Video-based vibration location and analysis method

Abstract

A method for locating and analyzing vibration signals based on video, which collects video of the observed target, decomposes the video image by frame and converts it into the frequency domain, and then performs spectrum analysis on the frequency domain image to locate the preliminary vibration area; the vibration direction is determined by performing multi-directional frequency filtering on the time domain signal superimposed in different frames of the frequency domain image and comparing the vibration intensity of the signal; after filtering according to the vibration direction on the preliminary vibration area of ​​the frequency domain image, the amplitude intensity in the filtered frequency domain and spatial domain is compared to obtain the precise vibration area; the phase value of the point in the precise vibration area on the frequency domain image in the precise vibration area is analyzed by power spectrum in the time domain to obtain the final vibration signal. The present invention solves the problems of the prior art that only single-point measurement and local testing can be performed, and non-contact measurement requires the posting of marking points. The above method realizes non-contact, fully automatic measurement without marking points, and the vibration signal frequency analysis result is highly consistent with the accelerometer, while being able to achieve multi-area measurement and accurate vibration positioning.

Description

Technical Field

[0001] The present invention relates to a technology in the field of noise monitoring, in particular to a vibration positioning and analysis method based on video. Background Art

[0002] In the vibration monitoring and fault diagnosis of buildings and production line equipment, it is necessary to locate and analyze vibration signals. Traditional contact measurement methods, such as accelerometers, can only perform contact single-point measurements. Non-contact measurement methods, such as laser vibrometers, are expensive and can only perform point-to-point measurements with limited measurement distances.

[0003] Video analysis of vibration displacement, as a non-contact measurement method, has the advantages of high automation, low labor cost, and wide measurement range compared to traditional contact measurement. Existing video analysis methods mainly monitor marker points and analyze vibration conditions by changing the coordinates of the marker pixel points. This method requires high-frequency instrument monitoring, is greatly affected by environmental factors, and requires high instrument resolution. Summary of the invention

[0004] The present invention aims at the problems that the existing vibration analysis methods are greatly affected by environmental factors, are contact-based, can only perform single-point measurement and local testing, and require marker points for non-contact measurement. A video-based vibration signal positioning and analysis method is proposed. The method can locate, orient and determine the corresponding vibration signal by collecting videos of the observation target, selecting frequency domain features and analyzing time domain signals of the video images. By collecting videos of the observation target, the video images are converted from the spatial domain to the frequency domain, and different directional features are found through directional frequency domain filtering, and then the vibration signal is positioned and analyzed from the time domain, which effectively improves the applicability of the vibration signal.

[0005] The present invention is achieved through the following technical solutions:

[0006] The invention relates to a vibration signal positioning and analysis method based on video. The method comprises the following steps: performing video acquisition on an observation target, decomposing a video image by frame and converting it into a frequency domain, and then performing power spectrum analysis on the frequency domain image to locate a preliminary vibration area; performing multi-directional frequency filtering on a time domain signal superimposed in different frames of the frequency domain image and comparing the vibration intensity of the signal to determine the vibration direction; performing filtering on the preliminary vibration area of ​​the frequency domain image according to the vibration direction, and comparing the amplitude intensity in the filtered frequency domain and space domain to obtain a precise vibration area; and performing power spectrum analysis on the time domain for the phase values ​​of points in the precise vibration area on the frequency domain image in the precise vibration area to obtain a final vibration signal.

[0007] The video acquisition is preferably performed by photographing the vibrating object at a sampling rate that is 2.56 times or more of the vibration frequency.

[0008] The frame-by-frame decomposition refers to decomposing the captured video into single-frame images.

[0009] The conversion refers to: performing image color space decomposition and conversion on a single frame image after frame decomposition, obtaining signal strengths under different color spaces, and then converting the corresponding image from the spatial domain to the frequency domain.

[0010] The color space includes but is not limited to HSV, YIQ, YUV, YCbCr and other color spaces.

[0011] The preliminary vibration region is obtained by performing power spectrum analysis on the frequency domain feature mean within the grid of the frequency domain image and according to the distribution characteristics of different regions at different frequencies.

[0012] The vibration direction is determined by:

[0013] ① Set the vibration direction to [0°, 180°), divide it into multiple vibration direction filters according to the vibration direction accuracy requirements and obtain the corresponding Garbor filter parameters;

[0014] ②According to the parameters obtained in step ①, perform Garbor filtering on the preliminary vibration area to obtain frequency characteristics in different directions;

[0015] ③ Take the time domain signal of the frequency domain characteristics and determine the vibration direction through the maximum value of the signal vibration intensity.

[0016] The precise vibration range is obtained by:

[0017] i) Performing Garbor filtering in a specified direction on the preliminary vibration area to extract frequency domain features;

[0018] ii) taking the amplitude and phase of the frequency domain features, and taking the amplitude intensity as the specified percentage area;

[0019] iii) taking the amplitude specified region for the amplitude and phase of the spatial domain image texture of the vibration region;

[0020] iv) Intersect the areas screened by spatial and frequency domain intensities to obtain the area with stronger signals, i.e., the precise vibration area.

[0021] The final vibration signal is obtained by performing Garbor filtering of a specified vibration direction on the precise vibration area of ​​each frame of the frequency domain image, taking the average of the filtered results, and superimposing the average results of each frame to obtain a signal in the time domain, which is the final vibration signal.

[0022] Technical Effects

[0023] The present invention solves the problems of the prior art that only single-point measurement and local testing are possible, and non-contact measurement requires marking points. The method can realize non-contact, fully automatic measurement without marking points. The vibration signal frequency analysis result is highly consistent with the accelerometer, and multi-region measurement and accurate vibration positioning can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flow chart of the present invention;

[0025] Figure 2 It is a schematic diagram of the system of the present invention;

[0026] Figure 3a and Figure 3b It is a schematic diagram of the vibration table and the speaker;

[0027] Figure 4 a and Figure 4 b is a schematic diagram of an embodiment scenario;

[0028] Figure 5 This is a diagram of the speaker vibration analysis result of the embodiment;

[0029] In the figure: (a) is the speaker vibration map; (b) is the speaker region of interest (ROI); (c) is the 7x7 grid signal diagram of the speaker ROI area; (d) is the enlarged diagram of the signal map in (c); (e) is the 7x7 grid of the speaker ROI area; (f) is the precise positioning of the vibration in the speaker ROI area; (g) is the vibration signal and power spectrum result;

[0030] Figure 6 The frequency domain phase and amplitude diagram of the embodiment;

[0031] In the figure: (a) is the phase image of the frequency domain image at 0°; (b) is the phase image of the frequency domain image at 90°; (c) is the amplitude image of the frequency domain image at 0°; (d) is the amplitude image of the frequency domain image at 90°;

[0032] Figure 7 It is a schematic diagram of the effect of the embodiment. DETAILED DESCRIPTION

[0033] like Figure 1-Figure 5As shown, the present embodiment relates to a video-based vibration positioning and analysis system, comprising: a video acquisition unit, a vibration orientation unit, a vibration positioning unit and a vibration analysis unit, wherein: the video acquisition unit acquires the vibration video of the object to be measured and outputs it to the vibration orientation unit in the format of a single-frame image; the vibration orientation unit converts the image into the frequency domain, performs multi-directional frequency domain feature extraction, and determines the vibration direction by comparison; the vibration positioning unit performs grid drawing and power spectrum analysis according to the frequency domain feature information of the specified direction to determine the vibration area; the vibration analysis unit performs regional signal averaging and power spectrum calculation according to the vibration area and vibration direction information to obtain the vibration signal analysis result.

[0034] This embodiment takes the vibration and positioning analysis experiment of a loudspeaker as an example. The object to be tested is a loudspeaker connected to the . This embodiment is performed as follows Figure 3a The shaking table shown and Figure 3b The specified excitation frequency signal experiment of the speaker shown in the figure sends the excitation voltage of the specified frequency and amplitude as the vibration signal, and the accuracy of vibration positioning and vibration signal analysis can be obtained by analysis.

[0035] like Figure 4 As shown, the video acquisition unit adopts a Basler_acA720-520um high-speed camera, sets the frame rate to 200FPS, the resolution to 720×540, and collects 5s of data.

[0036] like Figure 1 As shown, this is the vibration positioning and analysis method based on the above system in this embodiment.

[0037] Step 1) Decompose the video data by frame to obtain 1000 720×540 images;

[0038] Step 2) To improve the processing speed, select the region of interest Roi = [313, 155, 548, 401], such as Figure 5 As shown in (b), the RGB image is converted to YIQ space and normalized:

[0039] Step 3) Select the Y channel and use Fourier transform to transform the spatial domain into a frequency domain image to obtain a [42, 78, 1000] three-dimensional array I (x, y, t), where: t∈0, 1000);

[0040] Step 4) Perform Garbor filtering in two directions on each frame of the frequency domain image. The filtering results of multiple frames form a signal data Signal with a dimension of [2,42,78,1000] raw , including:

[0041] Step 4.1) Construct a two-dimensional Garbor filter, specifically:

[0042] Where: x and y are the coordinates of the frequency domain image point, λ is the filter scale, θ is the filter direction, ψ is the phase offset, σ is the filter variance, and γ is the filter shape.

[0043] Step 4.2) Perform Garbor filtering in two directions on each frame of the 1000 frequency domain images: Get the original (raw) signal data Signal raw In this embodiment, g 0° (x, y; 5, 0, 0, 2π, 0.5) and g 90° (x, y; 5, 90, 0, 2π, 0.5).

[0044] Step 5) Filter the signal in the time domain raw Perform power spectrum analysis to obtain the frequency characteristics at different frequencies and draw the power spectrum as Figure 5 (c) and Figure 5 (d) as shown in the table below, including:

[0045] Step 5.1) In this embodiment, the grid is set to 7×7 according to the image size, and the area is divided into 49 blocks. The mean value of the frequency domain phase signal in each grid area is calculated as the grid point signal, and the signal data Signal with a dimension of [2, 7, 7, 1000] is obtained. roi7×7 ;

[0046] Step 5.2) Signal data roi7×7 Calculate the cross power spectrum density matrix S between the 49 grid point signals. The dimension of this matrix is ​​related to the ROI and frame rate, that is, [49, 49, 100]. The matrix elements are Where: i, j are the matrix coordinates, i.e., the order of the grid point signal, w∈(0, 100] is the frequency index between 0 and half the sampling rate, R ij is the autocorrelation function of the ith signal and the jth signal, H i is the Fourier transform result of the i-th signal, is the conjugate of the Fourier transform result of the jth signal.

[0047] Step 5.3) Perform SVD decomposition on the cross power spectrum density matrix S at w to obtain the power spectrum of the i-th grid signal point at frequency w. The results of a total of 49 grid signal points (0 to 48) at 100 frequencies (1 to 100 Hz) are used to obtain the distribution of grid points at different frequencies through frequency domain decomposition (FDD) FDD , the dimension is [49, 100].

[0048] Step 6) Compare the frequency distribution signal of different grid points FDD , the peak value at the frequency of 20Hz is larger, and the visualization effect of superimposed light spots is as follows Figure 5 (e)

[0049] Step 7) Among the 49 regions (7×7), determine that the vibration region No. 31 has the largest frequency peak, i.e., the preliminary vibration region Roi init =[95,130,126,163];

[0050] Step 8) In area 31, multi-directional Garbor filtering is performed with the direction [0, 180°), and the bisection approximation principle is used to determine that the vibration direction is the horizontal direction and the vibration angle on the image is around θ=10°;

[0051] Step 9) According to the 31st grid area Roi init = [95, 130, 126, 163], perform Garbor filtering with a vibration angle of θ = 10°, decompose the filtering result to obtain a phase map and an amplitude map, and take the first 30% of the amplitude map as the Roi 频域 ;

[0052] Step 10) According to the 31 grid area Roi init = [95, 130, 126, 163], calculate the texture amplitude map of the Y space of the spatial domain YIQ image, and take the first 30% of the area as the Roi 空域 ;

[0053] Step 11) Take the intersection of the two regions Roi 频域 ∩Roi 空域 , and perform morphological processing on the intersection area, by first corroding and then dilating, and then connecting the domain to take the area with the largest area, and get the following Figure 5 The final region Roi shown in (f) 定位 ;

[0054] Step 12) In the final area Roi 定位 On the surface, perform Garbor filtering when θ = 10° θ (x, y; 5, 0, 0, 2π, 0.5), get the multidimensional array Signal l疨st ;

[0055] Step 13) According to the final area Roi 定位 The coordinates of each point in the time domain are calculated, and the power spectrum results of each point are calculated. The mean of the power spectrum results is taken as the final signal spectrum analysis result. The phase signal and power spectrum analysis results are shown in Figure 5 (g) as shown.

[0056] like Figure 6 and Figure 7 As shown in the figure, by performing frequency response analysis with the accelerometer, setting different excitation frequencies, and comparing the peak values ​​of the analysis results at different frequencies, the effect is shown in the figure. The frequency response curves of the accelerometer and video vibration measurement are consistent, ensuring the stability of the test results.

[0057] Compared with the existing technology, the accuracy of the peak frequency analyzed by this method in different excitation frequency experiments is about 98%; comparing the frequency response curves of the accelerometer and video vibration measurement, the curve correlation is above 95%, and the correlation curve is plotted as follows Figure 7 shown.

[0058]

[0059]

[0060] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The protection scope of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Each implementation scheme within its scope shall be subject to the constraints of the present invention.

Claims

1. A method for locating and analyzing vibration signals based on video, characterized in that: The video of the observed target is collected, and after the video image is decomposed by frame and converted into the frequency domain, the frequency domain image is subjected to power spectrum analysis to locate the preliminary vibration area; the vibration direction is determined by performing multi-directional frequency filtering on the time domain signal superimposed in different frames of the frequency domain image and comparing the vibration intensity of the signal; after filtering according to the vibration direction on the preliminary vibration area of ​​the frequency domain image, the amplitude intensity in the frequency domain and the spatial domain after filtering is compared to obtain the precise vibration area; the power spectrum analysis is performed in the time domain on the phase value of the point in the precise vibration area on the frequency domain image within the precise vibration area to obtain the final vibration signal; The preliminary vibration range is obtained by: 1) Perform Garbor filtering in the horizontal and vertical directions on the frequency domain images of all frames to extract the frequency domain features in both directions; 2) According to the size of the frequency domain image, set a reasonable grid size and divide the frequency domain image into grid areas; 3) Take the average of the frequency domain features in different regions, perform power spectrum analysis on the signals in the time domain, and obtain the distribution characteristics of different regions at different frequencies. The region with the maximum vibration frequency amplitude is taken as the preliminary vibration region; The vibration direction is determined by: ① Set the vibration direction to [0°, 180°), divide it into multiple vibration direction filters according to the vibration direction accuracy requirements and obtain the corresponding Garbor filter parameters; ②According to the parameters obtained in step ①, perform Garbor filtering on the preliminary vibration area to obtain frequency characteristics in different directions; ③ Take the time domain signal of the frequency domain characteristics and determine the vibration direction by the maximum value of the signal vibration intensity; The precise vibration range is obtained by: i) Performing Garbor filtering in a specified direction on the preliminary vibration area to extract frequency domain features; ii) taking the amplitude and phase of the frequency domain features, and taking the amplitude intensity as the specified percentage area; iii) taking the amplitude specified region for the amplitude and phase of the spatial domain image texture of the vibration region; iv) Intersect the areas screened by spatial and frequency domain intensities to obtain the area with stronger signals, i.e., the precise vibration area.

2. The method for locating and analyzing vibration signals based on video according to claim 1, characterized in that: The conversion refers to: performing image color space decomposition and conversion on a single frame image after frame decomposition, obtaining signal strengths under different color spaces, and then converting the corresponding image from the spatial domain to the frequency domain.

3. The method for locating and analyzing vibration signals based on video according to claim 1, characterized in that: The final vibration signal is obtained by performing Garbor filtering of a specified vibration direction on the precise vibration area of ​​each frame of the frequency domain image, taking the average of the filtered results, and superimposing the average results of each frame to obtain a signal in the time domain, which is the final vibration signal.

4. The method for locating and analyzing vibration signals based on video according to claim 1, characterized in that: The multi-directional frequency filtering specifically includes: constructing a two-dimensional Garbor filter, specifically: Where: x and y are the coordinates of the frequency domain image point, λ is the filter scale, θ is the filter direction, ψ is the phase offset, σ is the filter variance, and γ is the filter shape; perform Garbor filtering in two directions on each frame in the frequency domain image: Get the original signal data Signal raw .

5. The method for locating and analyzing vibration signals based on video according to claim 1, characterized in that: The power spectrum analysis specifically includes: dividing the grid according to the image size, calculating the frequency domain phase signal mean in each grid area respectively, and obtaining signal data; calculating the cross power spectrum density matrix between grid point signals, performing SVD decomposition on the cross power spectrum density matrix to obtain the power spectrum, and obtaining the distribution of grid points at different frequencies by frequency domain decomposition.

6. A video-based vibration location and analysis system for implementing the method described in any one of claims 1 to 5, characterized in that: include: A video acquisition unit, a vibration orientation unit, a vibration positioning unit and a vibration analysis unit, wherein: the video acquisition unit acquires the vibration video of the object to be measured and outputs it to the vibration orientation unit in the format of a single frame image; The vibration orientation unit converts the image into the frequency domain, extracts multi-directional frequency domain features, and determines the vibration direction by comparison; the vibration positioning unit performs grid drawing and spectrum analysis based on the frequency domain feature information in the specified direction to determine the vibration area; the vibration analysis unit takes the regional signal average and calculates the power spectrum based on the vibration area and vibration direction information to obtain the vibration signal analysis results.

Images