JP2018185312A - Position detector, model learning device, position detection system, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect a signal generation position if there is a complexity in the propagation of signals in a target region.SOLUTION: The position detector 20 of a position detection system 10 is provided in a plurality of parts of a target region, and acquires sensor values detected by each of sensors 10 detecting signals. The position detector 20 detects a signal generation position in the target region on the basis of a learned model learned in advance from learning data which represents the place where a learning signal generates in the target region and a learning sensor value detected by each sensor 10 when a signal generates and propagates in the target region and on the basis of each acquired sensor value.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a position detection device, a model learning device, a position detection system, and a program.

  Conventionally, position sensors that detect the position of an object are known. For example, a system for identifying the presence and posture of a vehicle object is known (for example, Patent Document 1). This system includes weight sensors, capacitive sensors, inductive sensors, ultrasound, optics, electromagnetics, motion, infrared and radar. Also, the algorithm inherent in the system's processor generates an output indicating the current occupancy of the sheet upon entry of the data set. The algorithm of this system may be a neural network or a neural fuzzy algorithm created by an appropriate algorithm creation program.

  In addition, a system for identifying a physical event using a vibration signal is known (for example, Patent Document 2). The system identifies an actual physical event by receiving a vibration signal associated with an actual physical event and comparing the vibration signal with a vibration signal associated with a known physical event. This system is specifically applied to an electric distribution system, a water distribution system, a gas distribution system, or the like.

  Further, based on the voltage generated by the deformation of the piezoelectric film, the position information of the user's finger is estimated according to the sum of squares of the error of each voltage ratio between the stored voltage pattern and the measured voltage pattern. A touch input device is known (for example, Patent Document 3).

  In addition, it is known that the position of a terminal is estimated according to reception level information that can be acquired in a base station (for example, Non-Patent Document 1).

Japanese translation of PCT publication No. 2004-522932 JP 2012-154928 A Japanese Patent Laid-Open No. 2006-6683

Noboru Kanazawa, Atsushi Nagata, “A Study on Location Estimation Using Machine Learning Using Neural Networks”, 2017, Proceedings of Communication Conference 1, IEICE General Conference

  However, in the techniques described in Patent Document 1, Patent Document 2, and Patent Document 3, it is necessary to arrange a large number of sensors at predetermined positions in order to obtain desired sensor information. For example, in the system described in Patent Document 1, a large number of sensors are arranged at predetermined positions in consideration of the space in the vehicle.

  Further, in the technique described in Non-Patent Document 1, reception level information that can be acquired at the base station is used, but it is simply a model of a signal that propagates in space and a position. It is not possible to cope with the case where the propagation complexity of

  The present invention has been made in view of the above circumstances, and a position detection apparatus, a model learning apparatus, and the like that can detect the generation position of a signal even when the signal has a complexity of signal propagation in a target region. An object is to provide a position detection system and a program.

  In order to achieve the above object, a position detection device according to the present invention is provided in a plurality of locations in a target region, and an acquisition unit that acquires sensor values detected by each of a plurality of sensors that detect signals, Learning learned in advance from learning data representing a learning sensor value detected by each of a plurality of sensors when the signal is generated in the target region and the signal is generated and propagated through the target region. A detection unit that detects a generation position of a signal in the target region based on the completed model and each of the sensor values acquired by the acquisition unit.

  The detection unit may detect a position of an object that generates the signal according to a generation position of the signal.

  The acquisition unit is configured to acquire voltage values detected by each of a plurality of piezoelectric sensors that are provided at a plurality of locations in the target region and detect the force as the signal, and the learning force in the target region is acquired. A learning model previously learned from learning data representing a learning voltage value detected by each of a plurality of piezoelectric sensors when the force is generated and propagated through the target region, and the acquisition unit The force generation position in the target region is detected based on each of the voltage values acquired by the above, and the position of the target object is detected according to the force generation position.

  The learned model may be a neural network model previously learned from the learning data by deep learning.

  Further, the neural network model may be the neural network model having a larger number of neurons of the neural network model as the target region is larger.

  The neural network model may be the neural network model in which the number of neurons in the neural network model increases as the number of the plurality of sensors increases.

  The signal may be at least one of force, light, electron beam, heat, magnetic signal, and electrical signal.

  The plurality of sensors are mounted on each of the plurality of moving bodies, and the detection unit determines the learning signal generation positions and the learning sensor values according to the movement forms of the plurality of moving bodies. The generation position of the signal in the target area can be detected based on the learned model learned in advance from the learning data to be represented and each of the sensor values acquired by the acquisition unit.

  Further, the acquisition unit further acquires power obtained from the plurality of sensors, and outputs the generation position of the signal detected by the detection unit using the power acquired by the acquisition unit. Can be further provided.

  When the target region is a two-dimensional space, the plurality of sensors can be at least three.

  When the target region is a three-dimensional space, the plurality of sensors can be at least four.

  Further, the acquisition unit acquires the sensor value at each time, and the detection unit, based on the learned model and each of the sensor values at each time acquired by the acquisition unit, The generation position of the signal at each time in the region can be detected.

  The detection unit may further detect a locus of the signal generation position based on the signal generation position at each time in the target region.

  In addition, the detection unit can predict the generation position of the signal at the next time based on the generation position of the signal up to the current time in the target area.

  The plurality of sensors are installed at each location of a three-dimensional object as the target region, and when the three-dimensional object is divided into each region by a surface, at least one of the volume and shape of each region is It can be installed at each location on the three-dimensional object forming the different surfaces.

  In addition, when the plurality of sensors are installed at each location of a two-dimensional region as the target region and the two-dimensional region is divided into each region by a line, at least one of the area and shape of each region is It can be installed at each location on the two-dimensional region forming the different lines.

  The model learning device of the present invention is a learning that represents a learning signal value detected by each of a plurality of sensors when a signal for learning in a target region is generated and the signal is generated and propagates through the target region. A learning data acquisition unit for acquiring data; and a learning unit for learning a model for detecting a signal generation position from a plurality of sensor values based on the learning data acquired by the learning data acquisition unit. It consists of

  The position detection system of the present invention includes a plurality of sensors that are provided at a plurality of locations in a target area and that detect signals, an acquisition unit that acquires sensor values detected by each of the plurality of sensors, and a target area. A learning model that is learned in advance from learning data representing a learning sensor value detected by each of a plurality of sensors when the signal for learning is generated and the signal is generated and propagates through the target region. And a detection unit that detects a signal generation position in the target region based on each of the sensor values acquired by the acquisition unit.

  The program of the present invention includes a computer provided at a plurality of locations in a target area, and an acquisition unit that acquires sensor values detected by each of a plurality of sensors that detect signals, and the signal for learning in the target area A learning model pre-learned from learning data representing a sensor value for learning detected by each of a plurality of sensors when the signal is generated and propagates through the target region, and the acquisition unit It is a program for functioning as a detection unit that detects a generation position of a signal in the target region based on each of the acquired sensor values.

  As described above, according to the position detection device, model learning device, position detection system, and program of the present invention, the generation position of the signal is detected even when the signal propagation complexity is present in the target region. The effect that it can do is acquired.

It is a figure which shows an example of schematic structure of the position detection system of the 1st Embodiment of this invention. It is explanatory drawing for demonstrating the position detection of this embodiment. It is explanatory drawing for demonstrating the generation position of the force which generate | occur | produces on the top plate of a table. It is a flowchart which shows the content of the learning process routine which concerns on the 1st Embodiment of this invention. It is a flowchart which shows the content of the position detection process routine which concerns on the 1st Embodiment of this invention. It is explanatory drawing for demonstrating the 2nd Embodiment of this invention. It is explanatory drawing for demonstrating the modification of embodiment of this invention. It is explanatory drawing for demonstrating the modification of embodiment of this invention. It is explanatory drawing for demonstrating the modification of embodiment of this invention. It is explanatory drawing for demonstrating the modification of embodiment of this invention. It is explanatory drawing for demonstrating arrangement | positioning of a sensor in case a target area | region is two-dimensional. It is explanatory drawing for demonstrating arrangement | positioning of a sensor in case a target area | region is three-dimensional. It is explanatory drawing for demonstrating the modification of embodiment of this invention. It is explanatory drawing for demonstrating the modification of embodiment of this invention. It is explanatory drawing for demonstrating the modification of embodiment of this invention. It is explanatory drawing for demonstrating the modification of embodiment of this invention. It is explanatory drawing for demonstrating the modification of embodiment of this invention. It is explanatory drawing for demonstrating the modification of embodiment of this invention.

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

<First Embodiment>
FIG. 1 is a block diagram illustrating a position detection system 100 according to the first embodiment. The position detection system 100 includes a plurality of sensors 10 that detect signals, a reception unit 12, and a position detection device 20.

  In the present embodiment, a signal generation position in the target region is detected according to the sensor values detected by the plurality of sensors 10, and the position of the target object is detected according to the signal generation position. For example, as shown in FIG. 2, when four sensors 10 are arranged in the target area A, the position detection device 20 of the position detection system 100 detects the target according to the signal S generated from the target X. The position of X is detected.

  Specifically, first, the position detection system 100 according to the present embodiment learns the position of the object X in the target area A and the sensor values detected by the plurality of sensors 10 when the object X exists at the position. Collect as data.

  Next, the position detection system 100 learns a neural network model, which is an example of a model, by deep learning based on the collected learning data, and obtains a learned neural network model.

  Then, the position detection system 100 detects the signal generation position in the target region based on the sensor values detected by the plurality of sensors 10 and the learned neural network model, and the target object according to the signal generation position. The position of is detected.

  In the first embodiment, the case where the sensor is a piezoelectric sensor and the signal generated from the object is force will be described as an example. The position detection system 100 has a learning function and a position detection function. Hereinafter, each functional unit corresponding to each function will be described.

[Learning function]
First, the learning function in the position detection system 100 will be described.

  The plurality of sensors 10 are installed at a plurality of locations in the target area. The plurality of sensors 10 detect learning sensor values when a signal is generated from the learning object and propagates through the target region.

  For example, as shown in FIG. 3, when the target region is a table T that is a rigid body, the sensor 10 that is a piezoelectric sensor is installed on the ground surface of the leg portion that is a support member of the table T and the floor surface. The The plurality of sensors 10 has, as learning sensor values, sensor values when a learning object is present at a predetermined position in the object region and force is applied from the object and propagates through the top plate and the leg portion. To detect. For example, when an object is placed at the generation position P of the table T, a force S is applied to the top plate surface at the generation position P and propagates through the top plate and legs of the table T. Then, the force applied to the top plate propagates through the legs of the table T, and sensor values are detected by the plurality of sensors 10. When the sensor 10 is a piezoelectric sensor, the sensor value is a voltage value.

  In this case, in order to detect the generation position P of the force S from the sensor value detected by the sensor 10, the shape and material of the table T that is the target region, the arrangement of the sensor, and the like are taken into account. It is necessary to model the propagation of force from the generation position P of the force S to the sensor. However, an actual table is difficult to model because the material is not uniform or the shape may vary. Also, since there are variations in detection characteristics of sensor values for a plurality of sensors, modeling is difficult.

  Therefore, in the present embodiment, the neural network model is learned by deep learning using the fact that sensor value patterns detected by the plurality of sensors 10 differ according to the position of the force applied to the top plate of the table T. Then, the position of the object is detected by detecting the force generation position using the learned neural network model obtained. Accordingly, it is possible to detect the position where the force is generated and to detect the position of the object according to the position where the force is generated without considering variations in the shape and material of the table T, which is the target area, and variations in sensor characteristics. it can.

  The reception unit 12 receives a generation position of a learning force in the target region when a force is actually applied to the target region for learning. Specifically, the reception unit 12 receives a generation position of a force generated by an object. The reception unit 12 is realized by a keyboard, a mouse, and the like, for example. When the reception unit 12 is realized by a keyboard or the like, the reception unit 12 receives a generation position of a force input manually.

  The position detection device 20 is configured by a computer including a CPU, a RAM, and a ROM storing a program for executing each processing routine, and is functionally configured as follows.

  As shown in FIG. 1, the position detection device 20 includes an information acquisition unit 22, a learning unit 24, a learned model storage unit 26, a detection unit 28, and an output unit 30. The information acquisition unit 22 is an example of an acquisition unit and a learning data acquisition unit.

  The information acquisition unit 22 acquires the generation position of the learning force received by the reception unit 12. Further, the information acquisition unit 22 acquires a sensor value for learning detected by each of the plurality of sensors 10 when a force is actually applied to the target region for learning.

  Then, the information acquisition unit 22 associates the learning sensor value detected by each of the plurality of sensors 10 when the learning force generation position and the learning force are generated and propagates through the target region. Then, it is stored in a storage area (not shown) as learning data. A storage area (not shown) stores a plurality of learning data.

  The learning unit 24 learns a neural network model for detecting a signal generation position from a plurality of sensor values based on the learning data acquired by the information acquisition unit 22. In the present embodiment, the learning unit 24 learns the neural network model by deep learning. Then, the learning unit 24 stores the learned neural network model obtained by deep learning in the learned model storage unit 26.

  The learned model storage unit 26 stores a learned neural network model obtained by the learning unit 24.

[Position detection function]
Next, the position detection function in the position detection system 100 will be described.

  The information acquisition unit 22 acquires each of the sensor values detected by each of the plurality of sensors 10 when the target object exists in the target region.

  The detection unit 28 detects the force generation position in the target region based on the learned neural network model stored in the learned model storage unit 26 and each of the sensor values acquired by the information acquisition unit 22. Since the force generation position and the position of the target object correspond to each other, the position of the corresponding target object is also detected by detecting the force generation position.

  Specifically, the detection unit 28 inputs each of the sensor values acquired by the information acquisition unit 22 to the learned neural network model, and acquires the position of the object output from the learned neural network model.

  The output unit 30 outputs the position of the object detected by the detection unit 28 as a result.

<Operation of Position Detection System 100>
Next, the operation of the position detection system 100 of this embodiment will be described. In the position detection device 20, a learning processing routine for learning a neural network model based on the learning data and a position detection processing routine for detecting the position of the target object based on the learned neural network model are executed.

<Learning processing routine>
First, when a force is actually applied to the target region for learning, the sensor values for learning are detected by the plurality of sensors 10 of the position detection system 100, and the learning unit generates the learning force. When the position is accepted, the information acquisition unit 22 acquires the generation position of the learning force and each of the learning sensor values. Then, the information acquisition unit 22 stores the learning force generation position and each of the learning sensor values in association with each other in a storage area (not shown) as learning data.

  And the position detection apparatus 20 will perform the learning process routine shown in FIG. 4, if the instruction | indication signal of learning process execution is received.

  In step S100, the learning unit 24 acquires a plurality of learning data stored in a storage area (not shown).

  In step S102, the learning unit 24 learns the neural network model by deep learning based on the plurality of learning data acquired in step S100.

  In step S104, the learning unit 24 stores the learned neural network model obtained in step S102 in the learned model storage unit 26, and ends the learning processing routine.

<Position detection processing routine>
When the learned neural network model is stored in the learned model storage unit 26, when the position detection device 20 receives an instruction signal for execution of the position detection process, the position detection process routine shown in FIG. The position of the target object existing in is detected.

  In step S <b> 200, the information acquisition unit 22 acquires each of sensor values corresponding to the force generated from the target when the target exists in the target region.

  In step S202, the detection unit 28 determines the force in the target area based on the learned neural network model stored in the learned model storage unit 26 in the learning process routine and each of the sensor values acquired in step S200. Detects the occurrence position. And the detection part 28 detects the position of the target object according to the generation | occurrence | production position of force.

  In step S204, the position of the object detected in step S202 is output as a result, and the position detection processing routine ends.

  As described above, according to the position detection system according to the first embodiment, the generation position of the learning signal in the target area and the detection by the plurality of sensors when the signal is generated and propagates through the target area. By learning the neural network model by deep learning based on the learning data representing the learned sensor value, it is possible to easily obtain a model that takes into account the complexity of signal propagation in the target region. Thus, the target area can be converted into a position sensor only by installing a plurality of sensors in the target area.

  Further, based on a learned neural network model learned in advance by deep learning and each of sensor values detected by each of a plurality of sensors, a signal generation position in the target region is detected, and the signal generation position Accordingly, by detecting the position of the object, the position of the object can be detected without considering the target area.

  In addition, for example, in a conventional position detection system using piezoelectric elements, a matrix sensor in which a large number of piezoelectric elements are spread is used, but in the position detection system of the present embodiment, several piezoelectric elements are used. Only with this, position detection with high accuracy becomes possible. As a result, the position detection system can be easily realized at low cost.

  In addition, the position detection system of the present embodiment can be mounted without learning the shape or scale of the target region by learning a plurality of learning data. In addition, the characteristics of the piezoelectric element are also incorporated into the learning, and the characteristic difference between the elements can be ignored.

  In addition, even when signal propagation is difficult in the target region, the signal generation position can be detected by a plurality of sensors.

<Second Embodiment>
Next, a second embodiment will be described. In addition, since the structure of the position detection system which concerns on 2nd Embodiment becomes a structure similar to 1st Embodiment, it attaches | subjects the same code | symbol and abbreviate | omits description.

  The second embodiment is different from the first embodiment in that a plurality of sensors are mounted on each of a plurality of moving bodies, and a neural network model is learned according to a moving form of the plurality of moving bodies.

  For example, as illustrated in FIG. 6, a case where the sensor 10 is mounted on a drone 50 that is an example of a moving body will be described as an example. The sensor 10 detects an electric signal.

  In 2nd Embodiment, as FIG. 6 shows, it is made to fly so that each positional relationship of the drone 50 carrying the sensor 10 may become predetermined | prescribed positional relationship. Then, when the electric signal S is actually generated in the target region for learning, the generation position P of the learning electric signal S corresponding to the flight form of the drone 50 and the sensors 10 mounted on the plurality of drones 50 are provided. Learning data representing each of the sensor values detected by is acquired. Then, the neural network model is learned from the learning data by deep learning to obtain a learned neural network model. Thereby, the plurality of drones 50 can be converted into position sensors using the learned neural network model. This will be specifically described below.

[Learning function]

  As shown in FIG. 6, the plurality of sensors 10 of the position detection device 20 according to the second embodiment are mounted on each of the drones 50 that are examples of the moving body. The plurality of sensors 10 mounted on each of the plurality of drones 50 detect sensor values for learning when an electric signal S is generated from the object X and propagates through the target region. Note that the target area is a three-dimensional space based on a plurality of drones 50 having a predetermined positional relationship, and the target area moves as the plurality of drones 50 fly. Further, the occurrence position is a position in the target area with the plurality of drones 50 as a reference.

  In addition, the reception unit 12 of the second embodiment receives the generation position P of the learning electrical signal S in the target region. At this time, the reception unit 12 acquires a positional relationship between the plurality of drones 50 and the generation position P of the electric signal S according to the flight form of the plurality of drones 50.

  The information acquisition unit 22 of the position detection device 20 of the second embodiment acquires the positional relationship between the plurality of drones 50 and the generation position P of the electric signal S acquired by the reception unit 12. In addition, a sensor value for learning detected by each of the plurality of sensors 10 is acquired.

  The information acquisition unit 22 is mounted on the drone 50 when the generation position P of the learning electric signal S corresponding to the flight form of the drone 50 and the electric signal S for learning are generated and propagated through the target region. The learning sensor values detected by each of the plurality of sensors 10 are associated with each other and stored as learning data in a storage area (not shown).

[Position detection function]
The information acquisition unit 22 acquires each of the sensor values detected by each of the plurality of sensors 10 when the target object X exists in the target region.

  Based on the learned neural network model stored in the learned model storage unit 26 and each of the sensor values acquired by the information acquisition unit 22, the detection unit 28 determines the generation position P of the electrical signal S in the target region. To detect. Since the generation position P of the electric signal S corresponds to the position of the object X, the position of the corresponding object X is also detected by detecting the generation position P of the electric signal S.

  In addition, about the other structure and effect | action of the position detection system concerning 2nd Embodiment, since it is the same as that of 1st Embodiment, description is abbreviate | omitted.

  As described above, according to the position detection system according to the second embodiment, learning is performed in advance from learning data representing learning signal generation positions and learning sensor values corresponding to a plurality of drone flight modes. By detecting the generation position of the signal in the target area based on the learned model and each of the sensor values, the position of the target can be detected without considering the target area.

  Further, by moving a plurality of sensors, the position of a wide range of objects can be detected using a small number of sensors. In addition, although the example of the drone which moves in the air was demonstrated, it is applicable also to the moving body underwater or underground.

<Third Embodiment>
Next, a third embodiment will be described. In addition, since the structure of the position detection system which concerns on 3rd Embodiment becomes a structure similar to 1st Embodiment, it attaches | subjects the same code | symbol and abbreviate | omits description.

  The third embodiment is different from the first or second embodiment in that the position detection function detects the generation position of the signal at each time based on the sensor value at each time.

  For example, in the first embodiment, when two points of force are applied to the top plate of the table T, the two barycentric points are detected as signal generation positions. Therefore, in the third embodiment, the sensor value at each time detected by each of the plurality of sensors 10 is stored in a storage area (not shown), and the generation position of the signal at each time is determined from the sensor value at each time. Detect individually. Thereby, even if there are a plurality of signal generation positions, the generation positions of the signals are detected without detecting the center of gravity of the generation positions of the plurality of signals as the positions, and the positions of the respective objects are detected.

  Specifically, the information acquisition unit 22 acquires each sensor value at each time detected by each of the plurality of sensors 10, and associates the acquired sensor value with the time to store a storage area (not shown). To store.

  Based on the learned neural network model stored in the learned model storage unit 26 and each of the sensor values at each time stored in the storage area (not shown), the detection unit 28 detects each time in the target area. Detect the signal generation position. Since the signal generation position corresponds to the position of the target object, the position of the corresponding target object at each time is also detected by detecting the signal generation position at each time.

  In addition, about the other structure and effect | action of the position detection system which concern on 3rd Embodiment, since it is the same as that of 1st Embodiment, description is abbreviate | omitted.

  As described above, according to the position detection system according to the third embodiment, generation of a signal at each time in the target region based on a learned model learned in advance and each sensor value at each time. By detecting the position, the position of the object can be detected even when there are a plurality of signal generation positions.

  Note that the present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention.

  For example, the signal is a force in the first embodiment, and the signal is an electric signal in the second embodiment. However, the present invention is not limited to this. For example, the signal may be any one of light, electron beam, heat, and magnetic signal. Further, a plurality of types of sensors may be combined, and a signal generation position may be detected based on a plurality of types of different signals. Multiplexing of sensor elements is easy, and detection accuracy of a signal generation position can be improved by combining a plurality of types of sensors.

  For example, since force and heat are generated from the fingertip as signals, the accuracy of finger position detection can be achieved by installing a stress sensor and heat sensor in the target area so that the force and heat generated by the finger can be detected. Can be improved. In addition, by combining a plurality of types of sensors, the position detection accuracy can be set to a factory even if the signal includes noise.

  In the first embodiment, the target area is a table, and in the second embodiment, the target area is a space. However, the present invention is not limited to this.

  For example, as illustrated in FIG. 7, the target area A may be a fabric, and the force generation position may be detected based on the tension generated by the force S.

  Also, for example, as shown in FIG. 8, when the target area A is a screen, the laser is irradiated by the laser pointer Y, and the signal S is the reflected light of the laser, the laser irradiated on the screen is reflected. You may make it detect the generation | occurrence | production position P with which the laser is irradiated based on the light to generate | occur | produce. The laser may be an electron beam.

  Further, as shown in FIG. 9, the target area A may be a three-dimensional space, and the signal S may be magnetism generated by the electronic device X. Further, as shown in FIG. 10, the target area A may be a board, and the signal S may be magnetism generated by a magnet X. Further, the signal S shown in FIG. 10 may be heat generated by the heat source X.

  As shown in FIG. 11, when the target region is a two-dimensional space, two-dimensional coordinates can be detected as the position of the target object by installing at least three sensors. Also, as shown in FIG. 12, when the target region is a three-dimensional space, three-dimensional coordinates can be detected as the position of the target object by installing at least four sensors.

  In each of the above embodiments, only the position of the object is detected.For example, the power obtained from a plurality of sensors is further acquired, and the generated position of the detected signal is determined using the acquired power. You may make it output outside. Thereby, the position detection of a target object and transmission of a detection result can be performed without a power source.

  For example, the position detection of the object and the transmission of the detection result can be performed without a power source by generating electric power from the piezoelectric element using vibration power generation. This method can be implemented on an IoT (Internet of Things) device or the like because the wiring of the device and the battery can be reduced. In the case of using a conventional matrix sensor, it is very complicated and inefficient to add electric power. However, according to the present embodiment, for example, the force is concentrated on several piezoelectric elements. And a simple non-power sensor can be realized.

  Further, it is conceivable that the accuracy of position detection decreases as the target area increases. For example, when a piezoelectric element is used as a sensor, the applied force is used for the deformation of the rigid body as the target region is larger, and the sensor value detected by each piezoelectric element may be reduced. When using a light wave sensor, the larger the target area, the longer the distance between the target and the sensor and the longer it takes to receive light. Therefore, the number of layers in the neural network model may be increased, and the number of neurons in the neural network model may be increased as the target region is larger. Further, the greater the number of sensors, the greater the number of neurons in the neural network model.

  Moreover, although the case where the moving body is a drone has been described as an example in the second embodiment, the present invention is not limited to this. For example, the moving body may be a vehicle or an underwater moving body. When the moving body is an underwater moving body, the target region is underwater, and the position of a signal generated in the water is detected.

  In the above embodiment, the case where the learning function and the position detection function are configured as one device has been described as an example. However, the learning function and the position detection function may be configured as separate devices. In this case, for example, a model learning device having a learning function and a position detection device having a position detection function may be configured.

  In addition, various objects can be considered as application examples of the above embodiment. For example, the floor surface can be converted into a position sensor by installing a plurality of sensors on the floor surface. Thereby, for example, since the position of the person is detected, the position of the people and the flow line management can be performed.

  Further, for example, by installing a plurality of sensors on the table legs, the table top is converted into a position sensor. Thereby, for example, TV, illumination, an air conditioner, and a PC can be operated by applying a force to a specific position of the table top.

  In the third embodiment, the case where the generation position of the signal at each time is detected based on the sensor value at each time has been described as an example. However, the present invention is not limited to this. For example, the detection unit 28 may further detect the locus of the signal generation position based on the signal generation position at each time in the target region. For example, as illustrated in FIG. 13, the detection unit 28 generates the signal generation position (x1, x1) based on the sensor values acquired by the plurality of sensors 10 provided in the target object and the learned model. y1), the generation position (x2, y2) of the signal at time t2, and the generation position (x3, y3) of the signal at time t3 are detected, and the trajectory Tr is detected based on these generation positions. May be.

  Further, the detection unit 28 may predict the signal generation position at the next time based on the signal generation position up to the current time in the target region. For example, as illustrated in FIG. 13, the detection unit 28 determines a predetermined point on the extension line between the generation position (x2, y2) of the signal at time t2 and the generation position (x3, y3) of the signal at time t3, You may make it estimate as a generation position (xp, yp) of the signal of the next time.

  Further, when detecting the position of the signal generated on the surface of the three-dimensional object, it is necessary to install a plurality of sensors at predetermined locations. Specifically, when a three-dimensional object is divided into areas by a surface, a plurality of points are formed in each of the three-dimensional objects to form a surface in which at least one of the volume and the shape of each area is different It is necessary to install a sensor.

  For example, as shown in FIG. 14, when a plurality of piezoelectric sensors 10S0, 10S1, and 10S2 are installed on the back side of the surface of the three-dimensional object A1, the position of the pressure generated on the surface of the three-dimensional object A1 is detected. think of.

  In this case, the piezoelectric sensor 10S0, 10S1, and 10S2 installed on the three-dimensional object A1 shown in FIG. 14 cannot distinguish between the force generated at the position T1 and the force generated at the position T2. This is because the piezoelectric sensors 10S0, 10S1, and 10S2 constitute the symmetry plane D1.

  On the other hand, in the three-dimensional object A2 shown in FIG. 14, the volume of each region or the shape of each region divided by the surface formed by the piezoelectric sensors 10S0, 10S1, and 10S2 is different, and the piezoelectric sensors 10S0, 10S1, and 10S2 does not constitute a symmetry plane (forms an asymmetric plane D2). For this reason, the force generated at the position T1 can be distinguished from the force generated at the position T2. In addition, since there is no symmetric position with respect to the asymmetric surface D2 in the three-dimensional object A2, in principle, all observation points on the surface of the three-dimensional object A2 can be distinguished by three sensors. .

  FIG. 15 is an explanatory diagram for explaining a case where the three-dimensional object is a sphere. As shown in FIG. 15, the piezoelectric sensors 10S0, 10S1, and 10S2 are installed on the back side of the surfaces of the sphere A3 and the sphere A4. For example, in the sphere A3 shown in FIG. 15, the piezoelectric sensors 10S0, 10S1, and 10S2 are installed on a plane that forms a symmetrical plane. For this reason, the force generated at the position T1 cannot be distinguished from the force generated at the position T2.

  On the other hand, in the sphere A4 shown in FIG. 15, the piezoelectric sensors 10S0, 10S1, and 10S2 are installed on an asymmetric surface. For this reason, the force generated at the position T1 can be distinguished from the force generated at the position T2. In addition, since there is no symmetric position with respect to the asymmetric surface in the sphere A4, in principle, all observation points on the surface of the sphere A4 can be distinguished by the three sensors.

  14 and 15 show an example in which the force generation position on the surface of the three-dimensional object is detected by the minimum number of sensors. However, in order to improve detection accuracy, four or more sensors are installed. You may do it. For example, as shown in FIG. 16, six sensors (10S0, 10S1, 10S2, 10S3, 10S4, 10S5) may be installed to improve detection accuracy.

  Also, as shown in FIG. 17, for a three-dimensional object that is topologically in phase with a plane, the position of the force generated on the surface of the three-dimensional object can be detected by three sensors in the same manner as the plane. For example, the single connected curved surface A5 and the spherical surface A6 with a hole can be converted into a two-dimensional plane A7 by appropriate coordinate conversion, and therefore the position of the force generated on the surface by three sensors (10S0, 10S1, 10S2). Can be detected.

  The same applies to forming a two-dimensional region with a one-dimensional object. In this case, a plurality of sensors are installed at each location in the two-dimensional area as the target area. In addition, when a two-dimensional region is divided into lines by a line, a plurality of sensors are attached to each part on the two-dimensional region that forms a line in which at least one of the area of each region and the shape of each region is different. Installed.

  For example, as shown in FIG. 18, it is assumed that a two-dimensional region is formed by connecting an end E1 and an end E2 of a one-dimensional object A8. In this case, the sensor 10S0 and the sensor 10S1 are installed on the one-dimensional object A8.

  When the two-dimensional region A9 is formed by connecting the end E1 and the end E2 of the one-dimensional object A8 shown in FIG. 18, the axis formed by the sensor 10S0 and the sensor 10S1 is the target axis. The force generated at T1 cannot be distinguished from the force generated at position T2.

  On the other hand, when the two-dimensional region A10 is formed, since the axis formed by the sensors 10S0 and 10S1 is an asymmetric axis, the force generated at the position T1 and the force generated at the position T2 can be distinguished.

  In the present specification, the embodiment has been described in which the program is installed in advance. However, the program can be provided by being stored in a computer-readable recording medium.

DESCRIPTION OF SYMBOLS 10 Sensor 12 Reception part 20 Position detection apparatus 22 Information acquisition part 24 Learning part 26 Learned model memory | storage part 28 Detection part 30 Output part 50 Drone 100 Position detection system

Claims (19)

An acquisition unit that is provided at a plurality of locations in the target region and acquires sensor values detected by each of a plurality of sensors that detect signals;
Learning learned in advance from learning data representing a learning sensor value detected by each of a plurality of sensors when the signal is generated in the target region and the signal is generated and propagated through the target region. A detection unit that detects a generation position of a signal in the target region based on a completed model and each of the sensor values acquired by the acquisition unit;
A position detection device comprising: The detection unit detects a position of an object that generates the signal according to a generation position of the signal.
The position detection device according to claim 1. The acquisition unit is provided in a plurality of locations of the target region, and acquires voltage values detected by each of a plurality of piezoelectric sensors that detect force as the signal,
A learning position representing the learning voltage value detected by each of the plurality of piezoelectric sensors when the force is generated in the target region and the force is generated and propagated through the target region. Based on the learned model and each of the voltage values acquired by the acquisition unit, a force generation position in the target region is detected, and the position of the target is detected according to the force generation position. To
The position detection device according to claim 2. The learned model is a neural network model previously learned from the learning data by deep learning.
The position detection apparatus of any one of Claims 1-3. The neural network model is the neural network model in which the larger the target area, the greater the number of neurons of the neural network model.
The position detection device according to claim 4. The neural network model is the neural network model in which the greater the number of sensors, the greater the number of neurons in the neural network model.
The position detection device according to claim 4 or 5. The signal is at least one of force, light, electron beam, heat, magnetic signal, and electrical signal.
The position detection apparatus of any one of Claims 1-6. The plurality of sensors are mounted on each of the plurality of moving bodies,
The detection unit includes a learning model that has been learned in advance from learning data representing a generation position of the learning signal and a sensor value for learning according to a movement form of the plurality of moving objects, and the acquisition unit. Detecting the generation position of the signal in the target region based on each of the acquired sensor values;
The position detection apparatus of any one of Claims 1-7. The acquisition unit further acquires power obtained from the plurality of sensors,
An output unit that outputs the generation position of the signal detected by the detection unit using the power acquired by the acquisition unit;
The position detection apparatus of any one of Claims 1-8. When the target region is a two-dimensional space, the plurality of sensors are at least three.
The position detection apparatus of any one of Claims 1-9. When the target region is a three-dimensional space, the plurality of sensors are at least four.
The position detection apparatus of any one of Claims 1-10. The acquisition unit acquires the sensor value at each time,
The detection unit detects a generation position of a signal at each time in the target region based on the learned model and each of the sensor values at each time acquired by the acquisition unit.
The position detection device according to any one of claims 1 to 11. The detection unit further detects a locus of the signal generation position based on the signal generation position at each time in the target region.
The position detection device according to claim 12. The detection unit predicts the generation position of the signal at the next time based on the generation position of the signal up to the current time in the target area.
The position detection device according to claim 12. The plurality of sensors are installed at each location of a three-dimensional object as the target region, and when the three-dimensional object is divided into each region by a surface, at least one of the volume and shape of each region is different. Installed at each location on the three-dimensional object forming the plane.
The position detection apparatus of any one of Claims 1-14. The plurality of sensors are installed in each part of a two-dimensional region as the target region, and when the two-dimensional region is divided into each region by a line, at least one of the area and shape of each region is different. Installed at each location on the two-dimensional area forming the line.
The position detection apparatus of any one of Claims 1-14. A learning data acquisition unit that acquires learning data representing a learning sensor value detected by each of a plurality of sensors when the signal is generated in the target region and the signal is generated and propagates through the target region. When,
A learning unit that learns a model for detecting a signal generation position from a plurality of sensor values based on the learning data acquired by the learning data acquisition unit;
A model learning apparatus comprising: A plurality of sensors that are provided at a plurality of locations in the target region and that detect signals; and an acquisition unit that acquires sensor values detected by each of the plurality of sensors;
Learning learned in advance from learning data representing a learning sensor value detected by each of a plurality of sensors when the signal is generated in the target region and the signal is generated and propagated through the target region. A detection unit that detects a generation position of a signal in the target region based on a completed model and each of the sensor values acquired by the acquisition unit;
A position detection system including a position detection device. Computer
An acquisition unit that acquires sensor values detected by each of a plurality of sensors that detect signals and is provided at a plurality of locations in the target region, and the generation position of the signal for learning in the target region and the signal are generated. A learned model previously learned from learning data representing sensor values for learning detected by each of a plurality of sensors when propagating through the target region, and each of the sensor values acquired by the acquisition unit Based on the above, a program for functioning as a detection unit for detecting a generation position of a signal in the target region.