KR101598863B1 - Added weight value applying system and method combined with association of continuous event - Google Patents

Abstract

The present invention relates to a system and a method for weighting in a situation deduction coupled to a consecutive event. More specifically, the present invention is to increase a quality of information by embedding an event function algorithm (software), and mutually cooperatively managing a plurality of heterogeneous sensors.

Description

[0001] The present invention relates to a method and system for weighting in contextual reasoning,

The present invention relates to a system and method for weighting in contextual reasoning in which there is a connection and combination of continuous events, and more particularly, To improve the quality of the image.

A situation is a combination of related events that occur according to the passage of time, and the situation itself occurs through the process of precedence of the situation, accompanied by evidence, and interlocking.

In the process of deducing such situations, it is important to recognize the preceding signs or the process of collecting or linking the evidences presented.

It is necessary to develop a plan to recognize the occurrence of a serious situation in advance and to prepare for the occurrence of a serious situation by detecting the occurrence of the related events continuously.

For example, much of the accidents that occur with student transport vehicles can be negligent on the part of the driver. In this regard, if the technology for detecting and responding to these events is available, the incidence can be reduced.

As a concrete method for preventing the above-described safety accidents, there may be a method of arranging sensors capable of detecting related risk factors at predetermined intervals and judging whether or not the events reported by the detection sensor are interlocked.

It is a method to infer a dangerous situation by recognizing whether a sensor is arranged in advance and a risk event that each sensor senses occurs in order.

In the first place, it detects the occurrence of a risk event and then determines whether the risk events are linked or related to each other.

The risk of each sensor is analyzed and evaluated to determine whether there is a correlation between the risk factors.

This method has advantages in adaptability and application to various risk aspects.

With respect to such technologies, it is possible to control the obstacle detection response speed by varying the perceived area for detecting the obstacle located in front of the vehicle based on the running speed of the vehicle, while controlling the heterogeneous sensor A technique for integrating sensing data generated from a plurality of sensors by providing a synchronizing signal for enabling temporal synchronization to be provided to a plurality of sensors is disclosed in Korean Patent Laid-Open Publication No. 2014-0098598 (Vehicle Speed Adaptive Obstacle Detecting Apparatus and Method ).

However, the above-mentioned technique has a problem that the quality of information to be acquired deteriorates because the consecutively occurring events are not considered to be correlated with each other, and the ability to recognize beforehand the occurrence of a specific situation is poor.

SUMMARY OF THE INVENTION An object of the present invention is to solve the problem described above and to improve the quality of information by incorporating an event operation function algorithm (software) in a control unit to cooperatively operate a plurality of heterogeneous sensors.

Another object of the present invention is to solve the above problem, and it is possible to grasp highly reliable data by analyzing correlation of successively related events and assessing approach to a dangerous situation.

In order to achieve the above object, there is provided a weighting system in a contextual reasoning system having a combination of continuous events and a combination of continuous events, A plurality of sensors (210, 220) for detecting and converting electrical signals into electrical signals; A preprocessor 110 for receiving and filtering signals output from the plurality of sensors 210 and 220; A first weight calculation module 121a for receiving a signal output from the preprocessing unit 110 and detecting a temporally continuous and spatially continuous event and assigning a first weight w ₁ ; A second weight calculation module 121a that detects an event at the same position in the same time zone and assigns a second weight w ₂ , and a hierarchical calculation of events sensed by different sensors of different kinds, A weight computing module 121 having a third weight computing module 121a for allocating a weight w ₃ ; And a data fusing module 122 for displaying each weight output from the weight computing module 121 as a weight vector of the three-dimensional space and outputting the risk according to the length of the vector.

Here, the plurality of sensors 210 and 220 are at least one ultrasonic sensor and an infrared sensor.

이며 Further, the first weight (w ₁₎ is the i-th when one or more events are detected in time _{i + 1 w 1 (i +} 1) = σ (w 1i + 1, 1) in a second time zone and aneumyeon event is detected w _{1 (i + 1) = σ} (w 1i - 1, 1) , and the modified sigmoid function σ (x, y) = 1 / is defined as (e ^-xy + 1), the second weight (w ₂ ) Is defined as w _2i = σ (w _{2 (i-1)} + 1, f _{en (} D _i) ) if another type of sensor is detected at the same position in the i th time zone and w _2i = σ (w _{2 (i-1)} - 1, f _{en (} D _i) ) where f _{en (} D _i) i is the number of all events in the dataset of the i time zone, and the third weight w ₃ is w _3i = σ (m (TS _i ), 1) in the i th time zone

And

Here a representation of a TS _i is type all of the data in _{time, m (D (l, k} ) i) is a set of all the events detected by the same sensor at the same position in one time zone, a _k is an arbitrary k-th Event.

Also, the risk according to the length of the vector is represented by a three-dimensional final weight vector (w _1i , w _2i , w _3i) in each time slot using first, second and third weights, 1,1), the higher the risk.

In order to accomplish the above object, there is provided a weighting method in contextual reasoning, which is a combination of continuous events and a combination of continuous events, (S10) of obtaining an event using the heterogeneous sensors; Obtaining an event operation function for computing the event (S20); Generating events according to time-space continuity to generate a first weight using the acquired event arithmetic function, generating events detected by different sensors at the same position in the same time zone for generating a second weight, (S30) of each weight according to the generation of an event through a multiplication of a hierarchical operation, which is detected by different sensors of different kinds to generate the event; And combining the respective weights to obtain a final weight (S40).

The weighting system and method in contextual reasoning, which have a combination of continuous events according to the present invention, have the effect of enhancing the quality of information acquired by cooperatively operating and using heterogeneous sensors having various functions and performances .

The weighting system and method in contextual reasoning, which has a combination of continuous events according to the present invention, proposes a reasonable risk weight calculation method by improving the situation information to detect and acquire successive occurrences of the related events, It has a different effect of being able to recognize in advance the occurrence.

The system and method of weighting in contextual reasoning that have a combination and association of consecutive events according to the present invention has the further effect of being able to evaluate the approaching of a risk situation by analyzing the correlation of successive events.

1 is a configuration diagram of a weighting system in contextual reasoning with association and association of consecutive events according to the present invention;
2 is a flowchart of a weighting method in contextual reasoning with association and association of consecutive events according to FIG. 1;
Figure 3 is a vector diagram of final weights according to Figure 1;

Hereinafter, a system and method for weighting in contextual reasoning in which the continuous events of the present invention are linked and combined will be described in detail with reference to the drawings so that those skilled in the art can easily carry out the present invention.

FIG. 1 is a configuration diagram of a weighting system in contextual reasoning with association and combination of consecutive events according to the present invention.

Referring to FIG. 1, the weighting system in the contextual reasoning, which is a combination and connection of continuous events according to the present invention, includes a plurality of heterogeneous sensors 210 and 220 and a controller 100, The control unit 100 includes a preprocessing unit 110 and a control unit 120. The control unit 120 includes a weight calculation module 121 and a data fusion module 122. [

The weight calculation module 121 includes a first weight calculation module 121a, a second weight calculation module 121b, and a third weight calculation module 121c.

The plurality of different types of sensors 210 and 220 may be provided on a moving object to detect an external moving object or a fixed object.

Here, the moving body can arrange a small amount of sensors as a small moving body such as a vehicle, a motorcycle, a bicycle or a pedestrian.

Although the ultrasonic sensor and the infrared sensor are described as an example, the present invention is not limited thereto.

The sensors 210a, 210b and 210c are ultrasonic sensors and the sensors 220a, 220b and 220c are infrared sensors. The sensors of the same kind are distributed at predetermined intervals for sensing range.

First, if the data sensed by the sensor is DS = <TS ₁ , TS ₂ , TS ₃ , ---, TS _n > Is assumed.

This means the data stream and TS ₁ to TS _n is Means a set of all the data obtained at each time 1 ~ n.

Where _{TS i = <D (1,1)} i, D (1,2) i, ---, D (3,2) i> i as a naturally has a value from 1 to n, all of the data in the time zone .

For example, when installed in a vehicle, D _{(1, k)} means the sensor on the right side of the vehicle body, D _{(2, k)} means the rear surface of the vehicle body _, and D .

Here, k means the sensing value of the ultrasonic sensor when k = 1, the sensing value of the infrared sensor when k = 2, and i means the time zone number.

For example, say D _{(2,1) 3} Bodywork Refers to the data obtained at the third time in the rear ultrasonic sensor.

In the above situation, the weight of the risk that is detected using the input data may be represented by a three-dimensional vector (see FIG. 3).

This will be described in more detail below.

The preprocessing unit 110 filters out signals output from a plurality of different types of sensors 210 and 220 to remove noise and outputs the signals. In addition, the preprocessing unit 110 performs a role of converting an analog signal into a digital signal, And may further include a multiplexer function.

In the first weight computing module 121a, the first basis of the three-dimensional vector is related to the generation of events according to space-time continuity, and the event operation function is as follows.

Define f _en (A) as the number of events detected in set A (event number).

_{if (f en (TS i)} ≠ 0 and f en (TS (i + 1)) ≠ 0)

_{∧ (f en (D (l} , k) i) ≠ 0 and f en (D (l + 1, k) (i + 1) ≠ 0)

? W _{1 (i + 1)} =? (W _1i + 1, 1)

if not

? W _{1 (i + 1)} =? (W _1i -1, 1)

If one event is detected at TS _i (i-th time) and one event is detected at TS _{i + 1} (i + 1-th time) at the next time (∧ means and) , D _{(l, k) i} spatially from (spatially l th sensor present in the position in the i times) from the event is detected and D _{(l + 1, k) (i + 1)} (i + 1 time l + If the event is detected in the first position (that is, if the event is continuously detected temporally and continuously in spatially)

→ The weight at the next time (i + 1) is the result of computing the value of (w _1i + 1,1) by the modified sigmoid function described below.

If the above if conditional statement is not satisfied (that is, if no temporally continuous events are detected)

→ The weight of the next time is the result of computing the value of (w _1i - 1,1) into the following transformation sigmoid function.

w _1i is Is the weight at time i for the first base in the vector space.

is defined as σ (x, y) = 1 / (e ^-xy + 1) and has a value between 0 and 1.

For example, assuming that the initial weight 0 (w ₁₁ = 0) is satisfied, the condition of space-time continuity is satisfied in the corresponding time zone.

→ w _{1 (i + 1)} = σ (w _1i + 1,1)

w ₁₂ =? (0 + 1,1) = 1 / (e ^-1 +1)? 0.7311.

Now suppose that the next moment does not satisfy the continuity condition of time and space.

w ₁₃ = ? (0.7311-1,1) = 1 / (e ^{- (-} 0.2689 ⁾ +1)? 0.5668.

In the second weight calculation module 121b, the third base second basis is the basis of the sensor overlay as follows.

If f _en (D _{(l, 1) i} ) ≠ 0 ∧ f _en (D _{(l, 2) i} ≠ 0

? _W2i =? ( _{W2 (i-1)} +1, _fen (Di))

Or else

? _W2i =? ( _{W2 (i-1)} -1, _fen (Di))

w _2i is The second This is the weight at time i for the base.

The expression is interpreted as follows.

Ten thousand and one D _{(l, 1) i (l} ultrasonic sensor in position in the time zone i) an event is detected on the D _{(l, 2) i (same} time zone (I.e., an infrared sensor at the same position) detects an event (i.e., when a different sensor detects the same event occurring at the same time and at the same time).

→ The weight for the second basis in the vector space is a function of the transformation sigmoid function

(w2 (i-1) +1, _fen (Di)).

w2 (i-1) +1 is the same concept as the weight at the first basis described above.

Here, f _en (Di) is newly appeared in the equation. This means that it is related to the overlap of events and is multiplied by the number of all events in the dataset of the corresponding time zone (i) and input to the sigmoid function.

That is, the larger the number of events, the more nesting is considered to have occurred.

Otherwise, a value of (w _{2 (i-1)} -1, f _en (Di)) is input to the modified sigmoid function.

In the third weight calculation module 121c, the third base of the three-dimensional vector is the basis of the hierarchy of the event as follows.

First, think about the _{D (l, k) i =} <a 1, a 2, a 3, ---, a n> data set (Data set) represented by.

It is a collection of all events detected on the same sensor at the same location in one time zone.

Let m (D) be the intermediate operator needed to calculate the weight in one dataset.

That is, if no event is detected, the intermediate value is output as 0. If the event is detected, the number of events is divided by the absolute value of data of each event multiplied by the number of events.

The larger the number of events, the larger the value.

This is based on event density.

Now, weights for hierarchy are defined as follows.

(1 - m) (D _{(l, k) i} ) is smaller than 1 (we subtract 1 again later to obtain the resultant value, The bigger the sex, the bigger it gets).

That is, to calculate a weight for the hierarchy of events detected by the homogeneous sensor.

을 통해 서로 다른 위치에 존재하는 센서들 간의 계층적 계산을 한다.It performs a hierarchical calculation through the product symbol pi. first

To perform hierarchical calculations between sensors at different locations.

를 통해 이기종의 서로 다른 센서들 간의 계층적 연산을 곱을 통해 한다.Also

Through the multiplication of hierarchical operations between different sensors of different types.

Finally, we subtract the computed value from 1 to obtain the weight.

1 is to prevent the case where the total weight is affected by the case where m (D _{(l, k) i} =

Now the third basis is w3i = σ (m (TS _i ), 1).

That is, it outputs (m (TS _i ), 1) to the modified sigmoid function. The input to the sigmoid function is to make an element equal to the previous two weights by performing the same operation as the two previous weights.

Finally, based on the hierarchy of the data set in which the risk is detected, it is expressed as between 0 and 1.

The data fusing module 122 may be expressed as a weight vector w _1i , w _2i , w _3i in each time slot using three bases and can be illustrated as shown in FIG.

The closer the vector is to the (1,1,1) direction, the higher the risk.

Through this, w _i (final weight) =

(norm of a = ∥a∥) of the final weight vector as a value of a single parameter (i means time zone).

The present invention models a method of deducing or recognizing a situation by connecting a small amount of sensors to a small mobile body and correlating sensed values.

In situ awareness research using multiple sensors, it is important to operate and utilize sensors with various functions and capabilities cooperatively to improve the quality of acquired information.

To this end, the present invention proposes a reasonable risk weight calculation method to improve the situation information to detect and acquire related events continuously, and to recognize the occurrence of a specific situation in advance.

We analyze the correlation of successive events and calculate the risk by analyzing the linkage and combination of successive events in order to evaluate the approach to risk situation.

For this purpose, sensors were installed to detect the events of related events and it was judged whether or not the risk level of the events reported from the deployed sensors was increased.

At this time, weights are given according to reasonable measures to be considered in risk calculation.

As a weighting criterion, we present the results of overlapping, time - space interlocking, and hierarchical convergence processing among heterogeneous sensors. Weights determined using these factors are determined by using computation in three - dimensional space.

Fig. 2 is a flow chart of a weighting method in contextual reasoning with associations and associations of consecutive events according to Fig. 1;

Referring to FIG. 2, a weighting method in contextual reasoning, which is a combination and association of continuous events according to the present invention, is a weighting method in contextual reasoning in which a continuous event is linked and combined using a sensor provided in a moving object (S10) of obtaining an event using a plurality of heterogeneous sensors (210, 220); (S20) of acquiring an event operation function for calculating the event from the memory (not shown) of the controller 120 by the weight calculation module 121; The weight computing module 121 generates events in accordance with the space-time continuity in order to generate the first weight w ₁ using the obtained event operation function, and generates the second weight w ₂ in the same time zone for generating the second weight w ₂ A step of assigning respective weights according to generation of an event through multiplication of a hierarchical operation by detecting a different kind of sensor to generate an event occurrence and a third weight w _{3 which} are sensed by different sensors at the same position (S30); And a data fusing module 122 combining the respective weights to obtain a final weight (S40).

이며 If the first weight w ₁ is w _{1 (i + 1)} = σ (w _1i + 1, 1) in the i + 1 th time zone when one or more events are detected in the i th time zone _{w 1 (i + 1) =} σ (w 1i - 1, 1) , and the modified sigmoid function σ (x, y) = 1 / is defined as (e ^-xy + 1), the second weight (w ₂ ) is w _2i = σ (w _{2 (i-1)} + 1, f _{en (} D _i) ) if another type of sensor is detected at the same position in the i th time zone and w _2i =? (w _{2 (i-1)} - 1, f _{en (} D _i) ), where f _{en (} D _i) i is the number of all events in the dataset of the i time zone, and the third weight w ₃ is w _3i = σ (m (TS _i ), 1) in the i th time zone

And

Here a representation of a TS _i is type all of the data in _{time, m (D (l, k} ) i) is a set of all the events detected by the same sensor at the same position in one time zone, a _k is an arbitrary k-th Event.

Each of the weight calculation modules in the weight calculation module 121 can embed an event calculation function for calculating each weight without reading the event calculation function software from a memory (not shown).

110:
120:
210, 220:

Claims (6)

A weighting system in a situation inference in which a continuous event is linked and combined using a sensor provided in a moving object,
A plurality of sensors (210, 220) for detecting the event and converting it into an electrical signal;
A preprocessor 110 for receiving and filtering signals output from the plurality of sensors 210 and 220;
A first weight calculation module 121a for receiving a signal output from the preprocessing unit 110 and detecting a temporally continuous and spatially continuous event and assigning a first weight w ₁ ; A second weight calculation module 121a that detects an event at the same position in the same time zone and assigns a second weight w ₂ , and a hierarchical calculation of events sensed by different sensors of different types, A weight computing module 121 having a third weight computing module 121a for allocating a weight w ₃ ; And
And a data convergence module (122) for displaying each weight output from the weight computing module (121) as a weight vector of the three-dimensional space and outputting a risk according to the length of the vector. A weighting system in contextual reasoning with an association.
The method according to claim 1,
Wherein the plurality of sensors (210, 220) are at least one ultrasonic sensor and an infrared sensor.
The method according to claim 1,
The first weight (w ₁₎ is the i-th when one or more events are detected in time i + 1, w _{1 (i + 1)} in the second time zone _{= σ (w 1i + 1,} 1) and aneumyeon event is detected w _{1 (i + 1)} = σ - and (w _1i 1, 1), is defined as a modified sigmoid function σ (x, y) = 1 / (e -xy + 1), the second weight (w ₂₎ is the i-th when the time zone other types of sensors is detected at the same position in the _{w 2i = σ (+ 1,} w 2 (i-1) f en (D i)) and no other types of sensors is not detected w _2i = σ _{(w 2 (i-1)} - 1, f en (D i)) , where f _{en (D i)} is i is the number of all events in the dataset of the i time zone, and the third weight w ₃ is w _3i = σ (m (TS _i ), 1) in the i th time zone

And

Here a representation of a TS _i is type all of the data in _{time, m (D (l, k} ) i) is a set of all the events detected by the same sensor at the same position in one time zone, a _k is an arbitrary k-th Wherein the event is an event, and the event is a combination and a combination of consecutive events.
The method of claim 3,
The risk according to the length of the vector is expressed as a three-dimensional final weight vector (w _1i , w _2i , w _3i) in each time slot using the first, second and third weights, and the final weight vector is , 1) a weighting system in contextual reasoning with linkage and combination of continuous events characterized by a higher risk as the direction is closer.
A method of weighting in contextual reasoning that has a combination of continuous events and a combination of events using a sensor provided in a moving object,
Wherein the control unit includes a pre-processing unit and a control unit, and the control unit is configured to include a combination of continuous events including a weight calculation module and a data fusion module, We use a weighting system in contextual reasoning,
(S10) of obtaining an event using a plurality of heterogeneous sensors;
Acquiring an event operation function for calculating an event from a memory provided in the control unit (S20);
The weight calculation module generates an event according to the space-time continuity to generate the first weight using the acquired event arithmetic function, generates an event generated by the different sensors at the same position in the same time zone for generating the second weight, And (S30) assigning respective weights according to the occurrence of the event through the multiplication of the hierarchical operation, the third weight being sensed by a different kind of sensor to generate the third weight; And
And a data fusion module combining the weights to obtain final weights (S40). &Lt; Desc / Clms Page number 20 &gt;
6. The method of claim 5,
In the step S30,
The first weight (w ₁₎ is the i-th when one or more events are detected in time i + 1, w _{1 (i + 1)} in the second time zone _{= σ (w 1i + 1,} 1) and aneumyeon event is detected w _{1 (i + 1)} = σ - and (w _1i 1, 1), is defined as a modified sigmoid function σ (x, y) = 1 / (e -xy + 1), the second weight (w ₂₎ is the i-th when the time zone other types of sensors is detected at the same position in the _{w 2i = σ (+ 1,} w 2 (i-1) f en (D i)) and no other types of sensors is not detected w _2i = σ _{(w 2 (i-1)} - 1, f en (D i)) , where f _{en (D i)} is i is the number of all events in the dataset of the i time zone, and the third weight w ₃ is w _3i = σ (m (TS _i ), 1) in the i th time zone

And

Here a representation of a TS _i is type all of the data in _{time, m (D (l, k} ) i) is a set of all the events detected by the same sensor at the same position in one time zone, a _k is an arbitrary k-th Wherein the event is a combination of continuous events and a combination of continuous events.

Images