KR20220152060A - A method for detecting interior noise location of vehicles and system - Google Patents

Abstract

The present invention relates to a method for detecting an interior noise location of a vehicle which can identify the characteristics of constant and major in-vehicle noise which gives people emotional discomfort, and a system using the same. The method comprises the steps of: a) collecting noise data frame by frame unit for a predetermined time at a two-dimensional location (i, j) in the vehicle; b) determining whether noise has occurred more than a reference number of times based on the frame at the two-dimensional location (i, j) in the vehicle, and deriving a set having the location at which the noise occurred more than the reference number of times as an element; and c) applying density-based spatial clustering of applications with noise (DBSCAN), which is a density-based clustering analysis method, to the set derived in step b).

Description

Vehicle interior noise location detection method and system {A method for detecting interior noise location of vehicles and system}

The present invention relates to a method and system for locating noise inside a vehicle.

Automobile noise is a factor that greatly affects the emotional quality of automobiles, and automobile NVH (Noise, Vibration, Harshness) has been an important research subject for automobile manufacturers for a long time. Noise and noise are reported as the most problems in the initial quality of automobiles, and recently, interior noise such as BSR (Buzz, Squeak, Rattle) has emerged as a major quality issue.

Studies dealing with interior noise in automobiles are largely divided into three categories. First, in the design stage, the acoustic characteristics of the noise generated from the interior module of the vehicle are analyzed and sound quality indicators are developed. Since the interior noise is a kind of sound, sound quality analysis for the noise source is performed using various sound quality analysis techniques. do. For example, an evaluation index for sound quality of indoor noise may be developed by analyzing indices representing characteristics of sound, such as loudness, sharpness, and roughness. It is a very important study in the design stage of the automobile manufacturing process as it analyzes the occurrence of interior noise such as BSR in the interior module of the vehicle such as the instrument panel.

Second, research on predicting the type of indoor noise through machine learning and deep learning is currently being conducted. After converting the noise data into an acoustic fingerprint, STFT (Short-time Fourier Transform) noise map, etc., classification techniques of machine learning and lip learning are used to predict the type of indoor noise such as BSR. It is known that deep learning-based indoor noise classification has superior prediction accuracy compared to conventional acoustic approaches. It is believed that this is due to the expression learning of deep learning, which automatically finds non-linear patterns necessary for classification in raw data.

Finally, this study explores the location of indoor noise by using noise data from multiple locations in the car interior. Research in this field uses noise data from multiple locations measured by acoustic cameras to explore the location of indoor noise by utilizing the difference in intensity and time-series characteristics of noise.

Since the above study is a method using a simple algorithm that compares the mean-difference of noise based on the simple noise level, it is necessary to identify the characteristics of continuous and major in-vehicle noise that gives people emotional discomfort. There was a difficult problem.

Korean Patent Laid-open Publication No. 1995-0033128 ("A device for fixing a microphone for measuring interior noise in a vehicle", published on December 22, 1995)

The present invention has been made to solve the above problems, and the purpose of a method and system for locating noise inside a car is to identify the characteristics of continuous and major noise in a car that gives people emotional discomfort An object of the present invention is to provide a method and system for locating noise inside a vehicle.

In order to solve the above problems, a method for locating noise inside a vehicle of the present invention includes the steps of: a) collecting noise data frame by frame at a two-dimensional location (i, j) in the vehicle for a predetermined time; b) determining whether noise is generated more than a reference number of times based on a frame at a two-dimensional location (i, j) in the vehicle, and deriving a set having as an element a location where noise is generated more than a reference number of times; and c) applying Density-based spatial clustering of applications with noise (DBSCAN), which is a density-based clustering analysis method, to the set derived in step b).

In addition, in step a), a-1) deriving a dataset X(t) by collecting noise levels for each frame t at a two-dimensional position (i, j) in the vehicle for a predetermined time; and a-2) deriving Y(t), which is two-dimensional location information (i, j) for each frame (t) in which noise greater than or equal to a reference value occurs in the vehicle, using the X(t) do.

In step a-2), when the minimum value of X(t) in a specific frame t is m(t), the Y(t) is derived according to the formula below.

는 상기 Y(t)의 원소, 상기

는 상기 X(t)의 원소, 상기 g는 상기 a-2) 단계에서 사용하는 기준치)(remind

Is an element of the Y (t), the

is an element of X(t), and g is a reference value used in step a-2))

In addition, DBSCAN performed in step c) is characterized by performing cluster analysis using two parameters, eps and MinPts.

를 원소로 하는 데이터세트인 F를 도출하고, 상기 F와 상기 Y(t)를 이용해 특정 시점의 어느 위치에서 소음이 발생했는지를 판단하는 것을 특징으로 한다.In addition, step c) is information about whether the location (i.j) in the car is included in the dense cluster of the DBSCAN.

It is characterized by deriving F, which is a data set having as an element, and determining at which position at a specific time point noise is generated using the F and the Y(t).

A vehicle interior noise location locating system according to the present invention includes a noise collecting means for collecting noise data at a two-dimensional position (i, j) in the vehicle frame by frame for a predetermined time; and receiving data from the noise collection means to determine whether noise has occurred more than a reference number of times based on a frame at a two-dimensional location (i, j) in the vehicle, and deriving a set having the location where noise has occurred more than a reference number of times as an element. and a control unit for applying Density-based spatial clustering of applications with noise (DBSCAN), which is a density-based clustering analysis method, to the derived set.

In addition, the control unit collects the noise level for each frame (t) at the two-dimensional position (i, j) in the vehicle for a predetermined time, derives a dataset X(t), and uses the X(t) , Y(t), which is the two-dimensional location information (i, j), for each frame (t) in which noise is greater than the standard value in the vehicle is derived, and a set having the location where the noise is generated as an element is derived.

In addition, when the minimum value of X(t) in a specific frame t is m(t), the control unit is characterized in that the Y(t) is derived according to the formula below.

는 상기 Y(t)의 원소, 상기

는 상기 X(t)의 원소, 상기 g는 기준치)(remind

Is an element of the Y (t), the

is an element of X(t), and g is a reference value)

In addition, the DBSCAN is characterized by performing cluster analysis using two parameters, eps and MinPts.

를 원소로 하는 데이터세트인 F를 도출하고, 상기 F와 상기 Y(t)를 이용해 특정 시점의 어느 위치에서 소음이 발생했는지를 판단하는 것을 특징으로 한다.In addition, the control unit is information on whether the location (i.j) in the vehicle is included in the dense cluster of the DBSCAN.

It is characterized by deriving F, which is a data set having as an element, and determining at which position at a specific time point noise is generated using the F and the Y(t).

According to the method and system for locating noise inside a vehicle according to various embodiments of the present invention as described above, the process of human noise perception is modeled and noise data of multiple locations extracted through a noise collecting means, which is an omnidirectional acoustic camera, is utilized. It has the effect of searching for the location and time of occurrence of major noises in the car interior, which are detected by the difference in loudness.

1 is a schematic diagram of a vehicle to which a method and system for locating noise inside a vehicle according to an embodiment of the present invention are applied.
2 is a graph visualizing noise data of a first frame and a 1592 th frame among noise data collected in step a) of a method for locating noise inside a vehicle according to an embodiment of the present invention.
3 and 4 are visualization graphs of application results obtained by performing steps a), b) and c) of the method for locating noise inside a vehicle according to an embodiment of the present invention.

Hereinafter, a method and system for locating noise inside a vehicle according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

1 is a schematic diagram of a vehicle to which a vehicle interior noise location locating system according to an embodiment of the present invention is applied.

1 is a state viewed from the top of the vehicle. As shown in FIG. 1, based on the state in which the car 10 is viewed from above, in the present invention, the interior of the car 10 can be partitioned into two-dimensional areas, and the two-dimensional areas are partitioned in a grid pattern. It can be. That is, with reference to FIG. 1 , the position of a specific area surrounded by a grid can be expressed as (i, j), where i is the position relative to the car's width and j is the position relative to the car's length. 1, i has a value of 1, 2, 3, ..., i from left to right, and j has a value of 1, 2, 3, ..., j from top to bottom has a value

In the present invention, heterogeneous and major interior noises that cause discomfort to the driver in terms of emotion are considered. Drivers cannot remember all the interior noise generated while driving a car for several tens of minutes to an hour or two. Among numerous noises, noises generated continuously and densely in many areas remain in the driver's memory after driving and can be recognized as important. The main indoor noise dealt with in the present invention refers to noise that is louder in the ear because it is louder than ambient sound, and does not occur temporarily and sporadically, but continuously in time and densely and continuously in one place in space. In the present invention, the noise recognition process is classified into three stages of detection, memory, and recognition, and modeling is used to propose an algorithm to search for major indoor noise.

A method for locating noise inside a vehicle according to an embodiment of the present invention may largely include steps a), b), and c).

Step a) is a sensing step, which simulates a situation in which a person perceives noise auditorily. When a sound from a specific location is heard louder than surrounding sounds, a person auditoryally detects the sound from that location. When the difference in volume is greater than a certain level, the sound can be distinguished, and in step a), noise data can be collected frame by frame for a predetermined time at the two-dimensional position (i, j) in the vehicle 10.

The noise data collected in step a) can be expressed in the form of a set. For example, assuming that the two-dimensional position inside the car 10 is divided into a total of 1,200 pieces of 60 horizontally and 20 vertically, and noise data is collected for 1 minute at 25 frames per second, noise data for a total of 1,500 frames Since it is collected, a total of 1,500 sets of 60 × 20 are created for each frame, and one set can have noise at the position (i, j) of the corresponding frame as an element. The element of the set for each frame may be 1,200, which is the total number of two-dimensional positions, and the element means the size of noise data and may be expressed in dB.

Step a) may be performed by a noise collecting means including a stereoscopic microphone and a control unit. To this end, the vehicle interior noise location locating system according to an embodiment of the present invention may include a noise collecting means and a control unit. .

The noise collecting means may be installed inside the vehicle 10 as a device such as a sound microphone, and during step a), the process of collecting noise may be performed by the noise collecting means, in addition to collecting noise. , the process of processing the data can be performed by the control unit. The control unit may be an electronic device including a kind of arithmetic device, and the noise collecting unit and the control unit may be wired or wirelessly connected to receive information from the noise collecting unit.

Step a) may be divided into steps a-1) and steps a-2) in detail.

First, step a-1) is a step of collecting the noise level for each frame (t) for a predetermined time at the two-dimensional position (i, j) in the vehicle, and deriving a set X(t) in the controller, a- In step 2), the control unit uses the set X(t) described above to obtain Y(t), which is a set having two-dimensional location information (i, j) for each frame (t) in which noise is greater than the standard value in the vehicle as an element. derive That is, Y(t) is obtained by applying a kind of filter to the above-mentioned X(t). Y(t) can be derived by assigning 1 to i, j) and assigning 0 to the corresponding position (i, j) if noise greater than or equal to the reference value does not occur.

Describing step a-2) in more detail, the control unit derives Y(t) according to the formula below when m(t) is the minimum value of elements in a specific frame t of the derived X(t) .

는 상기 Y(t)의 원소, 상기

는 상기 X(t)의 원소, 상기 g는 상기 a-2) 단계에서 사용하는 기준치)(remind

Is an element of the Y (t), the

is an element of X(t), and g is a reference value used in step a-2))

In the above formula, g may be a value arbitrarily designated by a user. In the present embodiment, the reason why Y(t) is obtained using the reference value g and the minimum value m(t) in the frame t as described above is that the noise inside the vehicle 10 is not the noise inside the vehicle 10, but the noise coming from the outside. It takes into account the situation in which the size of the noise increases.

Step b) performed after step a) is a storage step. In the second memory stage, it reflects the fact that people feel uncomfortable with noise in terms of emotion only when the driver's experience of noise remains in memory after driving. Generally, we drive our cars for minutes to hours, so we forget the experience of detecting noise over time. I can't remember all of my experiences of perceiving a sound louder than my surroundings in a particular location. It is expected that the experience of frequently detected noise among various noises will be mainly remembered.

In the step b), it is determined whether noise has occurred more than a reference number of times based on the frame at the two-dimensional position (i, j) in the vehicle, a set having the position where noise has occurred more than a reference number of times as an element is derived, and the control unit is carried out For example, assuming that a total of 10 noises occur at the position (1, 1) in FIG. 1, and a total of 30 noises occur at the position (3, 3), the reference value used in step b) is 20 In the case of times, step b) may derive a set having location information of (3, 3) as an element. The reference value used in step b) is referred to as c, and the above process can be expressed as a formula as follows.

로 표기하고 있으며,

을 원소로 갖는 행렬을 S로 나타낸다. 위치(i, j)에서 감지된 소음에 대한 변수를

라고 했을 때,

는 다음과 같이 표현할 수 있다.The above formula calculates the number of times noise is detected at position (i, j) during total T frames.

is denoted by

A matrix having as an element is represented by S. A variable for noise detected at location (i, j)

When I said

can be expressed as:

를 원소로 갖는 행렬을 Z로 나타내며, b) 단계가 수행된 이후에는 Z가 출력된다.of the above formula

A matrix having as an element is represented by Z, and after step b) is performed, Z is output.

Step c) is a recognition step, which is a process of finding dense noise sources that are recognized as important by the driver. Since noise is a kind of sound, the sound spreads around from the point where the noise is generated. If it is a noise that a person remembers and recognizes as important, it is expected that the location of the noise is gathered within a certain area. As described above, since the interior of the car 10 is subdivided into 1,200 locations, the driver perceives noise generated in a dense manner as important rather than sporadic noise generated at a few locations, c) step considers sporadic noise as an outlier and removes it.

=1인 위치 (i, j)를 원소로 하는 2차원 행렬을 H로 정의하며, c) 단계는 H에 밀도기반 군집 분석 방법인 DBSCAN(Density-based spatial clustering of applications with noise)을 적용한다. c) 단계에서 사용되는 DBSCAN은 노이즈(noise)를 포함하며 군집의 형상이 구가 아닌 복잡한 형태를 갖는 데이터에도 군집분석이 효과적이라고 알려져 있다. In the matrix Z described above,

A two-dimensional matrix having a position (i, j) of = 1 as an element is defined as H, and step c) applies Density-based spatial clustering of applications with noise (DBSCAN), a density-based clustering analysis method, to H. DBSCAN used in step c) contains noise, and it is known that cluster analysis is effective for data with complex shapes other than spheres.

In DBSCAN used in step c), there are two parameters, eps and MinPts. Step c) performs density-based clustering analysis using eps and MinPts. eps means the radius of a neighbor, and two points whose distance is less than eps on the coordinate plane are regarded as neighbors and grouped into the same cluster. MinPts is the minimum size of a dense cluster, and a cluster is considered dense only when the number of elements in the cluster is greater than or equal to MinPts.

The aforementioned eps and MinPts may be set by the user, or may be set to appropriate values by the controller according to circumstances.

로, 그렇지 않으면

로 정의한다. 아울러,

를 원소로 갖는 행렬을 F로 나타낸다. 최종적으로 행렬 F에서

인 위치 (i, j)는 주변보다 소음이 크고, 빈번하게 발생되며, 밀집되어 발생하는 주요 소음의 발생 위치를 나타낸다.If the specific position (i, j) in step c) is included in one of the dense clusters of DBSCAN,

as, otherwise

is defined as together,

A matrix having as an element is denoted by F. Finally in matrix F

The in position (i, j) represents a main noise generation position that is louder than the surroundings, frequently generated, and generated densely.

이면, 해당 위치 (i, j)가 어느 프레임 때 소음이 발생했는지를 Y(t)를 통해 확인할 수 있다. 즉,

이면, Y(t)에서 해당 위치 (i, j)의 원소가 1인 프레임을 확인해, 언제 소음이 발생했는지를 판단할 수 있다.Also, in step c), the occurrence point of the main indoor noise can be searched by utilizing the analysis result of the occurrence location of the main indoor noise. More specifically, when the above-described matrix F and the Y(t) are compared, the elements of F

, it is possible to check through Y(t) which frame the noise occurred at the corresponding position (i, j). in other words,

, it is possible to determine when noise is generated by checking a frame in Y(t) where the element at the position (i, j) is 1.

A process of performing steps a), steps b) and steps c) described above will be described.

First, as described above, in step a), noise data can be collected by the noise collecting unit at each of 1,200 locations inside the vehicle divided into 60 horizontally and 20 vertically. Noise data is collected for a total of 2 minutes, and since noise data is collected at 25 frames per second, the noise data consists of a total of 3,000 frames.

2 is a visualization of noise data of the first frame and the 1592nd frame among the noise data collected in step a).

As shown in FIG. 2, the noise data can be shaped in three dimensions, and as shown in FIG. 2A, the pattern of noise intensity in the first frame is gentle, so it can be visually confirmed that no heterogeneous and major indoor noise has occurred. can On the other hand, as shown in FIG. 2B, since the pattern of noise intensity is severely curved in the 1592nd frame, it can be visually confirmed that major noise is generated in the room.

In order to search for indoor noise, the reference value used in the steps a) to c) is set as follows.

3 is a visualization of the result of applying each of steps a), b) and c).

In FIG. 3, through steps a), b), and c), the location of noise that is not cognitively important is removed, and finally, the location of the main indoor noise can be found. In FIG. 3A, the number inside the graph indicates the number of detected noises in step a), the horizontal axis indicates the horizontal position of the car, and the vertical axis indicates the vertical position of the car. FIG. 3B visualizes this as a contour line.

Applying the above equation, the occurrence time at the main location inside the vehicle can be analyzed as shown in FIG. 4 . 4 is the analysis result of the 2629th frame, FIG. 4a is the result of detecting noise in the 2629th frame, that is, Y (2629), FIG. 4b is the location F where the indoor noise is generated, and FIG. Applying (2629) to the above equation, the finally recognized noise generation position is shown, and FIG. 4D is a visualization of the noise generation position. As shown in FIG. 4D , since the pattern of indoor noise intensity is highly curved and heterogeneous, it is possible to confirm the possibility that main indoor noise has occurred.

As a result of visually evaluating the process as described above for all 3,000 frames, when the reference value g used in step a) is 6, 7, 8, and 9, the main performance of the indoor noise occurrence time (frame) search result The results of analyzing the scales of accuracy, sensitivity, specificity, and precision are shown in the table below.

In the above table, the highest accuracy when g is 6 means that the evaluator's visual judgment about the curve of the noise intensity pattern is made on the basis of approximately 6dB. In addition, in the above table, specificity and precision increase as the noise detection reference value increases, and the false positive rate decreases as the noise detection reference value increases, so that the prediction that indoor noise has occurred is accurate.

The present invention is not limited to the above embodiments, and the scope of application is diverse, and various modifications and implementations are possible without departing from the gist of the present invention claimed in the claims.

10: car

Claims (6)

a) collecting noise data frame by frame for a predetermined time at a two-dimensional position (i, j) in the vehicle;
b) determining whether noise is generated more than a reference number of times based on a frame at a two-dimensional position (i, j) in the vehicle, and deriving a set having as an element a location where noise is generated more than a reference number of times; and
c) applying Density-based spatial clustering of applications with noise (DBSCAN), which is a density-based clustering analysis method, to the set derived in step b);
According to claim 1,
In step a),
a-1) deriving a dataset X(t) by collecting noise levels for each frame (t) at a two-dimensional position (i, j) in the vehicle for a predetermined period of time; and
a-2) deriving Y(t), which is 2-dimensional location information (i, j) for each frame (t) in which noise is greater than a reference value, using the X(t);
An automobile interior noise location search method comprising:
According to claim 2,
In step a-2),
Among the X(t), when the minimum value in a specific frame (t) is m(t), the Y(t) is derived according to the formula below.

(remind

Is an element of the Y (t), the

is an element of X(t), and g is a reference value used in step a-2))
According to claim 3,
Step c) is information on whether the position (i.j) in the car is included in the dense cluster of the DBSCAN.

Derive a dataset F, which has as an element,
A method for locating noise in a vehicle interior, characterized in that by using the F and the Y(t), it is determined at which location the noise is generated at a specific time point.
Noise collecting means for collecting noise data in frame units for a predetermined time at the two-dimensional position (i, j) in the vehicle; and
Receiving data from the noise collection means, determining whether noise has occurred more than a reference number of times based on a frame at a two-dimensional position (i, j) in the vehicle, and deriving a set having the location where noise occurred more than a reference number of times as an element, , A control unit for applying Density-based spatial clustering of applications with noise (DBSCAN), which is a density-based clustering analysis method, to the derived set.
According to claim 5,
The control unit,
The size of the noise for each frame (t) is collected for a predetermined time at the two-dimensional position (i, j) in the car to derive the dataset X(t), and using the X(t), the noise level exceeds the standard value in the car. A vehicle interior noise localization system characterized by deriving Y(t), which is two-dimensional location information (i, j) for each frame (t) where noise occurs, to derive a set having the location where noise occurs as an element.

Images