JPH05172522A - Method for detecting position and bearing of carrier vehicle - Google Patents

Abstract

PURPOSE:To prevent the deterioration of the detecting accuracy of a position and a bearing by correcting the detected values of the position and the bearing so as to minimize the errors between the detected values of the position and the bearing based on the aberration of a lens and the error of the setting position of a mark and the expected values. CONSTITUTION:A TV camera 10 photographs images in all directions around a carrier vehicle 1. The image data as the result of the photographing are imparted into a position/bearing detecting part 11. In the detecting part 11, the present position and bearing of the carrier vehicle 1 are detected based on the position data of marks 2 and 2 on the image by using a specified geometrical method based on the given image data. The data of the position/bearing detected with the detecting part 11 are imparted into a position/bearing correcting part 12. The correcting part 12 corrects the position/bearing data of the carrier vehicle 1 detected with the detecting part 11 by using a back propagation rule. The corrected result is imparted into a running control system 13 for controlling the running such as the steering of the carrier vehicle 1 based on the position/bearing data.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for detecting the position / orientation of a carrier vehicle.

[0002]

2. Description of the Related Art In an unmanned transportation system, a plurality of markers for detecting the position and direction of a light emitter are installed at a predetermined position such as a ceiling or a wall surface in a traveling area of a transportation vehicle, and a wide-angle lens such as a fisheye lens is mounted on the transportation vehicle. The mounted TV camera is mounted, the surroundings of the transport vehicle are imaged by the TV camera, and the current position / direction of the transport vehicle is geometrically determined based on the position and direction of the sign on the captured image. There are some that are designed to detect. The current position of such a carrier
The azimuth detection method has an advantage in that the television camera has a wide field of view and the sign can be reliably imaged, and thus the probability that the position / direction cannot be detected due to the non-imaging of the sign is low.

[0003]

However, the above-described position / orientation detecting method has a drawback in that the position / orientation detecting accuracy is deteriorated due to the aberration of the lens and the error in the installation position of the sign. As a method to solve this difficulty,
A method of previously determining a correction formula for correcting the error and using the correction formula to correct the error to improve detection accuracy can be considered. However, the correction formula becomes non-linear and complicated, and is practical. There was a problem that it was impossible to use.

The present invention has been made in view of the above circumstances, and it is possible to suppress the deterioration of the detection accuracy of the position / direction due to the aberration of the lens and the error of the installation position of the marker, and the position of the carrier vehicle. -The purpose is to provide a direction detection method.

[0005]

According to the method of detecting the position / orientation of a carrier vehicle according to the present invention, an image of a traveling area in which a plurality of markers are provided at predetermined positions is provided from a vehicle traveling in the traveling area to a lens. In the method of detecting the position / azimuth of the transport vehicle by a geometric method based on the position information of the sign on the captured image, the position / azimuth is detected by the geometric method. Based on the value and its expected value, learning for correcting the detected value so as to minimize these deviations is performed in advance on a predetermined neural network, and the learned neural network performs the geometrical analysis. It is characterized in that the detected values of the position / orientation by the dynamic method are corrected.

[0006]

In the present invention, the neural network minimizes the error between the detected value of the position / orientation and its expected value, which is generated based on the aberration of the lens and the error in the installation position of the sign. Learning is performed so as to correct the detection value of, and the detection value of the position and orientation by the geometrical method is corrected based on the learning result, so even if there is an error in the lens aberration and the installation position of the sign. Position / direction detection accuracy does not decrease.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below in detail with reference to the drawings showing the embodiments. FIG. 1 is an explanatory diagram showing a schematic configuration of a carrier system used for carrying out a position / orientation detection method for a carrier vehicle according to the present invention.

In the figure, reference numeral 1 denotes an unmanned carrier, and the carrier 1
The vehicle can run on the floor F in any direction by driving wheels (not shown) provided on the lower surface of the vehicle. Also,
At a predetermined position such as a ceiling or a wall surface in the space S within the traveling range of the carrier 1, a pair of markers 2 and 2 made of a light emitter and serving as an index for detecting the position and direction of the carrier 1 are installed.

A television camera 10 equipped with a fish-eye lens is provided on the upper portion of the transport vehicle 1, and the television camera 10 captures images in all directions around the transport vehicle 1 by the action of the fish-eye lens. This transport vehicle 1 is a TV camera
Based on the positions and orientations of the signs 2 and 2 on the image captured by 10, the geometrical detection of the current position and orientation of the carrier vehicle,
Based on the detection result, the vehicle travels on a predetermined course.

Next, the structure of the traveling control system of the carrier 1 will be described. FIG. 2 is a block diagram showing the configuration of the position / direction detection system of the carrier vehicle 1. The image data which is the image pickup result of the television camera 10 is given to the position / direction detecting unit 11. The position / direction detecting unit 11 detects the current position / direction of the carrier 1 from the position data of the markers 2 and 2 on the image by using a predetermined geometrical method based on the given image data. The position / direction data detected by the position / direction detecting unit 11 is supplied to the position / direction correcting unit 12. The position / orientation data detected by the position / orientation detector 11 is data including the aberration of the fish-eye lens and the installation position error of the markers 2 and 2.

The position / direction correction unit 12 corrects the position / direction data of the carrier 1 detected by the position / direction detection unit 11 using a known back propagation rule (error back-propagation learning rule). Is a neural network that performs the correction, and the correction result is based on the position / direction data.
Is given to the traveling control system 13 that controls traveling such as steering.

Next, the configuration of the neural network of the position / orientation correction unit 12 will be described. FIG. 3 is a block diagram showing the configuration of the neural network of the position / orientation correction unit 12. The neural network of the position / orientation correction unit 12 has a hierarchical structure including three layers of an input layer I, an intermediate layer M, and an output layer O, and the input layer I has three neurons I.
₁ , I ₂ , I ₃ , the intermediate layer M is composed of j neurons M ₁ , M ₂ , M ₃ , ... M _j , and the output layer O is composed of 3 neurons O ₁ , O ₂ , O _3. Is becoming

In the input layer I, the first neuron I ₁
Indicates x of the position data detected by the position / orientation detection unit 11.
The axis coordinate data X ₁ is input, the y-axis coordinate data X _{2 of} the position data detected by the position / orientation detector 11 is input to the second neuron I ₂ , and the third neuron I ₃ is The direction data X ₃ detected by the position / direction detecting unit 11 is input. Each neuron I ₁ , I ₂ , I _{3 of the} input layer I
Is all neurons M ₁ , M ₂ , M ₃ , ... M of the hidden layer M
They are coupled to _j with different coupling coefficients.

The neurons M ₁ , M ₂ , M ₃ , of the intermediate layer M
The output value Z _j (output values Z ₁ , Z ₂ , Z ₂ ) of M _j is based on the total value of the inputs from the input layer I (see the following equation (1)).
₃ , ... (general term of Z _j ) is determined. This output value is determined by
It is calculated by the sigmoid function (see the formula (2) below). Further, each neuron M ₁ , M ₂ , M ₃ , of the intermediate layer M,
... M _j is all neurons O ₁ , O ₂ , O ₃ of the output layer O
Are coupled with different coupling coefficients. The neurons O ₁ , O ₂ , and O ₃ of the output layer O are based on the total value of the inputs from the intermediate layer M (see the following equation (1)), and their output values Y _k (output values Y ₁ , Y ₂ , general term for Y ₃₎ is determined. This output value is also determined by a sigmoid function (see the equation (2) below). However, in the output layer O, from the first neuron O ₁ value obtained by correcting the x-axis coordinate data X ₁ (Y ₁₎ is output, the correction in the y-axis coordinate data X ₂ from the second neuron O ₂ The output value (Y ₂ ) is output and the third
A value (Y ₃ ) obtained by correcting the azimuth data X ₃ is output from the neuron O _{3 of} .

The concrete neurons M ₁ and M of the intermediate layer M
₂ , M ₃ , ... M _j and the neuron O _{1 of} the output layer O,
The output values of O ₂ and O ₃ are determined by the linear addition value ne of the input value a _i with the coupling coefficient w _i as the weight shown in the following formula (1).
An output value b obtained by obtaining t and substituting the value into a sigmoid function as shown in the following equation (2) is determined as an output value of each neuron.

[0016]

[Equation 1]

[0017]

[Equation 2]

Next, a learning method based on the back propagation rule will be described. Learning of the back propagation rule is divided into forward calculation and backward calculation.

In the forward calculation, first, the input data X
_i (general term for input data X ₁ , X ₂ , X ₃ ) is input from the input layer I, and the output value Z _j of each neuron of the intermediate layer M is obtained from the equations (1) and (2). And this output value Z
The output value Y _k of each neuron of the output layer O is obtained from _j by the above equations (1) and (2).

In the backward calculation, the intermediate layer M and the input layer I are arranged so that the error E between the teacher data t _k expressed by the following equation (3) and the output value Y _k of the output layer O is successively reduced. and it changes the coupling coefficient w _kj between the coupling coefficient w _ji and an output layer O and the intermediate layer M between.

[0021]

[Equation 3]

When the error E is partially differentiated by the coupling coefficient w _kj between the output layer O and the intermediate layer M, the following equation (4) is obtained.

[0023]

[Equation 4]

[0024] Therefore, the correction amount Δw _kj of the coupling coefficient w _kj is,
It is expressed as the following equation (5) using the learning coefficient η which is a parameter for controlling the magnitude of the correction amount Δw _kj .

[0025]

[Equation 5]

Similarly, when the error E is partially differentiated by the coupling coefficient w _ji between the intermediate layer M and the input layer I, the following equation (6) is obtained.

[0027]

[Equation 6]

[0028] Therefore, the correction amount Δw _ji of the coupling coefficient w _ji is,
The learning coefficient η, which is a parameter for controlling the magnitude of the correction amount Δw _ji , is used to represent the following expression (7).

[0029]

[Equation 7]

In this way, the correction amount Δ of the coupling coefficient w _kj
The correction amount Δw _ji of w _kj and the coupling coefficient w _ji is calculated to correct the coupling coefficient w _kj and the coupling coefficient w _ji . However, in practice, considering the convergence of learning, the previous correction amounts Δw _kj (t-1), Δ
The w _ji (t-1) is stored and the correction amounts Δw _kj (t) and Δw _ji (t) are obtained by using the following equations (8) and (9).

[0031]

[Equation 8]

[0032]

[Equation 9]

In the back propagation rule as described above, the forward calculation and the backward calculation as described above are repeated for all learning patterns, and learning is performed until the error E becomes sufficiently small.

Next, a method of detecting the position / direction of the carrier 1 based on the learning based on the back propagation rule as described above will be described. First, before the actual traveling of the transport vehicle 1, a travel for learning a method of correcting the detection result by the position / orientation detection unit 11 is performed on the actual travel route of the transport vehicle 1. This learning is performed in the position / orientation correction unit 12.

FIG. 4 is a flow chart showing the learning procedure of the correction method of the detection result of the position / direction. First, the number of learning data is reset to "0" (step S1), and the operator moves the carrier 1 to a known position / orientation (step S1).
S2).

Then, at that position, the television camera
Based on the positions / directions of the signs 2 and 2 on the image captured by 10 the position / direction detection unit 11 geometrically detects the current position / direction of the carrier vehicle 1 (step S3), The detection data is given to the position / orientation correction unit 12. Then, the operator inputs the correct position / orientation data at that position to the position / orientation correction unit 12 as teacher data (step S4). The position / orientation correction unit 12 stores the detection data and the teacher data regarding the position / orientation as one set of learning data (step S5).

Then, the number of learning data is increased by 1 (step S6), and it is determined whether or not the stored learning data exceeds a predetermined number (step S7).

If it is determined in step S7 that the number of stored learning data does not exceed the predetermined number,
Returning to step S2, the processing as described above is repeated, and the learning data is collected until the number exceeds the predetermined number. On the other hand, in step S7, when it is determined that the stored learning data exceeds the predetermined number, the back propagation rule as described above is used based on the collected learning data, and the set of the detection data and the teacher data is set. The learning is performed so that the neural network outputs the correct output from (step S8), and the traveling for the learning ends.

When the learning as described above is completed, the actual traveling for conveying the conveyed object is performed. FIG. 5 is a flowchart showing a procedure for detecting the position / direction of the carrier vehicle 1 during actual traveling.

First, the position / direction detector 11 detects the current position / direction of the carrier 1 (step S11). The detected values of the position / direction are information including the aberration of the fisheye lens and the installation position error of the markers 2 and 2.

Next, in the position / azimuth correcting unit 12, the detected values of the position / azimuth of the position / azimuth detecting unit 11 are corrected using a neural network which has been learned for correction as described above (step S12), and outputs the corrected detected value of the position / azimuth to the travel control unit 13 as the final detected value of the position / azimuth (step S13).

As described above, in the position / azimuth correction unit 12, the position / direction detection value of the position / direction detection unit 11 is preliminarily learned so as to be corrected so that the deviation from the expected value is minimized. Since the detection value of the position / orientation detecting unit 11 is corrected by using the neural network, even if there is an aberration of the fisheye lens and an installation position error of the markers 2 and 2, the output of the position / orientation correcting unit 12 of the carrier vehicle 1 is detected. The accuracy of the corrected detection value of the current position / direction does not decrease.

[0043]

As described above in detail, in the method of the present invention, the neural network minimizes the error between the detected value of the position / orientation and its expected value, which is caused based on the aberration of the lens and the error of the installation position of the sign. It is learned to correct the detected values of the position and azimuth so that the detected values of the position and azimuth by the geometrical method are corrected based on the learning result. Further, the present invention has excellent effects such as reduction of the detection accuracy of the position / orientation even if there is an error in the installation position of the sign.

[Brief description of drawings]

FIG. 1 is an explanatory diagram showing a schematic configuration of a transportation system used for implementing a position / orientation detecting method for a transportation vehicle according to the present invention.

FIG. 2 is a block diagram showing a configuration of a position / orientation detection system of a carrier vehicle.

FIG. 3 is a block diagram showing a configuration of a neural network of a position / azimuth correction unit.

FIG. 4 is a flowchart showing a learning procedure of a correction method of a position / azimuth detection result.

FIG. 5 is a flowchart showing a procedure for detecting the position / direction of a carrier vehicle in actual traveling.

[Explanation of symbols]

 1 Transport Vehicle 2 Sign 10 TV Camera 11 Position / Direction Detector 12 Position / Direction Corrector I Input Layer M Intermediate Layer O Output Layer S Space

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁵ Identification number Office reference number FI technical display location G06F 15/18 8945-5L 15/62 415 9287-5L G06G 7/60

Claims (1)

[Claims] 1. An image of a traveling area in which a plurality of signs are respectively provided at predetermined positions is imaged through a lens from a transportation vehicle traveling in the traveling area, and position information of the sign on the imaged image. In the method of detecting the position / azimuth of the guided vehicle by a geometrical method based on the above, based on the detected value of the position / azimuth by the geometrical method and its expected value, these deviations are minimized. Therefore, learning for correcting the detected value is performed in advance for a predetermined neural network, and the detected value of the position / orientation by the geometrical method is corrected by the learned neural network. Car position / direction detection method.