JPH03132467A - Moving object identifying device - Google Patents

Abstract

PURPOSE:To identify a moving object easily and certainly by detecting the image or sound of/from the object which is in transient through a specified place, analyzing a plurality of characteristic items, and selecting the classification items for the object. CONSTITUTION:When a moving object previously furnished with a plurality of classification items such as type, category, and identification number is going to pass a certain place, this point of passage is detected by a means 1, while another means 2 detects the image of the passing object or sound generated therefrom. Further, a means 3 analyzes a plurality of previously specified characteristic items on the basis of the detected image or sound. On the basis of the result therefrom a neural network means 4 specifies the classification items to which the moving object having passed the specified place belongs. A means 5 outputs the identification information on the specified classification items. Identification of the moving object is thereby performed easily and certainly.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Field of Industrial Application) The present invention relates to a mobile object identification device for identifying moving objects such as vehicles in various transportation systems.

(Prior art) In transportation systems such as railways, in order for a central control center to manage the operation of each train, it checks the operation number and other various information for each vehicle passing through a predetermined location. There is a need.

As a method for carrying out this confirmation, firstly, there is a system in which a transmitting device such as an oscillator or a resonator is installed in each vehicle, and a receiving device is installed at a predetermined location. According to this system, the receiving device can determine not only the type of vehicle but also the vehicle identification number based on the data of the transmission signal from the transmitting device.

In addition, as a second measure, there is a system that identifies only the type of train (for example, whether it is a limited express train or a local train) by installing a weight sensor at a specific location and measuring the weight of the passing train. . However, this system is limited to cases where it is sufficient to identify only the type of train.

As a third measure, a camera or microphone is installed at a predetermined location, and the vehicle identification number captured by the camera is
Alternatively, there is a system that obtains the necessary information by processing the sound etc. captured by a microphone based on a certain algorithm.

(Problem to be Solved by the Invention) However, in the case of the first method, the cost increases because a transmitting device must be installed in each vehicle, and the operational burden associated with transmitting and receiving increases. becomes. In particular, when managing sections that are operated by private railways affiliated with various companies, it is not a big problem if the transmitting device is installed on the company's own vehicles, but it is not necessary to force this installation on other companies' vehicles. There are also circumstances that make it difficult.

In addition, in the case of the second measure, it is not necessarily easy to secure installation space for the weight sensor, and there is also a condition that the weight of each car type of each train is sufficiently different to be distinguishable. Is required.

In the case of the third measure, there is a problem in that the identification accuracy varies greatly depending on the natural environment where the camera and microphone are installed. For example, image conditions differ greatly between daytime and nighttime, or between sunny and rainy weather, and depending on the ambient noise level, it may not be possible to clearly distinguish between these noises and the sound of a passing train.

The present invention has been developed in view of the above circumstances, and does not require the installation of a transmitting device on a vehicle or the installation of a weight sensor in a specific location, and moreover provides stable identification accuracy regardless of the natural environment. The aim is to provide a mobile object identification device that can

[Structure of the invention]

(Means for Solving the Problems) The present invention provides, as a means for solving the above problems, type 1
an image/sound detection means for detecting an image of the moving object or a sound generated from the moving object when the moving object whose type 1 identification number and other classification items are preset passes a predetermined location; a feature item analysis unit that analyzes a plurality of preset feature items based on the image or audio detected by the image/audio detection unit; and a plurality of analysis signals obtained by analysis of the feature item analysis unit. A configuration comprising: a neural network means for specifying the classification item to which the moving object that has passed the predetermined location belongs based on the above; and an identification information output means for outputting information regarding the classification item specified by the neural network means. It's Toshide.

(Function) When a moving object passes a predetermined location, the image/sound detection means detects an image of the moving object or a sound generated as the moving object passes.

The feature item analysis means analyzes a plurality of preset feature items based on the image or sound obtained by this detection, and outputs the analysis signal to the neural network means.

Based on this analysis signal, the neural network means selects the classification item to which the moving object that has passed through the predetermined location is most likely to belong from among a plurality of predetermined classification items, and moves the selected classification item. Specify as a body classification item.

The identification information output means outputs information regarding the classification item specified in this way.

(Example) Hereinafter, an example of the present invention will be described based on FIGS. 1 to 6.

FIG. 1 is a block diagram showing a schematic configuration of the present invention. That is, the passage time detection means 1 detects the time when the moving body passes a predetermined point, and sends a detection signal to the image/audio detection means 2.
It is designed to be sent to

When this detection signal is input, the image/sound detection means 2 starts detecting the image or sound of the moving body, and outputs the image signal or the sound signal to the characteristic item analysis means 3.

The feature item analysis means 3 analyzes a plurality of predetermined feature items for this image signal or audio signal, and outputs an analysis signal indicating the feature amount to the neural network means 4. For example, in the case of an image signal, the area of the target image, the number of corners, the shading value of each color component, etc. are determined with respect to the image quality, the intensity of each color, etc. Further, in the case of a sound 7N signal, the average frequency etc. of the pitch etc. of the sound are determined.

The neural network means 4 inputs these analytical signals to the human power layer, and passes them through the intermediate layer to the passing point detection means 1.
The model to which the moving object detected belongs.

For classification items such as company regulations and identification numbers, one of the two is estimated and specified.

Then, the identification information output means 5 outputs information about the specified classification item to a display device such as a display or other peripheral devices.

Note that the neural network means 4 has a learning function called a so-called pack propagation function. If the classification item specified by the neural network means 4 is found to be incorrect, the data given from the training data human power means 6 is used to perform the inference that led to the output of this incorrect classification item. The process can now be corrected. Therefore, from now on, even if similar analysis signals are input to the input layer of the neural network means 4, the correct classification item can be specified.

The control section 7 is designed to control the overall operation of each of the above-mentioned means 1 to 6.

FIG. 2 is a block diagram showing the schematic configuration shown in FIG. 1 in more detail.

In this figure, the passage point detection means 1 is composed of a sensor 1g that detects the passage of a moving object, and a sensor input section 1b that sends a sensor signal from the sensor la to the control section 7 via the control bus 8. has been done. This sensor 1a is
It consists of a sensor element that uses ultrasonic waves and light, and a track circuit often used in railways.

The image/audio detection means 2 includes a camera 2 that photographs a moving object.
a and a video input section 2b for inputting the video signal thereof;
It is composed of a microphone 2C that detects the voice of a moving body and a sound input section 2d that inputs the voice signal. The video human power section 2b is connected to the control bus 8 and the image bus 9, and the audio human power section 2d is connected to the control bus 8 and the audio data bus 10.

Feature item solution) The h means 3 is a control bus 8 and an image bus 9.
It is comprised of an image memory 3a and an image processing section 3b connected to the control bus 8 and an audio data bus 10, and an audio memory 3c and an audio processing section 3d connected to the control bus 8 and the audio data bus 10. The image memory 3a has two areas 3al''a2, and the audio memory 3c also has two areas 3cl''c2.
have.

The neural network means 4 has a neural network means 4 a connected to the control bus 8 .

The identification information output means 5 has an output interface 5a connected to the control bus 8, and this output interface 5a is connected to a peripheral system 11 having a display device such as a CRT.

The learning data input means 6 includes an input/output interface 6a connected to the control bus 8, and an input/output interface 6a connected to the control bus 8.
The input/output device 6b is connected to the input/output device 6b.

The control unit 7 includes an arithmetic control unit 7a connected to a control bus 8.
have.

Next, the operation of this embodiment configured as described above will be explained. Now, as shown in FIG. 3, n vehicles E1.
E2. ..., consider a train consisting of En as a moving object,
Assume that the passage of the leading vehicle E1 is detected by the sensor 1a. Then, this sensor signal is sent to the control section 7 via the sensor input section 1b and the connector box 12.

Immediately upon receiving this sensor signal, the control section 7 sends an operation command signal to the camera 2a to cause the camera 2a to photograph the second vehicle E2.

The video signal from the camera 2a is converted into an image signal by the video input section 2b, and an image as shown in FIG. 4(a) is stored in areas 3 and 1 of the image memory 3a. This stored image is analyzed for predetermined characteristic items by the image processing section 3b. These characteristic items include, for example, the density of each color component (red, green, blue), the area of the object when it is binarized, and the number of corners.

In the case of the train in FIG. 4(a), the number of red components is strong. FIG. 4(b) is a binarized image of FIG. 4(a), which shows that the area of the subject is large and the number of corners is relatively small. The image shown in FIG. 4(b) is stored in area 3a2 of image memory 3a.

Further, FIG. 5(a) is an image showing a case where a freight train passes, and in this case, the number of all components is strong.

FIG. 5(b) is a binarized image of FIG. 5(a).
According to the image, the area of the subject is small and the number of corners is relatively large.

Next, the image data of FIGS. 4(a) and 4(b) stored in the area 3at''a2 of the image memory 3a is processed by the neural network calculation section 4a of the neural network means 4.
The classification item to which the passing train is thought to belong is specified based on a predetermined feature amount.

That is, as shown in FIG. 6, first, the feature amount for a predetermined feature item is extracted from the area 3al''a2 of the image memory 3a at each point P1, P2, . . . in the input layer in the neural network calculation unit 4a. , Px.

Then, from each point Pt, P2, . . . , Pn, each point Q, , Q2 . . . , Q, a signal is outputted at a predetermined level to each point. For example, if the level of the signal input to point P1 is 1, the levels of the signals U1□, U1°, . . . Ul, output from point P are 0. 2. 0. 3゜...,0
．． 05 etc. (However, U11+U...+
U-1). Similarly, 12 is input to point P2.
If the level of the +y signal is 3, the signals U2□, U2°, . . . , U2 . The signal level of each is 0.45. 0. 2. ...,
0. 08 etc. (However, U2□+U22・
...+U2. -3).

Each point Q1 of the middle layer. Q2. . . , Q, in this way correspond to each point P1 . P2. ..., based on the total amount of signals sent from Px, each point R,,R of the output layer
2, . . . outputs a signal at a predetermined level to R7.

Then, each signal level at each point Rt, R2,..., R7 of the output layer is determined, and each signal level at each point R,, R2°...
. , R, point R2 has the highest level. Then, the neural network calculation unit 4a calculates the train Ei whose passing was detected by the sensor 1a,
It is specified that R2, . . . , Eo are express trains.

The train E1゜E2 that passed through such a predetermined place,
..., the information that En is an express train is displayed on the display device of the peripheral system 11 via the output interface 5a of the identification information output means 5.

By the way, although the specification in the neural network calculation unit 4a is very accurate, it is not impossible to think of a case where this specification is incorrect. In such a case, when the operator notices a specific error, the operator operates the input/output device 6b of the learning data input means 6 and sends the learning data to the neural network calculation section 4a via the input/output interface 6a. Let them input. As a result, the pack propagation function in the neural network calculation unit 4a is activated, the inference process from the human power layer to the output layer is corrected, and henceforth, correct identification is performed even under similar environments and conditions. It becomes like this.

Trains E,,E,,...,E as explained above.

With the above identification method, it is not necessary to attach a transmitting device such as an oscillator or a resonator to the vehicle, and there is no problem in terms of space when installing a weight sensor. Then, based on the analyzed predetermined feature items, the neural network calculates the trains Et and R2.

Since the type of .

Furthermore, since there is no need to have a unique algorithm for each type of target vehicle, adding or changing new target vehicle types can be easily done using the learning function of the neural network.

Note that the present invention also includes the following embodiments.

(b) The image/sound detection means 2 in the configuration shown in FIG. 2 is capable of detecting both images and sounds;
It may be possible to detect only either images or sounds.

(b) In Fig. 2, the control path 81, image bus 9, and audio data bus 10 are shared to connect each means, but a one-to-one configuration using a serial interface etc. It may also be possible to connect only certain points.

(C) In FIG. 2, the arithmetic control section 7a and the neural network arithmetic section 4a are connected to the control bus 8 independently from each other, but the configuration is such that the neural network arithmetic section 4a is provided inside the arithmetic control section 7a. Alternatively, both may be connected to the back end.

(d) Image bus 9 in Figure 2. Audio data bus 1
0 through the human output interface, and connect the image monitor device and the audio monitor device to the area 3a in the image memory 3a.
l' 3a2 and area 3゜1 in audio memory 3C,
It is also possible to monitor images and sounds stored in 3o2.

(E) In the above example, a train is assumed as a moving object,
Although we have described the case where this method is used for railroad crossing control and operation control, it is also effective for identifying the type of vehicle at expressway toll gates, etc.

Furthermore, the present invention can be expected to be applied not only to land vehicles but also to air and sea vehicles such as airplanes and ships.

〔Effect of the invention〕

As described above, according to the present invention, the neural network means detects the image or sound of a moving object passing a predetermined location, and identifies the classification item of the moving object based on the analysis of a plurality of preset feature items. However, since the configuration is configured to output information on the identified classification items, there is no need to attach a transmitting device to the moving object or install a weight sensor in a specific location, and moreover, it can be used at any time regardless of the natural environment. It becomes possible to identify moving objects with stable accuracy.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a schematic configuration of an embodiment of the present invention;
FIG. 2 is a schematic diagram showing an example of connection of each means in FIG. 1;
3 to 6 are explanatory diagrams of the operation of the embodiment of the present invention. 2... Image/audio detection means, 3... Feature item analysis means, 4... Neural network means, 5... Identification information output means.

Claims (1)

[Claims] When a moving object whose model, type, identification number, and other classification items are set in advance passes through a predetermined location, an image of the moving object or sound generated by the moving object is detected. an image/audio detection means for analyzing a plurality of preset feature items based on the image or audio detected by the image/audio detection means; and an analysis of the feature item analysis means. neural network means for identifying the classification item to which the moving object that has passed the predetermined location belongs, based on a plurality of analytical signals obtained by the method; and an identification device for outputting information regarding the classification item specified by the neural network means. A mobile object identification device comprising: an information output means;