JP7371856B2 - Mobile body flow line detection system - Google Patents

Description

The present invention relates to a moving body flow line detection system that detects the flow line of a moving body in a predetermined traffic area.

Patent Document 1 discloses a travel route tracking system that tracks the travel route of a moving object such as a pedestrian. The system of Patent Document 1 includes a plurality of pressure sensors embedded in the floor of a passageway at a predetermined density. The pressure sensor detects and outputs the pressure of the pedestrian walking along the passageway, and based on this output, the movement route of the pedestrian is calculated by the movement route calculating means.

Japanese Patent Application Publication No. 2006-164020

However, in Patent Document 1, for example, if there are variations in the characteristics etc. between multiple sensors, the reliability of the output value of each sensor when a pedestrian (moving object) moves decreases, and even when the pedestrian (moving object) moves, There is also a problem in that the accuracy in route calculation also decreases.

The present invention has been made in view of these points, and an object of the present invention is to provide a moving body flow line detection system that can improve the accuracy of detecting the flow line of a moving body.

A mobile body flow line detection system according to one aspect of the present invention is a mobile body flow line detection system that includes a plurality of sensor units installed side by side in a moving area and detects a moving body flow line in the traffic area. Each of the sensor units includes a floor forming member, and a sensor that supports the floor forming member and outputs a detected value according to a force applied to the floor forming member due to movement of a moving body, a trained model for position detection learned in advance from learning data representing the generation position of the force for learning in the plurality of sensor units and a detection value for learning output from the sensor when the force propagates; a position detection section that detects a position where the force is generated in the plurality of sensor units based on the detection value output from the sensor; a flow line detection section that detects a flow line of the moving object in the plurality of sensor units, the movement line detection section detecting the position of the moving object for learning in the plurality of sensor units and the movement line of the moving object in the plurality of sensor units. a trained model for flow line detection that has been learned in advance from learning data representing a flow line and a learning detection value output from the sensor when the force due to the movement of the moving object is propagated; The present invention is characterized in that the flow line of the moving body is detected based also on a detection value output from a sensor .
In addition, a moving body flow line detection system according to one aspect of the present invention includes a plurality of sensor units installed in parallel in a moving area, and detects a moving line of the moving body in the traffic area. In the system, each of the sensor units includes a floor forming member and a sensor that supports the floor forming member and outputs a detected value according to a force applied to the floor forming member due to movement of a moving body. a trained model for position detection learned in advance from learning data representing the generation position of the force for learning in the plurality of sensor units and a detection value for learning output from the sensor when the force propagates; and a position detection unit that detects a position where the force is generated in the plurality of sensor units based on a detection value outputted from the sensor; and a flow line detection unit that detects the flow line of the moving object in the plurality of sensor units, and the detection value of the sensor is an output waveform of the entire time domain in the state in which the force is generated. .

According to the present invention, the position detection unit detects the position where force is generated due to the movement of the moving object based on a trained model for position detection that has been trained in advance. Also, the detection accuracy of the occurrence position can be maintained satisfactorily. Thereby, it is possible to improve the accuracy of the movement line of the moving body detected by the movement line detection unit based on the detection result of the position detection unit.

FIG. 1 is a diagram showing the configuration of a moving body flow line detection system in an embodiment. FIG. 1 is an exploded perspective view schematically showing a sensor unit according to an embodiment. FIG. 2 is a partial vertical cross-sectional view of a sensor unit according to an embodiment. FIG. 1 is a schematic exploded perspective view of a partial configuration of a sensor unit according to an embodiment. FIG. 3 is a plan view showing an example of a traffic area according to the embodiment. It is a graph which shows an example of the output waveform of a piezoelectric element when a pedestrian walks. 3 is a flowchart showing the contents of a learning processing routine according to an embodiment. 3 is a flowchart showing the contents of a position detection processing routine according to the embodiment. It is a flow chart which shows the contents of a flow line detection processing routine concerning an embodiment. 3 is a flowchart showing the contents of an attribute detection processing routine according to the embodiment. It is a top view of the traffic area concerning the 1st modification. It is a top view of the traffic area concerning a 2nd modification. It is a top view of the traffic area concerning a 3rd modification. It is a top view of the traffic area concerning the 4th modification. It is a top view of the traffic area concerning the 5th modification. It is a top view of the traffic area concerning a 6th modification. It is a top view of the traffic area concerning the 7th modification. It is a top view of the traffic area concerning the 8th modification. It is a top view of the traffic area concerning a 9th modification.

Embodiments of the present invention will be specifically described below with reference to the drawings. Note that the present invention is not limited to the embodiments described below, and can be implemented with appropriate modifications within the scope without changing the gist thereof. In the following figures, some configurations may be omitted for convenience of explanation.

FIG. 1 is a diagram showing the configuration of a moving object flow line detection system in an embodiment. As shown in FIG. 1, the mobile body flow line detection system 1 includes a plurality of sensor units 10 and a detection device 20. The moving object flow line detection system 1 detects moving objects such as pedestrians and wheelchairs (hereinafter simply referred to as "pedestrians") in traffic areas such as the floors of buildings such as stations and outdoor paved surfaces. Target for detection. In addition, examples of moving objects include animals, robots, and the like.

A plurality of sensor units 10 are arranged in a row along the floor or the ground in a pedestrian traffic area. The plurality of sensor units 10 convert the force applied to the traffic area due to the pedestrian's movement into a detected value (voltage value) as an electric signal, and output the detected value as associated with the pedestrian's own ID information. Further, the sensor unit 10 similarly outputs the force applied to the traffic area due to the movement of a pedestrian for learning purposes as a detected value. The specific configuration of the sensor unit 10 will be described later.

The detection device 20 is constituted by a personal computer (PC) or the like, and includes a calculation means such as a CPU, a ROM that stores a program executed by the CPU, and a RAM that constitutes a work area of the CPU. Each function of the detection device 20 will be described later.

Next, the specific configuration of the sensor unit 10 will be described with reference to FIGS. 2 to 4.

FIG. 2 is an exploded perspective view schematically showing the sensor unit according to the embodiment. As shown in FIG. 2, the sensor unit 10 includes a plurality of (four in this embodiment) piezoelectric elements 11 as sensors, a lower protective member 12 on which each piezoelectric element 11 is placed, and a lower protective member 12. and an upper protective member 13 that covers it from above. The sensor unit 10 is disposed on an installation surface F such as a floor surface, and in this embodiment is formed into a square planar shape.

The lower protection member 12 and the upper protection member 13 are combined with each other and house each piezoelectric element 11 in a horizontal direction. The lower protection member 12 and the upper protection member 13 are made of insulating rubber.

The sensor unit 10 further includes a plate-shaped floor forming member 14 placed on the upper surface of the upper protection member 13 . In this embodiment, the floor forming member 14 is formed in a square shape when the sensor unit 10 is viewed from above. The floor forming member 14 has a top plate part 14a made of stone, and a back plate part 14b provided on the lower surface of the top plate part 14a. The back plate portion 14b is made of metal such as iron, and is provided to reinforce the front plate portion 14a.

FIG. 3 is a partial longitudinal sectional view of the sensor unit according to the embodiment. FIG. 4 is a schematic exploded perspective view of a partial configuration of the sensor unit according to the embodiment. As shown in FIG. 3, the piezoelectric element 11 is formed by laminating a pair of upper and lower electrode layers 11a and 11b, and a piezoelectric ceramic layer 11c located between the upper and lower electrode layers 11a and 11b. Become. The piezoelectric ceramic layer 11c is made of a circular, plate-shaped piezoelectric ceramic. The lower electrode layer 11b is a circular plate-shaped electrode, and a piezoelectric ceramic layer 11c is formed on the upper surface. The upper electrode layer 11a is formed in a film shape on the upper surface of the piezoelectric ceramic layer 11c. The material of the upper electrode layer 11a and the lower electrode layer 11b is selected from at least one of highly electrically conductive gold, silver, copper, brass, iron, stainless steel, and palladium. 11a is made of silver, and the lower electrode layer 11b is made of stainless steel. In the piezoelectric element 11, the diameter dimension increases in the order of the upper electrode layer 11a, the piezoelectric ceramic layer 11c, and the lower electrode layer 11b.

A copper foil tape (not shown) is attached to each of the upper surface of the upper electrode layer 11a and the lower surface of the lower electrode layer 11b to electrically connect them. The copper foil tape is connected to the detection device 20 (see FIG. 1) via a substrate and wiring (none of which are shown), so that the detected value (voltage value) of the piezoelectric element 11 can be output to the detection device 20. .

A receiving portion 15 is formed on the upper surface of the lower protective member 12 at each installation position of each piezoelectric element 11 . The receiving part 15 is formed by recessing the upper surface of the lower protective material 12 into a planar shape that is approximately the same as that of the piezoelectric element 11, and functions as a position regulating part that regulates the movement of the received piezoelectric element 11 in the horizontal direction. (See Figure 5). Further, a mounting portion 15a having an annular shape in plan view is formed on the outer peripheral side of the interior of the receiving portion 15. The mounting portion 15a is formed horizontally at a position slightly lower than the upper surface of the lower protection member 12 and higher than the bottom of the receiving portion 15, and supports the peripheral portion of the piezoelectric element 11 from below.

The upper protection member 13 is formed into a lid shape including a top portion 16 and a peripheral wall portion 17 that hangs down from the outer periphery of the top portion 16 . The peripheral wall portion 17 is disposed at a position surrounding the lower protective member 12 in a state where the lower protective member 12 and the upper protective member 13 are combined with each other. Regulates relative movement in direction. A protrusion 18 that protrudes downward is formed on the lower surface of the top portion 16 at a position corresponding to the installation position of each piezoelectric element 11 . Each protrusion 18 is circular in plan view, and its outer diameter is set smaller than the outer diameter of the upper electrode layer 11a (see FIG. 5).

The upper protection member 13 is supported by the protrusion 18 placed on the upper surface of the piezoelectric element 11 . Therefore, the upper protection member 13 and the floor forming member 14 placed thereon are supported from below by the piezoelectric element 11 and the lower protection member 12. In other words, the piezoelectric element 11 is provided at a position that supports the upper protection member 13 and the floor forming member 14.

In the sensor unit 10, when a pedestrian moves on the floor forming member 14 by walking or the like, force is applied to the floor forming member 14 due to the movement. The upper protection member 13 is deformed so as to be pushed down by this force, and the pressure generated at this time is applied to the piezoelectric element 11 by the protrusion 18 . Then, since the peripheral portion of the piezoelectric element 11 is supported by the mounting portion 15a, the center of the piezoelectric element 11 is deformed to be concave, and the piezoelectric ceramic layer 11c is bent downward, resulting in a large strain. This strain generates a voltage in each electrode layer 11a, 11b, and a detected value (voltage value) corresponding to the generated voltage is output to the detection device 20.

FIG. 5 is a plan view showing an example of a traffic area according to the embodiment. In this embodiment, the position of the pedestrian in the traffic area, the line of movement of the pedestrian, and the attribute of the pedestrian are detected according to the detection values output by the plurality of sensor units 10. For example, as shown in FIG. 5, when a pedestrian W moves in a traffic area A in which a total of nine sensor units 10 are arranged in three rows in the vertical and horizontal directions, each piezoelectric element 11 A detected value corresponding to the force applied by the movement of the pedestrian W is output to the detection device 20. Then, the detection device 20 detects the position, flow line, and attributes of the pedestrian W according to the detected values. In this embodiment, the position of the pedestrian W is the center of gravity of the pedestrian W on the sensor unit 10.

Specifically, first, in the mobile body flow line detection system 1, the detection values detected by the piezoelectric elements 11 in the plurality of sensor units 10 when the pedestrian W moves (walks) in the traffic area A are used as learning data. Collect as.

Next, the mobile body flow line detection system 1 trains a neural network model, which is an example of a model, by deep learning based on the collected learning data to obtain a trained neural network model. As trained neural network models, a trained model for position detection, a trained model for flow line detection, and a trained model for attribute detection are obtained.

Then, the mobile body flow line detection system 1 detects the position, flow line, and attributes of the pedestrian W in the traffic area A based on the detection values detected by the plurality of sensor units 10 and each learned model. .

The mobile body flow line detection system 1 has a learning function, a position detection function, a flow line detection function, and an attribute detection function. Each function will be explained below. First, the learning function in the mobile body flow line detection system 1 will be explained.

[Learning function]
The plurality of sensor units 10 detect detection values for learning according to the force propagated to the traffic area A due to the movement of the pedestrian W for learning.

In the passage area A shown in FIG. The detected value when the signal propagates to is detected as the learning detected value (voltage value). For example, when a learning pedestrian W walks along the footprints shown in FIG. , a plurality of piezoelectric elements 11 output detected values.

In order to detect the position, flow line, and attributes of the pedestrian W from such detected values, the shape and material of the sensor unit 10, the arrangement of the piezoelectric element 11, etc. are considered, and the position of the footprint of the pedestrian W (force generation It is necessary to model the propagation of force from the piezoelectric element 11 (position) to the piezoelectric element 11. However, in the actual sensor unit 10, the materials and structures of the protective members 12 and 13 (see FIG. 2) vary, and the conditions of the installation location (floor surface, etc.) also vary, making modeling difficult. Further, since there are variations in detection characteristics of detected values, deterioration states, etc. of the plurality of piezoelectric elements 11, modeling becomes difficult.

Therefore, in the present embodiment, a neural network model is developed by deep learning by utilizing the fact that the patterns of detected values in the plurality of piezoelectric elements 11 differ depending on the position, flow line, and attributes of the pedestrian W in the sensor unit 10. Let them learn. Then, the position, flow line, and attributes of the pedestrian W are detected using the obtained trained neural network models (the trained model for position detection, the trained model for flow line detection, and the trained model for attribute detection). As a result, the position (center of gravity position) of the pedestrian W applying force to the floor forming member 14 can be detected without considering the factors that make modeling difficult, such as variations in detection characteristics in the piezoelectric element 11. The flow line and attributes of the pedestrian W can be detected depending on the position and the like.

The input unit 22 inputs the position, flow line, and attributes of the pedestrian W for learning in the traffic area A when the pedestrian W actually walks in the traffic area A for learning. Examples of the input unit 22 include a keyboard, a mouse, and the like. When the input unit 22 is a keyboard or the like, the position, flow line, and attributes of the pedestrian W are input manually.

Next, the functions of the detection device 20 will be explained. As shown in FIG. 1, the detection device 20 includes an acquisition section 24, a position detection learning section 25, a flow line detection learning section 26, an attribute detection learning section 27, a learned model storage section 28 for position detection, and a flow line detection learning section 26. It includes a detection trained model storage section 29, an attribute detection trained model storage section 30, a position detection section 32, a flow line detection section 33, an attribute detection section 34, and an output section 35.

The acquisition unit 24 acquires a learning detection value detected by each of the plurality of sensor units 10 when a pedestrian W actually walks in the traffic area A for learning. Moreover, the position, flow line, and attribute of the pedestrian W for learning corresponding to the detection value for learning are input to the acquisition unit 24 by the input unit 22 .

The acquisition unit 24 then associates the position, flow line, and attributes of the learning pedestrian W with the learning detection values detected by each of the plurality of sensor units 10 as the learning pedestrian W walks. , are stored in a storage area (not shown) as learning data representing them. A plurality of pieces of learning data are stored in this storage area.

The position detection learning unit 25 previously learns a neural network model for detecting the position of the pedestrian W from a plurality of detection values based on the learning data acquired by the acquisition unit 24. In other words, such a neural network model is based on the learning detection value output from the piezoelectric element 11 of the sensor unit 10 included in the learning data and the learning position of the pedestrian W input through the input section 22. be learned.

The flow line detection learning unit 26 previously learns a neural network model for detecting the flow line of the pedestrian W from a plurality of detection values, based on the learning data acquired by the acquisition unit 24. In other words, this neural network model uses the learning detection value output from the piezoelectric element 11 of the sensor unit 10 included in the learning data, and the learning position and flow line of the pedestrian W input by the input unit 22. It is learned based on

The attribute detection learning unit 27 previously learns a neural network model for detecting the attributes of the pedestrian W from a plurality of detection values, based on the learning data acquired by the acquisition unit 24. In other words, such a neural network model uses the learning detection value output from the piezoelectric element 11 of the sensor unit 10 included in the learning data, the learning position and flow line of the pedestrian W input through the input section 22. and attributes. In this embodiment, each of the learning units 25 to 27 trains a neural network model by deep learning.

Then, the position detection learning unit 25 stores the trained neural network model obtained by deep learning in the position detection trained model storage unit 28 as a position detection trained model. The flow line detection learning unit 26 stores the trained neural network model obtained by deep learning in the flow line detection learned model storage unit 29 as a trained model for flow line detection. The attribute detection learning unit 27 stores the trained neural network model obtained by deep learning in the attribute detection trained model storage unit 30 as a trained attribute detection model.

The trained model storage unit for position detection 28 stores the trained model for position detection obtained by the learning unit 25 for position detection. Further, the trained model storage unit 29 for flow line detection stores the trained model for flow line detection obtained by the learning unit 26 for flow line detection, and the trained model storage unit 30 for attribute detection stores the trained model for flow line detection obtained by the learning unit 26 for flow line detection. The trained model for attribute detection obtained by the detection learning unit 27 is stored.

[Position detection function]
Next, the position detection function in the mobile body flow line detection system 1 will be explained.

When a pedestrian W walks in a traffic area A where a plurality of sensor units 10 are installed, the acquisition unit 24 acquires each of the detection values detected by each of the piezoelectric elements 11 in the plurality of sensor units 10. The position detection unit 32 detects the pedestrian W in the traffic area A based on the learned model for position detection stored in the learned model storage unit 28 for position detection and each of the detection values acquired by the acquisition unit 24. Detect the position of.

Specifically, the position detection unit 32 inputs each of the detection values acquired by the acquisition unit 24 into a trained model for position detection, and calculates the position of the pedestrian W output from the trained model for position detection. get. The output unit 35 outputs the position of the pedestrian W detected by the position detection unit 32 as a result.

[Flow line detection function]
Next, the flow line detection function in the mobile body flow line detection system 1 will be explained.

The flow line detection unit 33 determines the traffic area based on the position of the pedestrian W detected by the position detection unit 32 and the learned model for flow line detection stored in the learned model storage unit 29 for flow line detection. The flow line of the pedestrian W at A is detected. Furthermore, the flow line detection unit 33 detects the flow line based on each of the detection values acquired by the acquisition unit 24 when the pedestrian W walks in the traffic area A.

In other words, the flow line detection unit 33 inputs the position of the pedestrian W detected by the position detection unit 32 and each of the detection values acquired by the acquisition unit 24 to the learned model for flow line detection. Then, the flow line detection unit 33 acquires the flow line of the pedestrian W output from the trained model for flow line detection. The output unit 35 outputs the flow line of the pedestrian W detected by the flow line detection unit 33 as a result.

[Attribute detection function]
Next, the attribute detection function in the mobile body flow line detection system 1 will be explained.

The attribute detection unit 34 uses each of the detection values acquired by the acquisition unit 24 when the pedestrian W walks in the traffic area A and the learned model for attribute detection stored in the learned model storage unit 30 for attribute detection. Based on this, the attributes of the pedestrian W in the traffic area A are detected. Further, the attribute detection unit 34 detects the attribute based on the position of the pedestrian W detected by the position detection unit 32 and the flow line of the pedestrian W detected by the flow line detection unit 33. Ru.

In other words, the attribute detection unit 34 detects each of the detection values acquired by the acquisition unit 24, the position of the pedestrian W detected by the position detection unit 32, and the flow line detected by the flow line detection unit 33. Input to trained model for attribute detection. Then, the attribute detection unit 34 acquires the attribute of the pedestrian W output from the trained model for attribute detection. The output unit 35 outputs the attribute of the pedestrian W detected by the attribute detection unit 34 as a result. Here, the attributes include the gender, age, weight, stride length, physical condition (drunk, injury), presence or absence of a disability of the pedestrian W, as well as the type of moving object including the pedestrian W (e.g., wheelchair, robot, etc.). I can give an example.

In the above-described detection by each of the detection units 32 to 34, it is preferable to use an RNN (Recurrent Neural Network) algorithm in order to improve the accuracy.

FIG. 6 is a graph showing an example of an output waveform of a piezoelectric element when a pedestrian walks. In the graph of FIG. 6, the horizontal axis represents time, and the vertical axis represents the detected value (pressure value) of the piezoelectric element 11. As shown in FIG. 6, the detected value output from the sensor unit 10 (piezoelectric element 11) in a state where force is generated in the sensor unit 10 due to the walking of the pedestrian W is a predetermined period of time (several hundred milliseconds to several seconds, In this embodiment, it changes between t1 and t2). Here, the detected value acquired by the acquisition unit 24 may be a detected value at an arbitrary timing between t1 and t2, or an average value between t1 and t2, but may be an average value between t1 and t2. The output waveform can be a full time domain output waveform. This also makes it possible to improve the detection accuracy in each of the detection units 32 to 34.

<Operation of mobile body flow line detection system 1>
Next, the operation of the mobile body flow line detection system 1 according to the embodiment will be explained. The detection device 20 executes a learning processing routine for learning a neural network model based on learning data, and a position detection processing routine for detecting the position of the pedestrian W based on the trained model for position detection. In addition, the detection device 20 executes a flow line detection processing routine for detecting the flow line of the pedestrian W based on the trained model for flow line detection, and detects the attributes of the pedestrian W based on the learned model for attribute detection. An attribute detection processing routine to be detected is executed.

<Learning processing routine>
First, in the traffic area A, when a pedestrian W is actually walking in the traffic area A for learning and a force is applied to the sensor unit 10, the piezoelectric elements 11 of the plurality of sensor units 10 are used for learning. A detection value is detected. Moreover, when the position of the pedestrian W for learning is inputted by the input unit 22, the acquisition unit 24 acquires the position of the pedestrian W for learning and each of the detection values for learning. Then, the acquisition unit 24 associates the position of the pedestrian W for learning with each of the detected values for learning, and stores them in a storage area (not shown) as learning data.

Then, upon receiving the instruction signal to execute the learning process, the detection device 20 executes the learning process routine shown in FIG. 7 . FIG. 7 is a flowchart showing the contents of the learning processing routine according to the embodiment.

In step S01, each of the learning units 25 to 27 acquires a plurality of learning data stored in a storage area (not shown).

In step S02, each of the learning units 25 to 27 trains a neural network model by machine learning based on the plurality of learning data acquired in step S01.

In step S03, each of the learning units 25 to 27 stores the trained neural network model obtained in step S02 to each of the trained model storage units 28 to 30, and ends the learning processing routine.

<Position detection processing routine>
When the detection device 20 receives an instruction signal to execute the position detection process after completing the learning process routine, it executes the position detection process routine shown in FIG. To detect. FIG. 8 is a flowchart showing the contents of the position detection processing routine according to the embodiment.

In step S11, when the pedestrian W walks in the traffic area A, the acquisition unit 24 acquires the detected value of the piezoelectric element 11 according to the force applied from the pedestrian W to the sensor unit 10.

In step S12, the position detecting unit 32 determines whether traffic is detected based on the learned model for position detection stored in the learned model storage unit 28 in the learning process routine and the detected value acquired in step S11. The position of the pedestrian W in area A is detected.

In step S13, the position of the pedestrian W detected in step S12 is output as a result, and the position detection processing routine is ended.

<Flow line detection processing routine>
After the learning process routine is finished, when the detection device 20 receives an instruction signal to execute the flow line detection process, it executes the flow line detection process routine shown in FIG. Detect flow lines. FIG. 9 is a flowchart showing the contents of the flow line detection processing routine according to the embodiment.

In step S21, the acquisition unit 24 acquires a detection value of the piezoelectric element 11 according to the force applied from the pedestrian W to the sensor unit 10 when the pedestrian W walks in the traffic area A.

In step S22, the position of the pedestrian W detected by the position detection unit 32 is output to the flow line detection unit 33.

In step S23, the flow line detection unit 33 uses the learned model for flow line detection stored in the learned model storage unit 29 for flow line detection in the learning process routine to determine the flow line of the pedestrian W in the traffic area A. Detect. Such detection is performed based on the learned model for flow line detection, the detection value acquired in step S21, and the position of the pedestrian W output in step S22.

In step S24, the flow line of the pedestrian W detected in step S23 is output as a result, and the flow line detection processing routine is ended.

<Attribute detection processing routine>
When the detection device 20 receives an instruction signal to execute attribute detection processing after completing the learning processing routine, it executes the attribute detection processing routine shown in FIG. To detect. FIG. 10 is a flowchart showing the content of the attribute detection processing routine according to the embodiment.

In step S31, when the pedestrian W walks in the traffic area A, the acquisition unit 24 acquires the detected value of the piezoelectric element 11 according to the force applied from the pedestrian W to the sensor unit 10.

In step S32, the position of the pedestrian W detected by the position detection unit 32 is output to the attribute detection unit 34.

In step S33, the flow line of the pedestrian W detected by the flow line detection unit 33 is output to the attribute detection unit 34.

In step S34, the attribute detection unit 34 detects the attribute of the pedestrian W in the traffic area A using the learned model for attribute detection stored in the learned model storage unit 30 for attribute detection in the learning processing routine. In addition to the trained model for attribute detection, this detection uses the detection value acquired in step S31, the position of the pedestrian W output in step S32, and the movement of the pedestrian W output in step S33. It will be carried out based on the line.

In step S35, the attribute of the pedestrian W detected in step S34 is output as a result, and the attribute detection processing routine is ended.

As described above, in this embodiment, the position detection unit 32 detects the pedestrian W using the learned model for position detection obtained as described above and the detected values in the plurality of sensor units 10 (piezoelectric elements 11). The location can be detected. Thereby, even if there are variations in characteristics among the piezoelectric elements 11 provided in the plurality of sensor units 10, it is possible to incorporate the variations into learning and ignore the characteristic differences between the plurality of piezoelectric elements 11. Therefore, the influence of variations in the characteristics of the piezoelectric elements 11, etc. can be suppressed, and the accuracy of detecting the position of the pedestrian W can be improved. In addition, as well as variations in the characteristics of the piezoelectric element 11, there are also differences in the state of deterioration of the piezoelectric element 11, differences in the materials and structures of the protective materials 12 and 13 (see Fig. 2), and the condition of the floor surface where the piezoelectric element 11 is installed. The influence of differences can also be suppressed.

In the present embodiment, the flow line detection unit uses the result of the position detection of the pedestrian W, the learned model for flow line detection obtained as described above, and the detection values of the plurality of sensor units 10. 33, the flow line of the pedestrian W can be detected. Therefore, in this embodiment, not only the accuracy of detecting the position of the pedestrian W but also the accuracy of detecting the line of movement of the pedestrian W can be improved.

Furthermore, in the present embodiment, the attribute detection unit 34 detects walking based on the detection results of the position and flow line of the pedestrian W, the learned model for attribute detection described above, and the detection values of the plurality of sensor units 10. The attributes of the person W can be detected. Therefore, detection accuracy can also be improved regarding attributes such as the gender and stride length of the pedestrian W.

Furthermore, in this embodiment, by simply installing a plurality of sensor units 10 side by side, it is possible to not only turn the traffic area A into a position sensor, but also to make it a sensor capable of detecting the flow line and attributes of pedestrians W. can. Furthermore, the various types of detection described above are possible by using only a few piezoelectric elements 11 in the sensor unit 10, and compared to a structure in which many piezoelectric elements are installed in the traffic area A at intervals smaller than the footprint size, the overall system is can be constructed easily and at low cost.

Furthermore, when this embodiment is compared with the conventional method of specifying pedestrian flow lines and attributes using camera photography and image processing, there are advantages described below. In this embodiment, it is possible to detect areas that are in the shadow and cannot be photographed with a camera or the like, and it is also possible to perform specialized detection only in areas where detection results are required. Furthermore, in this embodiment, since the output is from the piezoelectric element 11, the data size can be made smaller than in image processing, and it is also possible to detect flow lines, etc. in combination with the above-mentioned conventional method.

Note that the present invention is not limited to the above embodiments, and can be implemented with various modifications. In the embodiments described above, the size, shape, direction, etc. illustrated in the accompanying drawings are not limited to these, and can be changed as appropriate within the scope of achieving the effects of the present invention. Other changes can be made as appropriate without departing from the scope of the invention.

For example, the arrangement of the plurality of sensor units 10 in the traffic area A and the arrangement of the piezoelectric elements 11 in the sensor unit 10 can be changed in various ways, as exemplified in the first to ninth modifications below. .

FIG. 11 is a plan view of the traffic area according to the first modification. FIG. 12 is a plan view of the traffic area according to the second modification. FIG. 13 is a plan view of the traffic area according to the third modification. In the first to third modifications shown in FIGS. 11 to 13, the width of the traffic area A is the same as the length (L) of one side of the square sensor unit 10, and the traffic area A is It is extending. Although three sensor units 10 are shown in each figure, a plurality of sensor units 10 may be arranged side by side in the horizontal direction.

In the first modification of FIG. 11, when the sensor unit 10 is viewed from above, the installation location of the piezoelectric element 11 is similar to the floor forming member 14, and is a virtual square (indicated by the two-dot chain line in the figure) with the same center position. ). When the length of one side of the floor forming member 14 is L, the length of one side of the virtual square is approximately half of that length, 2/L. Further, in the lateral direction (the direction in which the sensor units 10 are arranged), the distance between the piezoelectric elements 11 that are adjacent in the lateral direction across the sensor units 10 is also 2/L. Therefore, when looking at the entire traffic area A, the piezoelectric elements 11 can be arranged at each intersection of meshes lined up in the horizontal direction, with the minimum unit being a virtual square whose side length is 2/L.

In the second modification of FIG. 12, when the sensor unit 10 is viewed from above, the piezoelectric elements 11 are installed at each vertex of a virtual equilateral triangle (indicated by two-dot chain lines in the figure). When the length of one side of the floor forming member 14 is L, the height dimension of the virtual equilateral triangle is half of that length, 2/L. In the virtual equilateral triangle, the side extending in the vertical direction in the figure is 4/L apart in the horizontal direction from the left side in the figure of the floor forming member 14, and the vertex located on the right side in the figure is 4/L apart in the horizontal direction from the right side in the figure. /L leave. With such an arrangement, the piezoelectric element 11 can be arranged at each intersection or each vertex of the mesh arranged in the horizontal direction, with the minimum unit being a virtual equilateral triangle with a height of 2/L.

In the third modification of FIG. 13, when the central sensor unit 10 of the three sensor units 10 is viewed from above, the piezoelectric elements 11 are installed at each vertex of a virtual regular hexagon (indicated by a chain double-dashed line). And it is said to be the center of it. The center of the virtual regular hexagon is located at the same position as the center of the floor forming member 14, and the size of the virtual regular hexagon is as shown in the figure. In the other two sensor units 10, the virtual regular hexagons are arranged adjacent to each other in the horizontal direction, and piezoelectric elements 11 are installed at each vertex and center of each of the virtual regular hexagons. With such an arrangement, the piezoelectric element 11 can be arranged at each intersection or each vertex of a mesh lined up in a two-dimensional direction with the minimum unit being a virtual equilateral triangle smaller than the second modification.

FIG. 14 is a plan view of the passage area according to the fourth modification. FIG. 15 is a plan view of the passage area according to the fifth modification. FIG. 16 is a plan view of the passage area according to the sixth modification. In the fourth to sixth modifications shown in FIGS. 14 to 16, sensor units 10 are installed in a line in the traffic area A in two orthogonal directions, a vertical direction and a horizontal direction in the figure. Although nine sensor units 10 are shown in each figure, a plurality of sensor units 10 may be further arranged in the vertical and horizontal directions.

In the fourth modification example of FIG. 14, piezoelectric elements 11 are arranged at each vertex of a virtual square (indicated by a two-dot chain line) similar to the first modification example. In the fifth modification of FIG. 15, piezoelectric elements 11 are arranged at each vertex and center of a virtual regular hexagon (indicated by a two-dot chain line) similar to the central sensor unit 10 of the third modification. In the sixth modification of FIG. 16, when the sensor unit 10 is viewed from above, the piezoelectric elements 11 are installed at each vertex of a virtual regular octagon (indicated by a chain double-dashed line) and its center. The center of the virtual regular octagon is located at the same position as the center of the floor forming member 14. Note that the piezoelectric element 11 is installed at a location of another virtual regular n-gon or n-gon (n is a natural number of 3 or more) formed at the same central position as the floor forming member 14 when the sensor unit 10 is viewed from above. Each vertex and center may be used.

FIG. 17 is a plan view of the passage area according to the seventh modification. In the seventh modification shown in FIG. 17, when the sensor unit 10 is viewed from above, the sensor unit 10 and the floor forming member 14 are formed into a square shape, and the piezoelectric element 11 is installed at the apex of the square. Therefore, the integrated piezoelectric element 11 is disposed across the vertices of the four sensor units 10, and the force applied to the four sensor units 10 can be detected.

FIG. 18 is a plan view of the passage area according to the eighth modification. In the eighth modification of FIG. 18, when the sensor unit 10 is viewed from above, the sensor unit 10 and the floor forming member 14 are formed into an equilateral triangular shape, and are laid side by side so as to be adjacent to each other. A piezoelectric element 11 is installed at the apex of an equilateral triangle forming the sensor unit 10 and the floor forming member 14. Therefore, the integrated piezoelectric element 11 is disposed across the vertices of the six sensor units 10, and the force applied to the six sensor units 10 can be detected.

FIG. 19 is a plan view of a traffic area according to the ninth modification. In the ninth modification shown in FIG. 19, when the sensor unit 10 is viewed from above, the sensor unit 10 and the floor forming member 14 are formed into a regular hexagonal shape, and are laid side by side so as to be adjacent to each other. A piezoelectric element 11 is installed at the apex of a regular hexagon that forms the sensor unit 10 and the floor forming member 14. Therefore, the integrated piezoelectric element 11 is disposed across the vertices of the three sensor units 10, and the force applied to the six sensor units 10 can be detected. In the seventh to ninth modifications, the number of piezoelectric elements 11 installed can be reduced compared to the first to sixth modifications.

According to the first to ninth modifications described above, the piezoelectric elements 11 can be arranged in a well-balanced manner with regularity in the installation locations of the plurality of sensor units 10, and it is possible to better demonstrate the accuracy of various detections. .

Further, in the above embodiment, the case where the piezoelectric element 11 is used as the sensor has been described, but the sensor is not particularly limited as long as it is a pressure-sensitive sensor; It is also possible to use a vibration power generation element other than the element.

Furthermore, the detection of the flow line of the pedestrian W by the flow line detection unit 33 is not limited to the above processing, and various changes are possible. For example, without using the learned model for flow line detection in the above embodiment, in addition to the detection value of the sensor unit 10 and the detection result of the position detection section 32, the detection of the pedestrian W is performed based on the time information at the time of detection. Flow lines may also be detected.

Further, the detection of the attribute of the pedestrian W by the attribute detection unit 34 is not limited to the above processing, and various changes are possible. For example, in detecting the attributes of a pedestrian W, the information input to the learned model for attribute detection is at least one of the detection value of the sensor unit 10, the detection result of the position detection section 32, and the detection result of the flow line detection section 33. can be selected. In this case, the learning information acquired for learning the learned model for attribute detection can also be selected in the same way.

INDUSTRIAL APPLICABILITY The present invention is useful in that it can improve the accuracy of detecting the flow line of moving objects such as pedestrians in traffic areas such as the floors of buildings such as stations and outdoor paved surfaces.

1 Moving object flow line detection system 10 Sensor unit 11 Piezoelectric element (sensor)
14 Floor forming member 15 Receiving part (position regulating part)
32 Position detection unit 33 Flow line detection unit 34 Attribute detection unit A Traffic area W Pedestrian (moving object)

Claims (7)

A moving object flow line detection system that includes a plurality of sensor units installed in parallel in a moving area, and detects the moving line of the moving object in the moving area,
Each of the sensor units includes a floor forming member, and a sensor that supports the floor forming member and outputs a detected value according to a force applied to the floor forming member due to movement of the movable body,
Position detection learned in advance from learning data representing the position of the moving body for learning in the plurality of sensor units and the detection value for learning output from the sensor when the force due to the movement of the moving body is propagated. a position detection unit that detects the position of the mobile object in the plurality of sensor units based on the trained model and the detection value output from the sensor;
a flow line detection unit that detects a flow line of the mobile body in the plurality of sensor units based on the position of the mobile body detected by the position detection unit ,
The flow line detection unit is
the position of the moving object for learning in the plurality of sensor units, the line of movement when the moving object moves, and the learning detection output from the sensor when the force due to the movement of the moving object is propagated; a trained model for flow line detection that has been learned in advance from learning data representing the value;
A moving body flow line detection system , characterized in that the moving body flow line is detected based also on a detection value output from the sensor . further comprising an attribute detection unit that detects an attribute of the moving object,
The attribute detection unit is
The position of the moving object for learning in the plurality of sensor units, the flow line when the moving object moves, the attribute of the moving object, and the output from the sensor when the force due to the movement of the moving object is propagated. a trained model for attribute detection learned in advance from learning data representing the detected value for learning;
a detection value output from the sensor;
the position of the moving body detected by the position detection unit;
The moving body flow line detection system according to claim 1 , wherein attributes of the moving body are detected based on the flow line of the moving body detected by the flow line detection unit. A moving object flow line detection system that includes a plurality of sensor units installed in parallel in a moving area, and detects the moving line of the moving object in the moving area,
Each of the sensor units includes a floor forming member, and a sensor that supports the floor forming member and outputs a detected value according to a force applied to the floor forming member due to movement of the movable body,
Position detection learned in advance from learning data representing the position of the moving body for learning in the plurality of sensor units and the detection value for learning output from the sensor when the force due to the movement of the moving body is propagated. a position detection unit that detects the position of the mobile object in the plurality of sensor units based on the trained model and the detection value output from the sensor;
a flow line detection unit that detects a flow line of the mobile body in the plurality of sensor units based on the position of the mobile body detected by the position detection unit ,
A moving body flow line detection system characterized in that the detected value of the sensor is an output waveform of the entire time domain in a state in which the force is generated . When the sensor unit is viewed from above, the floor forming member is formed in a square shape, and the installation locations of the sensors are each vertex of a virtual square formed in a similar shape to the square and at the same central position. The moving object flow line detection system according to any one of claims 1 to 3 , characterized in that: 5. The mobile body flow line detection system according to claim 4 , wherein the length of one side of the virtual square is set to about half the length of one side of the floor forming member. When the sensor unit is viewed from above, the sensor is installed at each vertex of a virtual n-gon (n is a natural number of 3 or more) formed at the same central position as the floor forming member. A mobile body flow line detection system according to any one of claims 1 to 3 . The moving object flow line detection system according to any one of claims 1 to 6 , wherein the sensor unit includes a position regulating section that regulates horizontal movement of the sensor.

Images