Notification system and sound volume estimation program

Abstract

To prevent people from being surprised by alam sounds and falling over.SOLUTION: A notification system comprises: a first speaker which is provided on an industrial vehicle and outputs a notification sound V to a person H; a second speaker which is located in a facility and outputs the notification sound V to the person H; a learned model having a sound volume estimation part; and a sound volume change part which changes sound volume of both or any one of the first speaker and the second speaker 20. The sound volume estimation part previously performs machine learning by teacher data that uses a notification distance from the first speaker or the second speaker 20 to the person H and sound volume of the notification sound of the speaker as input data and uses a surprise degree of the person H who has been notified as output data. When the notification distance of the first speaker or the second speaker is inputted to the sound estimation part, the sound estimation part estimates such a sound volume of the speaker that the person H who is notified becomes a prescribed surprise degree or less. The sound volume change part changes sound volume of the speaker to the estimated sound volume or less.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an alarm system and a volume estimation program. [Background technology]

[0002] In recent years, the proportion of workers aged 60 or older among those killed or injured in industrial accidents has been increasing. One of the major workplace accidents involving elderly people is falling while working. Therefore, it is necessary to create a working environment that reduces the number of falls among elderly people.

[0003] Industrial vehicles such as forklifts are equipped with speakers that warn of their approach while driving, and the warning sound helps to avoid contact with pedestrians (see, for example, Patent Documents 1 and 2). However, many elderly people have hearing impairments and tend to have difficulty hearing the warning sound. As a result, it is possible to make the warning sound louder, but if the warning sound is too loud, it may startle pedestrians and cause them to fall over. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 6-149373 [Patent Document 2] Patent Publication No. 2022-139495 Summary of the Invention [Problem to be solved by the invention]

[0005] Therefore, an object of the present invention is to provide an alarm system that can prevent a person from being startled by an alarm sound and falling over. [Means for solving the problem]

[0006] In order to solve the above problems, the notification system according to the present invention comprises: A first speaker that is provided in the industrial vehicle and outputs an alarm sound to a person; a second speaker that is arranged in the facility and outputs an alarm sound to people; A trained model having a volume estimation unit; a volume change unit that changes the volume of the first speaker and / or the second speaker; the volume estimation unit performs machine learning in advance using teacher data in which the notification distance from the first or second speaker to a person and the volume of the notification sound of the speaker are input data and the surprise level of the person who is notified is output data, and when the notification distance of the first or second speaker is input, the volume estimation unit estimates the volume of the speaker at which the surprise level of the person who is notified is equal to or lower than a predetermined level; The volume change unit changes the volume of the speaker to be equal to or lower than the estimated volume.

[0007] The notification system preferably comprises: A speaker selection unit is further provided, The speaker selection unit determines, based on the notification distance, whether the notification sound is to be output from the first speaker or the second speaker.

[0008] The notification system preferably comprises: When the notification distance between the first speaker and the person is equal to or shorter than a predetermined notification distance, the speaker selection unit determines to output the notification sound from the first speaker regardless of the notification distance between the second speaker and the person.

[0009] The notification system preferably comprises: A speaker selection unit is further provided, The speaker selection unit determines that the notification sound is to be output from the second speaker when the person to be notified is not in the traveling direction of the industrial vehicle and is within a predetermined distance from the industrial vehicle.

[0010] The notification system preferably comprises: A speaker selection unit is further provided, The second speaker is a plurality of speakers, The speaker selection unit determines, based on the notification distance, from which speaker of the plurality of second speakers the notification sound is to be output.

[0011] The notification system preferably comprises: A first camera provided in the industrial vehicle; A second camera disposed near the second speaker, The trained model further includes a distance estimation unit, The first and second cameras capture images of a person to generate a person image; the distance estimation unit performs machine learning in advance using teacher data in which the human image generated by the first camera is input data and the notification distance from the first speaker to the person is output data, and when the human image generated by the first camera is input, the distance estimation unit estimates the notification distance from the first speaker to the person; Furthermore, the distance estimation unit performs machine learning in advance using teacher data in which the human image generated by the second camera is input data and the reported distance from the second speaker to the person is output data, and when the human image generated by the second camera is input, the distance estimation unit estimates the reported distance from the second speaker to the person.

[0012] The notification system preferably comprises: A first camera provided in the industrial vehicle; a second camera disposed within the facility; The trained model further includes a posture estimation unit, The first and second cameras capture images of a person to generate a person image; The posture estimation unit performs machine learning in advance using training data in which a human image is input data and the type of posture of the person is output data, and when a human image is input, the posture estimation unit estimates the type of posture of the person, When the type of posture of the person is input, the volume estimation unit estimates a volume that will be equal to or lower than a predetermined surprise level based on the type of posture of the person.

[0013] The notification system preferably comprises: A first camera provided in the industrial vehicle; a second camera disposed within the facility; The trained model further includes an orientation estimation unit, The first and second cameras capture images of a person to generate a person image; The orientation estimation unit performs machine learning in advance using teacher data in which a human image is input data and the orientation of the human face is output data, and when a human image is input, the orientation estimation unit estimates the orientation of the human face, When the direction of the person's face is input, the volume estimation unit estimates a volume that will be equal to or lower than a predetermined surprise level based on the direction of the person's face.

[0014] The notification system preferably comprises: An environmental sound acquisition unit for acquiring environmental sound is further provided, When the environmental sound is input, the volume estimation unit estimates a volume that will be equal to or lower than a predetermined surprise level based on the environmental sound.

[0015] The notification system preferably comprises: A first camera provided in the industrial vehicle; a second camera disposed within the facility; The trained model further includes a surprise level estimation unit, The first and second cameras capture images of a person to generate a person image; the surprise level estimation unit performs machine learning in advance using teacher data in which the image of the notified person is input data and the surprise level is output data, and when the image of the person notified by the alarm sound is input, the surprise level estimation unit estimates the surprise level; The volume estimation unit re-learns the volume that will be below a predetermined surprise level based on the image of the person before the alert, the alert distance from the first or second speaker to the person, the volume of the alert sound when it is given, and the surprise level of the person who is alerted.

[0016] In order to solve the above problems, a volume estimation program according to the present invention includes: A volume estimation program for use in an alarm system, comprising: The notification system is a system including an industrial vehicle having a first speaker for outputting a notification sound to people, and a second speaker arranged in a facility for outputting a notification sound to people, the system including a computer, The volume estimation program is The volume estimation unit performs machine learning in advance using teacher data in which the notification distance from the first or second speaker to a person and the volume of the notification sound from the speaker are input data and the level of surprise of the person who is notified is output data.When the notification distance of the first or second speaker is input, the volume estimation unit estimates the volume of the speaker at which the person who is notified will be surprised to a predetermined level or less. Effect of the Invention

[0017] The alarm system according to the present invention can prevent a person from falling over due to being startled by the alarm sound while still achieving the effect of the alarm sound. [Brief description of the drawings]

[0018] [Figure 1] 1 is a schematic diagram showing an alarm system according to an embodiment of the present invention, illustrating a state in which a second speaker outputs an alarm sound. [Diagram 2] FIG. 2 is a block diagram of a control unit of the notification system. [Diagram 3] 2 is a schematic diagram showing the notification system shown in FIG. 1, illustrating a state in which a first speaker outputs a notification sound. [Figure 4] A diagram illustrating the operation of each estimation unit of a trained model. [Diagram 5] FIG. 1 is a diagram showing examples of human postures. [Figure 6] FIG. 13 is a diagram illustrating machine learning of a volume estimation unit. [Figure 7] FIG. 11 is a diagram illustrating an operation of a volume estimation unit. [Figure 8] FIG. 11 is a flow chart showing an operation of the industrial vehicle to notify a person of its approach. [Figure 9] FIG. 11 is a schematic diagram showing a notification system according to a modified example of the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0019] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an embodiment of a notification system and a volume estimation program according to the present invention will be described with reference to the accompanying drawings.

[0020] Fig. 1 is a schematic diagram showing an alarm system S according to an embodiment of the present invention. As shown in Fig. 1, the alarm system S includes an industrial vehicle 1, a second speaker 20 and a second camera 21 arranged in a facility, and a management device 3.

[0021] <Industrial vehicles> The industrial vehicle 1 is a forklift and is configured to be able to communicate with a management device 3. As shown in Fig. 1, the industrial vehicle 1 is traveling along an aisle R, and a pedestrian H1 (hereinafter sometimes referred to as "person H") is walking ahead of the industrial vehicle 1, and a worker H2 is working beside the aisle R.

[0022] The industrial vehicle 1 is an autonomously traveling unmanned guided vehicle, but this is merely one example, and the industrial vehicle 1 according to the present invention may be a manned industrial vehicle. The industrial vehicle 1 does not have to be a forklift, and may be, for example, an in-plant transport vehicle or an in-plant towing vehicle that does not have the fork 11, and is not particularly limited.

[0023] The industrial vehicle 1 includes a vehicle body 10, a fork 11, a first speaker 12 (see FIG. 2), and a first camera 13 (see FIG. 2). The fork 11 is configured to be capable of ascending and descending, thereby lifting and lowering a load.

[0024] The first speaker 12 has directionality and outputs an alarm sound V to the person H as shown in FIG. 3. The first speaker 12 is composed of a plurality of parametric speakers and is configured to be able to expand and contract the range of directivity according to the distance from the person H. This makes it possible to prevent the first speaker 12 from outputting the alarm sound V to anyone other than the target person H. Therefore, it is possible to prevent the alarm sound V from being output to a worker H2 or the like who is not located in the traveling direction of the industrial vehicle 1, and to prevent the worker H2 from being startled. Since the industrial vehicle 1 moves forward and backward, the first speaker 12 is configured to be able to output the alarm sound V at least in the forward and backward directions of the industrial vehicle 1. Note that the number of the first speakers 12 and the configuration of the directionality of the first speakers 12 are not particularly limited.

[0025] There is no particular limitation on the specific sound of the notification sound V. For example, the notification sound V may be a buzzer sound, a specific melody, or a human voice.

[0026] The first speaker 12 may not output the alarm sound V unless the person H is detected by the first camera 13. Alternatively, the first speaker 12 may be configured to output the alarm sound V at a predetermined volume only at an intersection of the passage R, for example, even if the person H is not detected.

[0027] Moreover, the industrial vehicle 1 further includes a melody output unit (not shown) that outputs a melody around the vehicle while it is traveling. The volume of the melody is set to a volume that will not startle the person H.

[0028] The first camera 13 is configured to be able to capture an image of the periphery of the vehicle body 10, captures an image of a person H, and generates a person image HI. The first camera 13 is made up of a plurality of cameras, and is configured to be able to capture images at least in the forward and backward directions of the industrial vehicle 1. The person image HI generated by the first camera 13 is transmitted to the management device 3.

[0029] <Second speaker and second camera> 1, the second speaker 20 and the second camera 21 are arranged in the facility and are configured to be able to communicate with the management device 3. The number of the second speaker 20 and the second camera 21 may be more than one, and the number of the second speakers 20 and the second cameras 21 arranged is not particularly limited. In addition, the specific location of the second speaker 20 is not particularly limited, and may be, for example, the ceiling, the wall of the passage R, the top surface or side surface of a rack, etc.

[0030] The second speaker 20 has directionality similar to the first speaker 12, and outputs an alarm sound V to the person H as shown in Fig. 1. Also, the second speaker 20 is configured with a plurality of parametric speakers similar to the first speaker 12, and is configured so that the range of directionality can be expanded or contracted according to the distance from the person H. However, the configuration of the second speaker 20 is not limited to this.

[0031] The specific voice of the alarm sound V from the second speaker 20 may be different from that of the first speaker 12. This allows the person H to recognize whether the alarm sound V is the voice of the second speaker 20 or the voice of the first speaker 12. For example, the voice of the second speaker 20 may be a synthesized voice such as "An industrial vehicle 1 is approaching. Please be careful."

[0032] The second camera 21 is disposed near the second speaker 20 and configured to be able to capture an image of the surroundings of the second speaker 20. The second camera 21 captures an image of a person H, similar to the first camera 13, and generates a person image HI. The person image HI generated by the second camera 21 is transmitted to the management device 3.

[0033] <Management device> The management device 3 is a so-called server computer having a storage device, a memory, and a calculation device, and the storage device stores a volume estimation program. The volume estimation program causes the computer to execute a speaker selection unit 30, a trained model 31, and a volume change unit 32.

[0034] The speaker selection unit 30 determines which of the first and second speakers 12 and 20 is to output the notification sound V based on the notification distance DN from the first speaker 12 or the second speaker 20 to the person H. Specifically, the speaker selection unit 30 determines which of the first speaker 12 and the second speaker 20 is closer to the person H to be notified. Therefore, when the second speaker 20 is closer to the person H, the speaker selection unit 30 determines to output the notification sound V from the second speaker 20 as shown in FIG. 1, and when the first speaker 12 is closer to the person H, the speaker selection unit 30 determines to output the notification sound V from the first speaker 12 as shown in FIG. 3. However, when the notification distance DN between the first speaker 12 and the person H is equal to or less than a predetermined notification distance DN, the speaker selection unit 30 determines to output the notification sound V from the first speaker 12 regardless of the notification distance DN between the second speaker 20 and the person H. This predetermined notification distance DN may be, for example, 10 m.

[0035] Furthermore, when a plurality of second speakers 20 are provided, the speaker selection unit 30 may determine that the notification sound V is to be output from the second speaker 20 that is closer to the person H.

[0036] When the speaker selection unit 30 determines which speaker is to output the informative sound V, the speaker selection unit 30 transmits the determination to the volume estimation unit 315 .

[0037] <Configuration of trained model> The trained model 31 includes a surprise level estimation unit 311, a distance estimation unit 312, a posture estimation unit 313, a direction estimation unit 314, and a volume estimation unit 315.

[0038] The surprise degree estimation unit 311 is, for example, a neural network, and previously learns the correlation by machine learning using teacher data having the human image HI of the notified person H as input data and the surprise degree Q as output data. The machine learning method may be, for example, a machine learning method that learns the correlation between the difference between the human image HI of the person H before being notified and the human image HI of the person H after being notified and the surprise degree Q. Alternatively, the surprise degree estimation unit 311 may learn by machine learning a face image expressing an emotion such as surprise or impatience as teacher data, and may be configured to estimate the surprise degree Q of the notified person H when the human image HI of the notified person H is input. By such machine learning, the surprise degree estimation unit 311 estimates the surprise degree Q of the notified person H when the human image HI of the person H notified by the alarm sound V is input, as shown in FIG. 4A. The surprise degree Q may be output as a numerical value between 0.01 and 1.0, for example.

[0039] The "predetermined surprise level" RQ in the present invention is a surprise level Q that can prevent a user from falling over, dropping something they are holding, or making a mistake in their work due to being startled. This predetermined surprise level RQ may be a value of the surprise level Q, such as 0.7.

[0040] The distance estimation unit 312 is, for example, a neural network, and is configured to perform machine learning in advance using teacher data having a human image HI generated by the first camera 13 as input data and a notification distance DN from the first speaker 12 to the person H as output data, and to estimate the notification distance DN from the first speaker 12 to the person H when the human image HI generated by the first camera 13 is input. Furthermore, the distance estimation unit 312 is configured to perform machine learning in advance using teacher data having a human image HI generated by the second camera 21 as input data and a notification distance DN from the second speaker 20 to the person H as output data, and to estimate the notification distance DN from the second speaker 20 to the person H when the human image HI generated by the second camera 21 is input. As a result, when the human image HI is input, the distance estimation unit 312 estimates the notification distance DN from the first speaker 12 or the second speaker 20 to the person H, as shown in FIG. 4B. The estimated notification distance DN is output to the volume estimation unit 315.

[0041] The posture estimation unit 313 is, for example, a neural network, and has previously learned the correlation between these by machine learning using teacher data having the human image HI as input data and the posture type AP of the person H as output data. As a result, when the posture estimation unit 313 receives the human image HI as input data, it estimates the posture type AP of the person H (hereinafter, may be simply referred to as the "posture AP of the person H") as shown in FIG. 4C. The estimated posture AP of the person H is output to the volume estimation unit 315.

[0042] Fig. 5 is a diagram showing examples of postures AP of a person H. As shown in Fig. 5, types of postures AP of the person H include, for example, a standing posture (see Fig. 5A), a posture carrying a load (see Fig. 5B), a walking posture (see Fig. 5C), and a posture performing work (see Fig. 5D).

[0043] The orientation estimation unit 314 is, for example, a neural network, and has previously undergone machine learning using teacher data in which the human image HI is input data and the facial orientation DF of the person H is output data. As a result, when the orientation estimation unit 314 receives the human image HI as input, it estimates the facial orientation DF of the person H, as shown in Fig. 4D. The facial orientation DF of the person H is, for example, the direction of the industrial vehicle 1, the opposite direction to the industrial vehicle 1, the direction of a load that the person H is carrying, the direction of a work tool when working, etc. The estimated facial orientation DF of the person H is output to the volume estimation unit 315.

[0044] The volume estimation unit 315 is, for example, a neural network, and as shown in Fig. 6, the notification distance DN, the posture AP of the person H, the facial direction DF of the person H, and the volume (hereinafter referred to as "notification volume") db of the notification sound V are input as input data, and the correlation between them is machine-learned in advance using teacher data having the surprise level Q as output data. As a result, when the notification distance DN, the posture AP of the person H, and the facial direction DF estimated by the distance estimation unit 312, the direction estimation unit 314, and the posture estimation unit 313 are input as shown in Fig. 7, the volume estimation unit 315 estimates the notification volume db that is equal to or less than a predetermined surprise level RQ. The volume estimation unit 315 outputs the estimated notification volume db related to the speaker that outputs the notification sound V determined by the speaker selection unit 30 to the volume change unit 32.

[0045] In this embodiment, the volume estimation unit 315 does not output the notification volume db to the volume change unit 32 unless the notification distance DN is input (i.e., unless the person H is detected). Note that the volume estimation unit 315 may be configured to estimate, for example, the notification volume db that is equal to or less than a predetermined surprise level RQ as zero when the notification distance is equal to or less than a predetermined notification distance DN. This makes it possible to prevent the notification sound V from being output when the distance between the industrial vehicle 1 and the person H is too close.

[0046] The volume change unit 32 changes the notification volume db of the speaker that outputs the notification sound V determined by the speaker selection unit 30 to be equal to or lower than the notification volume db estimated by the volume estimation unit 315. By outputting the notification sound V at a volume equal to or lower than the estimated notification volume db, the volume change unit 32 can prevent the person H from being startled by the notification sound V and falling over.

[0047] Furthermore, when the alarm sound V is actually output, the volume estimation unit 315 uses the alarm distance DN, the posture AP of the person H, the facial orientation DF of the person H, the actually output alarm volume db, and the surprise level Q estimated by the surprise level estimation unit 311 at that time as teacher data, and re-learns the correlations among them. That is, the volume estimation unit 315 in this embodiment uses the alarm distance DN, the posture AP of the person H, the facial orientation DF, and the actually output alarm volume db estimated by the distance estimation unit 312, the orientation estimation unit 314, and the posture estimation unit 313 as input data, and uses the surprise level Q estimated by the surprise level estimation unit 311 as teacher data, and re-learns the correlations among them, thereby making it possible to make more accurate estimation from the next time.

[0048] <Operation of the notification system> Next, the operation of the notification system S will be described again with reference to FIG.

[0049] (1) The notification system S outputs a melody by the melody output unit while the industrial vehicle 1 is traveling (see S1 (step 1) in FIG. 8). The notification system S sets the notification volume db to zero until the person H is photographed by the first camera 13 (see S2 in FIG. 8).

[0050] (2) When the first camera 13 captures an image of a person H (i.e., when the person H is detected in the traveling direction of the industrial vehicle 1) (Yes in S3 in FIG. 8), the alarm system S estimates the alarm distance DN, the posture AP, and the face direction DF of the person H based on the human image HI generated by the first and second cameras 13 and 21 using the trained model 31 (see S4 in FIG. 8). In addition, the alarm system S determines, based on the alarm distance DN, which of the first and second speakers 12 and 20 is to output the alarm sound V using the speaker selection unit 30 (see S5 in FIG. 8). Note that, even when the second speaker 20 is composed of a plurality of speakers, the alarm system S similarly determines, based on the alarm distance DN, using the speaker selection unit 30, to output the alarm sound V to the speaker closest to the person H.

[0051] (3) Next, the alarm system S estimates an alarm volume db for the speaker that outputs the alarm sound V based on the estimated alarm distance DN, the posture AP of the person H, and the facial direction DF, which will be less than a predetermined surprise level RQ (see S6 in Figure 8).

[0052] (4) Next, the notification system S changes the notification volume db of the first speaker 12 or the second speaker 20 by the volume change unit 32 so that the notification volume db is equal to or lower than the estimated notification volume db (see S7 in FIG. 8).

[0053] (5) Next, the notification system S outputs the notification sound V from the first speaker 12 or the second speaker 20 (see S8 in FIG. 8).

[0054] When the first camera 13 is capturing an image of the person H (i.e., when the person H continues to be detected in the traveling direction of the industrial vehicle 1), the notification system S repeats steps S4 to S8 in Fig. 8, and automatically changes the speaker for issuing the notification and the notification volume db as the notification distance DN, the posture AP of the person H, and the facial direction DF change. Furthermore, the notification system S changes the speaker for issuing the notification as needed by repeating step S5 in response to fluctuations in the notification distance DN between the speakers due to the movement of the person H and the industrial vehicle 1. Then, when the first camera 13 no longer captures an image of the person H (i.e., when the person H is no longer in the traveling direction of the industrial vehicle 1) (No in S3 in Fig. 8), the notification volume db is changed to zero again (see S2 in Fig. 8).

[0055] The notification system S has the above configuration, and can prevent the person H from falling over due to being startled by the notification sound V. In addition, the notification system S learns not only the correlation between the notification distance DN, the notification volume db, and the surprise level Q, but also the correlation between the posture AP and the face direction DF of the person H by using the learned model 31, so that it is possible to prevent the notification system S from outputting an excessively loud notification volume db to the person H concentrating on work or the person H looking in the opposite direction of the industrial vehicle 1, causing the person H to fall over due to being startled. Moreover, the notification system S estimates the notification distance DN, the posture AP, and the face direction DF of the person H based on the person image HI, so that there is no need to add a special sensor. Furthermore, when the distance between the industrial vehicle 1 and the person H is relatively far, the notification system S can generate a more appropriate person image HI by photographing the person H with the second camera 21 arranged in the facility, and can output the notification sound V from the second speaker 20 to notify the person H with a smaller notification sound V.

[0056] Although an embodiment of the notification system and the volume estimation program of the present invention has been described above, the present invention is not limited to the above embodiment. For example, the notification system and the volume estimation program of the present invention may be implemented by appropriately combining the following modified examples.

[0057] <Modification> The notification system S may include a known sensor capable of detecting the direction of the person H and each notification distance DN. In this case, the volume estimation unit 315 acquires each notification distance DN from the sensor, and the trained model 31 does not need to include the distance estimation unit 312.

[0058] The human image HI when each component of the trained model 31 undergoes machine learning and the human image HI when each estimation is performed may be input data consisting of a plurality of human images HI. In other words, the human image HI according to the present invention is a concept that includes a moving image.

[0059] The input data of the teacher data when the volume estimation unit 315 performs machine learning may include at least the notification distance DN and the notification volume db. In this case, when the notification distance DN is input, the volume estimation unit 315 estimates the notification volume db that is equal to or less than a predetermined surprise level RQ.

[0060] The notification system S may further include an environmental sound acquisition unit that acquires environmental sound. In this case, the volume estimation unit 315 may further learn correlations between the environmental sound, the notification distance DN, the posture AP of the person H, the facial direction DF of the person H, and the notification volume db by machine learning in advance using teacher data having the environmental sound, the notification distance DN, the posture AP of the person H, and the facial direction DF as output data, and estimate the notification volume db that is equal to or less than a predetermined surprise level RQ when the environmental sound, the notification distance DN, the posture AP, and the facial direction DF of the person H are input. In this case, the volume estimation unit 315 may learn correlations between the environmental sound, the notification distance DN, and the notification volume db by machine learning in advance using teacher data having the surprise level Q as output data, and estimate the notification volume db that is equal to or less than a predetermined surprise level RQ when the environmental sound and the notification distance DN are input.

[0061] The volume change unit 32 and the trained model 31 in the management device 3 may also be provided in the industrial vehicle 1. In this case, the industrial vehicle 1 further has a computer, which operates as a surprise level estimation unit 311, a distance estimation unit 312, a direction estimation unit 314, a posture estimation unit 313, a volume estimation unit 315, and a volume change unit 32 by a volume estimation program, and changes the notification volume db of the first speaker 12 to a notification volume db that is equal to or less than a predetermined surprise level RQ.

[0062] The notification system S may further include a position detection unit that detects the position of the industrial vehicle 1. In this case, the position of the person H may be detected by the second camera 21, and the speaker selection unit 30 may determine to output the notification sound V from the second speaker 20 when the person H to be notified is not in the traveling direction of the industrial vehicle 1 and is within a predetermined distance from the industrial vehicle 1, as shown in Fig. 9. This makes it possible to notify people H in the blind spot of the industrial vehicle 1 that the industrial vehicle 1 is approaching. The predetermined distance in this case may be, for example, 20 m. [Explanation of symbols]

[0063] S Notification System 1. Industrial vehicles (forklifts) 10. Body 11. Fork 12 1st Speaker 13 Camera 1 20 2nd Speaker 21 Second Camera 3 Management device 30 Speaker selection section 31 Pre-trained models 311 Surprise Level Estimation Unit 312 Distance Estimation Unit 313 Posture estimation section 314 Orientation Estimation Unit 315 Volume Estimation Unit 32 Volume change section Types of AP posture DF Face Direction DN broadcast distance H People (pedestrians, workers) HI People Images Q Surprise Level R aisle RQ Predetermined surprise level V Notification sound

Claims

1. a first speaker provided in the industrial vehicle and configured to output an alarm sound to a person; a second speaker that is arranged within the facility and outputs the notification sound to people; A trained model having a volume estimation unit; a volume change unit that changes the volume of the first speaker and / or the second speaker, the volume estimation unit performs machine learning in advance using teacher data in which the notification distance from the first speaker or the second speaker to a person and the volume of the notification sound of the speaker are input data and the surprise level of the person who is notified is output data, and when the notification distance of the first speaker or the second speaker is input, the volume estimation unit estimates the volume of the speaker at which the surprise level of the person who is notified is equal to or lower than a predetermined level; The volume change unit changes the volume of the speaker to be equal to or lower than the estimated volume.

2. A speaker selection unit is further provided, The notification system according to claim 1 , wherein the speaker selection unit determines which of the first speaker and the second speaker is to output the notification sound based on the notification distance.

3. The alarm system of claim 2, wherein the speaker selection unit determines to output the alarm sound from the first speaker regardless of the alarm distance between the second speaker and the person when the alarm distance between the first speaker and the person is equal to or less than a predetermined alarm distance.

4. A speaker selection unit is further provided, The alarm system according to claim 1, wherein the speaker selection unit determines that the alarm sound is to be output from the second speaker when the person to be notified is not in the direction of travel of the industrial vehicle and is within a predetermined distance from the industrial vehicle.

5. A speaker selection unit is further provided, The second speaker is a plurality of speakers, The notification system according to claim 1 , wherein the speaker selection unit determines which of the second speakers is to output the notification sound based on the notification distance.

6. A first camera provided on the industrial vehicle; A second camera disposed near the second speaker, The trained model further includes a distance estimation unit, the first and second cameras capture images of people to generate person images; the distance estimation unit performs machine learning in advance using teacher data in which the human image generated by the first camera is input data and the notification distance from the first speaker to a person is output data, and when the human image generated by the first camera is input, the distance estimation unit estimates the notification distance from the first speaker to a person; Furthermore, the distance estimation unit performs machine learning in advance using teacher data in which the human image generated by the second camera is input data and the notification distance from the second speaker to a person is output data, and when the human image generated by the second camera is input, the distance estimation unit estimates the notification distance from the second speaker to a person.

7. A first camera provided on the industrial vehicle; and a second camera disposed within the facility, The trained model further includes a posture estimation unit, the first and second cameras capture images of people to generate person images; the posture estimation unit performs machine learning in advance using teacher data in which the human image is input data and a type of posture of the person is output data, and when the human image is input, the posture estimation unit estimates a type of posture of the person; The notification system according to claim 1 , wherein the volume estimating unit, upon receiving the type of posture of the person, estimates a volume that will be equal to or lower than a predetermined level of surprise based on the type of posture of the person.

8. A first camera provided on the industrial vehicle; and a second camera disposed within the facility, The trained model further includes an orientation estimation unit, the first and second cameras capture images of people to generate person images; the orientation estimation unit performs machine learning in advance using teacher data in which the human image is input data and the orientation of the person's face is output data, and when the human image is input, estimates the orientation of the person's face; The notification system according to claim 1 , wherein the volume estimating unit, upon receiving the direction of the person's face, estimates a volume that will be equal to or lower than a predetermined surprise level based on the direction of the person's face.

9. An environmental sound acquisition unit for acquiring environmental sound is further provided, The notification system according to any one of claims 1 to 8, wherein the volume estimation unit, when receiving the environmental sound, estimates a volume that will be equal to or lower than a predetermined level of surprise based on the environmental sound.

10. A first camera provided on the industrial vehicle; and a second camera disposed within the facility, The trained model further includes a surprise level estimation unit, the first and second cameras capture images of people to generate person images; the surprise level estimation unit performs machine learning in advance using teacher data in which the human image of the notified person is input data and a surprise level is output data, and when the human image of the person notified by the alarm sound is input, the surprise level estimation unit estimates the surprise level; The notification system of claim 1, wherein the volume estimation unit re-learns a volume that is below a predetermined level of surprise based on the human image of the person before being notified, the notification distance from the first or second speaker to the person, the volume of the notification sound at the time of notification, and the level of surprise of the person who is notified.

11. A volume estimation program for use in an alarm system, comprising: The notification system is a system including an industrial vehicle having a first speaker that outputs a notification sound to people, and a second speaker that is disposed within a facility and outputs the notification sound to people, the system including a computer, The volume estimation program includes: A volume estimation program that performs machine learning in advance using teacher data in which the notification distance from the first or second speaker to a person and the volume of the notification sound from the speaker are input data and the level of surprise of the person who is notified is output data, and that is executed as a volume estimation unit that, when the notification distance from the first speaker or the second speaker is input, estimates the volume of the speaker at which the level of surprise of the person who is notified is below a predetermined level.

Images