JP2007280407A - Traffic terminal device and accident detection system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a traffic terminal device and an accident detection system having high detection accuracy. <P>SOLUTION: The accident detection system is constructed of: terminal devices 100a and 100b installed in a plurality of places; and a control center managing the terminal devices 100a and 100b. When a shock sound is inputted via microphones 106a and 106b in the terminal devices 100a and 100b, an accident determination parts 120a and 120b with a learning function determine whether the shock sound is caused by an accident. The shock sound is sent to an accident determination part 41 with a learning function in the control center. The accident determination part 41 with a learning function on the control center side has a neural network, and carries out learning based on the sent sound data. The learning result is sent to the accident determination parts 120a and 120b with a learning function in the terminal devices 100a and 100b. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a traffic terminal device and an accident detection system, and more particularly to a traffic terminal device that detects an accident by inputting voice generated near a road and determining whether the voice is generated by an accident. It relates to an accident detection system.

  2. Description of the Related Art Conventionally, a traffic accident automatic recording device is known for efficiently collecting data immediately before and after the occurrence of a traffic accident, and elucidating the occurrence mechanism of the accident based on the data. The following documents are known as techniques relating to such an apparatus.

  In Japanese Patent Laid-Open No. 4-338900, a traffic accident situation recording device that determines the presence or absence of an accident by detecting an impact sound and records video and sound before and after the occurrence of the accident together with time, location, and traffic signal information. Is disclosed.

  Japanese Patent Application Laid-Open No. 8-263785 discloses an apparatus for determining whether or not an accident has occurred by determining sound pressure and recording video and sound before and after the accident. This document discloses that data is transmitted to a control center, a single line of sound and video is used, and a warning is issued when the remaining amount of video tape is low.

  Japanese Patent Application Laid-Open No. 10-142022 discloses an accident sound detection circuit that obtains a power spectrum of acoustic data and determines whether or not an accident has occurred according to the pattern. This is a revision of the conventional accident determination algorithm based only on sound pressure.

  However, when an accident is detected by the above-described algorithm, it is pointed out that there are many cases in which an event other than an accident is determined as an accident, and there is a problem in the accident detection accuracy (for example, “by the traffic accident automatic recording device (TAAMS) "Study on the mechanism of the occurrence of traffic accidents"; Science Traffic Research Institute Report Traffic Vol.38, No.2, p.12-29, July 1997).

  The cause of the above-mentioned problem is that the accident determination method is based on a comparison between a pattern such as a measured acoustic power spectrum or autocorrelation and a model pattern of the accident occurrence. That is, sufficient detection accuracy cannot be obtained only by determination based on a linear pattern comparison as in the prior art.

  Therefore, an object of the present invention is to provide a traffic terminal device and an accident detection system with high detection accuracy.

  In order to achieve the above object, according to one aspect of the present invention, the traffic terminal device is determined to be an impact sound based on the sound pressure of the input means for inputting the sound data and the input sound data. A frequency distribution extracting unit that extracts a frequency distribution of sound; a determination unit that determines a generation factor of input audio data based on the extracted frequency distribution; a video input unit that inputs video data; An endless recording unit that records an input video for a predetermined time endlessly and a storage unit that accumulates video data. The endless recording unit determines whether the generation factor determined by the determination unit is a vehicle collision sound and a sudden In the case of at least one of the brake sounds, video data before and after the occurrence of at least one of the vehicle collision sound and the sudden brake sound is recorded in the storage means. And, judging means comprises a neural network, the neural network determines the cause of the audio data corresponding to the extracted frequency distribution.

  Preferably, the traffic terminal device further includes signal status acquisition means for acquiring the status of the signal display displayed by the traffic light, the endless recording means further records the signal display data endlessly, and the storage means further includes the signal display status. The endless recording means stores the indication data in the storage means when the cause determined by the determination means is at least one of a vehicle collision sound and a sudden brake sound. Signal display data before and after the occurrence of at least one of the brake sounds is recorded.

  Preferably, the traffic terminal device further includes character creation means for converting the acquired signal display status into character data, and video composition means for synthesizing the input video data and the created character data, and an endless recording means. Records the synthesized video.

  Preferably, the endless recording means records the input voice data endlessly, and accumulates when the generation factor determined by the determination means is at least one of a collision sound and a sudden brake sound of the vehicle. The means records voice data before and after the occurrence of at least one of the collision sound and sudden braking sound of the vehicle.

  Preferably, the traffic terminal device can rewrite software constituting the neural network or parameters for controlling the neural network from the outside.

  According to another aspect of the present invention, an accident detection system includes any of the traffic terminal devices described above and a control center device connected to the traffic terminal device, and the traffic terminal device is determined by a determination unit. When the generated cause is at least one of a collision sound and a sudden braking sound of the vehicle, images before and after the occurrence of at least one of the collision sound and the sudden braking sound of the vehicle in the control center device Send data.

  Preferably, in the accident detection system, software constituting the neural network or parameters for controlling the neural network can be rewritten from the device of the control center.

  According to these inventions, since it is determined whether or not an accident has occurred based on learning, it is possible to provide a traffic terminal device and an accident detection system with high detection accuracy.

<First Embodiment>
FIG. 1 is a perspective view showing the configuration of an accident detection system according to the first embodiment of the present invention. Referring to the drawing, the accident detection system includes a plurality of terminal devices 100a and 100b provided at different locations (intersections different from each other in this embodiment), and control center 4 connected thereto. .

  Signal devices 102a and 102b are connected to the terminal devices 100a and 100b, respectively. The terminal devices 100a and 100b include cameras 104a and 104b and microphones 106a and 106b for inputting video and audio near the intersection, respectively.

  The control center 4 includes a communication line modem 43, a control device 42, and an accident determination unit 41 with a learning function including a neural network.

  In the accident detection system according to the present embodiment, the data of accidents occurring at various points (various intersections) and various times are centrally managed by the control center 4, and the accident detection system is learned while making the system intelligent. Increasing the sophistication improves accident determination accuracy. The learning system employed in the present embodiment learns an event that has occurred at an intersection from a linear or non-linear association, and determines whether an accident has occurred. In this embodiment, a neural network is used as a learning system. However, a system using chaos or fuzzy can be used as long as the system has a learning function.

  FIG. 2 is a block diagram for explaining the details of the terminal devices 100a and 100b. Referring to the figure, signal devices 102a and 102b are connected to each terminal device, and each terminal device 100a and 100b has cameras 104a and 104b, microphones 106a and 106b, character creation units 108a and 108b, and composition unit 110a. 110b, endless recording / playback units 112a and 112b, data storage units 114a and 114b, sound pressure determination units 116a and 116b, data processing units 118a and 118b, and an accident determination with a learning function including a neural network (NN) Parts 120a and 120b.

  The accident determination unit with learning function 120a, 120b receives the learning result from the accident determination unit with learning function 41 of the control center. Further, the accident determination units with learning function 120a and 120b determine whether or not the input voice is a voice caused by an accident (a specific cause), and the endless recording / reproducing units 112a and 112b with the determination result as a determination signal. Send to. When it is determined that an accident has occurred, the data storage units 114a and 114b read and record data from the endless recording / reproducing units 112a and 112b.

  The character creation units 108a and 108b convert the data of the time and place at the intersection and the situation (signal display) of the traffic lights 102a and 102b at that time into character data and output them.

  The synthesizing units 110a and 110b synthesize the video from the cameras 104a and 104b and the output results of the character creation units 108a and 108b corresponding to the video on the video.

  The endless recording / playback units 112a and 112b collect voices (traffic sounds) from the microphones 106a and 106b, and record them endlessly while updating the latest data for a certain period of time. Similarly, the endless recording / reproducing units 112a and 112b record the video output from the synthesizing units 110a and 110b constantly and endlessly while updating data for a certain past time.

  The sound pressure generators 116a and 116b determine whether or not the sound data (input sound data) from the microphones 106a and 106b is an impact sound, and when it is determined that the sound is an impact sound, select the impact sound data. The data is output to the data processing units 118a and 118b.

  The data processing units 118a and 118b process the impact sound data output from the sound pressure determination units 116a and 116b into a frequency distribution of sound signals, time-series sound pressure data, and the like, and send them to the accident determination units 120a and 120b with a learning function. Output. The data processing units 118a and 118b transmit the processed data to the accident determination unit 41 with a learning function of the control center 4 through a line.

  The accident determination unit 41 with a learning function in the control center 4 performs learning based on the transmitted processing data. The accident determination units with learning function 120a and 120b included in the terminals 100a and 100b input the learning result of the accident determination unit with learning function 41 in the control center 4 using a line. Thereby, a learning result is reflected in each terminal 100a, 100b. Specifically, the software of the accident determination unit with learning function 41 in the control center 4 is copied to each terminal device 100a, 100b at regular intervals.

  The accident determination units 120a and 120b with learning function of the terminal devices 100a and 100b input data from the data processing units 118a and 118b and determine the cause of the impact sound. Based on the determination result, the accident determination unit with learning function 120a, 120b gives a determination signal to the endless recording / reproducing units 112a, 112b.

  Based on the determination signal, the endless recording / playback units 112a and 112b transmit video data and audio data before and after the impact sound is generated to the data storage units 114a and 114b.

  FIG. 3 is a flowchart showing a learning procedure of the accident detection system. Hereinafter, processing performed in each step will be described.

[Step S1]
First, data of impact sounds at intersections prerecorded in the accident detection system are sequentially input. Examples of the input data include a waveform of frequency distribution data obtained by frequency analysis of acoustic data. The accident detection system learns when a human gives a teacher signal as to what causes the collision sound. For example, a human inputs a signal such as the input impact sound data is a vehicle collision sound, a horn sound, a siren sound, or a sudden brake sound.

[Step S3]
In step S1, a system that has learned a certain amount of data is introduced into the field.

[Steps S5 and S7]
The Nth impact sound data sensed by the accident detection system introduced in the actual field is determined by the learning function-equipped accident determination units 120a and 120b for any cause. For example, the accident determination unit with learning function 120a, 120b determines whether the sound is a collision sound of a vehicle, a horn sound, a siren sound, or a sudden brake sound.

[Step S9]
The correspondence between the impact sound data and the determination result is automatically learned by the learning function of the accident determining unit 41 with the learning function of the control center 4.

[Step S11]
In the (N + 1) th impact sound data inputted thereafter, it is determined by the accident detection system after learning the Nth data which factor the impact sound is. The correspondence between the impact sound data and the determination result is automatically learned by the learning function of the accident detection system.

  As described above, in the present embodiment, each impact sound data is determined by the accident determination unit with a learning function, and at the same time, the correspondence of the determination result of the data is learned to make the accident determination unit intelligent. Thus, the accident determination accuracy can be improved.

  As described above, the learning function-equipped accident determination units 120a and 120b in the terminal devices 100a and 100b determine what the impact sound is. Moreover, the accident determination part 41 with a learning function of the control center 4 learns an impact sound. When a certain period of time elapses, the software of the accident determination unit with learning function 41 of the control center 4 is copied to the accident determination units with learning function 120a and 120b in the terminal devices 100a and 100b.

FIG. 4 is a flowchart showing the copying operation.
Referring to the figure, in step S101, acoustic data is collected by the terminal devices 100a and 100b. In step S103, the impact sound is detected by the learning function-equipped accident determination units 120a and 120b on the terminal devices 100a and 100b side. It is judged whether it is due to. The impact sound data is transferred to the control center in step S105, and the accident determination unit with learning function 41 of the control center 4 learns acoustic data in step S107.

The operations of steps S101 to S107 are continued for a certain period and repeated.
If the predetermined time has elapsed, in step S109, the accident determination software in the accident determination unit 41 with learning function of the control center 4 is copied to the accident determination units 120a and 120b with learning function of the terminal devices 100a and 100b. .

  What is copied is the accident detection software itself. In addition, for example, a method of copying only the numerical values of parameters used for controlling the software is also conceivable. Thereby, the reference | standard for determination of an impact sound is changed.

  Furthermore, after the next fixed period has passed (S111 to S117), the accident determination software is copied from the control center side to each terminal device (S119).

  Such an operation is repeated every certain period. The purpose of copying the software is to make the judgment of events to be recorded in each terminal (selection of video and audio data). Specifically, this determination is made in accordance with a standard such that recording is performed when it is determined that a vehicle collision sound or a sudden braking sound has occurred, and recording is not performed for a horn sound or siren sound.

  When the accident determination unit with a learning function learned by such a procedure performs sufficient learning and it is determined that further accuracy improvement is not necessary, copying from the control center 4 to each terminal device 100a, 100b is canceled. You may make it do. Note that it is conceivable to determine whether or not the accuracy needs to be improved in a situation where the fluctuation value of the parameter value used by the accident determination software is equal to or less than a predetermined reference value.

  As described above, in the accident detection system according to the present embodiment, learning in determination of acoustic data is performed by the accident determination unit with a learning function, so that a system with high accident detection accuracy can be provided.

<Second Embodiment>
In the first embodiment described above, a plurality of terminal devices are prepared, and the acoustic data obtained from the plurality of terminal devices is centrally managed and learned at the control center 4, but the second embodiment In the embodiment, the terminal device operates alone and collects and learns acoustic data independently.

  FIG. 5 is a block diagram illustrating a configuration of the terminal device according to the second embodiment. Referring to the figure, as in the first embodiment, the terminal device includes a traffic light 102, a character creation unit 108, a camera 104, a synthesis unit 110, an endless recording / playback unit 112, a data storage unit 114, and a microphone. 106, a sound pressure determination unit 116, a data processing unit 118, and an accident determination unit 120 with a learning function.

  In the present embodiment, the accident determination unit 120 with a learning function not only determines an accident as in the first embodiment, but also learns based on the collected acoustic data.

Also in this embodiment, it is possible to detect an accident with high accuracy by learning.
<Third Embodiment>
The accident detection system according to the third embodiment is characterized in that accident determination and learning are not performed at the terminal device side but at the control center side. If such a method is adopted, it is necessary to transfer a large amount of data to the control center instantly. Such a system is feasible if the capacity and speed of data transmission are increased along with the advancement of data communication technology in the future, and humans will not have to take out accident data from the data storage unit of the terminal device. It is considered that the video before and after the accident can be confirmed at the control center immediately after the accident.

  FIG. 6 is a block diagram showing a configuration of an accident detection system according to the third embodiment of the present invention. On the terminal device 100a, 100b side, traffic lights 102a, 102b, cameras 104a, 104b, microphones 106a, 106b, character creation units 108a, 108b, synthesis units 110a, 110b, endless recording / playback units 112a, 112b, The point that the sound pressure determination units 116a and 116b are provided is the same as that in the first embodiment. However, the present embodiment is different in that the data storage units 150a and 150b and the acoustic data processing units 150a and 150b are provided on the control center side.

  The operation of the accident detection system in the third embodiment will be described below. First, the acoustic data collected in advance is input to the accident determination unit 41 with a learning function of the control center, and the correspondence between the acoustic data and the generation factor is learned. The specific procedure for learning is as described above.

  When the acoustic data collected by the terminal device is determined to be due to the impact sound by the sound pressure determination units 116a and 116b, the determination signal is sent to the endless recording / playback units 112a and 112b. Then, together with the acoustic data before and after the occurrence of the impact sound, video data in which the information on the point, time, and traffic lights 102a and 102b is synthesized is sequentially transmitted to the control center.

  Of the transmitted data, the acoustic data is processed by the acoustic data processing unit 152a. The data format after processing is the same as in the first embodiment. The processed acoustic data is input to the accident determining unit 41 with a learning function. And the accident determination part 41 with a learning function determines what kind of factor caused the impact sound. The data is selected according to the determination result, and for those determined to be stored, the accident determination unit with learning function 41 outputs a storage command to the data storage units 150a and 150b. Based on this command, the audio data and the video data are collectively stored in the data storage units 150a and 150b. The criteria for selection are as described above.

  Furthermore, the accident determination unit with learning function 41 automatically learns the correspondence between the acoustic data and the determination result. The learning method is also as described above.

  In the present embodiment, determination and learning about an accident can be performed at a control center in a concentrated manner. Thereby, the hardware configuration of the terminal device can be simplified.

  As described above, in the present invention, a learning function is provided in the accident detection system. The determination is performed not only by the simple pattern recognition but also by taking into account the non-linear association (for example, determination by a neural network).

  Further, in the first embodiment and the third embodiment, the acoustic data collected from a plurality of points is concentrated and managed in the control center, so that the learning effect can be enhanced.

  The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

It is a perspective view which shows the structure of the accident detection system in the 1st Embodiment of this invention. It is a block diagram of an accident detection system. It is a flowchart which shows the learning procedure of an accident detection system. It is a flowchart for demonstrating the procedure of a software copy from a control center to a terminal device. It is a block diagram which shows the structure of the terminal device in the 2nd Embodiment of this invention. It is a block diagram which shows the structure of the accident detection system in the 3rd Embodiment of this invention.

Explanation of symbols

  41 accident determination unit with learning function, 100 terminal device, 102 traffic light, 104 camera, 106 microphone, 112 endless recording / playback unit, 114 data storage unit, 116 sound pressure determination unit, 118 data processing unit, 120 accident determination unit with learning function .

Claims (7)

Input means for inputting voice data;
A frequency distribution extracting means for extracting a frequency distribution of a sound determined to be an impact sound based on the sound pressure of the input voice data;
Determination means for determining a generation factor of the input voice data based on the extracted frequency distribution;
Video input means for inputting video data;
Endless recording means for recording the input video for a predetermined time in the past in an endless manner;
Storage means for storing video data,
The endless recording means, when the generation factor determined by the determination means is at least one of a vehicle collision sound and a sudden brake sound, stores the vehicle collision sound and sudden brake sound in the storage means. Record the video data before and after the occurrence of at least one of
The determination means includes a neural network,
The said neural network is a traffic terminal device which determines the generation | occurrence | production factor of the said audio | voice data according to the extracted frequency distribution. It further comprises signal status acquisition means for acquiring the status of signal display displayed by the traffic light,
The endless recording means further records the signal display data endlessly,
The storage means further stores the signal display data,
The endless recording means, when the generation factor determined by the determination means is at least one of a vehicle collision sound and a sudden brake sound, stores the vehicle collision sound and sudden brake sound in the storage means. The traffic terminal device according to claim 1, wherein signal display data before and after occurrence of at least one of the signals is recorded. Character creation means for converting the acquired signal display status into character data;
Video synthesis means for synthesizing the input video data and the created character data;
The traffic terminal apparatus according to claim 2, wherein the endless recording unit records the synthesized video.   The endless recording unit records the input voice data together in an endless manner, and when the generation factor determined by the determination unit is at least one of a vehicle collision sound and a sudden brake sound, The traffic terminal apparatus according to any one of claims 1 to 3, wherein voice data before and after at least one of a collision sound and a sudden braking sound of the vehicle is recorded in the storage means.   The traffic terminal device according to any one of claims 1 to 4, wherein software constituting the neural network or parameters for controlling the neural network can be rewritten from the outside. The traffic terminal device according to any one of claims 1 to 5,
A control center device connected to the traffic terminal device,
When the generation factor determined by the determination means is at least one of a vehicle collision sound and a sudden brake sound, the traffic terminal device transmits the vehicle collision sound and a sudden sound to the control center device. An accident detection system that transmits video data before and after the occurrence of at least one of the brake sounds.   The accident detection system according to claim 6, wherein software constituting the neural network or parameters for controlling the neural network can be rewritten from an apparatus of the control center.

Images