Pinch monitoring device

Abstract

To provide a pinching monitoring device capable of preventing excessive detection such as pinching from the inside of a moving body by analyzing an image outside the moving body.SOLUTION: A pinching monitoring device 11 monitors pinching of an object in a side sliding door opened and closed between the inside and the outside of a vehicle. The pinching monitoring device 11 includes a pinching detection part 8A for detecting the pinching of the object existing outside the side sliding door on the basis of a detection result of a door pinching detection device 8 for detecting the presence of the pinching of the object in the side sliding door and an analysis result of image analysis devices 8B and 11a for analyzing a photographed image outside the car. The obstruction detection unit 11a detects obstruction of an object present in the area to which approach of an object is restricted. The entrapment monitoring device 11 includes a detection result transmission unit 11a that transmits the detection result of the entrapment detection unit 11b.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a pinch monitoring device that monitors whether an object has been pinched from the outside of a moving body by a door that opens and closes between the inside and outside of the moving body. [Background technology]

[0002] In order to detect door entrapment and dragging, which are difficult to detect with current side sliding doors, many technologies for installing sensors in the door edge rubber installed in side sliding doors have been studied. A conventional entrapment detection device (hereinafter referred to as Prior Art 1) includes a waveguide that elastically deforms when an object is entrapped between a first member and a second member, a first signal acquisition unit that converts sound waves propagating through this waveguide into an electrical signal and outputs a response signal, and a signal processing unit that determines that an object is entrapped when there is a peak in the position response of this response signal (see, for example, Patent Document 1). Prior Art 1 reduces the dead zone by performing processing to remove trend components from the response signal.

[0003] A conventional railway vehicle door entrapment detection device (hereinafter referred to as Prior Art 2) includes an acceleration acquisition unit that acquires the acceleration of the side sliding door in the sliding direction, and a determination unit that determines that door entrapment has occurred when the absolute value of the acceleration of the side sliding door is less than a predetermined threshold value in the latter stage of the closing operation of the side sliding door (see, for example, Patent Document 2). Prior Art 2 detects the occurrence of door entrapment by detecting a decrease in the speed of the side sliding door immediately before the side sliding door reaches a fully closed position.

[0004] A conventional door entrapment detection device (hereinafter referred to as Prior Art 3) includes a sensor that detects the rotation of a rotor at the door tip of a vehicle door, a determination unit that determines whether or not a door entrapment state has occurred in the vehicle door based on an output signal from the sensor, and an operation confirmation unit that confirms whether or not the sensor is operating based on the output signal from the sensor (see, for example, Patent Document 3). Prior Art 3 suppresses erroneous determination of door entrapment due to malfunction of the sensor by checking the operation of the sensor outside of a predetermined period when the sensor is not detecting door entrapment.

[0005] A conventional drag detection device (hereinafter referred to as Prior Art 4) comprises a rotor that is rotatable at the end of the sliding door and slidable in the opening direction of the sliding door, a rotation detection means that detects the rotation of the rotor, a drag state determination means that determines the occurrence of a drag state based on an output signal from the rotation detection means, and a slide detection means that detects the sliding of the rotor (see, for example, Patent Document 4). In Prior Art 4, it is determined that a drag state has occurred when sliding of the rotor is detected after the sliding door has entered a closed state.

[0006] Furthermore, technology for detecting passengers approaching a train on a platform from images is also being studied. A conventional passenger approach detection system (hereinafter referred to as Prior Art 5) includes an imaging unit that continuously acquires images of a predetermined range of the train and the edge of the platform, a determination unit that analyzes the continuous images from the imaging unit to determine whether the train has departed the platform and started moving, and a detection unit that detects the presence of approaching passengers located within a predetermined range of the edge of the platform when the determination unit determines that the train has started moving (see, for example, Patent Document 5). Prior Art 5 uses image analysis using machine learning to detect passengers approaching a train when the train is departing and the platform is crowded.

[0007] A conventional station platform monitoring system (hereinafter referred to as Prior Art 6) comprises multiple cameras on the side of a vehicle, a dangerous event detection means that analyzes images taken by the multiple cameras to detect dangerous events, a determination means that performs logical operations on the detection results of the dangerous event detection means to determine the area where the dangerous event has occurred, and a notification means that notifies nearby people of the occurrence of the dangerous event based on the determination results of the determination means (see, for example, Patent Document 6). Prior Art 6 alerts the train driver by detecting dangerous events on the station platform.

[0008] A conventional train operation support device (hereinafter referred to as Prior Art 7) includes an event detection unit that detects two events with different pre-specified priorities using the results of camera photography, and an operation reception means that displays the two events on a display device in a distinguishable manner and allows each event to be confirmed (see, for example, Patent Document 7). Prior Art 7 efficiently notifies monitors of events with high priority.

[0009] A conventional object detection device (hereinafter referred to as Prior Art 8) is equipped with a machine learning unit that receives an image of the area surrounding the door clamp and outputs information on the presence or absence of a foreign object contained in the image (see, for example, Patent Document 8). Prior Art 8 trains a learning model using the image of the area surrounding the door clamp and information on the presence or absence of a foreign object contained in the image as training data.

[0010] A conventional detection device (hereinafter referred to as Prior Art 9) includes a ToF sensor that detects the distance to an object based on the time between emitting light and receiving reflected light, and a control unit that detects whether an object is caught in an elevator door based on a distance image generated by the ToF sensor (see, for example, Patent Document 9). Prior Art 9 detects an object protruding from an elevator door based on a distance image of an area including the elevator door of the railway vehicle generated by a ToF sensor attached to the railway vehicle. [Prior art documents] [Patent documents]

[0011] [Patent Document 1] Japanese Patent Application Publication No. 2019-035630

[0012] [Patent Document 2] Japanese Patent Application Publication No. 2020-104688

[0013] [Patent Document 3] Japanese Patent Application Publication No. 2019-108753

[0014] [Patent Document 4] Japanese Patent Publication No. 2020-041410

[0015] [Patent Document 5] Japanese Patent Publication No. 2022-089208

[0016] [Patent Document 6] Japanese Patent Publication No. 2022-032611

[0017] [Patent Document 7] Japanese Patent Application Publication No. 2023-148909

[0018] [Patent Document 8] Patent Publication No. 2021-037904

[0019] [Patent Document 9] Patent Publication No. 2021-116029 Summary of the Invention [Problem to be solved by the invention]

[0020] In Prior Art 1 to 4, when a pinch is detected by the sensor door edge rubber, it is impossible to distinguish whether it is detected from inside the vehicle or from outside the vehicle (platform). Therefore, Prior Art 1 to 4 have the problem of overdetecting even when, for example, a passenger's clothing is trapped a few centimeters from inside the vehicle and there is no problem with operation. Furthermore, Prior Art 5 to 7 can detect approaching the vehicle by image detection using a vehicle-mounted camera, but have the problem of not being able to detect the actual state of door entrapment. Prior Art 8 uses machine learning to determine the presence or absence of a foreign object around the door pinch area from captured images, which can result in a false determination that a foreign object is present around the door pinch area even when there is no foreign object around the door pinch area. Prior Art 9 may detect a door entrapment when a passenger pulls an object trapped in a boarding / alighting door toward the inside of the vehicle, resulting in the problem of notifying the driver or conductor of the door entrapment even when the train is able to safely operate.

[0021] An object of the present invention is to provide a pinch monitoring device that can prevent overdetection of pinching from the inside of a moving object by analyzing an image of the outside of the moving object. [Means for solving the problem]

[0022] The present invention solves the above problems by the means described below. Although the reference numerals corresponding to the embodiments of the present invention are given in parentheses, the present invention is not limited to these embodiments. The invention of claim 1 is an entrapment monitoring device (11) that monitors whether an object (O2) is caught in a door (5R, 5L) that opens and closes between the inside and outside of a moving body (2), as shown in Figures 1 to 4 and 6 to 10, and is characterized in that it comprises an entrapment detection unit (11a) that detects (S130) whether an object is caught in the door based on the detection result of a door entrapment detection device (8) that detects (S110) whether the object is caught in the door and the analysis result of an image analysis device (10) that analyzes (S130) an image taken of the outside of the moving body.

[0023] The invention of claim 2 is a pinch monitoring device according to claim 1, characterized in that, as shown in Figures 1 and 2, the pinch detection unit detects the pinch of an object that is present in an area (A1) where access of the object is restricted while the moving body is moving.

[0024] The invention of claim 3 is the entrapment monitoring device according to claim 1, characterized in that it further comprises a detection result transmission unit (11b) that transmits (S140, S150, S170) the detection result of the entrapment detection unit, as shown in Figs. 4, 6, 7 and 9.

[0025] The invention of claim 4 is a pinch monitoring device according to claim 3, characterized in that the detection result transmission unit transmits the detection result of the pinch detection unit to a notification device (12) that notifies an attendant of the detection result of the pinch detection unit.

[0026] The invention of claim 5 is a pinch monitoring device according to claim 1, characterized in that, as shown in FIG. 1, the pinch detection unit detects pinch of an object present outside the door based on the analysis results of images captured by the imaging devices (9A, 9B) on the moving body (2) side and / or the fixed body (1) side.

[0027] The invention of claim 6 is the entrapment monitoring device of claim 1, characterized in that, as shown in Figures 4, 6, 7 and 9, when the door entrapment detection device detects entrapment (S110), the image analysis device changes the criteria for determining whether or not an object is present outside the door (S120), and the entrapment detection unit detects entrapment of an object present outside the door based on the analysis result of the image analysis device using the changed criteria and the detection result of the door entrapment detection device (S130).

[0028] The invention of claim 7 is a pinch monitoring device according to claim 1, characterized in that, as shown in Figs. 4 to 6, the image analysis device comprises a determination unit (10b) that determines (S130) that an object exists outside the door when the probability of whether or not an object exists outside the door exceeds threshold values ​​(Th1, Th2), and the determination unit, when the door pinch detection device detects pinch (S110), changes the threshold value (Th1) to determine (S130) whether or not an object exists outside the door, and the pinch detection unit detects pinch of an object existing outside the door based on the determination result of the determination unit using the changed threshold value (Th2) and the detection result of the door pinch detection device.

[0029] The invention of claim 8 is the entrapment monitoring device of claim 7, characterized in that, as shown in Figures 5 and 6, when the door entrapment detection device detects entrapment (S110), the determination unit lowers the threshold value (Th1) and determines whether or not the object is present outside the door (S130).

[0030] The invention of claim 9 is a pinch monitoring device according to claim 1, wherein, as shown in FIGS. 6 to 8, the image analysis device includes a determination unit (10d) that determines (S130) whether or not an object (O1, O2) to be analyzed exists outside the door based on an analysis object (AO1, AO2) analyzed by the image analysis device and a captured image of the outside of the moving body, and the determination unit, when the door pinch detection device detects pinch (S110), changes the analysis object (AO1) and determines (S130) whether or not the object to be analyzed exists outside the door, and the pinch detection unit detects pinch of an object existing outside the door based on the determination result of the determination unit using the changed analysis object (AO2) and the detection result of the door pinch detection device (S130).

[0031] The invention of claim 10 is a pinch monitoring device according to claim 9, characterized in that, as shown in Figures 6 and 8, when the door pinch detection device detects pinch (S110), the determination unit changes the analysis target from a person to the person and an object used by the person, and determines whether or not a person and an object used by the person are present outside the door (S130).

[0032] The invention of claim 11 is a pinch monitoring device according to claim 1, characterized in that, as shown in Figures 6, 9 and 10, the image analysis device comprises an analysis area (A3, A4) a predetermined distance away from the outside of the moving body, and a determination unit (10f) that determines (S110) whether or not the object is present within this analysis area based on a captured image of the outside of the moving body, and the determination unit, when the door pinch detection device detects pinch (S110), changes the analysis area (A3) to determine (S130) whether or not the object is present outside the door, and the pinch detection unit detects pinch of an object present outside the door based on the determination result of the determination unit using the changed analysis area (A4) and the detection result of the door pinch detection device.

[0033] The invention of claim 12 is the entrapment monitoring device of claim 11, characterized in that, as shown in Figures 6 and 10, when the door entrapment detection device detects entrapment (S110), the determination unit expands the analysis area (A3) and determines whether or not the object is present outside the door (S130). [Effects of the Invention]

[0034] According to the present invention, by analyzing an image of the outside of a moving object, it is possible to prevent overdetection such as pinching from the inside of the moving object. [Brief explanation of the drawings]

[0035] [Figure 1]1A and 1B are schematic diagrams illustrating a platform and a vehicle in which pinching of an object is monitored by a pinch monitoring device according to a first embodiment of the present invention, where (A) is a plan view and (B) is a side view. [Figure 2] 1A and 1B are schematic diagrams showing an example of an object being pinched, monitored by the pinch monitoring device according to the first embodiment of the present invention, where (A) is a schematic diagram showing a case where passengers are on the platform, and (B) is a schematic diagram showing a case where passengers are inside the train. [Figure 3] 1A and 1B are cross-sectional views schematically showing the entrapment of an object monitored by the entrapment monitoring device according to the first embodiment of the present invention, in which (A) is a cross-sectional view schematically showing a state in which an object is entrapped in the side sliding door, and (B) is a cross-sectional view schematically showing a state in which no object is entrapped in the side sliding door. [Figure 4] 1 is a configuration diagram of a pinch monitoring system including a pinch monitoring device according to a first embodiment of the present invention. [Figure 5] 4 is a timing chart schematically showing a determination process by an image analyzer of the pinch monitoring system including the pinch monitoring device according to the first embodiment of the present invention. [Figure 6] 4 is a flowchart for explaining the operation of the pinch monitoring system including the pinch monitoring device according to the first embodiment of the present invention. [Figure 7] FIG. 10 is a configuration diagram of a pinch monitoring system including a pinch monitoring device according to a second embodiment of the present invention. [Figure 8] 10A and 10B are schematic diagrams for explaining the operation of changing the analysis target by the image analysis device of the pinch monitoring system including the pinch monitoring device according to the second embodiment of the present invention, where FIG. 10A is a schematic diagram showing the operation before changing the analysis target, and FIG. 10B is a schematic diagram showing the operation after changing the analysis target. [Figure 9] FIG. 10 is a configuration diagram of a pinch monitoring system including a pinch monitoring device according to a third embodiment of the present invention. [Figure 10]10A and 10B are schematic diagrams illustrating the operation of changing the analysis area by the image analyzer of the pinch monitoring system including the pinch monitoring device according to the third embodiment of the present invention, where (A) is a schematic diagram showing the operation before the analysis area is changed, and (B) is a schematic diagram showing the operation after the analysis area is changed. DETAILED DESCRIPTION OF THE INVENTION

[0036] (First embodiment) A first embodiment of the present invention will be described in detail below with reference to the drawings. The platform 1 shown in Figure 1 is a facility (boarding and alighting area) provided along the tracks for the purpose of passengers getting on and off. The platform 1 is provided at a station or other stop where trains 2 are stopped to allow passengers to get on and off. The platform 1 has a platform floor 1a on which passengers walk and a platform edge 1b that forms the edge on the track side.

[0037] The restricted area A1 is an area where passenger entry is restricted. The restricted area A1 is, for example, a dangerous area outside the boundary line L, where passenger entry is restricted in principle, but passenger entry is temporarily permitted when passengers are getting on or off. The non-restricted area A2 is an area where passenger entry is permitted. The non-restricted area A2 is, for example, a safe area inside the boundary line L, where passengers are generally allowed to pass freely at all times. The boundary line L is a line that separates the restricted area A1 from the non-restricted area A2. The boundary line L is located outside the white line, which is a row of white tiles laid near the edge of the platform 1b to prevent passengers from coming into contact with the moving vehicle 2, or outside the warning blocks (guidance blocks) laid near the edge of the platform 1b to assist visually impaired people in walking.

[0038] Objects O1 and O2 shown in FIG. 2 are intervening objects (foreign objects) caught between the side sliding doors 5R and 5L. Object O1 is a person (passenger) getting on or off the vehicle 2. Object O1 is, for example, the body, foot, hand, or finger of a male or female adult, a child, or a mobility-restricted person such as a visually impaired person. Object O2 is an object used by a person. Object O2 is, for example, clothing, a bag, shoes, an umbrella, a handbag, a backpack, a suitcase, a carry-on bag, a skateboard, a snowboard, a walking stick, a white cane, a walker, a stroller, or a wheelchair. The following describes an example in which object O2 is caught between the side sliding doors 5R and 5L as shown in FIG. 2, and object O1 or object O2 is located outside the side sliding doors 5R and 5L.

[0039] The vehicle 2 shown in Figures 1 and 2 is a moving object that travels along a railroad track. The vehicle 2 is, for example, a railway vehicle such as an electric train, a diesel railcar, or a passenger car. One or more vehicles 2 make up a train. The vehicle 2 is manned by a driver who operates the vehicle 2 and a conductor who handles signals necessary for driving the vehicle 2 and operates the conductor switch that opens and closes the side sliding doors 5R, 5L. In the case of one-man operation, the conductor's duties are omitted and only the driver who also performs the conductor's duties is on board as a full-time staff member, or in the case of automatic operation, only staff who respond to emergencies or abnormalities are on board. The vehicle 2 is equipped with a car body 3, which is a structure for loading and transporting passengers.

[0040] The side entrance 4 shown in Figure 1 is an entrance used by passengers when getting on and off. The side entrance 4 is formed on a side surface that constitutes the side structure (side body structure) of the car body 3. For example, one side entrance 4 is provided on each side in the case of an express car, three on each side in the case of a suburban car, and four on each side in the case of a commuter car. The side sliding doors 5R, 5L shown in Figures 1 and 3 are side doors that open and close between the inside and outside of the car 2. The side sliding doors 5R, 5L are two double sliding doors that can move back and forth between a closed position and an open position and open and close in opposite directions to open and close the side entrance 4.

[0041] The door edge rubbers 6R and 6L shown in FIGS. 1 and 3 are rubber components that form the leading ends of the side sliding doors 5R and 5L. The door edge rubbers 6R and 6L absorb the impact that occurs when an object O2 is caught between the door edge rubbers 6R and 6L and also tightly fit the door edge rubbers 6R and 6L so that no gaps are formed between the door edge rubbers 6R and 6L when the side sliding doors 5R and 5L are closed. As shown in FIG. 3, the door edge rubbers 6R and 6L are hollow members with a roughly U-shaped cross section when cut horizontally, and are attached continuously along the height direction of the side sliding doors 5R and 5L as shown in FIG. 1(B). As shown in FIG. 3(B), the door edge rubbers 6R and 6L tightly fit the leading ends of the door edge rubbers 6R and 6L in a slightly elastically deformed state so that no gaps are formed between the leading ends of the door edge rubbers 6R and 6L when the side sliding doors 5R and 5L are closed. The door edge rubbers 6R and 6L are formed, for example, by coating the surface of natural rubber, which is a base material, with highly durable chloroprene rubber.

[0042] The entrapment monitoring system 7 shown in FIG. 4 is a system that monitors whether an object O2 is pinched in the side sliding doors 5R, 5L of a vehicle 2. The entrapment monitoring system 7 monitors whether an object O2 is pinched in the side sliding doors 5R, 5L from outside the side sliding doors 5R, 5L. The entrapment monitoring system 7 includes a door entrapment detection device 8 shown in FIGS. 3 and 4, photographing devices 9A and 9B shown in FIGS. 1, 2, and 4, an image analysis device 10 shown in FIG. 4, an entrapment monitoring device 11, and a notification device 12. The entrapment monitoring system 7 determines whether an object O2 is pinched in the side sliding doors 5R, 5L as shown in FIG. 2, and analyzes photographed images of the outside of the side sliding doors 5R, 5L to monitor whether an object O1 or O2 is present outside the side sliding doors 5R, 5L.

[0043] The door entrapment detection device 8 shown in Figures 3 and 4 is a device that detects whether an object O2 is caught in the door edge rubbers 6R, 6L. As shown in Figure 3(A), the door entrapment detection device 8 detects whether an object O2 is caught between the door edge rubbers 6R and 6L by detecting changes in the door edge rubbers 6R, 6L when the side sliding doors 5R, 5L are closed. The door entrapment detection device 8 detects whether an object O2 is caught between the door edge rubbers 6R, 6L, for example, by converting the elastic deformation of the door edge rubbers 6R, 6L into an electric signal using the piezoelectric effect, by detecting whether an object is caught based on the rotation of a rotatable rotor at the tip of the door edge rubbers 6R, 6L when an object is caught, by detecting whether an object is caught based on changes in the acceleration of the side sliding doors 5R, 5L, or by detecting sound waves generated by a waveguide in the door edge rubbers 6R, 6L when an object is caught.

[0044] The door entrapment detection device 8 determines whether an object O2 is trapped between the door edge rubbers 6R and 6L as shown in FIG. 3(A) or whether an object O2 is trapped between the door edge rubbers 6R and 6L as shown in FIG. 3(B). The door entrapment detection device 8 can detect, for example, an object O2 that exceeds a predetermined thickness, but cannot detect an object O2 that is thinner than the predetermined thickness. For example, even if the object O2 is thinner than the predetermined thickness, the door entrapment detection device 8 can detect the entrapment when the object O2 is pulled out and a load is applied to the door edge rubbers 6R and 6L. The door entrapment detection device 8 outputs a door entrapment detection signal corresponding to whether or not the object O2 is trapped between the door edge rubbers 6R and 6L to the image analysis device 10 and the entrapment monitoring device 11 via a communication network, wired, or wireless.

[0045] The photographing devices 9A and 9B shown in Figures 1, 2, and 4 are devices that photograph the outside of the vehicle 2. The photographing devices 9A and 9B are, for example, video cameras that capture video of a predetermined area outside the side sliding doors 5R and 5L of the vehicle 2, or still cameras that capture still images. The photographing devices 9A and 9B photograph an area that includes the platform edge 1b so that objects O1 and O2 present near the platform edge 1b can be recognized. As shown in Figures 1 and 2, the photographing device 9A photographs the outside of the vehicle 2 from the platform 1 side, and the photographing device 9B photographs the outside of the vehicle 2 from the vehicle 2 side.

[0046] As shown in FIG. 1, the camera device 9A is disposed in a single position at the center of the platform 1 in the longitudinal direction, or in multiple positions spaced apart along the platform 1, and captures images of the exterior of all of the side sliding doors 5R, 5L of the vehicle 2. For example, the camera device 9A starts its photographing operation before the first train enters the platform 1, continues its photographing operation thereafter, and ends its photographing operation after the last train leaves the platform 1. As shown in FIG. 1, the camera device 9B is disposed on both sides of the car body 3 near the front and rear ends in the traveling direction, and captures images of the exterior of all of the side sliding doors 5R, 5L of the vehicle 2. For example, the camera device 9B starts its photographing operation when the vehicle 2 enters the platform 1 and passes a predetermined position, continues its photographing operation while the vehicle 2 is parked at the platform 1, and ends its photographing operation when the vehicle 2 leaves the platform 1 and reaches a predetermined speed. The camera devices 9A, 9B output the photographed images of the exterior of the vehicle 2 as image data to the image analysis device 10 via a communication network, wired connection, or wireless connection.

[0047] The image analysis device 10 shown in FIG. 4 is a device that analyzes captured images of the outside of the vehicle 2. The image analysis device 10 analyzes image data output by the image capture devices 9A and 9B to determine whether objects O1 and O2 are present outside the vehicle 2. The image analysis device 10 analyzes the captured images to determine whether objects O1 and O2 are present outside the side sliding doors 5R and 5L using machine learning. Here, machine learning is a type of artificial intelligence, a computer algorithm that automatically improves by learning from experience. Machine learning can identify or predict data by learning from the data and grasping its characteristics. For example, as shown in FIG. 2, the image analysis device 10 determines whether objects O1 and O2 are present outside the side sliding doors 5R and 5L of the vehicle 2 within the restricted area A1 of the platform 1. When the door entrapment detection device 8 detects entrapment, the image analysis device 10 changes the criteria for determining whether objects O1 and O2 are present outside the side sliding doors 5R and 5L. The image analysis device 10 includes an accuracy calculation unit 10a and a determination unit 10b.

[0048] The accuracy calculation unit 10a is a means for calculating the accuracy of whether or not objects O1 and O2 are present outside the side sliding doors 5R and 5L based on captured images of the outside of the vehicle 2. The accuracy calculation unit 10a repeatedly learns from the image data output by the image capture devices 9A and 9B, finds patterns of objects O1 and O2 hidden in this image data, and predicts objects O1 and O2 according to these patterns by applying the learned patterns to new image data. The accuracy calculation unit 10a learns using data called learning data, training data, or training data, and uses the learning results (trained models) to calculate the accuracy of whether or not objects O1 and O2 are present outside the side sliding doors 5R and 5L.

[0049] The accuracy calculation unit 10a calculates the accuracy of whether or not objects O1 and O2 are present outside the side sliding doors 5R and 5L using deep learning (DL). Deep learning is a machine learning technique that combines advanced computing power and multi-layered neural networks to learn complex patterns hidden in large amounts of image data. The accuracy calculation unit 10a verifies and confirms the analysis, for example, using cross-validation. Cross-validation is a statistical technique in which standard data is divided, a portion is analyzed, and the analysis results are tested on the remaining portion to verify and confirm the validity of the analysis and evaluate generalization performance. Examples of cross-validation include holdout validation, which divides data into training data and evaluation data in advance, and k-fold cross-validation, which divides the training data and evaluation data multiple times and performs training and evaluation on each data. The accuracy calculation unit 10a, for example, divides image data into learning data (training data or training data) and evaluation data (test data), and divides verification data from the learning data to verify it. The accuracy calculation unit 10a, for example, learns using the learning data, evaluates the accuracy of the accuracy calculation model using the verification data, adjusts parameters of the accuracy calculation model to determine a learned accuracy calculation model, and evaluates this learned accuracy calculation model using test data. The accuracy calculation unit 10a applies the learned accuracy calculation model to unknown image data and calculates the accuracy (0 to 100%) of whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L.

[0050] The determination unit 10b is a means for determining that objects O1, O2 are present outside the side sliding doors 5R, 5L when the accuracy exceeds threshold values ​​Th1, Th2. As shown in FIG. 5 , when the door entrapment detection device 8 detects entrapment, the determination unit 10b changes threshold value Th1 and determines whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L. When the door entrapment detection device 8 detects entrapment, the determination unit 10b lowers threshold value Th1 and determines whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L. The determination unit 10b outputs the determination result, whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L, as a determination result signal to the entrapment monitor 11 via a communication network, wired or wireless.

[0051] The entrapment monitoring device 11 shown in FIG. 4 is a device that monitors whether an object O2 is caught in the side sliding doors 5R, 5L. The entrapment monitoring device 11 prevents the door entrapment detection device 8 from overdetecting a dangerous situation when it is not a dangerous situation, or the photography devices 9A, 9B from failing to detect a dangerous situation when it is not a dangerous situation. The entrapment monitoring device 11 functions as a safety device that prevents accidents caused by an object O2 being caught in the side sliding doors 5R, 5L, and also functions as an auxiliary or support device that notifies attendants of a dangerous situation caused by entrapment. The entrapment monitoring device 11 is located on the platform 1 side or the car 2 side so that entrapment can be monitored from the platform 1 side or the car 2 side. The entrapment monitoring device 11 includes an entrapment detection unit 11a and a detection result transmission unit 11b.

[0052] The entrapment detection unit 11a is a means for detecting the entrapment of objects O1, O2 that exist outside the side sliding doors 5R, 5L based on the detection results of the door entrapment detection device 8 and the analysis results of the image analysis device 10. The entrapment detection unit 11a detects the entrapment of objects O1, O2 that exist within the restricted area A1 while the vehicle 2 is moving. As shown in FIG. 2, the entrapment detection unit 11a detects the entrapment of objects O1, O2 that exist outside the side sliding doors 5R, 5L based on the analysis results of the image analysis device 10 using the changed criteria and the detection results of the door entrapment detection device 8. The entrapment detection unit 11a detects entrapment of objects O1, O2 present outside the side sliding doors 5R, 5L based on the judgment result of the judgment unit 10b using the changed threshold value Th2 shown in Figure 5 and the detection result of the door entrapment detection device 8.

[0053] The detection result transmission unit 11b is a means for transmitting the detection result of the entrapment detection unit 11a. The detection result transmission unit 11b transmits the detection result of the entrapment detection unit 11a to the notification device 12. The detection result transmission unit 11b transmits the entrapment of the objects O1, O2 present outside the side sliding doors 5R, 5L as an entrapment detection signal to the notification device 12 via a communication network, wired or wireless.

[0054] The notification device 12 is a device that notifies an attendant of the detection result of the entrapment detection unit 11a. The notification device 12 is arranged on the platform 1 side or the car 2 side so that it can notify the attendant on platform 1 or the attendant in car 2. The notification device 12 notifies the attendant by image or sound, for example, on a display device or audio generating device in the crew compartment where the attendant is on duty, or on a mobile terminal device carried by the attendant. The notification device 12 notifies the attendant, for example, to check which door number in which car of car 2 the attendant should check.

[0055] Next, the operation of the pinch monitoring device according to the first embodiment of the present invention will be described. In step (hereinafter referred to as S) 100, the image analysis device 10 determines whether a photographed image has been input from the photographing devices 9A and 9B. The image analysis device 10 executes image analysis processing according to an image analysis program. When the image analysis device 10 determines that image data has been input from the photographing devices 9A and 9B, it proceeds to S110. On the other hand, when the image analysis device 10 determines that no image data has been input from the photographing devices 9A and 9B, the image analysis device 10 repeats the determination until image data is input.

[0056] In S110, the door pinch detection device 8 determines whether a door pinch has been detected. The door pinch detection device 8 determines whether an object O2 is sandwiched between the door tip rubbers 6R and 6L as shown in Fig. 3(A), or whether no object O2 is sandwiched between the door tip rubbers 6R and 6L as shown in Fig. 3(B). When the door pinch detection device 8 determines that an object O2 is sandwiched, it proceeds to S120, and when the door pinch detection device 8 determines that no object O2 is sandwiched, it proceeds to S160.

[0057] In S120, the image analysis device 10 changes the determination criteria for image analysis. Based on the photographed image outside the vehicle 2, the accuracy calculation unit 10a calculates the probability of whether objects O1 and O2 exist outside the sliding doors 5R and 5L. As shown in Fig. 5, when the door pinch detection device 8 detects a pinch, the determination unit 10b changes the threshold Th1 to the threshold Th2 (Th2 < Th1). For example, when the probability calculated by the accuracy calculation unit 10a exceeds the threshold of 80%, and when the determination unit 10b normally determines that the object O1 is a person, and when the door pinch detection device 8 detects a pinch, and when it exceeds the threshold of 50%, the determination unit 10b changes the threshold Th1 from 80% to 50% so as to determine that the object O1 is a person. Similarly, for example, when the probability calculated by the accuracy calculation unit 10a exceeds the threshold of 80%, and when the determination unit 10b normally determines that the object O2 is an object used by a person, and when the door pinch detection device 8 detects a pinch, and when it exceeds the threshold of 50%, the determination unit 10b changes the threshold Th1 from 80% to 50% so as to determine that the object O2 is an object used by a person.

[0058] In S130, the entrapment monitor device 11 determines whether or not the objects O1 and O2 present outside the side sliding doors 5R and 5L are pinched. The entrapment monitor device 11 executes the entrapment monitoring process according to the entrapment monitoring program. For example, while object O1 could not previously be determined to be a person unless the accuracy exceeded a threshold of 80%, when entrapment is detected, object O1 is now determined to be a person if the accuracy exceeds a threshold of 50%. As a result, when there is a risk of object O2 being pinched and dragged, as shown in FIG. 2, the determination unit 10b can make a safe determination. If the entrapment detection unit 11a determines that object O2 present outside the side sliding doors 5R and 5L is pinched, the process proceeds to S140. If the entrapment detection unit 11a determines that object O1 and O2 present outside the side sliding doors 5R and 5L are not pinched, the process proceeds to S150.

[0059] In S140, the detection result transmitter 11b transmits the detection result. As a result, the notification device 12 notifies the attendant that the objects O1 and O2 present outside the side sliding doors 5R and 5L may be caught in the side sliding doors 5R and 5L. As a result, for example, the driver does not start the car 2, and the conductor operates the conductor switch to open the side sliding doors 5R and 5L.

[0060] In S150, the detection result transmitter 11b transmits the detection result. As a result, the notification device 12 notifies the attendant that there is a possibility that the object O2 is caught in the side sliding doors 5R, 5L, but that there is a possibility that the objects O1, O2 are not present outside the side sliding doors 5R, 5L. For example, if the hem or sleeve of a passenger's clothing is caught in the side sliding doors 5R, 5L, or if an umbrella or cane is caught in the side sliding doors 5R, 5L, a passenger inside the vehicle 2 can pull these out into the vehicle, so there is no object O2 present outside the side sliding doors 5R, 5L. Therefore, for example, the attendant performs a safety check to see if there are any objects O1, O2 outside the side sliding doors 5R, 5L.

[0061] In S160, the image analysis device 10 determines whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L. If the determination unit 10b determines that objects O1, O2 are present outside the side sliding doors 5R, 5L, the process proceeds to S170, and if the determination unit 10b determines that objects O1, O2 are not present outside the side sliding doors 5R, 5L, the series of entrapment monitoring operations ends.

[0062] In S170, the detection result transmission unit 11b transmits the detection result. As a result, the notification device 12 notifies the staff member that there is a possibility that the object O2 is not caught between the side sliding doors 5R, 5L, but that there is a possibility that the objects O1, O2 are present outside the side sliding doors 5R, 5L. For example, there may be a case where the object O2 is not caught between the side sliding doors 5R, 5L, but a person is present outside the side sliding doors 5R, 5L near the side sliding doors 5R, 5L. Therefore, for example, the staff member performs safety check to determine whether or not the objects O1, O2 are present outside the side sliding doors 5R, 5L.

[0063] The pinch monitoring device according to the first embodiment of the present invention has the following effects. (1) In the first embodiment, the entrapment detection unit 11a detects whether or not an object O1 or O2 is entrapped between the side sliding doors 5R or 5L based on the detection result of the door entrapment detection device 8, which detects whether or not an object O2 is entrapped between the side sliding doors 5R or 5L, and the analysis result of the image analysis device 10, which analyzes a captured image of the outside of the vehicle 2. The door entrapment detection device 8 has the advantage of being able to detect whether or not an object O2 is entrapped between the side sliding doors 5R or 5L, but has the disadvantage of being unable to detect whether or not an object O1 or O2 is entrapped between the side sliding doors 5R or 5L. For this reason, for example, if a passenger's clothing or other item is entrapped between the side sliding doors 5R or 5L but the passenger is able to safely remove the clothing or other item from between the side sliding doors 5R or 5L, the door entrapment detection device 8 may overdetect that the passenger is in danger even though the passenger is not in danger. On the other hand, the image analysis device 10 has the advantage of being able to analyze the presence of objects O1 and O2 outside the side sliding doors 5R and 5L, but the disadvantage of being unable to analyze the entrapment of an object O2 in the side sliding doors 5R and 5L. Therefore, for example, there is a problem in that the image analysis device 10 may erroneously determine that a passenger is in danger even when the passenger is merely present near the side sliding doors 5R and 5L and is not in danger. In this first embodiment, by compensating for the drawbacks of the door entrapment detection device 8 and the image analysis device 10, it is possible to prevent overdetection by the door entrapment detection device 8 and erroneous determination by the image analysis device 10. For example, because the exterior condition of the vehicle 2 can be monitored based on images captured by the image capture devices 9A and 9B, it is possible to prevent overdetection by the door entrapment detection device 8 of low-risk entrapment from inside the vehicle 2. Furthermore, for example, because the door entrapment detection device 8 can reliably detect entrapment, it is possible to prevent missed detection of entrapment that is difficult to detect based on images captured by the image capture devices 9A and 9B. As a result, the performance of the door entrapment detection system that detects entrapment in the side sliding doors 5R, 5L of the vehicle 2 can be improved.

[0064] (2) In the first embodiment, the entrapment detection unit 11a detects the entrapment of objects O1, O2 that are present within the restricted area A1, where the approach of objects O1, O2 is restricted, while the vehicle 2 is moving. Therefore, it is possible to detect a dangerous situation in which the vehicle 2 is traveling with objects O1, O2 that are present outside the side sliding doors 5R, 5L entrapped between the side sliding doors 5R, 5L.

[0065] (3) In the first embodiment, the detection result of the entrapment detector 11a is transmitted by the detection result transmitter 11b. This allows the staff in the carriage 2 and the staff on the platform 1 to be notified that the objects O1, O2 located outside the side sliding doors 5R, 5L have been entrapped between the side sliding doors 5R, 5L.

[0066] (4) In the first embodiment, the detection result transmission unit 11b transmits the detection result of the entrapment detection unit 11a to the notification device 12, which notifies the staff of the detection result of the entrapment detection unit 11a. Therefore, it is possible to notify the monitor device in the crew compartment of the car 2 or the portable terminal device carried by the staff on the platform 1 or in the car 2 that the objects O1, O2 located outside the side sliding doors 5R, 5L have been entrapped between the side sliding doors 5R, 5L.

[0067] (5) In the first embodiment, the entrapment detection unit 11a detects whether or not objects O1, O2 present outside the side sliding doors 5R, 5L are caught in the side sliding doors 5R, 5L based on the analysis results of the images captured by the camera devices 9A, 9B on the platform 1 side and the vehicle 2 side. Therefore, by using the existing camera device 9A installed on the platform 1 or the existing camera device 9B mounted on the vehicle 2, it is possible to detect whether or not objects O1, O2 present outside the side sliding doors 5R, 5L are caught in the side sliding doors 5R, 5L.

[0068] (6) In this first embodiment, when the door entrapment detection device 8 detects entrapment, the image analysis device 10 changes the criteria for determining whether or not the objects O1, O2 are present outside the side sliding doors 5R, 5L. In this first embodiment, the entrapment detection unit 11a detects the entrapment of the objects O1, O2 present outside the side sliding doors 5R, 5L based on the analysis results of the image analysis device 10 using the changed criteria and the detection results of the door entrapment detection device 8. Therefore, by changing the criteria when it is detected that the object O2 is entrapped in the side sliding doors 5R, 5L, it is possible to safely evaluate whether or not the objects O1, O2 are present outside the side sliding doors 5R, 5L.

[0069] (7) In this first embodiment, the accuracy calculation unit 10a calculates the accuracy of whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L based on a captured image of the outside of the vehicle 2, and when the accuracy exceeds a threshold value Th2, the determination unit 10b determines that objects O1, O2 are present outside the side sliding doors 5R, 5L. Also, in this first embodiment, when the door entrapment detection device 8 detects entrapment, the determination unit 10b changes the threshold value Th1 and determines whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L. Furthermore, in this first embodiment, the entrapment detection unit 11a detects entrapment of objects O1, O2 present outside the side sliding doors 5R, 5L based on the determination result of the determination unit 10b using the changed threshold value Th2 and the detection result of the door entrapment detection device 8. Therefore, by changing the threshold value Th1 when it is detected that an object O2 has been caught in the side sliding doors 5R, 5L, it is possible to safely evaluate whether or not an object O1, O2 is present outside the side sliding doors 5R, 5L.

[0070] (8) In this first embodiment, when the door entrapment detection device 8 detects entrapment, the determination unit 10b lowers the threshold value Th1 and determines whether or not the objects O1, O2 are present outside the side sliding doors 5R, 5L. Therefore, in a dangerous state in which the object O2 is detected to be entrapped in the side sliding doors 5R, 5L, the threshold value Th1 is lowered to the safe side, making it possible to safely evaluate whether or not the objects O1, O2 are present outside the side sliding doors 5R, 5L. For example, by changing the threshold value Th1, which is set in normal times to determine whether or not the objects O1, O2 are present by image analysis, to a lower value when the object O2 is entrapped, it is possible to prevent the objects O1, O2 from being overlooked and determine whether or not the object O2 is entrapped in a dangerous state.

[0071] (Second embodiment) In the following, the same parts as those shown in FIGS. 1 to 4 are denoted by the same reference numerals and detailed description thereof will be omitted. 7 includes an analysis target setting unit 10c and a determination unit 10d. When the door entrapment detection device 8 detects entrapment, the image analysis device 10 changes the criteria for determining whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L.

[0072] The analysis target setting unit 10c is a means for setting analysis targets AO1 and AO2 to be analyzed by the image analysis device 10. The analysis target setting unit 10c sets objects O1 and O2 that may be present outside the side sliding doors 5R and 5L as analysis targets AO1 and AO2 in advance. The analysis target setting unit 10c sets a person and objects used by the person as analysis targets AO1 and AO2. For example, as shown in FIG. 8(A), when a person is the analysis target AO1, the analysis target setting unit 10c sets the object O1 as the analysis target AO1. For example, as shown in FIG. 8(B), when a person and objects used by the person are the analysis target AO2, the analysis target setting unit 10c sets the objects O1 and O2 as the analysis target AO2. The analysis target setting unit 10c includes an analysis target generation unit that generates various objects O1 and O2 as analysis target data, and a storage device that stores the generated analysis target data in advance.

[0073] The determination unit 10d is a means for determining whether or not the objects O1 and O2 of the analysis objects AO1 and AO2 are present outside the side sliding doors 5R and 5L based on the captured image of the outside of the vehicle 2 and the analysis objects AO1 and AO2. The determination unit 10d repeatedly learns from the image data output by the image capture devices 9A and 9B, finds patterns of the analysis objects AO1 and AO2 hidden in this image data, and predicts the objects O1 and O2 of the analysis objects AO1 and AO2 according to these patterns by applying the learned patterns to new image data. The accuracy calculation unit 10a learns using data called learning data, training data, or training data, and uses the learning results (trained models) to determine whether or not the objects O1 and O2 of the analysis objects AO1 and AO2 are present outside the side sliding doors 5R and 5L.

[0074] The determination unit 10d determines, for example, whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L by deep learning (DL). For example, the determination unit 10d learns using learning data, evaluates the accuracy of a determination model using verification data, adjusts parameters of the determination model to determine a learned determination model, and evaluates this learned determination model using test data. The determination unit 10d applies the learned determination model to unknown image data to determine whether or not objects O1, O2, which are analysis targets AO1, AO2, are present outside the side sliding doors 5R, 5L. The determination unit 10d outputs the determination result, whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L, as a determination result signal to the entrapment monitoring device 11 via a communication network, wired, or wireless.

[0075] When the door entrapment detection device 8 detects entrapment, the determination unit 10d changes the analysis object AO1 and determines whether or not objects O1 and O2 of the analysis object AO2 are present outside the side sliding doors 5R and 5L. As shown in Fig. 8, when the door entrapment detection device 8 detects entrapment, the determination unit 10d changes the analysis object AO1 from a person to the person and an object used by the person and determines whether or not the person and an object used by the person are present outside the side sliding doors 5R and 5L. The entrapment detection unit 11a detects entrapment of objects O1 and O2 present outside the side sliding doors 5R and 5L based on the determination result of the determination unit 10b using the changed analysis object AO2 and the detection result of the door entrapment detection device 8.

[0076] Next, the operation of the pinch monitoring device according to the second embodiment of the present invention will be described. In the following, the processes corresponding to those shown in FIG. 6 are denoted by the same numbers and detailed descriptions thereof will be omitted. In S120 shown in FIG. 6, the image analysis device 10 changes the judgment criteria for image analysis. As shown in FIG. 8, when the door entrapment detection device 8 detects entrapment, the judgment unit 10d changes the analysis object AO1 to the analysis object AO2. The judgment unit 10d changes the object to be judged as to whether or not an object is present outside the side sliding doors 5R, 5L from a person to the person and an object used by the person. For example, as shown in FIG. 8(A), when the door entrapment detection device 8 does not detect entrapment, the judgment unit 10d changes the analysis object AO2 set by the analysis object setting unit 10c to the analysis object AO1. As a result, the judgment unit 10d changes the object to be judged as to whether or not an object is present outside the side sliding doors 5R, 5L to a person. On the other hand, as shown in FIG. 8(B), when the door entrapment detection device 8 detects entrapment, the judgment unit 10d changes the analysis object AO1 set by the analysis object setting unit 10c to the analysis object AO2. As a result, the determination unit 10d changes the objects to be determined as to whether or not they are present outside the side sliding doors 5R, 5L to a person and an object used by that person.

[0077] In S130, the entrapment monitor 11 determines whether or not an object O1, O2 outside the side sliding doors 5R, 5L is trapped. When the door entrapment detector 8 detects entrapment, the determination unit 10d expands the analysis object AO1 from a person to an analysis object AO2 that includes the person and the object used by the person, and makes a determination. For example, as shown in FIG. 8(A), when a person is approaching the side sliding doors 5R, 5L but the object used by the person is not trapped between the side sliding doors 5R, 5L, the entrapment monitor 11 prevents a false detection of a dangerous situation in which the person may be dragged. On the other hand, for example, as shown in FIG. 8(B), when a person is approaching the side sliding doors 5R, 5L and the object used by the person is not trapped between the side sliding doors 5R, 5L, a dangerous situation in which the person may be dragged is detected. Therefore, the determination unit 10d determines not only people but also objects such as a white cane, wheelchair, or stroller used by the person, expanded from the analysis object AO1 to the analysis object AO2. As a result, the entrapment monitor 11 determines whether or not objects O1 and O2 present outside the side sliding doors 5R and 5L are entrapped.

[0078] The pinch monitoring device according to the second embodiment has the following advantages in addition to the advantages of the first embodiment. (1) In this second embodiment, the analysis object setting unit 10c sets the analysis objects AO1 and AO2 to be analyzed by the image analysis device 10, and the determination unit 10d determines whether or not the objects O1 and O2 of the analysis objects AO1 and AO2 are present outside the side sliding doors 5R and 5L based on the captured image of the outside of the vehicle 2 and the analysis objects AO1 and AO2. Also in this second embodiment, when the door entrapment detection device 8 detects entrapment, the determination unit 10d changes the analysis objects AO1 and AO2 and determines whether or not the objects O1 and O2 of the analysis objects AO1 and AO2 are present outside the side sliding doors 5R and 5L. Furthermore, in this second embodiment, the entrapment detection unit 11a detects entrapment of the objects O1 and O2 present outside the side sliding doors 5R and 5L based on the determination result of the determination unit 10d using the changed analysis objects AO1 and AO2 and the detection result of the door entrapment detection device 8. Therefore, when pinching is detected, the range of the analysis objects AO1 and AO2 can be changed, and it can be determined in a wider range and in more detail whether or not the objects O1 and O2 of the analysis objects AO1 and AO2 are present outside the side sliding doors 5R and 5L.

[0079] (2) In this second embodiment, when the door entrapment detection device 8 detects entrapment, the determination unit 10d changes the analysis objects AO1 and AO2 from a person to the person and an object used by the person, and determines whether a person and an object used by the person are present outside the side sliding doors 5R and 5L. This allows for a more comprehensive determination of whether objects O1 and O2 present outside the side sliding doors 5R and 5L are entrapped, and allows for more detailed detection of entrapment of objects O1 and O2 present outside the side sliding doors 5R and 5L. For example, when entrapment in the side sliding doors 5R and 5L is detected, by expanding the analysis object AO1 to the analysis object AO2, it is possible to detect whether not only a person but also an object used by the person is entrapped. As a result, it is possible to prevent missed entrapment detection and to accurately determine whether entrapment occurs in a dangerous situation.

[0080] (Third embodiment) 9 includes an analysis area setting unit 10e and a determination unit 10f. When the door entrapment detection device 8 detects entrapment, the image analysis device 10 changes the criteria for determining whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L.

[0081] The analysis domain setting unit 10e is a means for setting analysis domains A3 and A4 that are a predetermined distance away from the outside of the vehicle 2. For example, the analysis domain setting unit 10e sets an analysis domain A3 that is a relatively short distance from the outside of the side sliding doors 5R and 5L, and an analysis domain A4 that is a relatively long distance from the outside of the side sliding doors 5R and 5L. For example, the analysis domain setting unit 10e sets a narrow analysis domain A3 that is approximately 10 cm away from the outside of the vehicle 2 as shown in FIG. 10(A), and a wide analysis domain A4 that is approximately 100 cm away from the outside of the vehicle 2 as shown in FIG. 10(B). The analysis domain setting unit 10e includes an analysis domain generation unit that generates analysis domain data for the analysis domains A3 and A4 according to distances from the outside of the vehicle 2, etc., that are required to set the analysis domains A3 and A4, when the distances from the outside of the vehicle 2, etc., that are required to set the analysis domains A3 and A4 are input in advance, and a storage device that stores the generated analysis domain data in advance.

[0082] The determination unit 10f is a means for determining whether or not objects O1 and O2 exist within the analysis areas A3 and A4 based on a captured image of the outside of the vehicle 2 and the analysis areas A3 and A4. The determination unit 10f, for example, cuts out areas corresponding to the analysis areas A3 and A4 from image data captured by the image capture devices 9A and 9B, and performs processing such as trimming and masking unnecessary portions from the cut-out image data to determine whether or not objects O1 and O2 exist within the analysis areas A3 and A4. The determination unit 10f iteratively learns image data corresponding to the analysis areas A3 and A4 cut out from the image data output by the image capture devices 9A and 9B, finds patterns of objects O1 and O2 hidden in this image data, and applies the learned patterns to new image data to predict objects O1 and O2 according to these patterns. 7, the determination unit 10f learns using data called learning data, training data, or training data, and uses the learning results (trained model) to determine whether or not the objects O1, O2 are present in the analysis areas A3, A4 outside the side sliding doors 5R, 5L by deep learning. The determination unit 10f outputs the determination result as to whether or not the objects O1, O2 are present outside the side sliding doors 5R, 5L as a determination result signal to the entrapment monitoring device 11 via a communication network, wired, or wirelessly.

[0083] When the door entrapment detection device 8 detects entrapment, the determination unit 10f expands the analysis area A3 and determines whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L. When the door entrapment detection device 8 detects entrapment, the determination unit 10d changes the narrow analysis area A3 to a wide analysis area A4 and determines whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L. The entrapment detection unit 11a detects entrapment of objects O1, O2 present outside the side sliding doors 5R, 5L based on the determination result of the determination unit 10b using the changed analysis area A4 and the detection result of the door entrapment detection device 8.

[0084] Next, the operation of the pinch monitoring device according to the third embodiment of the present invention will be described. In S120 shown in Fig. 6, the image analyzer 10 changes the judgment criteria for image analysis. As shown in Fig. 10, when the door entrapment detection device 8 detects entrapment, the judgment unit 10d changes the judgment criteria from analysis area A3, which is close to the side sliding doors 5R, 5L, to analysis area A4, which is farther from the side sliding doors 5R, 5L. For example, as shown in Fig. 10(A), when the door entrapment detection device 8 does not detect entrapment, the judgment unit 10d changes the judgment criteria to analysis area A3, which is a narrow range approximately 10 cm away from the outside of the vehicle 2. On the other hand, as shown in Fig. 10(B), when the door entrapment detection device 8 detects entrapment, the judgment unit 10d changes the judgment criteria to analysis area A4, which is a wide range approximately 100 cm away from the outside of the vehicle 2.

[0085] In S130, the entrapment monitoring device 11 determines whether or not objects O1 and O2 are trapped outside the side sliding doors 5R and 5L. For example, as shown in FIG. 10(A), the determination unit 10b normally determines whether or not objects O1 and O2 are present in a narrow analysis area A3. As shown in FIG. 10(B), when the door entrapment detection device 8 detects entrapment, the determination unit 10b determines whether or not objects O1 and O2 are present in a wide analysis area A4. Therefore, when entrapment is detected, the determination unit 10f expands the narrow analysis area A3 to a wide analysis area A4 and makes a determination. For example, as shown in FIG. 10(A), when the door entrapment detection device 8 detects entrapment and the determination unit 10f determines whether or not objects O1 and O2 are present in the narrow analysis area A3, the determination unit 10f may only identify a white cane and fail to identify a person at a distance who is present outside the analysis area A3. In this embodiment, when door entrapment detection device 8 detects entrapment, analysis area A3 is expanded to analysis area A4, so that determination unit 10f can distinguish not only nearby people with white canes but also people at a distance. Therefore, in a dangerous situation where entrapment is detected, analysis area A3 is expanded so that determination unit 10f can determine whether or not not only object O2 close to side sliding doors 5R, 5L but also object O1 far from side sliding doors 5R, 5L are present outside side sliding doors 5R, 5L. As a result, entrapment monitor 11 can determine over a wide range whether or not objects O1, O2 present outside side sliding doors 5R, 5L are entrapped.

[0086] The pinch monitoring device according to the third embodiment of the present invention has the following advantages in addition to the advantages of the first embodiment. (1) In this third embodiment, the analysis area setting unit 10e sets analysis areas A3 and A4 that are a predetermined distance away from the outside of the vehicle 2, and the determination unit 10f determines whether or not objects O1 and O2 are present within the analysis areas A3 and A4 based on the captured image of the outside of the vehicle 2 and the analysis areas A3 and A4. Also, in this third embodiment, when the door entrapment detection device 8 detects entrapment, the determination unit 10f changes the analysis area A3 and determines whether or not objects O1 and O2 are present outside the side sliding doors 5R and 5L. Furthermore, in this third embodiment, the entrapment detection unit 11a detects entrapment of objects O1 and O2 present outside the side sliding doors 5R and 5L based on the determination result of the determination unit 10f using the changed analysis area A4 and the detection result of the door entrapment detection device 8. Therefore, when entrapment is detected, the analysis area A4 is changed, making it possible to determine over a wide range whether or not objects O1 and O2 are present outside the side sliding doors 5R and 5L.

[0087] (2) In this third embodiment, when the door entrapment detection device 8 detects entrapment, the determination unit 10f expands the analysis area A3 and determines whether or not objects O1, O2 are present outside the side sliding doors 5R, 5L. Therefore, it is possible to determine whether or not a dangerous state exists not only for objects O1, O2 present in the analysis area A3 close to the side sliding doors 5R, 5L, but also for objects O1, O2 present in the analysis area A4 far from the side sliding doors 5R, 5L. For example, when an object longer than the normal analysis area A3 is entrapped, the analysis area A3 is expanded to A4, preventing the omission of detection of objects far from the side sliding doors 5R, 5L, and enabling a highly accurate determination of whether or not entrapment occurs in a dangerous state.

[0088] (Other embodiments) The present invention is not limited to the above-described embodiment, and various modifications and alterations are possible as described below, and these are also within the scope of the present invention. (1) In this embodiment, an object O2 used by a person is caught between the side sliding doors 5R and 5L. However, the present invention can also be applied to a case where an object O1 such as a person's body, hand, foot, finger, or hair is caught between the side sliding doors 5R and 5L. In addition, in this embodiment, an example where the door is a door of a vehicle 2 is described. However, the present invention can also be applied to doors of other means of transportation such as automobiles, airplanes, and ships, and doors of moving objects such as elevators. Furthermore, in this embodiment, an example where the door is a double sliding door such as the side sliding doors 5R and 5L is described. However, the present invention can also be applied to one or more sliding doors that open and close in one direction, a swing door that can be opened and closed by rotating, a folding door that can be folded into multiple parts, a plug door that fits into the side entrance 4, or an airtight partition door that can maintain a pressure difference between the inside and outside of the room.

[0089] (2) In this embodiment, the camera 9B is disposed on each side of the car body 3 near the front and rear ends in the direction of travel. However, the present invention can also be applied to cases where the camera 9B is disposed on each side of the car body 3 near the front or rear ends in the direction of travel, or where the camera 9B is disposed at all side entrances 4 of the car 2. Furthermore, in this embodiment, the image analyzer 10 changes the image analysis criteria. However, the present invention can also be applied to cases where the image analyzer 10 does not change the image analysis criteria and the entrapment detector 11a detects the entrapment of objects O1, O2 outside the side sliding doors 5R, 5L. Furthermore, in this embodiment, the entrapment monitor 11 and the notification device 12 are disposed on the platform 1 side or the vehicle 2 side. However, the present invention can also be applied to cases where the entrapment monitor 11 and the notification device 12 are disposed on the platform 1 side or the vehicle 2 side. [Explanation of symbols]

[0090] 1 Platform (fixed) 1a Platform floor 1b Platform edge 2. Vehicles (moving objects) 3. Body 4 side entrance 5R, 5L side sliding door (door) 6R,6L Door edge rubber 7. Entrapment monitoring system 8 Door Entrapment Detection Device 9A, 9B Imaging device 10 Image analysis device 10a Accuracy calculation section 10b Judgment part 10c Analysis target setting section 10d Judgment section 10e Analysis area setting section 10f Judgment section 11 Entrapment monitoring device 11a Entrapment detection unit 11b Detection result transmission unit 12 Notification device A1 restricted area A2 unrestricted area L border Th1, Th2 thresholds AO1 and AO2 analysis targets A3,A4 analysis area O1,O2 object

Claims

1. A pinch monitoring device that monitors whether an object is pinched in a door that opens and closes between the inside and outside of a moving body, a pinch detection unit that detects whether an object is pinched in the door based on the detection result of a door pinch detection device that detects whether the object is pinched in the door and the analysis result of an image analysis device that analyzes a captured image of the outside of the moving body; A pinch monitoring device characterized by:

2. The pinch monitoring device according to claim 1, the pinch detection unit detects pinch of an object that exists in an area where access of the object is restricted while the moving body is moving; A pinch monitoring device characterized by:

3. The pinch monitoring device according to claim 1, a detection result transmission unit that transmits the detection result of the pinch detection unit; A pinch monitoring device characterized by:

4. The pinch monitoring device according to claim 3, the detection result transmission unit transmits the detection result of the entrapment detection unit to a notification device that notifies an attendant of the detection result of the entrapment detection unit; A pinch monitoring device characterized by:

5. The pinch monitoring device according to claim 1, the pinch detection unit detects pinching of an object present outside the door based on an analysis result of an image captured by the image capture device on the moving body side and / or the fixed body side; A pinch monitoring device characterized by:

6. The pinch monitoring device according to claim 1, the image analysis device changes a criterion for determining whether or not an object is present outside the door when the door entrapment detection device detects entrapment, the entrapment detection unit detects entrapment of an object present outside the door based on the analysis result of the image analysis device using the changed determination criterion and the detection result of the door entrapment detection device; A pinch monitoring device characterized by:

7. The pinch monitoring device according to claim 1, The image analysis device includes a determination unit that determines that an object exists outside the door when a probability of whether or not an object exists outside the door exceeds a threshold value, the determination unit, when the door entrapment detection device detects entrapment, changes the threshold value and determines whether or not an object is present outside the door; the entrapment detection unit detects entrapment of an object present outside the door based on a determination result of the determination unit using the changed threshold value and a detection result of the door entrapment detection device; A pinch monitoring device characterized by:

8. The pinch monitoring device according to claim 7, the determination unit, when the door entrapment detection device detects entrapment, lowers the threshold value and determines whether the object is present outside the door; A pinch monitoring device characterized by:

9. The pinch monitoring device according to claim 1, The image analysis device includes a determination unit that determines whether or not an object to be analyzed exists outside the door based on an object to be analyzed by the image analysis device and a captured image of the outside of the moving body, the determination unit, when the door entrapment detection device detects entrapment, changes the object to be analyzed and determines whether or not the object to be analyzed is present outside the door; the entrapment detection unit detects entrapment of an object present outside the door based on a determination result of the determination unit using the changed analysis target and a detection result of the door entrapment detection device; A pinch monitoring device characterized by:

10. The pinch monitoring device according to claim 9, the determination unit, when the door entrapment detection device detects entrapment, changes the analysis target from a person to the person and an object used by the person, and determines whether or not a person and an object used by the person are present outside the door; A pinch monitoring device characterized by:

11. The pinch monitoring device according to claim 1, the image analysis device includes a determination unit that determines whether the object exists within an analysis area based on an analysis area spaced a predetermined distance from the outside of the moving object and a captured image of the outside of the moving object; the determination unit, when the door entrapment detection device detects entrapment, changes the analysis area and determines whether the object is present outside the door; the entrapment detection unit detects entrapment of an object present outside the door based on the determination result of the determination unit using the changed analysis area and the detection result of the door entrapment detection device; A pinch monitoring device characterized by:

12. The pinch monitoring device according to claim 11, the determination unit, when the door entrapment detection device detects entrapment, expands the analysis area and determines whether the object is present outside the door; A pinch monitoring device characterized by:

Images