ES2525517A1 - Method and system for measuring vibrations - Google Patents

Abstract

The invention relates to a method and system for measuring vibrations in an object (4, 9), said method comprising the following steps: capturing an image using a camera (1), said camera (1) having an acquisition frequency of at least double the vibration frequency to be registered selecting a region of interest within the captured image defining a scene (2) quantifying the number of pixels that vary in the scene (2) and determining the existence of a vibration pattern by applying a Fourier transform to the scene (2), as well as calculating the frequency of said vibration pattern.

Description

P201300498

12-09-2014

METHOD AND SYSTEM FOR MEASURING VIBRATIONS

The object of the present invention is a non-invasive method to measure vibrations, that is, a method where no type of contact is used to measure vibrations.

5 imperceptible to the human eye. This method uses a video camera, in such a way that by processing the recorded sequence it is possible to detect movements that, apparently, do not cause a change in the image, as well as to detect its frequency of movement.

10 State of the art

The detection of movements and vibrations is a topic of interest in different scientific fields. Thus, for example, in engineering and architecture, the study of vibrations is used to diagnose the state of health of the structures and forces to which they are

15 submitted.

The measurement of vibrations is possible by acoustic methods. Thus, in certain situations the vibrations produce audible effects and with a microphone it is possible to detect them. However, in most cases this is not possible and it is necessary to go

20 to more sophisticated techniques.

Another technique known in the state of the art is the use of accelerometers. These systems are connected to the surface or structure under study and record the acceleration suffered by it due to dynamic forces or impacts. Despite what

25 widespread use, has several drawbacks, since it is an invasive system that requires its own installation and wiring, as well as an economic cost to take into account [Nassif HN, Gindy M, Davis J. Comparison of laser Doppler vibrometer with contact sensors for monitoring bridge deflection and vibration. NDT & E Int. 2005; 38 (3): 213-128].

30 On the other hand, in many cases the use of accelerometers is not indicated since, for reasons of safety, inaccessibility or integrity of the object to be measured, it is not possible to fix the system and the wiring. On these occasions, the use of non-invasive distance measurement methods is required, such as Doppler interferometry.

35 Methods based on Doppler interferometry use the interference between a beam of

P201300498

12-09-2014

reference and one reflected on the surface to be studied to measure the speed of the displacement. Although it is a very precise method, it requires the use of a laser beam and a very short working distance. Additionally, the cost of this method makes it prohibitive for most applications [Castellini P, Martarelli M, Tomasini EP. Lasser Doppler

5 Vibrometry: Development of advanced solutions answering to technology’s needs. Mechanical systems and signal processing. 2006; 20: 1265-1285].

In recent years, motion and vibration detection techniques based on the detection of Speckle patterns have appeared [Valero, E., V. Micó, Z. Zalevsky, and J.

10 Garcia. "Depth sensing using coherence mapping." Opt. Commun. 283 no. 16 (2010): 3122-3128].

These systems have given rise to 3D detection systems, such as Kinect® [http://es.wikipedia.org/wiki/Kinect] and methods for capturing audio using techniques.

15 optics (as in WO / 2007/043036, US2009 / 096783 or US 2010/226543), as well as applications in biomedicine [A. Shenhav, Z. Brodie, Y. Beiderman, J. Garcia, V. Micó, and

Z. Zalevsky “Optical Sensor for remote estimation of alcohol concentration in blood stream” Opt. Commun. 289, 15, 149-157 (2013)].

Despite the advantages of these systems, these methods have the disadvantage of requiring the projection of a light pattern using low-power lasers, which limits their scope, precision and use.

Artificial vision techniques make it possible to detect and measure movement from a distance without

25 need to use lasers or contact probes. In general, these methods use convolutional pattern recognition, which makes the technique very sensitive to noise and inaccurate. In addition to these drawbacks, the detection of rapid movements

or high frequencies requires high-speed and high-resolution video systems, which significantly increases the final cost of the technique.

However, there are techniques for, from the detection and recognition of patterns determined in an object, to determine its position with great precision, even at sub-pixel resolutions.

35 Although the image resolution is limited by the quality of the optical sensor, in certain cases

P201300498

12-09-2014

conditions it is possible to obtain information above the limit of spatial resolution. These techniques are known as sub-pixel techniques and take advantage of information about the object that is known a priori in order to increase the capabilities of the system in detecting objects.

5

The most common technique for obtaining sub-pixel resolution is to introduce a target of known geometry into the scene. By recognizing shapes and adjusting the detected geometry to the real one, it is possible to locate the object with precision of up to 0.02 pixels [D. Mas, J. Espinosa, A. B. Roig, B. Ferrer, J. Pérez, C. Illueca “Measurement of

10 wide frequency range structural microvibrations with a pocket digital camera and sub-pixel techniques ”Applied Optics, 51, 14, 2664-2671 (2011)].

However, to apply this technique it is essential to adhere to the vibrating object a target foreign to the scene, which can sometimes be complicated or even unfeasible.

15 The main problem with standard video cameras is that the temporal resolution (frames per second -fps) is usually limited to 30-50 fps, which, according to Nyquist's criteria, does not allow its use to measure frequencies above 15-25 Hz. A vibration recording system should have an acquisition capacity above

20,500 fps in order to accommodate a wide range of applications (civil structures tend to vibrate below 200 Hz and the fundamental frequency of the human voice is around 250 Hz).

At this point, the limitations of electronics must be taken into account. Data flow

The 25 per second that a video camera can process and store depends on both the size of the frame (pixels) and the number of frames per second that the camera is capable of capturing. For a constant stream of data, increasing the acquisition speed means decreasing the frame size and vice versa.

30 Thus, the higher the acquisition speed of the camera, the lower its spatial resolution will be and therefore it will provide a lower image quality. As a consequence, increasing the data flow means increasing the price of the device significantly, which can make the final solution unprofitable for most applications.

35

P201300498

12-09-2014

Description of the invention

The object of the present invention is a non-invasive vibration measurement method, which does not require external elements, such as lasers or Speckle patterns and at a cost

5 reduced. For this, the present invention is based on an analysis of the lighting changes produced by a moving object, from which it is determined whether it vibrates and, consequently, its vibration frequency.

Preferably, the method of the invention comprises:

10 capturing a succession of images of said object with a camera, where said camera has an acquisition frequency of at least twice the frequency of the vibrations to be measured;

selecting a region of interest within the succession of images captured defining a scene; 15 identifying a plurality of pixels whose luminance varies in the scene;

performing a multilevel thresholding of the sequence within the region of interest, said thresholding being defined according to a predetermined luminance level, so that a sequence of binary images is obtained;

sum the active pixels in each image and represent the variation of the number of 20 pixels as a function of time; performing a Fourier transform on the representation obtained for each binary luminance level; compose the Fourier transforms obtained for each level as a scene characterization signal; Obtain the frequency of vibration of the object as the one with the greatest amplitude in the scene characterization signal.

The method object of the invention allows the measurement of vibrations in objects of a wide range of frequencies by capturing and subsequent processing of a sequence of

30 video. The invention allows measurements to be made from sequences with very low image quality and without the need for additional lighting systems or targets.

Thus, with respect to existing systems, the invention is applied without contact and, logically, without altering the structure or surface under study. Also, it is not

35 necessary a distance close to the object under study, since the distance to said object is

P201300498

12-09-2014

function of the video system used, allowing the study of inaccessible or unsafe structures. Another technical problem that is solved, derived from the use of image analysis, compared to other systems, is that it avoids the problem of signal attenuation as a function of distance.

5

On the other hand, the invention does not require special lighting and / or specific instrumentation, which also enables its implementation in low-cost cameras.

Apart from measurement, the invention allows direct visualization of the object and more analysis

10 sophisticated algorithms that can be implemented in the camera itself.

Finally, it should be noted that, as it is based on images, its use is possible for both macroscopic and microscopic objects.

Throughout the description and claims the word "comprise" and its variants are not intended to exclude other technical characteristics, additives, components or steps. For those skilled in the art, other objects, advantages and characteristics of the invention will emerge in part from the description and in part from the practice of the invention. The

The following examples and drawings are provided by way of illustration, and are not intended to restrict the present invention. Furthermore, the present invention covers all the possible combinations of particular and preferred embodiments indicated herein.

Brief description of the figures

A series of drawings that help to better understand the invention and that are expressly related to an embodiment of said invention that is presented as a non-limiting example thereof, will now be described very briefly.

FIG. 1 shows a diagram of the setup required to capture a scene.

FIG.2 shows three images where Fig.2a shows a captured scene and adaptation to the sensor grid in the camera. In Fig.2b the image detected by the digital sensor is shown. Finally, Fig. 2c shows the image detected by

35 the digital sensor after an original image shift of 0.25 pixels.

P201300498

12-09-2014

FIG. 3 shows four images where Fig. 3a shows an object captured with four levels of gray (0-3) on a uniform background. In Fig.3b an image with values greater than 0 is shown. In Fig.3c an image with values greater than 1 is shown

5 and, finally, in Fig. 3d an image with values greater than 2 is shown.

FIG. 4 shows an example of a gray scale region of interest within an image captured in a practical example to determine the vibration of a 440 Hz tuning fork.

FIG. 5 shows a sequence of thresholds from level 32 to 255 every 32 levels of the image in figure 4.

FIG. 6 shows the product of the Fourier transforms obtained from the variation of pixels in each threshold sequence of figure 5.

FIG. 7 shows the product of the Fourier transforms obtained from the variation of pixels in each threshold sequence for an example of a loudspeaker vibrating at 290 Hz.

twenty

Presentation of a detailed embodiment of the invention

The present invention is based on sub-pixel techniques where high precision motion detection is possible. For a better understanding of the invention,

25 considers a point object and it is detected that this object has moved when its detection on the detector matrix of the camera passes from one pixel to another pixel. Thus, in the image plane, the nominal resolution of the camera will be 1 pixel.

The object to be detected is considered below to be smaller than one pixel. In that

In this case, the camera will not be able to solve it and the entire pixel provides a response regardless of where the object is within that pixel. If the object below the pixel size (sub-pixel object) can be found to one side of the useful area of the camera's optical sensor, a small displacement will suffice to move to another area. Thus, motion detection does not require the

35 a full pixel shift.

P201300498

12-09-2014

The displaced point moves within the new pixel without the movement being detected and only when it changes pixel again, a new movement is detected. This causes that the detection of movement is not continuous, but jumps take place.

5

To avoid this problem, it must be ensured that a second point performs the detection once the first is amortized. With two points in different pixels, the first of them located in the right half of the sensor and the second in the left half, jumps of half a pixel can be detected. In general, you need as many additional points as

10 pixel fractions are to be detected.

In [D. Mas, B. Ferrer, J. T. Sheridan, J. Espinosa “Resolution limits to object tracking in subpixel accuracy” Opt. Letters, 53, 23 4877-4879 (2012)] it is theoretically demonstrated that the detection capacity of this system is of the order of the inverse of the number of

15 total pixels in the detector, being able to reach micro-pixel resolutions.

The invention that is presented deals with the use of the exposed theory for the measurement of vibrations. As has been explained, objects that allow sub-pixel resolution have specific properties and are not natural objects. Therefore, the use of this technique to its full potential requires the design of special targets that should adhere to the subject under study. However, the invention uses everyday objects, at the cost of losing resolution with respect to the theoretical limit, as shown in Figures 1 and

2.

More specifically, the invention comprises a first capture of an image of the object

(4) by means of a camera (1) whose acquisition frequency must be at least twice the vibration frequency to be recorded.

In the present invention, scene (2) is understood to be the image of a region of interest of a certain object (4) captured by the camera (1).

The objective (3) is variable, depending on the distance of the object (4) with respect to the camera (1), all in order to maximize the size of the object (4) or the area of interest on the sensor. the chamber (1).

35

5

10

fifteen

twenty

25

30

35

P201300498

12-09-2014

Real objects (4), in general, have complex shapes, which do not conform to the grid structure (5) of the digital camera sensor (generally CCD and / or CMOS). Thus, it is possible to find pixels completely occupied by the object and pixels in which the area is shared by the object and the background (6).

If, simplifying, we admit that the sensor will only give a white or black response, the contour pixel will be assigned to one of those values depending on whether the dominant area on the pixel belongs to the object or the background and the final image will no longer have a smooth but trimmed profile (7). If the object moves slightly, even by an amount less than one pixel, the proportion of areas in some of the contour pixels will change, resulting in a slight change in profile (8). This small change will only affect a few pixels but it will be enough to be detected. Displacements greater than one pixel can be easily detected and are not the subject of this patent.

Depending on the complexity of the object, a small movement will cause a change in the state of different pixels. Although simple movement is easy to detect, in a complete scene it is difficult to control which pixels have changed and which direction, so the proposed method cannot fully track the object.

However, if the object vibrates, it will cause a pixel shift in a definite pattern, turning certain pixels on and off at regular intervals. Simply by counting the number of pixels that change and performing a Fourier analysis of the captured image, it is possible to detect the existence of said vibration pattern and determine its frequency.

In any case, the objects are not pure black and white, nor do they present well-defined contours, but the transitions are smooth. Assuming a gray-toned image, a movement will be seen as a slight change in the luminance of some pixels due to a change in the dominant region in the detector area. These changes are often subtle and difficult to appreciate, making it difficult to directly apply the method explained above. In addition, the movement of the image can be subtle and not affect the outline of the object but the shadows or light diffused by the object, producing small alterations.

In any case, a movement will translate as changes in the distribution of

P201300498

12-09-2014

luminances in the object to be quantified. We can, however, transform the object into a binary object through multilevel thresholds.

In figure 3 a simple object is considered (9). This object has been considered to only

5 has 4 levels of gray. From the object we can obtain the different thresholds by determining the pixels whose value is greater than zero (10), greater than one (11) and greater than two (12). These profiles are binary images to which the method explained above is applied.

10 The idea is easily generalizable to a scene with N levels, from which N-1 binary sequences are obtained to analyze. Assuming that the object under study will move homogeneously, all levels will change in a similar way, so that, in general, it will not be necessary to study all levels.

15 A possible limitation of the method is that any variation in the scene due to movement, shadows or change in lighting will produce a change in the different levels of gray detected and, therefore, it will not be easy to discriminate which changes are due to movements and which are due to other causes.

20 As indicated, a vibrating object will periodically alter its luminance. These periodic alterations will be reflected in each binarized sequence and can be identified on the noise by means of a Fourier analysis. Additionally, having multiple binary sequences provides redundant information that can be used to enhance detection. The mean or the product of the transforms obtained

25 at each level will reinforce detection peaks and cancel image noise.

Example of measurement on a tuning fork at 440 Hz

The method explained has been used to determine the vibration of a 440 Hz tuning fork.

30 The camera used was a Casio EXFH100 recording at 1000 fps with a resolution of 224x64 pixels (width by height) located 75 cm from the fretboard. Figure 3 shows the tuning fork, that is, the object (9), on a uniform background. To illuminate the laboratory ambient light and a 50 W halogen lamp connected to a direct current transformer are used to avoid light oscillations due to alternating current (100 Hz).

35 The distance between the lamp and the tuning fork is 50 cm with an illumination angle of 45º

P201300498

12-09-2014

with respect to the camera-object axis.

After hitting the tuning fork with a rubber hammer, the scene was filmed. The first three seconds of filming are discarded to minimize the effect of camera shake produced when the shot is pressed. A sequence of 1 second (1000 frames) has been taken for analysis.

From the sequence, the study of vibration has focused on the upper part of a rod. A 15 x 12 pixel rectangle has been selected to which the method has been applied

10 described.

The procedure followed once the area of interest was established has been as follows: the image has been converted to gray levels and the dynamic range of the image has been extended by redefining black and white and linearly transforming the images.

15 luminances (figure 4). Next, 8 threshold levels have been established, every 32 gray levels, (figure 5) and the variation in the number of white pixels with respect to the first frame has been analyzed.

The variation obtained has been studied using Fourier analysis. The transformed

20 obtained for each thresholding have been multiplied in order to reinforce the peaks and cancel the noise. Figure 6 shows the product of the Fourier transforms with the frequency peak obtained, with an error of ± 1 Hz. It can be clearly seen that the method detects the frequency of movement of the tuning fork with great precision, providing a result of 441 Hz.

25

Example of measurement on a speaker at 290 Hz

Following the same method, the vibration of a loudspeaker membrane vibrating at 290 Hz has been studied. The distance between the chamber and the loudspeaker is 75 cm. To illuminate the

30 object has been used a 50 W halogen lamp connected to a direct current transformer. Since the loudspeaker membrane is matte-black in color, the luminaire was placed 50 cm in front of the object, forming an angle of 15º with respect to the imaginary line that joins the object and the camera.

35 As before a sequence of several seconds was taken at 1000 fps and 224x64

P201300498

12-09-2014

pixel resolution. The first three seconds were discarded by camera shake with 1000 frames being analyzed. The loudspeaker volume was set below 35 dB, so that the emitted sound was below the ambient noise.

5 A high contrast region of interest has been analyzed from the previous sequence in order to maximize the number of levels and thus ensure optimal detection.

Following exactly the process described for the tuning fork, the Fourier transform of the number of pixels at each threshold has been obtained. The product of the eight transforms 10 obtained is shown in figure 7, where it can be seen that the method provides satisfactory results with an error of ± 1 Hz.

The present method can be implemented in different ways, either on a personal computer or on a processor designed for this purpose and integrated into the camera itself. In any case, the system comprises at least one camera, one or more processors, a

or more memories and one or more programs in which the program (s) are stored in memory and configured to be executed by, at least, the processor (s), the programs including instructions for executing the described method.

20 Finally, the method has multiple possible uses without modifying its essential characteristics, such as:

-Measurement of small vibrations in small objects. If the object is small in size and irregular in shape, it is difficult to project a 25 speckle pattern or use contact methods. -Measurement of vibrations in civil structures. It is particularly interesting for measuring inaccessible structures, or whose proximity implies danger. -Interception of mobile conversations by reading the vibration of the terminal. 30 -Measurement of periodic variations in astronomical or satellite images, such as object tracking, pulsations, variations in brightness / reflectivity. -Mapping of vibrations of a scene through a single sequence and zonal analysis. -Measurement of vital signs (pulse and respiration) through webcam cameras. 35 -Measurement of periodic variations in biological images.

P201300498

12-09-2014

-

Detection of changes in low resolution images, such as resonances,

X-rays.

-

Monitoring of ultra-slow periodic variations by combining the

explained technique and time-lapse photography.

5

In general, since the method is based on image analysis, it can be applied to any sequence or video stack from any source where periodic change is suspected. The detection limit frequency will depend on the temporal sampling frequency, which may be a few tenths or hundredths of a Hertz (time-lapse photography) or

10 of several hundred Hertz (high speed video).

Claims (1)

P201300498 12-09-2014 1 - Method for measuring vibrations in an object (4,9) comprising: capturing a sequence of images of said object (4,9) with a camera (1), 5 where said camera (1) has an acquisition frequency of at least twice the frequency of the vibrations to be measured; selecting a region of interest within the sequence of captured images defining a scene (2); characterized in that it further comprises the steps of: identifying a plurality of pixels whose luminance varies in the scene (2); performing a multilevel thresholding of the sequence within the region of interest, said thresholding being defined according to a predetermined luminance level, so that a sequence of binary images is obtained; sum the active pixels in each image and represent the variation of the number of 15 pixels as a function of time; performing a Fourier transform on the representation obtained for each binary luminance level; compose the Fourier transforms obtained for each level as a characterization signal of the scene (2); Obtain the vibration frequency of the object (4.9) as the one with the greatest amplitude in the scene characterization signal (2). 2 - Method according to claim 3, where the composition of the Fourier transforms obtained for each level is carried out by calculating the mean of the applied Fourier transforms, and / or the multiplication of said transformed into each other. 3 - System for measuring vibrations in an object (4,9) comprising, at least one camera (1), at least one processor, at least one memory and at least one program 30 stored in memory and configured to be executed by the processor, wherein the program comprises instructions for executing the method according to any of claims 1-2. 14