, the positions of all the sensors 14 are adjusted. In this way, the position and angle of each sensor 14 arranged in a certain environment are recognized in the same coordinate system (global coordinate system). That is, the position and angle of the sensor 14 are calibrated.

However, if the average value of the position errors for all the sensors 14 exceeds a predetermined value (in this embodiment, 50 mm-100 mm), the human position is maintained until the average value of the position errors is equal to or less than the predetermined value. The process from the detection process to the sensor position adjustment process is repeatedly executed. Here, the position error of the sensor 14 means an error included in the estimated position of the sensor 14. Referring to FIG. 14, when the sensor 14a and the sensor 14c are arranged, these include the difference σ _AC and σ _CA (= 200 mm) when the constraint between the sensors is calculated. The position (14c) includes the error (position error). Next, in the case of arranging the sensor 14b, in addition to the positional error of the sensor 14a, a difference σ _AB (= 500 mm) of the restriction between the sensors 14a and 14b is included, so the position of the sensor 14b The error is 200 mm + 500 mm. In this way, the position error is increased each time the sensor 14 is arranged. The average value of the position errors is set to a predetermined value or less to optimize the positions of the plurality of sensors 14.

  FIG. 15 shows an example of a memory map 300 of the memory (RAM) 12a built in the computer 12 shown in FIG. As shown in FIG. 15, the memory 12 a includes a program storage area 302 and a data storage area 304. A calibration program is stored in the program storage area 302. The calibration program includes a main processing program 312, a background scanning program 314, a human position detection program 316, a trajectory generation program 318, a trajectory matching processing program 320, a shared person observation. A point list creation program 322, an inter-sensor constraint calculation program 324, an inter-sensor constraint average program 326, a sensor position estimation program 328, a sensor position adjustment program 330, and the like.

  The main processing program 312 is a program for processing the main routine of the calibration processing of this embodiment. The background scanning program 314 excludes background walls and pillars from the detection ranges from the outputs (sensor information) of the sensors 14a-14f, and sets a predetermined angle (in this embodiment, 0 for each sensor 14a-14f). .5 °) is a program for extracting the maximum detectable distance.

  The human position detection program 316 is a program for detecting a person from each output (sensor information) of the sensors 14a to 14f and detecting the position (human observation point OBS) using the result of the background scan. . However, the human position is represented by a distance and an angle in each of the local coordinate systems of the sensors 14a-14f, and as shown in FIG. 6A, a human observation point list created for each of the sensors 14a-14f. Registered (added).

  The trajectory generation program 318 refers to the human observation point list for each of the sensors 14a to 14f, and arranges the human observation points determined to be continuous human observation points in time series, thereby moving in the local coordinate system. This is a program for generating a trajectory and creating each local trajectory list. The trajectory matching processing program 320 determines whether or not the movement trajectories generated based on the outputs of the two different sensors 14 match based on the speed change of the movement trajectory, and associates the two corresponding movement trajectories with each other. It is a program for creating a table.

  The shared person observation point list creation program 322 determines that the human observation point OBS detected at the same time (same frame) among the two associated human observation points OBS on the local locus is the shared person observation point OBS. It is a program for creating a list of shared human observation points.

  The inter-sensor constraint calculation program 324 is a program for calculating the distance between the two sensors 14 and the relative angle between the two sensors 14 (the constraint between the sensors). The inter-sensor constraint average program 326 is a program for calculating an average and a variance of a plurality of constraints between the sensors 14 calculated according to the inter-sensor constraint calculation program 324, respectively. In this embodiment, as described above, the average of the constraints between the sensors 14 and the variance thereof are calculated using the Kalman filter.

  The sensor position estimation program 328 is a program for arranging all the sensors 14 in the same coordinate system (global coordinate system) using the average of the constraints between the sensors calculated according to the inter-sensor constraint average program 326 and the variance thereof. It is. The sensor position adjustment program 330 is a program for adjusting (refining) the positions and angles of all the sensors 14 arranged by the sensor position estimation program 328 by relaxation calculation.

  FIG. 16 is an illustrative view showing specific contents of the data storage area 304 shown in FIG. As shown in FIG. 16, the data storage area 304 includes sensor information data 350, background data 352, human observation point list data 354, local trajectory list data 356, association table data 358, shared human observation point list data 360, Restriction list data 362, sensor position list data 364, sensor position error data 366, and the like are stored.

  As shown in FIG. 4, the sensor information data 350 is sensor information (table) data describing distances detected at fixed angles in each frame, and is created and stored for each sensor 14. . The background data 352 is numerical data about the distance to the boundary or the maximum distance R for each constant angle, and is created and stored for each sensor 14.

  As shown in FIG. 6A, the human observation point list data 354 is data on a human observation point list describing a human position (human observation point OBS) detected based on sensor information. Created and stored for each sensor 14. As shown in FIG. 6B, the local trajectory list data 356 is a list of moving trajectories generated (detected) based on the human observation point list, and is created and stored for each sensor 14. .

  As shown in FIG. 9A, the association table data 358 includes two movement trajectories generated (detected) by different sensors 14 and each of the movement trajectories determined (identified) to match each other. This is data about the association table associated. As shown in FIG. 9B, the shared human observation point list data 360 includes the human observation points OBS detected simultaneously (in the same frame) among the two human observation points OBS on the associated movement trajectories. Is a list of data registered as a shared person observation point OBS. As shown in FIG. 12, the restriction list data 362 is data of a restriction list regarding restrictions between sensors (distance and relative angle between sensors 14) and dispersion of estimated values thereof.

  The sensor position list data 364 is information about a list indicating information (estimated position, angle, error between sensors, etc.) about the sensors 14 arranged on the virtual map when a plurality of sensors 14 are arranged on the virtual map. It is data. The sensor position error data 316 is numerical data indicating the position error for each of the plurality of sensors 14 arranged on the virtual map.

  Although not shown, the data storage area 304 stores other data necessary for execution of the calibration program, and is provided with a counter (timer) and a register.

  FIG. 17 is a flowchart showing calibration processing of the computer 12 (CPU) shown in FIG. As shown in FIG. 17, when starting the calibration process, the computer 12 executes a background scan process (see FIG. 18) described later in step S1, and a human position detection process (described later) (step S3). (See FIG. 19).

  Although illustration is omitted, detection processing of scan data from each sensor 14 is started at the same time or almost simultaneously with the start of the calibration processing. In this detection process, calibration processes are executed in parallel.

  Subsequently, in step S5, a local locus generation process (see FIG. 20), which will be described later, is executed. In step S7, a locus matching process, which will be described later (FIGS. 21 and 22), is executed. In step S9, a shared human observation point list creation process (see FIG. 23 and FIG. 24), which will be described later, is executed. In step S11, a calculation process for constraints between sensors (see FIG. 25), which will be described later, is executed. In step S13, it is determined whether the restriction between the sensors for connecting all the sensors 14 can be expressed. In other words, the computer 12 determines whether or not there is a sensor 14 for which a restriction between sensors is not calculated.

  If “NO” in the step S13, that is, if the restriction between the sensors for connecting all the sensors 14 cannot be expressed, the process returns to the step S3, and the processes of the steps S3-S11 are executed again. On the other hand, if “YES” in the step S13, that is, if the restriction between the sensors for connecting all the sensors 14 can be expressed, an arrangement process of all the sensors described later (FIGS. 26 and 26) is performed in a step S15. 27), and a sensor position adjustment process (see FIG. 28) described later is executed in step S17.

  Then, in step S19, it is determined whether or not the estimated average position error is within a threshold value (for example, 50 mm-100 mm). If “NO” in the step S19, that is, if the average of the estimated position errors is not within the threshold value, the process returns to the step S3. On the other hand, if “YES” in the step S19, that is, if the average of the estimated position errors is within the threshold value, the calibration process is ended.

  FIG. 18 is a flowchart of the background scanning process shown in step S1 of FIG. This background scanning process is executed for each sensor 14 by multitasking. When the background scanning process is started, the computer 12 (CPU) collects scan data as time passes in step S31. That is, the distance for every fixed angle is detected for every frame. In the next step S33, a representative value of the distance indicated by the scan data is calculated for each angle. In this embodiment, the representative value is a distance value indicated by the most frequently scanned data. In step S35, it is determined whether or not the number of scan data indicating the distance for the representative value is greater than or equal to a predetermined value (100 in this embodiment).

  If “NO” in the step S35, that is, if the number of scan data indicating the distance with respect to the representative value is less than the predetermined value, the process returns to the step S31. On the other hand, if “YES” in the step S35, that is, if the number of scan data indicating the distance with respect to the representative value is equal to or larger than the predetermined value, the numerical value data of the representative value for each angle is obtained as the background of the sensor 14 in a step S37. Data is stored in the data storage area 304, and the process returns to the calibration process shown in FIG. This background processing is executed for each sensor 14, and the background data 352 shown in FIG. 16 is generated.

  FIG. 19 is a flowchart of the human position detection process in step S3 shown in FIG. This human position detection process is also executed for each sensor 14 by multitasking. As shown in FIG. 19, when the computer 12 (CPU) starts human position detection processing, it collects scan data for one frame in step S51. That is, the distance of each angle in one frame is detected. In the next step S53, a difference from the background is detected. That is, the computer 12 refers to the background data 352 and calculates the difference between the distance of each angle indicated by the background data of the sensor 14 and the distance of each angle of the current frame detected in step S51.

  In the subsequent step S55, a segment different from the background is specified within the human width threshold value (in this embodiment, within the range indicated by Formula 2). That is, a human is specified. In the subsequent step S57, the position (human observation point) for the identified segment is registered (added) to the human observation point list, and the process returns to the calibration process shown in FIG. However, as described above, distances and angles are registered in the human observation point list as human observation points. By executing this human position detection process for each sensor 14, the human observation point list data 354 is created or updated.

  FIG. 20 is a flowchart of the local locus generation process in step S5 shown in FIG. This local locus generation process is also executed for each sensor 14 by multitasking. As shown in FIG. 20, when starting the local locus generation process, the computer 12 reads one human observation point OBS among the human observation points OBS detected this time in step S71.

  In the next step S73, it is determined whether or not the human observation point OBS detected this time is within 500 mm of any one of the human observation points OBS detected last time (in the frame before the previous frame). If “NO” in the step S73, that is, if the human observation point OBS detected this time is not within 500 mm of any of the previously detected human observation points OBS, the process proceeds to a step S81 as it is. On the other hand, if “YES” in the step S73, that is, if the human observation point OBS detected this time is within 500 mm of any one of the previously detected human observation points OBS, the previous detection is made in the step S75. It is determined whether the human observation point OBS is listed in the local trajectory list. At this time, the computer 12 refers to the local trajectory list data 356.

  If “YES” in the step S75, that is, if the previously detected human observation point OBS is listed in the local trajectory list, in the step S77, the human observation point OBS detected this time is changed to an existing local observation point. It adds to a locus | trajectory and progresses to step S81. That is, in step S77, the index of the human observation point OBS detected this time is described corresponding to the existing local ID in the local trajectory list indicated by the local trajectory list data 356.

  On the other hand, if “NO” in the step S75, that is, if the previously detected human observation point OBS is not listed in the local trajectory list, the previously detected human observation point OBS and the current detection point are detected. A new movement trajectory including the human observation point OBS is generated, and the process proceeds to step S81. That is, in step S79, the local ID is added to the local trajectory list indicated by the local trajectory list data 356, and the index of the human observation point OBS detected last time and the current detection are detected corresponding to the local ID. The index of the human observation point OBS is described.

  In step S81, it is determined whether or not the process (steps S73 to S79) for generating the local trajectory has been executed for all the observation points OBS detected this time. If “NO” in the step S81, that is, if there is a human observation point OBS that has not been subjected to the process for generating the local trajectory among the human observation points OBS detected this time, in a step S83, Of the human observation points OBS detected this time, the next human observation point OBS is read out, and the flow returns to step S73. In step S83, the human observation point OBS described next (below) in the human observation point list is read out. On the other hand, if “YES” in the step S81, that is, if the process for generating the local locus is executed for all the human observation points OBS detected this time, the process returns to the calibration process shown in FIG. To do.

  21 and 22 are flowcharts of the locus matching process in step S7 shown in FIG. This trajectory matching process is executed for each set of sensors 14 by multitasking. As shown in FIG. 21, when the computer 12 (CPU) starts the trajectory matching process, in step S101, the computer 12 initializes a variable m (m = 1). m individually identifies the local trajectory generated based on the output of one of the sensors 14 in the set of sensors 14. In the locus matching process, the local locus list for one sensor 14 is referred to in order from the first column.

  In the subsequent step S103, the variable n is initialized (n = 1). n individually identifies a local locus generated based on the output of the other sensor 14 in the set of sensors 14. In the locus matching process, the local sensor list for the other sensor 14 is referred to in order from the first column.

Then, in step S105, it is calculated according to Equation 3 the RMS error between the movement trajectory _{T m} and movement track _{T n.} In the next step S107, the speed of the movement track T _m and movement track T _n determines whether greater than a threshold of minimum speed. The threshold for the minimum speed is set to a speed for determining that a human is walking (moving). However, if the number of people in the observation area (environment) is large, the threshold for this minimum speed is set to a relatively small value. If the number of people is small or the sensor measurement results include a lot of noise. The threshold value for the minimum speed is set to be relatively large. As described above, when the number of humans is large, the threshold value of the minimum speed is reduced to make it easy to determine whether or not the human is moving. Conversely, when the number of humans is small, it can be determined whether or not humans are moving even if the minimum speed threshold is increased to some extent. In order not to be affected by noise, it is necessary to increase the minimum speed threshold for filtering.

If "NO" in the step S107, that is, if at least one of the speed of the movement track _{T m} and movement track _{T n} is less than or equal to the threshold value of the minimum speed, the process proceeds to step S113. On the other hand, if "YES" in the step S107, that is, if the speed of the movement track _{T m} and movement track _{T n} is larger than the threshold value of either minimum speed, at step S109, RMS error is error threshold (this embodiment In the example, it is determined whether it is smaller than 100 mm / s).

If “NO” in the step S109, that is, if the RMS error is not less than the error threshold value, the process proceeds to a step S113 as it is. On the other hand, if "YES" in the step S109, that is, the RMS error is less than the error threshold value, it is determined that the movement track _{T m} and movement track _{T n} can match, at step S 111, RMS Corresponding to the error, the movement trajectory _Tn is stored in a work area (not shown) of the memory 12a.

In step S113, the variable n is incremented by 1 (n = n + 1). In a succeeding step S115, it is determined whether or not the variable n exceeds the maximum number. That is, the computer 12 determines whether or not the process (determination process) for determining whether or not all the movement trajectories T _n described in the local trajectory list for the other sensor 14 match the movement trajectory T _m is performed. to decide.

If in step S115 "NO", that is the variable n does not exceed the maximum number, the flow proceeds to a determination process of the next movement track T _n. On the other hand, if "YES" in the step S115, that is, if the variable n exceeds the maximum number, in step S117 shown in FIG. 22, whether only movement track _{T n} that matches the movement locus _{T m} there to decide. That is, the computer 12 refers to the work area of the memory 12a and determines whether or not the movement locus _Tn is described corresponding to one RMS error.

If “NO” in the step S117, that is, if the movement locus _Tn is not described in the work area of the memory 12a corresponding to the RMS error, or if the movement locus Tn corresponds to each of two or more RMS errors. _{if n} is described, it is determined that there is no unique movement track T _n that matches the movement locus T _m is, the process proceeds to step S121. On the other hand, if “YES” in the step S117, that is, if the movement trajectory T _n is described corresponding to one RMS error in the work area of the memory 12a, it is the only one that matches the movement trajectory T _m. It is determined that there is a movement locus T _n , the movement locus T _m and the movement locus T _n are registered (added) in the association table, and the process proceeds to step S121.

In step S121, the variable m is incremented by 1 (m = m + 1). In a succeeding step S123, it is determined whether or not the variable m exceeds the maximum number. In other words, the computer 12, or for all of the movement track T _m described in the local trajectory list for one of the sensors 14, consistent with all movement track T _n described in the local trajectory list for the other sensor 14 It is determined whether or not the determination process is executed.

If “NO” in the step S123, that is, if the variable m is equal to or less than the maximum number, the process returns to the step S103 illustrated in FIG. That is, whether the computer 12, respectively of the movement track T _n described in the local trajectory list for the next local trajectory T _m and the other sensor 14 described in the local trajectory list for one of the sensors 14 matches The process proceeds to the determination process. On the other hand, if “YES” in the step S123, that is, if the variable m exceeds the maximum number, the process returns to the calibration process shown in FIG.

23 and 24 are flowcharts of the process for creating the shared human observation point list in step S9 shown in FIG. The shared human observation point list creation process is also executed for each set of sensors 14 by multitasking. As shown in FIG. 23, the computer 12 (CPU) starts the process of creating a shared human observation point list, in step S141, each of the first movement track T _m present in the association table and movement track T _n make up the human observation point OBS _(OBS v, OBS _w) read and the number of frames that detected the person observation point OBS.

In the next step S143, an initial value is set in the variable f _c (f _c = 1). However, _{f c} is the number of frames. In the next step S145, in this frame, to determine whether human observation points OBS _v constituting the movement track _{T m} there. That is, the computer 12 determines whether or not the human observation point OBS _v having the number of frames indicated by the variable f _c exists among the human observation points OBS _v constituting the movement locus T _m , and the human observation point list data 354 and the local locus. The determination is made with reference to the list data 356. This process is the same in the next step S147.

If in step S145 "NO", that is, in this frame, when human observation point OBS _v constituting the movement track _{T m} is not, the process proceeds to step S151. On the other hand, if "YES" in the step S147, that is, in this frame, when the human observation point OBS _v constituting the movement track _{T m} is present, in step S147, in this frame, constituting the movement trajectory _{T n} It is determined whether or not there is a human observation point OBS _w .

If "NO" in the step S147, that is, in this frame, when no human observation point OBS _w constituting the movement track _{T n} is directly proceeds to step S151. On the other hand, if “YES” in the step S147, that is, if there is a human observation point OBS _w constituting the movement trajectory T _n in this frame, the human observation points OBS _v , OBS _w in this frame in the step S149. Is registered (added) to the shared human observation point list as shown in FIG. 9B, and the process proceeds to step S151.

In step S151, it is determined whether this frame is the maximum frame (current final frame). If “NO” in the step S151, that is, if this frame is not the maximum frame, the variable f _c is incremented by 1 (f _c = f _c +1) in a step S153, and the process returns to the step S145. That is, the computer 12 executes a process for determining whether or not there is a shared human observation point OBS in the next frame.

On the other hand, if "YES" in step S151, the that is, if the frame is the maximum frame, in step S155 shown in FIG. 24, whether or not the next movement track _{T m} and movement track _{T n} exists in the association table Judging. In other words, the computer 12 refers to the table data 358 associated, in this set of sensors, than it is determined whether or not been another movement trajectory T _m and movement track T _n determined there to be consistent.

If "NO" in the step S155, that is, if and the movement track T _n and the next movement track T _m not in the table in association with, and then returns to the calibration process shown in FIG. 17. On the other hand, if "YES" in the step S155, i.e. when the the next movement track _{T m} and movement track _{T n} is in the association table, in step S157, the next movement track present in the table in association with _{T m} the movement track _{T n} and the constituting each human observation points OBS _(OBS v, OBS _w) and reads the number of frames detected that person observation point OBS, the flow returns to step S143 of FIG. 23.

FIG. 25 is a flowchart of the calculation processing of the constraint between the sensors in step S11 shown in FIG. The calculation processing of the constraints between sensors is executed for each set of sensors 14 by multitasking. As shown in FIG. 25, when the computer 12 (CPU) starts the calculation processing of the constraints between the sensors, in step S161, the latest shared human observation point OBS _p and one or more (for example, three) old ones. The shared human observation point OBS _q is extracted. However, as described above, the old shared human observation point OBS _q is extracted at random.

In the next step S163, the constraints (distance and angle) between the sensors are calculated using the triangle geometry. That is, the computer 12 calculates the constraints between the sensors according to Equation 4 to Equation 11. This is calculated for the latest shared human observation point OBS _p and each of one or more (three in this example) old shared human observation point OBS _q . In step S165, the average of the constraints between the plurality of sensors calculated in step S163 and the variance (error) of the estimated values are calculated by repeatedly using Equations 12 to 15. In step S167, the average of the constraints between the sensors and the variance of the estimated values are stored (added) in the constraint list shown in FIG. 11, and the process returns to the calibration process shown in FIG.

  26 and 27 are flowcharts of the arrangement processing of all the sensors in step S15 shown in FIG. As shown in FIG. 26, when the computer 12 (CPU) starts the arrangement process of all the sensors, the restriction list is rearranged in descending order of reliability in step S181. That is, the computer 12 rearranges the constraint list in ascending order of estimated value variance (error).

  In the next step S183, the reference sensor 14 and the pair of sensors 14 are registered in the sensor position list as shown in FIG. 14 according to the highest reliability constraint. At this time, as described with reference to FIG. 13A, the reference sensor 14 and the pair of sensors 14 are arranged on the virtual map. In the next step S185, the variance (error) of the estimated value for the constraint error between sensors, that is, the constraint with the highest reliability is set in the position error. Although illustration is omitted, at this time, the variance of the estimated value for the highest reliability constraint is registered as an error between sensors corresponding to the reference sensor 14 and the pair of sensors 14 in the sensor position list. That is, the variance of the estimated value for the highest reliability constraint is registered in the first and second lines of the item of error between sensors in the sensor position list shown in FIG. Therefore, position error data corresponding to each of the reference sensor 14 and the pair of sensors 14 is stored as sensor position error data 366.

  Subsequently, in step S187, the variable k is initialized (k = 1). Here, k is a variable for individually identifying each set of sensors 14 and sequentially executing the processing from step S189 on for each set of sensors 14. In the next step S189 shown in FIG. 27, it is determined whether there is an estimated position of the sensor i. Here, as described above, the sensor i means the sensor 14 to which the identification information described on the left side is assigned in the set of the sensors 14 shown in parentheses. The sensor j described later means a sensor 14 to which identification information described on the right side is assigned in a set of sensors 14 indicated in parentheses. In step S189, the computer 12 determines whether or not the estimated position of the sensor i is registered in the sensor position list. The same applies to step S191 described later.

  If “NO” in the step S189, that is, if there is no estimated position of the sensor i, the sensor i used for estimating the position of the sensor j is not arranged, and the process proceeds to a step S201 as it is. On the other hand, if “YES” in the step S189, that is, if there is an estimated position of the sensor i, it is determined whether or not there is an estimated position of the sensor j in a step S191. If “YES” in the step S191, that is, if there is an estimated position of the sensor j, it is not necessary to estimate the position of the sensor j, so the process proceeds to a step S201 as it is.

  On the other hand, if “NO” in the step S191, that is, if there is no estimated position of the sensor j, it is determined whether or not there is a restriction between the sensor i and the sensor j in the restriction list in a step S193. If “NO” in the step S193, that is, if there is no restriction between the sensor i and the sensor j in the restriction list, it is determined that the sensor j cannot be arranged, and the process proceeds to a step S201 as it is. On the other hand, if “YES” in the step S193, that is, if the restriction between the sensor i and the sensor j is in the restriction list, the position of the sensor j is estimated in a step S195 according to the restriction with the sensor i, Add the estimated position of sensor j to the sensor position list. Although illustration is omitted, at this time, the angle of the sensor j is also calculated and registered in the sensor position list.

In the subsequent step S197, an error between the sensors is added to the position error of the sensor i to set it as the position error of the sensor j. However, the error between the sensors is the deviation σ _ij of the estimated value between the sensors i and j as described above. The position error calculated in steps S185 and S197 is used in a sensor position adjustment process (see FIG. 28) described later.

  Subsequently, in step S199, it is determined whether the positions for all the sensors 14 have been estimated. That is, the computer 12 (CPU) determines whether or not the identification information (A to F in this embodiment) of all the sensors 14 is described in the placement sensor ID column of the sensor position list. Alternatively, the computer 12 determines whether or not the identification information described in the sensor position list has reached the total number of sensors 14 (6 in this embodiment).

  If “YES” in the step S199, that is, if the positions of all the sensors 14 are estimated, the process returns to the calibration process shown in FIG. On the other hand, if “NO” in the step S199, that is, if there is a sensor 14 whose position is not estimated, the variable k is incremented by 1 (k = k + 1) in a step S201. In step S203, it is determined whether or not the variable k exceeds the maximum number (15 in this embodiment). That is, it is determined whether or not the processing after step S189 has been executed for all the sets of sensors 14.

  If “YES” in the step S203, that is, if the variable k exceeds the maximum number, the process returns to the calibration process. On the other hand, if “NO” in the step S203, that is, if the variable k is equal to or less than the maximum number, the process returns to the step S189, and the process after the step S189 is executed for the next set of the sensors 14.

  FIG. 28 is a flowchart of the sensor position adjustment process in step S17 shown in FIG. As shown in FIG. 28, when the computer 12 starts the sensor position adjustment process, in step S211, the computer 12 executes an overall error calculation process (FIG. 29) described later. In the next step S213, the variable i is initialized (i = 1). This variable i is a variable for individually identifying the sensors 14.

Next, in step S215, the new position and position error of sensor i are calculated using the estimated positions of all other sensors j (j ≠ i) and the constraints between the sensors. In other words, the computer 12 calculates the position (x _i , y _i ) and the position error ν _i of the new sensor i according to the above-described equations 16 to 19. In the next step S217, the minimum error between the actual angle with respect to the estimated positions of all the other sensors j and the angle estimated by the constraint is set as the angle of the sensor i. That is, the computer 12 calculates the angle θ _{i of the} sensor i according to the equation 20 described above.

  Subsequently, in step S219, the variable i is incremented by 1 (i = i + 1). In step S221, it is determined whether or not the variable i exceeds the maximum number (6 in this embodiment). If “NO” in the step S221, that is, if the variable i is equal to or less than the maximum number, the process returns to the step S215, and the processes of the steps S215 and S217 are executed for the next sensor i. On the other hand, if “YES” in the step S221, that is, if the variable i exceeds the maximum number, a calculation process of the total error is executed in a step S223, and the total error calculated this time is calculated last time in a step S225. It is determined whether or not it is smaller than the overall error (the first time, the overall error calculated in step S211).

  If “YES” in the step S225, that is, if the total error calculated this time is smaller than the total error calculated last time, the process returns to the step S213. That is, the computer 12 further adjusts the sensor position. On the other hand, if “NO” in the step S225, that is, if the total error calculated this time is equal to or larger than the total error calculated last time, the sensor position calculated (estimated) in the previous time is adopted in the step S227, and FIG. Return to the calibration process shown in.

  29 and 30 are flowcharts showing the overall error calculation processing in steps S211 and S223 shown in FIG. As shown in FIG. 29, when the computer 12 (CPU) starts the calculation process of the total error, the computer 12 (CPU) sets the total error value to 0 (initial value) in step S241. In the subsequent step S243, the variable r is initialized (r = 1). In the next step S245, the variable s is initialized (s = 1). However, the variable r and the variable s are variables for identifying each sensor 14 individually.

  In the next step S247, it is determined whether or not the variable r and the variable s match. If “YES” in the step S247, that is, if the variable r and the variable s match, the process proceeds to a step S257 shown in FIG. 30 as it is. On the other hand, if “NO” in the step S247, that is, if the variables r and s do not match, it is determined whether or not the average sensor constraint between the sensor r and the sensor s is in the constraint list in a step S249. to decide.

If “NO” in the step S249, that is, if the average sensor constraint between the sensor r and the sensor s is not in the constraint list, it is determined that an error (d _est −d _cstr ) described later cannot be calculated. Then, the process proceeds to step S257 as it is. On the other hand, if “YES” in the step S249, that is, if the average of the sensor constraints between the sensor r and the sensor s is in the constraint list, the estimated position of the sensor r and the sensor s in the variable d _est in the step S251. Set the distance to the estimated position. In the subsequent step S253, the variable d _cstr is defined by the distance specified by the average of the sensor constraints between the sensor r and the sensor s. In step S255, an error (d _est −d _cstr ) is added to the overall error, and the process proceeds to step S257.

As shown in FIG. 30, in step S257, the variable s is incremented by 1 (s = s + 1). In step S259, it is determined whether the variable s exceeds the maximum number (6 in this embodiment). If “NO” in the step S259, the process returns to the step S247. That is, an error (d _est −d _cstr ) between the sensor r and the next sensor s is calculated, and the overall error is updated.

On the other hand, if “YES” in the step S259, the variable r is incremented by 1 (r = r + 1) in a step S261. In step S263, it is determined whether the variable r has exceeded the maximum number (6 in this embodiment). If “NO” in the step S263, the process returns to the step S245 shown in FIG. That is, an error (d _est −d _cstr ) between the next sensor r and each sensor s is calculated, and the overall error is updated. On the other hand, if “YES” in the step S263, that is, if the variable r exceeds the maximum number, it is determined that the calculation process of the entire error is finished, and the process returns to the sensor position adjustment process shown in FIG.

  According to this embodiment, since the position and angle of each distance sensor are estimated based on the outputs of a plurality of distance sensors arranged in a certain environment, the position and orientation of the distance sensor can be determined without giving an initial value. Can be calibrated. Therefore, even if the measurement environment is changed or the arrangement position of the distance sensor is changed, the user's hand is not bothered.

  In this embodiment, the average and variance (error) of the constraints between the sensors are calculated using the Kalman filter. However, the present invention is not limited to this. As another example, an arithmetic mean or a median filter may be used. When the arithmetic mean is used, the variance (error) can be obtained by calculating the standard variance. When a median filter is used, the variance (error) can be obtained as an approximate value by calculating the median absolute deviation. For example, whether to use a Kalman filter, an arithmetic average, or a median filter may be selected according to the environment (the number of people, the walking speed of people, etc.).

  In this embodiment, only the case where a laser range finder (laser distance meter) is used as the distance sensor has been described. However, the present invention is not limited to this. If the distance for each angle can be detected, a distance sensor using ultrasonic waves, infrared rays, microwaves, and sound waves (with directivity) can be applied.

  Furthermore, in this embodiment, only the position and angle of the sensor in the two-dimensional direction (XY plane) are calibrated. This is because the vertical direction (Z direction) can be adjusted independently of other sensors by providing an acceleration sensor or gyro sensor, for example.

  Furthermore, in this embodiment, by using a speed change as a feature of the movement trajectory, it is determined whether the movement trajectories detected by different sensors are the same based on the difference. It is not necessary to be limited to.

  For example, instead of speed, acceleration (acceleration for each frame) may be used to determine whether or not the movement trajectories detected by two different sensors are the same based on the difference. However, the acceleration can be obtained by differentiating the speed of the movement trajectory. Moreover, the determination method of whether a movement locus | trajectory is the same is the same as that of the above-mentioned Example.

  Moreover, you may determine whether the movement locus | trajectory detected by two different sensors is the same based on the curvature of a movement locus | trajectory. In such a case, a plurality of points constituting the movement trajectory are extracted using an algorithm such as ICP (Iterative Closest Point), and the curvature is calculated in the order of the moving trajectory from the extracted points. To do. Then, when the traveling direction is used as a reference, the direction (right or left) in which the movement locus is bent can be extracted (labeled) in time series. For example, as shown in FIG. 31A, a plurality of points Z1, Z2,..., Z15 are extracted from a certain movement locus detected by a certain sensor. However, the point Z1 is the oldest in time and the point Z15 is the newest in time. Using the extracted points Z1-Z15, the local curvature is calculated according to the time series, and the direction with respect to the traveling direction is determined. Therefore, the movement trajectory can be labeled as “right, right, right, right, left, left, right, left, left, right, left”. However, the curvature is noted in the order (Z1-Z3, Z2-Z4, Z3-Z5,..., Z13-Z15) in which three consecutive points out of a plurality of extracted points follow a time series, and attention is paid to this. It is calculated based on a circumscribed circle of a triangle formed by three points. Further, for example, as shown in FIG. 31A, the traveling direction is temporally from the oldest (first) point (here, the point Z1) among the extracted points. It is determined in the direction toward the newest (last) point (here, the point Z15). Similarly, other sensors are labeled with respect to the movement trajectory. When two different sensors have the same labeling movement locus, they can be determined to be the same movement locus.

  As in the above-described embodiment, when two or more different sensors have the same labeling movement locus, they are not determined as the same movement locus. It is because it cannot be said that it is a common movement locus.

  Furthermore, human behavior can be extracted (labeled) as a feature of the movement trajectory. In this case, it is possible to determine the human local behavior based on the time change of the human observation point included in the movement trajectory. As a method for determining local behavior, a method disclosed in Japanese Patent Application Laid-Open No. 2010-231359, which has been filed by the present applicant and has been published, can be employed. According to this prior art document, four classes of “how to walk”, “speed of walking”, “how to walk for a short time”, and “speed of walking for a short time” are defined, and each class includes a plurality of categories. Classified as (local action). In the “Walking” class (also the “Short Walking” class), “Straight”, “Turn right”, “Turn left”, “Wander”, “U-turn” and “Unknown” Are classified into six categories. Also, in the “walking speed” class (as well as the “short walking speed” class), “running”, “busy walking”, “slow walking”, “stopping” and “waiting” Are classified into five categories. Feature quantities (feature quantity data) are extracted from the movement trajectory, categories are labeled on the extracted feature quantity data, and movement is performed by causing the SVM (Support Vector Machine) to learn the labeled feature quantity data. The way a person walks is determined from the trajectory. And in two different sensors, when the movement locus | trajectory determined with the same way of walking exists only, it can determine with those being the same movement locus | trajectory.

  Even in such a case, as in the above-described embodiment, if there are three or more moving loci determined to be the same way of walking in two different sensors, the same moving loci are not determined.

  Furthermore, as a feature of the movement trajectory, it may be determined whether the movement trajectories detected by two different sensors are the same based on the size. For example, as shown in FIG. 31B, when a circle surrounding the movement trajectory is drawn and there is only one movement trajectory in which different circles are determined to have the same or substantially the same size, they are the same. It can be determined that the movement trajectory is. Although a detailed description is omitted, the circle surrounding the movement trajectory calculates the median value of a plurality of points constituting the movement trajectory, and among the movement trajectories, the median and the point farthest from this median value (basically Is a circle whose radius is the distance from one of the end points of the movement trajectory. However, the median is an average value of XY coordinates in the local coordinate system for a plurality of points. In addition, it is not necessary to actually draw a circle, and it may be determined whether or not the lengths of the radii are the same or substantially the same.

  Even in such a case, as in the above-described embodiment, if there are three or more moving loci in which two different sensors are determined to have the same or substantially the same size, the same It is not determined as a movement trajectory.

  However, as shown in FIG. 31C, it may be determined that the movement trajectories are the same even if the movement trajectories are different based on the size of the circle alone. For this reason, in order to consider not only the size of the moving track but also the dispersion of a plurality of points constituting the moving track, an ellipse that approximates the size of the moving track is drawn as shown in FIG. You may do it. For example, a median value is calculated, and a point (first point) farthest from the median value is detected. However, the median is obtained by the method described above. Further, a point (second point) that is symmetric with respect to the first point around the median is detected, and a line segment connecting the first point and the second point is determined as the major axis of the ellipse. Further, the distance from the long axis to the farthest point in the direction orthogonal to the long axis is detected, and a point (third point) that passes through the median and is separated by the distance in the direction orthogonal to the long axis is calculated. . Further, the third point and the point point (fourth point) are detected with the median as the center, and a line segment connecting the third point and the fourth point is determined as the short axis. In two different sensors, if there is only one moving locus in which the size of the ellipse, that is, the major axis and the minor axis are determined to be the same or substantially the same, the moving tracks are determined to be the same. The

  Even in such a case, as in the above-described embodiment, if there are three or more moving trajectories in which two different sensors are determined to have the same or substantially the same size, the same It is not determined as a movement trajectory.

  Further, as can be seen from FIG. 32A, when the size of the movement trajectory is approximated by an ellipse, the movement trajectory may not be contained within the ellipse, unlike the case where it is enclosed by a circle.

Furthermore, based on the shape of the movement trajectory, it may be determined whether the movement trajectories detected by two different sensors are the same. For example, as shown in the upper side of FIG. 32B, similarly to the case shown in FIG. 31A, a plurality of points Z1-Z15 are extracted from the movement locus using the ICP algorithm. Next, as shown on the lower side of FIG. 32 (B), using a plurality of extracted points, the distance between two adjacent points (here, d _Z1 , d _{Z2,. Z14} ) is calculated. The distance can be calculated using Equation 1 shown in the above-described embodiment. However, the coordinates of two adjacent points are (x1, y1) and (x2, y2). In two different sensors, the difference between the distances calculated according to the time series is obtained, and when there is only one moving locus whose accumulated value of the difference is closest, it is determined that the moving locus is the same. However, when calculating the distance between frames without using the ICP algorithm, it can also be obtained by integrating the speed of the movement trajectory. In such a case, since the distance between frames is calculated, the amount of calculation for obtaining the difference is enormous, but it is considered that the shapes of the movement trajectories can be compared correctly.

DESCRIPTION OF SYMBOLS 10 ... Measuring device 12 ... Computer 14 ... Distance sensor

Claims (7)

A distance sensor calibration device that is arranged in a certain environment and calibrates the position and orientation of a plurality of distance sensors that detect a distance at a certain angle.
Human position detecting means for detecting the position of a human in the certain environment for each distance sensor based on the outputs of the plurality of distance sensors;
A trajectory generating means for generating a human trajectory for each of the distance sensors based on the human position detected by the human position detecting means;
Trajectory identification means for identifying the movement trajectory generated for each of the distance sensors by the trajectory generation means between two different distance sensors based on the characteristics of the movement trajectory,
In the movement locus identified by the locus identification means, the sensor constraints on the distance between the two distance sensors and the relative orientation of the two distance sensors are based on the positions of two arbitrarily selected humans. A sensor constraint calculation unit that calculates each set of the two distance sensors that generated the identified movement locus, and the sensor constraint calculated by the sensor constraint calculation unit, A distance sensor calibration device comprising an estimation means for estimating the position and orientation of a distance sensor.   The sensor constraint calculation unit changes at least one of two arbitrarily selected human positions in the movement trajectory identified by the trajectory identification unit, for each set of the two distance sensors. The distance sensor calibration device according to claim 1, further comprising a sensor constraint averaging unit that calculates a plurality of the sensor constraints and averages the plurality of sensor constraints. A variance calculating means for calculating a variance of the plurality of sensor constraints;
The estimation means includes
Reference sensor placement means for placing two distance sensors having the smallest dispersion calculated by the dispersion calculation means on a virtual map according to an average of sensor constraints of the two distance sensors, and placed by the reference sensor placement means In addition to the two distance sensors, the sequential arrangement means sequentially arranges the distance sensors having an average of sensor constraints with the already arranged distance sensors on the virtual map according to the average of the sensor constraints in order of increasing dispersion. The distance sensor calibration device according to claim 2, comprising: A difference for calculating a difference between each distance of each pair of the distance sensors indicated by the sensor constraint and each distance of each pair of the distance sensors based on the positions of the plurality of distance sensors estimated by the estimation unit. Calculation means,
A minimum value determining means for determining whether or not a total value of differences calculated by the difference calculating means is a minimum value; and the plurality of distances until the total value of the differences becomes a minimum value by the minimum value determining means. 4. The apparatus according to claim 1, further comprising a position adjusting unit that adjusts the position of each sensor based on the position of each of the other plurality of distance sensors estimated by the estimating unit and the sensor constraint. The distance sensor calibration device described. Position error calculating means for calculating an error of the position of each of the plurality of distance sensors estimated by the estimating means based on the variance calculated by the dispersion calculating means;
An error average value calculating means for calculating an average value of errors of the respective positions of the plurality of distance sensors calculated by the position error calculating means;
Position error determination means for determining whether the average value calculated by the error average value calculation means is less than or equal to a predetermined value; and until the position error determination means determines that the average value is less than or equal to a predetermined value. 4. The distance sensor calibration apparatus according to claim 3, further comprising a repeat control unit that repeatedly functions the human position detection unit, the track generation unit, the track identification unit, the sensor constraint calculation unit, and the estimation unit. A calibration program for a calibration apparatus for a distance sensor, which is arranged in a certain environment and calibrates the positions and orientations of a plurality of distance sensors that detect a distance at a certain angle,
In the processor of the calibration device,
A human position detecting step for detecting a human position in the certain environment for each distance sensor based on outputs of the plurality of distance sensors;
A trajectory generating step for generating a human trajectory for each of the distance sensors based on the human position detected by the human position detecting step;
A trajectory identification step for identifying the movement trajectory generated for each of the distance sensors by the trajectory generation step between two different distance sensors based on the characteristics of the movement trajectory;
In the movement locus identified by the locus identification step, sensor constraints on the distance between the two distance sensors and the relative orientation of the two distance sensors are set based on the positions of two arbitrarily selected humans. A sensor constraint calculation step that calculates each set of the two distance sensors that generated the identified movement trajectory, and the sensor constraint calculated by the sensor constraint calculation step. A calibration program for executing an estimation step for estimating the position and orientation of a distance sensor. A distance sensor calibration method for calibrating positions and orientations of a plurality of distance sensors that are arranged in a certain environment and detect a distance at a certain angle,
(A) detecting the position of a person in the certain environment for each distance sensor based on the outputs of the plurality of distance sensors;
(B) Based on the position of the person detected in step (a), for each distance sensor, generate a movement trajectory of the person,
(C) identifying the movement trajectory generated for each of the distance sensors by the step (b) between two different distance sensors based on the characteristics of the movement trajectory;
(D) About the distance between the two distance sensors and the relative orientation of the two distance sensors based on the positions of two arbitrarily selected humans in the movement trajectory identified in the step (c). For each set of the two distance sensors that generated the identified movement trajectory, and
(E) A distance sensor calibration method comprising: estimation means for estimating the positions and orientations of the plurality of distance sensors in the certain environment using the sensor constraint calculated in the step (d).

Images