A fast sound source assessment and localization method based on clustering

Abstract

A method for rapid assessment and positioning of sound sources based on clustering ideas belongs to the technical field of sound target signal detection and parameter estimation, and solves the problem that sound sources cannot be quickly assessed and positioned under special conditions. The present invention targets sound source signals, firstly constrains the sound source position and eliminates abnormal points through trigger statistics; secondly, performs time delay statistics on the constrained sound source signal, and divides the data set according to the time delay; thirdly, calculates the time delay difference triggered between different sound detection modules, and excludes abnormal data points in combination with the distance between the modules; finally, performs grid division according to the geometric relationship of the arrangement of the sound detection modules, and traverses any node in the same data set based on the idea of ​​ray tracing, completes a straight line connection between any two points, counts the number of times all straight lines cross different grids, and then performs cluster analysis on the number of grid crossings to determine the sound source position. The present invention can improve the speed and accuracy of sound source positioning.

Description

Technical Field

[0001] The present invention belongs to the technical field of sound target signal detection and parameter estimation, and particularly relates to a sound source rapid assessment and positioning method based on clustering idea. Background Art

[0002] Currently, sound source localization methods mainly include traditional methods and deep learning intelligent methods. The time delay estimation method based on the time difference of arrival (TDOA) is a traditional sound source localization method commonly used both domestically and internationally. This method has a simple structure and fast calculation speed and is often used to estimate the three-dimensional coordinates of the sound source. In the process of solving the sound source location, the existing methods have the following shortcomings:

[0003] In embedded applications, direct calculation methods are often used to simplify calculations, but their anti-interference ability and positioning accuracy are not high.

[0004] Newton iteration method is also often used for position estimation, but its positioning accuracy depends on the initial value and the calculation time is long;

[0005] The spherical interpolation method has fast calculation speed and relatively high positioning accuracy and has been widely used. However, its noise resistance is weak and it is not suitable for some occasions where high positioning accuracy is required. Summary of the Invention

[0006] The main purpose of the present invention is to overcome the deficiencies in the prior art and solve the technical problem that traditional TDOA methods are difficult to quickly and accurately identify and estimate sound source signals. The present invention provides a sound source rapid assessment and positioning method based on clustering ideas, which utilizes time delay sorting, data set partitioning, and grid clustering analysis based on clustering ideas to complete the rapid assessment and positioning of parameter estimation of sound source signals.

[0007] The present invention is implemented through the following technical solution: a method for rapid sound source assessment and positioning based on clustering, comprising the following steps:

[0008] S1, collecting sound signals through the sound detection module, and triggering time sorting through the sound detection module;

[0009] S2. Divide the acoustic detection module data obtained in step S1 into several data sets based on the acquisition time in time periods T, and count the number of times the acoustic detection module is triggered in each data set;

[0010] S3, select the time domain T with the largest total number of times the acoustic detection module is triggered in step S2. a and adjacent time domain T a-1 、T a+1The acoustic detection module data in the time domain is less than 10ms, and the time domain T a-1 、T a 、T a+1 Merge into a new sound source signal dataset;

[0011] S4, repeating the above steps S2 to S3 until the sound source signal data of the required time length is obtained and the influence of abnormal signals is eliminated;

[0012] S5. Let “0” be the unknown sound source center, and the delay difference P between adjacent sensor nodes ki for:

[0013]

[0014] In formula (1), P i is the triggered acoustic sensor node, t k is the trigger delay of acquisition, i = 1, 2, ..., Q, k = 1, 2, ..., Q;

[0015] S6. Divide and cluster the delay differences between adjacent sensor nodes determined in step S5 according to time range to obtain a data set D:

[0016]

[0017] In formula (2), v is the speed of sound propagation in the medium, ε is the error of the acoustic signal trigger delay; R is the distance position of the triggered sensor;

[0018] S7. Arrange the data set D obtained in step S6 in descending order according to the delay size, and divide the data set according to the value of the error ε. The divided data set D' is:

[0019] D′=[A(x1,y1),B(x2,y2),C(x3,y3),D(x4,y4),E(x5,y5)]; Formula (3)

[0020] The size of the divided data set D' in the X and Y directions is divided into Δx and Δy respectively;

[0021] S8. Connect any two points A, B, C, D, or E in step S7 to get the straight line y between points i and j. ij (i≠j) is:

[0022]

[0023] S9, take the coordinate point of the grid Y axis in step S7 as the reference point, traverse all the straight lines in formula (4) in step S8, and get the straight line y ij (i≠j) Coordinates intersecting the X-axis for:

[0024]

[0025] According to formula (5), the grid position where the corresponding straight line passes is determined as:

[0026]

[0027] S10, traverse the grids that each straight line obtained by equation (6) passes through, and cluster analysis and count the number of times each grid passes through the straight line, and select the coordinates of the grid with the maximum number of times, which is the estimated location of the sound source

[0028] S11, use the shortest delay node to perform weighted averaging on all nodes, and adjust the sound source estimation position in step S10 to improve the sound source estimation accuracy; extract the minimum delay t1 of D' in formula (3), and select the corresponding reference point A(x1, y1), and the position difference between OA can be obtained as:

[0029]

[0030] S12, according to formula (7) the position of the sound source estimated in step S10 Adjust and estimate twice, we can get:

[0031]

[0032] In formula (8), l represents the number of points in the dataset D', r x 、r y Indicates the amount of adjustment of the sound source position;

[0033] S13. Select each data set in formula (3) as a reference point in turn, repeat the above steps S11 to S12, and perform mean processing on the obtained sound source positions to complete the sound source rapid assessment and positioning based on clustering thinking.

[0034] Furthermore, in step S1, the sound detection module includes a high threshold and a low threshold, the initial value of the high threshold is 120dB, and the initial value of the low threshold is 90dB. When the decibel value of the sound signal exceeds the high threshold or is lower than the low threshold, it is triggered and outputs a low-level signal.

[0035] Furthermore, in step S4, the required time length is 18 seconds.

[0036] Furthermore, in step S6, v is the speed of sound propagation in the air, v=340 m / s; the error ε of the acoustic signal trigger delay is determined according to the distance between the sensors.

[0037] Furthermore, in step S7, the values ​​of the grid sizes Δx and Δy are determined according to the distance resolution of the sampling nodes.

[0038] Furthermore, in step S12, the sound source position adjustment amount r x 、r y for:

[0039]

[0040] The beneficial effects of the present invention are:

[0041] 1. The present invention targets the sound source signal and triggers statistics through the detection module, which can realize the sound source location constraint and abnormal point elimination;

[0042] 2. The present invention can divide the sound source data set by time delay sorting for constrained sound source signals;

[0043] 3. The present invention completes grid division based on the geometric relationship of the detection module layout, and traverses the data set nodes based on the idea of ​​ray tracing to achieve accurate cluster analysis to determine the sound source location

[0044] In summary, the present invention can improve the speed and accuracy of sound source localization, and provide technical support for the detection and identification requirements of subsequent comprehensive test systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flowchart of the present invention;

[0046] Figure 2 Schematic diagram of dividing the sound source signal into three data sets in Example 1;

[0047] Figure 3 This is a schematic diagram of the acoustic sensor data obtained after traversing the data set nodes in Example 1;

[0048] Figure 4 This is a statistical diagram of the first time delay extracted by the acoustic sensor in the data set of Example 1;

[0049] Figure 5 Schematic diagram of the acoustic sensor triggering times statistical data set obtained by dividing by T=180s in Example 2;

[0050] Figure 6 Schematic diagram of the acoustic sensor triggering times statistical data set obtained by dividing by T=20s in Example 2;

[0051] Figure 7 Schematic diagram of valid data in the statistical data set of the number of acoustic sensor triggering obtained by dividing by T=600ms in Example 3;

[0052] Figure 8This is a schematic diagram of the first trigger delay of three acoustic sensors in Example 3;

[0053] Figure 9 Schematic diagram of valid data in the statistical data set of four groups of acoustic sensor triggering times obtained by dividing by T = 600ms;

[0054] Figure 10 This is a schematic diagram of the first trigger delay of four groups of acoustic sensors;

[0055] Figure 11 Schematic diagram of sound source localization for grid cluster analysis;

[0056] In each figure: Figure 2 、 Figure 5 and Figure 6 The horizontal axis is the dataset name, and the vertical axis is the number of times the acoustic sensor is triggered; Figure 3 、 Figure 4 、 Figure 7 、 Figure 8 、 Figure 9 and Figure 10 The horizontal axis is the acoustic sensor trigger time in milliseconds, and the vertical axis is the acoustic sensor number. DETAILED DESCRIPTION

[0057] The present invention will be described in further detail below with reference to the accompanying drawings and examples.

[0058] Example 1

[0059] like Figure 1 A method for rapid sound source assessment and positioning based on clustering is shown, comprising the following steps:

[0060] S1. Acquire sound signals through the sound detection module and trigger time sorting through the sound detection module; the sound detection module includes a high threshold and a low threshold. The initial value of the high threshold is 120dB, and the initial value of the low threshold is 90dB. When the decibel value of the sound signal exceeds the high threshold or falls below the low threshold, it is triggered and outputs a low-level signal;

[0061] S2. Divide the acoustic detection module data obtained in step S1 into several data sets based on the acquisition time in time periods T, and count the number of times the acoustic detection module is triggered in each data set;

[0062] S3, select the time domain T with the largest total number of times the acoustic detection module is triggered in step S2. a and adjacent time domain T a-1 、T a+1 The acoustic detection module data in the time domain is less than 10ms, and the time domain T a-1 、T a 、T a+1Merge into a new sound source signal data set, in this embodiment 1, a=11;

[0063] S4. Repeat steps S2 to S3 until the sound source signal data with a required time length of 18 seconds is obtained to eliminate the influence of abnormal signals;

[0064] S5. Let “0” be the unknown sound source center, and the delay difference P between adjacent sensor nodes ki for:

[0065]

[0066] In formula (1), P i is the triggered acoustic sensor node, t k is the trigger delay of acquisition, i = 1, 2, ..., Q, k = 1, 2, ..., Q;

[0067] S6. Divide and cluster the delay differences between adjacent sensor nodes determined in step S5 according to time range to obtain a data set D:

[0068]

[0069] In formula (2), v is the speed of sound propagation in the medium. In this embodiment 1, v is the speed of sound propagation in the air, v=340 m / s; ε is the error of the acoustic signal trigger delay, and the error ε of the acoustic signal trigger delay is determined according to the distance between the sensors; R is the distance position of the triggered sensor;

[0070] S7. Arrange the data set D obtained in step S6 in descending order according to the delay size, and divide the data set according to the value of the error ε. The divided data set D' is:

[0071] D′=[A(x1,y1),B(x2,y2),C(x3,y3),D(x4,y4),E(x5,y5)]; Formula (3)

[0072] The size of the divided data set D' in the X and Y directions is divided into grids Δx and Δy respectively, and the values ​​of the grid sizes Δx and Δy are determined according to the distance resolution of the sampling nodes;

[0073] S8. Connect any two points A, B, C, D, or E in step S7 to get the straight line y between points i and j. ij (i≠j) is:

[0074]

[0075] S9, take the coordinate point of the grid Y axis in step S7 as the reference point, traverse all the straight lines in formula (4) in step S8, and get the straight line y ij(i≠j) Coordinates intersecting the X-axis for:

[0076]

[0077] According to formula (5), the grid position where the corresponding straight line passes is determined as:

[0078]

[0079] S10, traverse the grids that each straight line obtained by equation (6) passes through, and cluster analysis and count the number of times each grid passes through the straight line, and select the coordinates of the grid with the maximum number of times, which is the estimated location of the sound source

[0080] S11, use the shortest delay node to perform weighted averaging on all nodes, and adjust the sound source estimation position in step S10 to improve the sound source estimation accuracy; extract the minimum delay t1 of D' in formula (3), and select the corresponding reference point A(x1, y1), and the position difference between OA can be obtained as:

[0081]

[0082] S12, according to formula (7) the position of the sound source estimated in step S10 Adjust and estimate twice, we can get:

[0083]

[0084] In formula (8), l represents the number of points in the dataset D', r x 、r y Indicates the amount of adjustment of the sound source position;

[0085] Sound source position adjustment amount r x 、r y for:

[0086]

[0087] S13. Select each data set in formula (3) as a reference point in turn, repeat the above steps S11 to S12, and perform mean processing on the obtained sound source positions to complete the sound source rapid assessment and positioning based on clustering thinking.

[0088] In this embodiment 1, the sound source signal is divided into T 10 、T 11 、T 12 Three datasets, such as Figure 2 shown. Figure 3 is the result of traversing the dataset nodes, Figure 4 To extract the first delay of the data set, grid cluster analysis is performed. Figure 3and Figure 4 The time period with the most concentrated acoustic signal triggering can be obtained, providing a data source for clustering diversity.

[0089] It can be seen that the present invention can complete grid division according to the geometric relationship of the detection module layout, traverse the data set nodes based on the idea of ​​ray tracing, and accurately perform cluster analysis to determine the sound source location.

[0090] Example 2

[0091] like Figure 1 A method for rapid sound source assessment and positioning based on clustering is shown, comprising the following steps:

[0092] S1. Acquire sound signals through the sound detection module and trigger time sorting through the sound detection module; the sound detection module includes a high threshold and a low threshold. The initial value of the high threshold is 120dB, and the initial value of the low threshold is 90dB. When the decibel value of the sound signal exceeds the high threshold or falls below the low threshold, it is triggered and outputs a low-level signal;

[0093] S2. For the acoustic detection module data obtained in step S1, the data is divided into several data sets based on the acquisition time, with the time period T1 = 180s as the unit, and the number of times the acoustic detection module is triggered in each data set is counted, such as Figure 5 As shown;

[0094] S3, select the time domain T with the largest total number of times the acoustic detection module is triggered in step S2. a and adjacent time domain T a-1 、T a+1 The acoustic detection module data in the time domain T a-1 、T a 、T a+1 Merge into a new sound source signal data set, a=5 in this embodiment 2;

[0095] S4, repeat the above steps S2 to S3, divide the data sets into several units with time period T2 = 20s, select the acoustic detection module data in the time domain T6 with the largest total number of acoustic detection module triggers and the adjacent time domains T5 and T7, and merge the time domains T5 to T7 into a new sound source signal data set, and count the number of times the acoustic detection module is triggered in each data set, such as Figure 6 As shown; until the sound source signal data with the required time length of 18s is obtained, the influence of abnormal signals is eliminated;

[0096] Subsequent steps S5 to S13 are the same as those in Example 1 and are not described in detail here, thereby completing the rapid assessment and positioning of the sound source based on the clustering concept.

[0097] This embodiment 2 is used to verify the effectiveness of the present invention in selecting valid sound source signals and eliminating abnormal sound source signals, and can obtain a valid sound source signal data set after eliminating abnormal signals.

[0098] Example 3

[0099] like Figure 1 A method for rapid sound source assessment and positioning based on clustering is shown, comprising the following steps:

[0100] S1. In this embodiment 3, three acoustic sensors are respectively arranged at 35 meters, 40 meters and 80 meters away from the sound source, and sound signals are collected by the acoustic detection module, and the triggering time of the acoustic detection module is sorted; the acoustic detection module includes a high threshold and a low threshold, the initial value of the high threshold is 120dB, and the initial value of the low threshold is 90dB. When the decibel value of the sound signal exceeds the high threshold or is lower than the low threshold, it is triggered and a low-level signal is output;

[0101] S2. Divide the acoustic detection module data obtained in step S1 into several data sets in time periods T, and count the number of times the acoustic detection module is triggered in each data set;

[0102] S3, select the time domain T with the largest total number of times the acoustic detection module is triggered in step S2. a and adjacent time domain T a-1 、T a+1 The acoustic detection module data in the time domain T a-1 、T a 、T a+1 Merge into a new sound source signal dataset;

[0103] S4, repeat the above steps S2 to S3 until the sound source signal data of the required time length of 600ms is obtained, eliminating the influence of abnormal signals, and the valid data in the sound sensor triggering number statistical data set obtained by dividing T = 600ms is as follows Figure 7 As shown;

[0104] S5. Let “0” be the unknown sound source center, and the delay difference P between adjacent sensor nodes ki for:

[0105]

[0106] In formula (1), P i is the triggered acoustic sensor node, t k is the trigger delay of acquisition, i = 1, 2, ..., Q, k = 1, 2, ..., Q;

[0107] Figure 8 To obtain the data after the first delay of the high-level threshold trigger signal, Figure 8It can be seen that the first delays for the three acoustic sensors to trigger are 7374146ms, 7374161ms, and 7374280ms, respectively. The delay differences are 15ms and 119ms, respectively. The distance differences obtained from the delay differences are 5.1m and 40.5m, respectively, which are basically consistent with the theoretical values. Therefore, the present invention is effective in calculating the sound source location by selecting the first delay for the acoustic sensor to trigger.

[0108] Subsequent steps S6 to S13 are the same as those in Example 1 and are not described in detail here, thereby completing the rapid assessment and positioning of the sound source based on the clustering concept.

[0109] Example 4

[0110] like Figure 1 A method for rapid sound source assessment and positioning based on clustering is shown, comprising the following steps:

[0111] S1. In this embodiment 4, four groups of acoustic sensors are arranged in the east, south, west and north directions of the sound source, respectively. Each group of sensors includes three sensors, which are arranged at 30, 40 and 60 meters from the sound source, respectively. Sound signals are collected by the acoustic detection module, and the trigger time of the acoustic detection module is sorted; the acoustic detection module includes a high threshold and a low threshold. The initial value of the high threshold is 120dB, and the initial value of the low threshold is 90dB. When the decibel value of the sound signal exceeds the high threshold or is lower than the low threshold, it is triggered and a low-level signal is output;

[0112] S2. Divide the acoustic detection module data obtained in step S1 into several data sets based on the acquisition time in time periods T, and count the number of times the acoustic detection module is triggered in each data set;

[0113] S3, select the time domain T with the largest total number of times the acoustic detection module is triggered in step S2. a and adjacent time domain T a-1 、T a+1 The acoustic detection module data in the time domain T a-1 、T a 、T a+1 Merge into a new sound source signal data set, a=11 in this embodiment 4;

[0114] S4, repeat the above steps S2 to S3 until the sound source signal data with the required time length of 600ms is obtained (such as Figure 9 As shown), eliminate the influence of abnormal signals;

[0115] S5. Let “0” be the unknown sound source center, and the delay difference between adjacent sensor nodes (such as Figure 10 (shown) ki for:

[0116]

[0117] In formula (1), P i is the triggered acoustic sensor node, t k is the trigger delay of acquisition, i = 1, 2, ..., Q, k = 1, 2, ..., Q;

[0118] Subsequent steps S6 to S13 are the same as those in Example 1 and are not described in detail here, thereby completing the rapid assessment and positioning of the sound source based on the clustering concept.

[0119] Figure 11 This is the sound source position obtained by grid cluster analysis. It can be seen that the present invention is effective in determining the sound source position by completing grid division.

[0120] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for rapid sound source assessment and positioning based on clustering, characterized in that: The following steps are involved: S1, collecting sound signals through the sound detection module, and triggering time sorting through the sound detection module; S2. Divide the acoustic detection module data obtained in step S1 into several data sets based on the acquisition time in time periods T, and count the number of times the acoustic detection module is triggered in each data set; S3, select the time domain T with the largest total number of times the acoustic detection module is triggered in step S2. a and adjacent time domain T a-1 、T a+1 The acoustic detection module data in the time domain is less than 10ms, and the time domain T a-1 、T a 、T a+1 Merge into a new sound source signal dataset; S4, repeating the above steps S2 to S3 until the sound source signal data of the required time length is obtained and the influence of abnormal signals is eliminated; S5. Let "0" be the unknown sound source center, and the delay difference P between adjacent sensor nodes ki for: In formula (1), P i is the triggered acoustic sensor node, t k is the trigger delay of acquisition, i = 1, 2, ..., Q, k = 1, 2, ..., Q; S6. Divide and cluster the delay differences between adjacent sensor nodes determined in step S5 according to time range to obtain a data set D: In formula (2), v is the speed of sound propagation in the medium, ε is the error of the acoustic signal trigger delay; R is the distance position of the triggered sensor; S7. Arrange the data set D obtained in step S6 in descending order according to the delay size, and divide the data set according to the value of the error ε. The divided data set D' is: D′=[A(x1,y1),B(x2,y2),C(x3,y3),D(x4,y4),E(x5,y5)]; Formula (3) The size of the divided data set D' in the X and Y directions is divided into Δx and Δy respectively; S8. Connect any two points A, B, C, D, or E in step S7 to get the straight line y between points i and j. ij , i≠j, is: S9, take the coordinate point of the grid Y axis in step S7 as the reference point, traverse all the straight lines in formula (4) in step S8, and get the straight line y ij , i≠j, the coordinate that intersects the X axis for: According to formula (5), the grid position where the corresponding straight line passes is determined as: S10, traverse the grids that each straight line obtained by equation (6) passes through, and cluster analysis and count the number of times each grid passes through the straight line, and select the coordinates of the grid with the maximum number of times, which is the estimated location of the sound source S11, using the shortest delay node to perform weighted averaging on all nodes, and adjusting the sound source estimation position in step S10 to improve the sound source estimation accuracy; Extract the minimum delay t1 of D' in equation (3) and select the corresponding reference point A(x1, y1), and the position difference between OA can be obtained as: S12, according to formula (7) the position of the sound source estimated in step S10 Adjust and estimate twice, we can get: In formula (8), l represents the number of points in the dataset D', r x 、r y Indicates the amount of adjustment of the sound source position; S13. Select each data set in formula (3) as a reference point in turn, repeat the above steps S11 to S12, and perform mean processing on the obtained sound source positions to complete the sound source rapid assessment and positioning based on clustering thinking.

2. The method for rapid sound source assessment and positioning based on clustering concept according to claim 1, characterized in that: In step S1, the sound detection module includes a high threshold and a low threshold. The initial value of the high threshold is 120dB, and the initial value of the low threshold is 90dB. When the decibel value of the sound signal exceeds the high threshold or is lower than the low threshold, it is triggered and outputs a low-level signal.

3. The method for rapid sound source assessment and positioning based on clustering concept according to claim 1, characterized in that: In step S4, the required time length is 18 seconds.

4. The method for rapid sound source assessment and positioning based on clustering concept according to claim 1, characterized in that: In step S6, v is the speed of sound propagation in the air, v=340 m / s; the error ε of the acoustic signal trigger delay is determined according to the distance between the sensors.

5. The method for rapid sound source assessment and positioning based on clustering concept according to claim 1, characterized in that: In step S7 , the values ​​of the grid sizes Δx and Δy are determined according to the distance resolution of the sampling nodes.

6. The method for rapid sound source assessment and positioning based on clustering concept according to claim 1, characterized in that: In step S12, the sound source position adjustment amount r x 、r y for:

Images