JP4595010B2 - escalator - Google Patents

Description

The present invention relates to an escalator over to give a sense of security to ensure the safety of passengers riding in the escalator.

  Conventionally, there has been an accident in which an escalator passenger is caught and injured by a corner portion between two intersecting escalators or a narrow angle portion between the escalator and the ceiling. This occurs when a passenger gets on the handrail for some reason, moves with the handrail as it is, and puts his head and body in the customer part.

  In order to prevent such an accident, a triangular acrylic protective plate is installed in the above-mentioned depression part, and the passenger hits this protective plate to make the passenger notice and avoid being caught in the narrow angle part. Is generally performed (for example, refer to Patent Document 1).

In addition, an infrared sensor or a millimeter wave sensor is attached along the handrail, and when a passenger gets on the handrail, it is detected and a warning broadcast is made (for example, see Patent Document 2). .)
Japanese Patent Laid-Open No. 7-309574 Japanese Patent Laid-Open No. 10-297865

  In the case of the said patent document 1, even if a passenger collides with an acrylic protection board when the speed of an escalator is high, there exists a danger that it will be pinched | interposed into a narrow-angle part as it is. In addition, there is a problem in that it is impossible to alert and escalator control until a collision occurs, causing passengers to feel painful.

  In the case of the above-mentioned Patent Document 2, it is possible to perform alerting and escalator control before colliding with the protective plate, but for the first time, for example, when a passenger is about to hold the upper body from the handrail, There was a problem that detection delay occurred. Moreover, since it is necessary to install a sensor along a handrail, there existed a problem that introduction cost and a maintenance cost started.

An object of the present invention is to provide a escalator over which to perform the necessary caution as soon as possible detect passengers abnormal state.

To achieve the above object, in claim 1 of the present invention, Oite the installation has been that escalator over protection plate侠角portion between the escalator and between the ceiling or two escalators crossing, A camera is provided on the protective plate, and image processing means for preliminarily detecting a passenger's abnormal boarding from the video of the camera, and when the abnormal boarding of the passenger is determined by the image processing means , The escalator is stopped or slowly stopped , and the image processing means detects a swing of the protective plate by a shake of the video of the camera, and includes a protection device that stops or slowly stops the escalator .

According to this configuration, prior to collision detection, the boarding state of passengers approaching from far away from the camera is monitored, and when a state in which the passenger is about to go on the handrail is detected, a warning broadcast is sent to the passengers. Recognize that the escalator intersection is approaching and prevent accidents. Moreover, the speed of an escalator can be made slow, or it can stop depending on the case, and a passenger can be evacuated by the time it contacts a protection board. In addition, when the passenger leans out immediately before the protective plate or moves the load and collides with the camera, the escalator is immediately stopped, so that damage to the passenger and the load can be reduced.

Further, in claim 2, when there is a preliminary detection of the passenger's abnormal boarding, the image processing means according to claim 1 has a sensitivity of shaking of the image of the camera as a preliminary detection of the passenger's abnormal boarding. It is characterized in that it is larger than when there is no .

According to this configuration, if preliminary detection is not performed, it is considered that the possibility of a passenger collision is low, and the detection sensitivity of collision detection is lowered to reduce false detection due to wind, earthquake, mischief or inadvertent contact .

Further, in claim 3, in claim 1, an unmanned background moving image that is not used by passengers used when performing person detection or motion detection in image processing is triggered by a signal at the time of starting or stopping an escalator. It is characterized by being photographed and used .

According to this configuration, it is possible to easily set an unattended background moving image in which no passenger is on board.

  According to the present invention, prior to the collision detection, the boarding state of passengers approaching from a distance from the camera is monitored, and when a state in which the passengers are about to go on the handrail is detected, a warning broadcast is sent to the passengers. Recognize that the escalator intersection is approaching and prevent accidents. Moreover, the speed of an escalator can be made slow or stopped depending on the case, and a passenger can be evacuated by the time it contacts a protection board.

Hereinafter will be described with reference to FIG embodiments of the escalator over according to the present invention.

Figure 1 is an overall configuration diagram showing an escalator over, according to an embodiment of the present invention, FIG. 2 is a functional block diagram of the image processing, FIG. 3 is a flowchart of the image processing, FIG. 4 is a passenger does not venture out them from the handrail FIG. 5 is a flow chart for learning the background state for detecting abnormal boarding, FIG. 6 is a flow chart for dangerous preliminary detection, and FIG. It is a flowchart when contacting.

  In FIG. 1, 1 is an escalator step, 2 is an escalator handrail, and 3 is an upper floor or ceiling. 4 is a movable protective plate for preventing collision of passengers, which is provided via a movable protective plate jig 5 at a corner of the upper floor or the ceiling 3 and the handrail 2. It is structured to shake when it collides. Reference numeral 6 denotes a video monitoring camera that monitors the riding state of the passenger 10 on the step. Reference numeral 7 is a warning broadcast speaker, which alerts the passenger when the passenger is riding abnormally or dangerously and makes the passenger aware of the danger. Reference numeral 8 denotes an image processing apparatus provided on the floor surface or ceiling 3, which processes an image photographed by the camera 6 to detect abnormal boarding or dangerous boarding. Reference numeral 9 denotes a video recording device which always records or records video only before and after the abnormal event using the abnormal event detected by the image processing device 8 as a trigger signal.

  The trigger signal output from the image processing device 8 is connected to an escalator control device (not shown), and can stop or slowly stop the escalator when an abnormal event occurs. The trigger signal output from the image processing apparatus 8 is sent as an alarm signal to the monitoring room via a wired or wireless network, and is used to increase the monitoring efficiency of the monitor. Also, such trigger signals and alarm signals are embedded in video data, or added to video data as separate files as metadata, and used when searching for video later offline for accident verification. It is done.

  Next, the internal functions of the image processing apparatus 8 will be described using the functional block diagram of FIG. In FIG. 2, first, an image is input to the image input unit 101 from the camera 6 attached to the movable protection plate 4 of FIG. 1. The input image is stored in the image memory 102 for a certain number of frames, and is sent out when an abnormal event is detected. Next, in the input image, the unmanned state determination unit 103 recognizes the presence or absence of a passenger in the camera view. Next, the preliminary detection unit 104 preliminarily detects an abnormal passenger ride before the passenger collides with the movable protection plate 4. Here, it is identified whether or not the passenger 10 is on the handrail 2. Further, the preliminary detection unit 104 detects abnormal boarding events such as reverse running, stepping through, and falling on the step in addition to the handrail 2 being launched, and alerts the speaker 7. Processing area settings and parameter settings for image processing performed by the preliminary detection unit 104 are input from the parameter input device 105. Alternatively, it is sent from the outside to the image processing apparatus 8 via the network. The collision detection unit 107 detects by image recognition that a passenger collided with the collision prevention protection plate 4 and the camera image was shaken. When the preliminary detection unit 104 and the collision detection unit 107 detect an abnormal boarding or a collision with a passenger's protective plate, a detection signal is sent to the image transmission unit 106 and stored in the image memory 102. The video before and after the abnormal event is appropriately encoded by the image sending unit 109 and sent to the video recording device 9. Furthermore, when the preliminary detection unit 104 and the collision detection unit 107 detect abnormal boarding or a collision with a passenger's protective plate, an alarm signal of an appropriate format is generated in the alarm signal transmission unit 108, for example, the speaker 7 Is sent to the escalator control section, and is sent to the escalator control unit to control the operation of the escalator, such as stopping or slow stopping.

  Next, the flow of processing of the image processing apparatus 8 will be described using the flowchart of FIG.

  The input image (S1) is determined whether or not there is a passenger in the camera field of view (S2). If it is determined that the passenger is unattended, the background image and the background feature amount are updated (S8). finish. Unattended state detection and background update will be described later with reference to FIG. On the other hand, when it is determined that there is a passenger, the image at that time is recorded (S3), and it is identified whether it is in a dangerous state or not (S4). In S4, the dangerous state is preliminarily detected before the passenger comes into contact with the collision protection plate 4. Based on the detection result of the preliminary detection, the preliminary detection state image is transmitted (S6) and the alarm signal is transmitted (S7). I do. In the contact detection presence / absence determination in S5, determination processing is performed regardless of the preliminary detection result in S4. If an abnormal boarding is preliminarily detected in S4, the detection sensitivity of contact detection is increased in S5 to reduce detection omissions. On the other hand, if the preliminary detection is not performed in S4, the detection sensitivity in S5 is lowered to reduce false detection caused by intentionally shaking the anti-collision protection plate such as wind, earthquake, or mischief. Details of the dangerous preliminary detection (S4) will be described later with reference to FIG.

  FIG. 4 is an explanatory view for explaining a normal time when the passenger does not get out of the handrail and an abnormal time when the passenger gets out of the handrail, and is a view when the escalator is viewed from the anti-collision protection plate 4.

  In FIG. 4, the image processing area 12 including step 1, the handrail 2, and the outer deck 11 is set from another PC via the parameter input device 105 or network of FIG. When the passenger 10 is present in the image processing area 12, the dangerous boarding determination module (S4 in FIG. 3) outputs a preliminary detection signal for abnormal boarding.

  Next, the flow of processing of unattended state detection (S2) and unattended state learning (S8) will be described using the flowchart of FIG.

  First, it is determined from the input image (S10) whether or not a person exists in the monitoring area (S11). Conventionally, a background difference process or an inter-frame difference process is basically used as a method for determining the presence or absence of a person in a monitoring area. Since these methods are processed on the premise that the background is stationary, they cannot be applied to a background that is constantly moving like an escalator step. There is a method that uses a human detector to detect a person from a single image without using background or motion, but currently it is vulnerable to changes in the appearance of the person silhouette and partial concealment of the person, and the processing cost is also high. Because it is expensive, it is not practical at present. Therefore, for example, using a spatio-temporal feature amount previously invented by the same applicant (see Japanese Patent Application No. 2007-296206), a dynamic background without a person is learned using both texture and motion information.

  For learning, for example, a mixed normal distribution model is used, and the distribution of the spatio-temporal feature amount in the background state is expressed by a plurality of normal distribution parameters. If the spatiotemporal feature in the input data is far from the learned distribution, it can be determined that a person is in the dynamic background. If a person is detected, the image input (S11) is repeated, and at the same time, the process proceeds to the risk preliminary detection process (S4 in FIG. 3). If no person is detected, N images are stored in the memory for learning the background state (S12). In the plurality of stored images, the spatiotemporal features in the entire image or in the monitoring region set in S104 of FIG. 2 are calculated (S13), and the feature amount distribution of the dynamic background is approximated by parameters (S14). These processes are not only performed in the entire image and in the monitoring area, but can also be calculated for each block by dividing the image and the monitoring area into a plurality of local blocks.

  In this case, the presence determination of the person is performed for each divided block. Alternatively, the presence / absence determination of a person is performed by majority decision of detection results of a plurality of blocks. By the way, a scene in which no person exists can be obtained from an escalator start / stop signal. In other words, there is no shopper in the distribution store immediately after starting the escalator before opening the store and immediately before stopping the escalator after closing the store. It is also possible to forcibly calculate and update the background feature.

  Next, the process flow of the preliminary risk detection (S4 in FIG. 3) will be described using the flowchart in FIG.

  First, a parameter input device (S105 in FIG. 2) or an area (12 in FIG. 4) for image processing is designated from the outside via a network (S20). Next, in the input image (S21), the spatiotemporal feature is calculated for each block or in the entire image processing area (S22). Next, the background spatio-temporal features obtained by unattended state learning (S2 in FIG. 3) are referred to in advance (S24), and compared with the spatio-temporal features (S22) in the image processing area (12 in FIG. 4). Thus, it is determined whether or not a person exists in the image processing area (12 in FIG. 4) for the entire image or for each local block (S23). When a local block is used for determination, it is determined that there is a person if there are more than a threshold number of blocks for each local block, that is, it is determined that a person protrudes from the handrail area as shown in FIG. S26).

  At this time, since there is a possibility that the protruding item is a baggage carried by the passenger or a shadow of the passenger is reflected, an abnormal operation detection determination is performed in the next step (S26). For abnormal operation detection, temporal changes in the spatio-temporal feature amount are analyzed, and abnormal boarding and normal boarding are identified. When the movement and texture have characteristics different from those of a person such as luggage and shadow, it is possible to identify the luggage and shadow as not being a person by the time-series change of the spatio-temporal characteristics. If an abnormal operation is detected, a warning broadcast is performed (S27), and a preliminary detection signal for setting the sensitivity of collision detection described later is transmitted (S28). As a type of operation that can be preliminarily detected, in addition to entering the handrail, it is possible to detect dangerous actions such as reverse running and running through, and abnormal events such as falls.

  Next, the flow of image processing for detecting that a passenger has collided with the collision prevention protection plate 4 will be described using the flowchart of FIG.

  If the difference between frames or the amount of motion is equal to or greater than a certain threshold by inter-frame difference processing or optical flow calculation processing of the input image (S30), it is determined that the camera is vibrating for some reason (S31). ). Next, depending on the presence or absence of the preliminary detection signal (S28) obtained by the flowchart processing of FIG. 6, the passenger sets the sensitivity of collision detection to the collision prevention protection plate (S33). That is, if it is determined that there is preliminary detection (S32), the sensitivity of vibration detection is increased to reduce detection omissions so as not to cause personal injury. Conversely, if preliminary detection is not performed, leave the detection sensitivity at the default value or lower the detection sensitivity so that camera vibration due to natural phenomena such as wind and earthquakes or intentional camera vibration due to mischief is not detected. Like that. If it is determined that a person has collided with the collision prevention protection plate under the above settings (S34), the escalator is stopped or slowly stopped in the next step (S35). Thereafter, an alarm signal for which a collision has been determined is subsequently sent (S36). For example, an alarm signal is sent to the monitoring center so that the monitor can directly observe the monitor, and the escalator can be forcibly stopped manually. At this time, a plurality of frames before and after the contact are transmitted along with the alarm signal transmission (S37), and the recording device 9 records it, so that it is possible to search the accident video later by offline processing and investigate the cause of the accident. it can.

Is an overall configuration diagram showing an escalator over, according to an embodiment of the present invention. It is a functional block diagram of image processing. It is a flowchart of an image process. It is explanatory drawing explaining the normal time when a passenger does not lean out from a handrail, and the time of abnormality which leaned out. It is a flowchart for learning the background state for abnormal boarding detection. FIG. 6 is a flowchart of dangerous preliminary detection. FIG. 7 is a flowchart when the passenger contacts the protective plate.

Explanation of symbols

1 Step 2 Handrail 3 Ceiling 4 Movable protective plate 5 Mounting jig 6 Camera 7 Speaker 8 Image processing device 9 Video recording device 10 Passenger 101 Image input unit 102 Image memory 103 Unmanned state determination unit 104 Preliminary detection unit 105 Parameter input device 106 Image transmission control unit 107 Collision detection unit 108 Alarm signal transmission unit 109 Image transmission unit

Claims (3)

Oite escalator and between the ceiling or installed by being escalator over protection plate侠角portion between the two escalators crossing,
A camera is provided on the protective plate, and image processing means for preliminarily detecting a passenger's abnormal boarding from the video of the camera, and when the abnormal boarding of the passenger is determined by the image processing means , An escalator comprising a protection device that stops or slowly stops operation of the escalator, and detects the shaking of the protection plate by the image processing means by the image processing means to stop or slowly stop the escalator. - . The image processing means is characterized in that when there is a preliminary detection of the passenger's abnormal boarding, the sensitivity of the camera image shake is larger than when there is no preliminary detection of the passenger's abnormal boarding. claim 1 escalator over description. An unmanned background moving image that is not used by passengers used when performing human detection or motion detection in image processing is captured and used as a trigger when an escalator is started or stopped. according to claim 1 or 2 of the escalator over.

Images