A Visual Rotational Speed ​​Measurement Method Based on Temporal Feature Analysis and Dual-Domain Refinement Estimation

Abstract

This invention provides a visual rotational speed measurement method based on temporal feature analysis and dual-domain refinement estimation, belonging to the field of computer vision and signal processing technology. Specifically, it includes the following steps: First, a high-frequency image sequence is obtained using a ROI dynamic cropping strategy; second, the images are converted to HSV space to extract multi-channel histograms and calculate the Bavarian coefficients to construct a temporal similarity signal resistant to illumination interference; then, the signal is segmented using a Hamming window, and a temporal interpolation-autocorrelation coupling algorithm is used to obtain time-frequency dual-domain frequency estimates in parallel; subsequently, the dual-domain results are adaptively fused based on confidence levels to obtain refined frequencies, and finally, the instantaneous rotational speed is calculated. This invention effectively overcomes hardware sampling limitations and environmental interference, achieving high-precision sub-Hertz-level rotational speed measurement under non-stationary conditions, providing an accurate time-frequency reference for the condition monitoring and fault diagnosis of rotating machinery.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and signal processing technology, and in particular relates to a visual rotation speed measurement method based on temporal feature analysis and dual-domain thinning estimation. Background Technology

[0002] As the core power source of modern industrial equipment, the monitoring of the operating status of rotating machinery is crucial for ensuring equipment safety. Instantaneous rotational speed is not only a core parameter characterizing the energy state of the rotor system, but also a key physical constraint for fault diagnosis, dynamic balancing analysis, and three-dimensional displacement field reconstruction of rotating machinery. Only by obtaining high-precision instantaneous rotational speed can the width of the motion fuzzy kernel be determined, and thus the true three-dimensional displacement field can be reconstructed from the degraded image sequence.

[0003] Currently, traditional speed measurement methods mainly rely on contact-based devices such as photoelectric encoders and magnetoelectric sensors. However, these contact-based measurement methods are limited by physical intrusion and installation environment constraints, making it difficult to meet the needs of modern high-end equipment for full-field, non-contact intelligent sensing. With the development of machine vision technology, non-contact measurement based on video image analysis has gradually become a research hotspot.

[0004] While machine vision technology offers a new approach to rotational speed measurement, significant technical bottlenecks remain in its transformation into a precision metrology method. On one hand, limited frame rates and exposure mechanisms of industrial cameras result in highly discrete and blurred motion characteristics of high-speed rotating targets over time, making it difficult to capture continuous motion details. On the other hand, complex lighting fluctuations and mechanical vibrations in industrial environments significantly reduce the signal-to-noise ratio (SNR) of image features, rendering traditional grayscale or RGB-space-based feature extraction methods insufficiently robust. Furthermore, at the frequency resolution level, traditional spectral analysis methods based on Fast Fourier Transform (FFT) are limited by the picket fence effect and spectral leakage, making it difficult to capture subtle sub-hertz-level instantaneous rotational speed fluctuations in short time series, resulting in measurement accuracy that fails to meet high-end diagnostic requirements. Therefore, a visual rotational speed measurement method that overcomes hardware sampling limitations, resists environmental interference, and possesses high frequency resolution is urgently needed. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a visual rotation speed measurement method based on temporal feature analysis and dual-domain refinement estimation, in order to solve the problems of poor robustness of existing visual measurement methods under non-uniform illumination interference and low frequency resolution due to limitations in discrete sampling.

[0006] This invention provides a visual rotational speed measurement method based on temporal feature analysis and dual-domain refinement estimation, comprising the following steps:

[0007] S1, acquire a continuous video image sequence of the side of the rotating machinery rotor, and crop the image according to the preset region of interest (ROI) parameters to obtain effective sampled frames;

[0008] S2, convert the continuous sampled frame image sequence into HSV color space, extract the multi-channel color histogram features of each frame image, calculate the color histogram Bach coefficient between adjacent frames, and obtain a one-dimensional temporal similarity signal reflecting the rotor rotation motion.

[0009] S3 standardizes the one-dimensional time-series similarity signal and uses a sliding window and Hamming window function to segment and weight the signal to obtain the weighted signal segment.

[0010] S4. For the weighted signal segment, on the one hand, the autocorrelation function is calculated in the time domain and cubic spline interpolation is performed to obtain the time domain frequency estimate; on the other hand, discrete Fourier transform and spectral resampling interpolation are performed in the frequency domain to obtain the frequency domain frequency estimate.

[0011] S5, calculate the confidence scores of the time-domain frequency estimate and the frequency-domain frequency estimate, and adaptively fuse the time-domain frequency estimate and the frequency-domain frequency estimate based on the confidence scores to obtain the refined frequency;

[0012] S6 calculates the instantaneous rotational speed of the rotating machinery using a rotational speed calculation model based on the refined frequency.

[0013] Preferably, the specific steps of S2 are as follows:

[0014] The first Frame image at pixel position The BGR components at the location are converted to HSV three-channel values, and a normalized histogram is constructed.

[0015] Calculate the single-channel similarity between adjacent frames:

[0016] ;

[0017] in, Indicates channel At any moment The similarity of color distribution between adjacent frames. Indicates color channels, Indicates the time of the current frame. This represents the statistical interval index of the histogram, with values ​​ranging from 0 to 255. Indicates time Time Channel The first in the histogram The quantified value corresponding to each statistical interval Indicates time Time Channel The first in the histogram The quantified value corresponding to each statistical interval Indicates the time interval between adjacent frames;

[0018] The global similarity signal is obtained by weighted averaging of the similarities of the H, S, and V channels. :

[0019] ;

[0020] in, This represents the final global similarity. Indicates the hue channel at time similarity, Indicates the saturation channel at time... similarity, Indicates the brightness channel at time The similarity.

[0021] Preferably, the specific steps of S3 are as follows:

[0022] For the original similarity signal Z-score normalization yields the standard normal distribution signal. ;

[0023] Using Hamming window function w [ n' ]right Perform a sliding cut to obtain the first... each signal segment :

[0024] ;

[0025] in, Indicates the first The signal segments are weighted, and n' represents the index of a local discrete time point within the signal segment. This represents the standardized signal sequence. Indicates the index of the signal segment. This indicates the step size of the sliding window. w [ n' ] represents the Hamming window function. Indicates the window length;

[0026] The Hamming window function w [ n' The expression for ] is:

[0027] ;

[0028] Where 0.54 and 0.46 are the constant coefficients of the Hamming window. Pin' For local discrete time point indexes, This is the window length.

[0029] Preferably, in step S4, the autocorrelation function is calculated in the time domain and cubic spline interpolation is performed to obtain the time-domain frequency estimate. The specific steps are as follows:

[0030] Calculate signal segment Discrete autocorrelation function And the peak-valley detection algorithm is used to locate the integer index of the main periodic peak. ;

[0031] exist Construct piecewise cubic spline interpolation functions within the local neighborhood :

[0032] ;

[0033] in, This represents the continuous autocorrelation function value after interpolation. Represents a continuous time delay variable. Indicates the first Integer delayed index, Indicates the first Integer delayed index, They represent the intervals respectively. The coefficients of the cubic polynomial over;

[0034] By solving the derivative equation Obtain refined non-integer peak delay ;

[0035] according to and sampling frequency Computation of time-domain refinement period Time domain frequency estimate :

[0036] ;

[0037] in, This represents the estimated frequency in the time domain. Indicates the time-domain fine-grained period. This indicates the camera's sampling frequency. This represents the refined non-integer peak delay.

[0038] Preferably, in step S4, performing discrete Fourier transform and spectral resampling interpolation in the frequency domain to obtain the frequency domain frequency estimate involves the following specific steps:

[0039] Perform a Discrete Fourier Transform on the signal segment to obtain the spectrum, and locate the integer index corresponding to the peak value of the amplitude spectrum. ;

[0040] exist Define frequency resampling grid within the neighborhood The original discrete Fourier transform spectrum is reconstructed using cubic B-splines:

[0041] ;

[0042] in, Indicates the resampling frequency point Interpolated spectral amplitude at the location, Indicates the index of the interpolation node. This indicates the total number of interpolation nodes. Indicates the first Interpolation coefficients, Indicates the first A B-spline basis function at frequency The value at;

[0043] For interpolated spectrum Extremum detection is performed to obtain frequency estimates in the frequency domain. .

[0044] Preferably, in step S5, the time-domain confidence level is first calculated according to the normalized peak salience criterion. and frequency domain confidence Then calculate the adaptive weighting factor. ,according to Time-domain frequency estimates and frequency domain frequency estimates Weighted fusion is performed to obtain refined frequencies. .

[0045] Preferably, the specific steps of S6 are as follows:

[0046] Refine frequency Convert to rotational speed in revolutions per minute. :

[0047] ;

[0048] in, The instantaneous rotational speed of rotating machinery is expressed in revolutions per minute (rpm). The refined frequency after dual-domain adaptive fusion is represented in Hertz; 60 is the time unit conversion factor.

[0049] Preferably, the specific process of S5 is as follows:

[0050] Calculate the time-domain confidence level according to the normalized peak prominence criterion. and frequency domain confidence The formula is as follows:

[0051] ;

[0052] ;

[0053] in, Indicates the confidence level in the time domain. Indicates the autocorrelation function at the main peak index The amplitude at that point, This represents the mean value within the neighborhood of the main peak of the autocorrelation function. This represents the standard deviation within the neighborhood of the main peak of the autocorrelation function; Indicates frequency domain confidence. Indicates the amplitude spectrum at the main peak index The amplitude at that point, This represents the mean value within the neighborhood of the main peak of the amplitude spectrum. This represents the standard deviation within the neighborhood of the main peak of the amplitude spectrum;

[0054] Calculate adaptive weighting factors The formula is as follows:

[0055] ;

[0056] in, Indicates the adaptive weighting factor. Indicates the confidence level in the time domain. Indicates frequency domain confidence;

[0057] according to Time-domain frequency estimates and frequency domain frequency estimates Weighted fusion is performed to obtain refined frequencies. :

[0058] ;

[0059] in, This represents the refined frequency after fusion. This represents the estimated frequency in the time domain. This represents the frequency estimate in the frequency domain.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] 1. This invention significantly improves the effective sampling frame rate while ensuring field of view coverage through a dynamic ROI cropping strategy, laying the data foundation for high-frequency sampling;

[0062] 2. This invention adopts a Bach similarity measurement model based on the HSV color space, which effectively improves the robustness of time-series signals to illumination interference through multi-channel feature fusion, and can still maintain high-precision feature extraction under non-uniform illumination.

[0063] 3. The temporal interpolation-autocorrelation coupling algorithm (TIACA) proposed in this invention overcomes the picket fence effect of the traditional FFT method through dual-domain collaborative processing and adaptive fusion, achieves sub-hertz-level frequency estimation resolution, and significantly improves the measurement accuracy of instantaneous rotational speed. Attached Figure Description

[0064] To more clearly illustrate the technical solutions of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the following description is only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 This is a flowchart illustrating the overall steps of the visual rotational speed measurement method based on temporal feature analysis and dual-domain refinement estimation of the present invention.

[0066] Figure 2 This is a schematic diagram of the experimental system principle of the visual rotation speed measurement system in an embodiment of the present invention.

[0067] Figure 3 This is a schematic diagram of the signal flow topology for multi-source data acquisition and control in an embodiment of the present invention.

[0068] Figure 4 This is a schematic diagram illustrating the dynamic cropping principle of the region of interest (ROI) in an embodiment of the present invention.

[0069] Figure 5 This is a schematic diagram illustrating the principle of Bach similarity measurement based on color histogram features in an embodiment of the present invention.

[0070] Figure 6 This is a roadmap for signal segmentation and standardization processing techniques in an embodiment of the present invention.

[0071] Figure 7 This is the overall logic flowchart of the temporal interpolation-autocorrelation coupling algorithm (TIACA) in this embodiment of the invention.

[0072] Figure 8 This is a schematic diagram illustrating the principle of time-domain branching cubic spline interpolation refinement in an embodiment of the present invention.

[0073] Figure 9 This is a schematic diagram illustrating the principle of frequency domain branch spectrum resampling and B-spline reconstruction in an embodiment of the present invention.

[0074] Figure 10 This is a schematic diagram of the instantaneous constant acceleration and variable speed measurement results in an embodiment of the present invention.

[0075] Figure 11 This is a schematic diagram of the instantaneous step speed measurement results in an embodiment of the present invention.

[0076] Figure 12 This is a schematic diagram of the instantaneous gradual speed measurement results in an embodiment of the present invention. Detailed Implementation

[0077] The present invention will be further described below with reference to embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0078] Please see Figure 1 The diagram illustrates a flowchart of a visual rotational speed measurement method based on temporal feature analysis and dual-domain refinement estimation provided by an embodiment of the present invention. The method includes the following steps:

[0079] S1. Build a vision acquisition system to acquire a continuous video image sequence of the side of the rotating machinery rotor, and crop the image according to the preset region of interest (ROI) parameters to obtain effective sampling frames.

[0080] Please see Figure 2 , Figure 2 A schematic diagram of the experimental system is shown. The system mainly consists of a rotor test bench, an industrial camera, a lighting system, and a computing workstation. The industrial camera is mounted on the side of a servo-motor-driven steel rotor to continuously capture the rotor's rotation process. By establishing specific observation positions on the side of the rotor, high frame rate video is used to capture the texture changes on the rotor surface. Please also refer to... Figure 3 , Figure 3 The signal flow topology of multi-source data acquisition is demonstrated. This embodiment uses an S7-1200 PLC as the control core to synchronously trigger camera acquisition and encoder data recording, ensuring time alignment between visual data and the reference true value.

[0081] Further reading Figure 4 , Figure 4 This demonstrates the principle of dynamic ROI cropping. Due to the large amount of data acquired in full-frame imaging, the frame rate is limited. This embodiment, while maintaining the original sensor physical size (3072*2048 pixels), sets the following parameters: Width = 928 pixels, Height = 680 pixels, OffsetX = 1064 pixels, OffsetY = 800 pixels. Figure 4As shown, the imaging area is locked in the region with significant rotor motion texture ( Figure 4 The strategy, which eliminates background interference (specifically region b), reduces the amount of data per frame by 81% and increases the acquisition frame rate to 120 FPS, ensuring high accuracy. High temporal resolution in milliseconds.

[0082] S2 converts the continuous video image sequence into the HSV color space, extracts the multi-channel color histogram features of each frame, calculates the Bach coefficient of the color histogram between adjacent frames, and obtains a one-dimensional temporal similarity signal reflecting the rotor rotation motion.

[0083] Please see Figure 5 , Figure 5 This invention demonstrates the principle of Bach similarity measurement based on color histogram features. To improve illumination robustness, this invention utilizes the HSV model to decouple luminance and chromaticity.

[0084] Specifically, firstly, the first Frame image at pixel position The BGR components at each location are normalized and converted to HSV space. Then, the single-channel similarity between adjacent frames is calculated according to the following formula:

[0085] ;

[0086] in, Indicates channel At any moment The similarity of color distribution between adjacent frames, here It can represent the H (hue), S (saturation), or V (lightness) channels; Indicates the time of the current frame; This represents the statistical interval index of the histogram, with a value range from 0 to 255; Indicates time Time Channel The first in the histogram The quantified values ​​corresponding to each statistical interval; Indicates time Time Channel The first in the histogram The quantified value corresponding to each statistical interval.

[0087] Finally, the weighted average of the similarities of the H, S, and V channels is calculated according to the following formula to obtain the global similarity signal. :

[0088] ;

[0089] in, Let represent the final global similarity, satisfying . .like Figure 5 As shown, this process transforms a two-dimensional video stream into a one-dimensional time series.

[0090] S3 standardizes the one-dimensional time-series similarity signal and uses a sliding window and Hamming window function to segment and weight the signal to obtain the weighted signal segments.

[0091] Please see Figure 6 , Figure 6 This demonstrates the technical approach to signal segmentation and normalization. Limited by discrete sampling, the original similarity signal... It is a non-stationary sequence.

[0092] Specifically, firstly regarding Z-score standardization yields a standard normal distribution signal with zero mean and one variance. Subsequently, a sliding window is used to capture the signal, and a Hamming window function is applied. w [ n' To smooth the signal segment boundaries. The first... each signal segment The calculation formula is as follows:

[0093] ;

[0094] in, Indicates the first A weighted signal segment; n' For local discrete time point indexes; The sliding step size (e.g., 24 frames); The window length (e.g., 120 frames).

[0095] The Hamming window function w[n'] The expression is:

[0096] ;

[0097] The Hamming window, through second-order cosine correction, achieved a sidelobe attenuation of -42.7 dB, effectively suppressing sidelobe reduction. Figure 6 The spectrum leakage phenomenon shown is illustrated.

[0098] S4. Perform time-domain interpolation-autocorrelation coupling processing on the weighted signal segment: on the one hand, calculate the autocorrelation function in the time domain and perform cubic spline interpolation to obtain the time-domain frequency estimate; on the other hand, perform discrete Fourier transform and spectral resampling interpolation in the frequency domain to obtain the frequency-domain frequency estimate.

[0099] Please see Figure 7 , Figure 7The overall logical flow of the Time-Domain Interpolation-Autocorrelation Coupled Algorithm (TIACA) is shown, including the time-domain processing branch (left) and the frequency-domain processing branch (right).

[0100] 1. Time-domain branching:

[0101] Please see Figure 8 , Figure 8 This demonstrates the principle of time-domain cubic spline interpolation. First, the signal segment is calculated. Discrete autocorrelation function And the peak-valley detection algorithm is used to locate the integer index of the main periodic peak. .like Figure 8 As shown, in Construct piecewise cubic spline interpolation functions within the local neighborhood :

[0102] ;

[0103] By solving the derivative equation Obtain refined non-integer peak delay Finally, the time-domain frequency estimate is calculated according to the following formula. :

[0104] .

[0105] 2. Frequency domain branching:

[0106] Please see Figure 9 , Figure 9 This demonstrates the principle of frequency domain resampling and reconstruction. A Discrete Fourier Transform (DFT) is performed on the signal segment to obtain the spectrum $S[j]$, and the integer indices corresponding to the peak values ​​of the amplitude spectrum are located. To overcome the fence effect, such as Figure 9 As shown, in A finely defined frequency resampling grid is defined within the neighborhood. Step length Can be set to The original discrete Fourier transform spectrum is reconstructed using cubic B-splines:

[0107] ;

[0108] For interpolated spectrum Extremum detection is performed to obtain frequency estimates in the frequency domain. .

[0109] S5 calculates the confidence levels of the time-domain frequency estimate and the frequency-domain frequency estimate, and adaptively fuses the time-domain frequency estimate and the frequency-domain frequency estimate based on the confidence levels to obtain the refined frequency.

[0110] It should be noted that, as Figure 7 As shown at the bottom, the adaptive fusion mechanism quantitatively balances the confidence levels of time-domain and frequency-domain estimates.

[0111] Specifically, the time-domain confidence level is calculated according to the normalized peak salience criterion. and frequency domain confidence :

[0112] ;

[0113] ;

[0114] in, and These are the amplitudes of the main peak; and These are the mean and standard deviation within the local neighborhood. Next, the adaptive weighting factor is calculated. Finally, the refined frequency is obtained. :

[0115] ;

[0116] S6 calculates the instantaneous rotational speed of the rotating machinery using a rotational speed calculation model based on the refined frequency.

[0117] Specifically, the refined frequency is obtained using the following formula. Convert to rotational speed in revolutions per minute. :

[0118] ;

[0119] Please see Figures 10 to 12 Measurement results under three instantaneous speed variation conditions are presented. As can be seen, the instantaneous speed curve (red) calculated by the method of this invention highly coincides with the encoder's true value (blue), maintaining excellent dynamic tracking capability during constant acceleration, step change, and gradual change processes, thus verifying the effectiveness of the algorithm.

[0120] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0121] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A visual rotational speed measurement method based on temporal feature analysis and dual-domain refinement estimation, characterized in that, Includes the following processes: S1, acquire a continuous video image sequence of the side of the rotating machinery rotor, and crop the image according to the preset region of interest (ROI) parameters to obtain effective sampled frames; S2, convert the continuous sampled frame image sequence into HSV color space, extract the multi-channel color histogram features of each frame image, calculate the color histogram Bach coefficient between adjacent frames, and obtain a one-dimensional temporal similarity signal reflecting the rotor rotation motion. S3 standardizes the one-dimensional time-series similarity signal and uses a sliding window and Hamming window function to segment and weight the signal to obtain the weighted signal segment. S4. For the weighted signal segment, on the one hand, the autocorrelation function is calculated in the time domain and cubic spline interpolation is performed to obtain the time domain frequency estimate; on the other hand, discrete Fourier transform and spectral resampling interpolation are performed in the frequency domain to obtain the frequency domain frequency estimate. S5, calculate the confidence scores of the time-domain frequency estimate and the frequency-domain frequency estimate, and adaptively fuse the time-domain frequency estimate and the frequency-domain frequency estimate based on the confidence scores to obtain the refined frequency; the specific process is as follows: Calculate the time-domain confidence level according to the normalized peak prominence criterion. and frequency domain confidence The formula is as follows: ； ； in, Indicates the confidence level in the time domain. Indicates the autocorrelation function at the main peak index The amplitude at that point, This represents the mean value within the neighborhood of the main peak of the autocorrelation function. This represents the standard deviation within the neighborhood of the main peak of the autocorrelation function; Indicates frequency domain confidence. Indicates the amplitude spectrum at the main peak index The amplitude at that point, This represents the mean value within the neighborhood of the main peak of the amplitude spectrum. This represents the standard deviation within the neighborhood of the main peak of the amplitude spectrum; Calculate adaptive weighting factors The formula is as follows: ； in, Indicates the adaptive weighting factor. Indicates the confidence level in the time domain. Indicates frequency domain confidence; according to Time-domain frequency estimates and frequency domain frequency estimates Weighted fusion is performed to obtain refined frequencies. : ； in, This represents the refined frequency after fusion. This represents the estimated frequency in the time domain. This represents the frequency estimate in the frequency domain. S6 calculates the instantaneous rotational speed of the rotating machinery using a rotational speed calculation model based on the refined frequency.

2. The visual rotational speed measurement method based on temporal feature analysis and dual-domain refinement estimation as described in claim 1, characterized in that: The specific steps of S2 are as follows: The first Frame image at pixel position The BGR components at the location are converted to HSV three-channel values, and a normalized histogram is constructed. Calculate the single-channel similarity between adjacent frames: ； in, Indicates channel At any moment The similarity of color distribution between adjacent frames. Indicates color channels, Indicates the time of the current frame. This represents the statistical interval index of the histogram, with values ​​ranging from 0 to 255. Indicates time Time Channel The first in the histogram The quantified value corresponding to each statistical interval Indicates time Time Channel The first in the histogram The quantified value corresponding to each statistical interval Indicates the time interval between adjacent frames; The global similarity signal is obtained by weighted averaging of the similarities of the H, S, and V channels. .

3. The visual rotational speed measurement method based on temporal feature analysis and dual-domain refinement estimation as described in claim 1, characterized in that: The specific steps of S3 are as follows: For the original similarity signal Z-score normalization yields the standard normal distribution signal. ; Using Hamming window function w [ n' ]right Perform a sliding cut to obtain the first... each signal segment : ； in, Indicates the first The signal segments are weighted, and n' represents the index of a local discrete time point within the signal segment. This represents the standardized signal sequence. Indicates the index of the signal segment. This indicates the step size of the sliding window. w [ n' ] represents the Hamming window function. Indicates the window length.

4. The visual rotational speed measurement method based on temporal feature analysis and dual-domain refinement estimation as described in claim 1, characterized in that: In step S4, the autocorrelation function is calculated in the time domain and cubic spline interpolation is performed to obtain the time-domain frequency estimate. The specific steps are as follows: Calculate the discrete autocorrelation function of the signal segment and use the peak-valley detection algorithm to locate the integer index of the main periodic peak ; ; exist Construct piecewise cubic spline interpolation functions within the local neighborhood : ； in, This represents the continuous autocorrelation function value after interpolation. Represents a continuous time delay variable. Indicates the first Integer delayed index, Indicates the first Integer delayed index, They represent the intervals respectively. The coefficients of the cubic polynomial over; By solving the derivative equation Obtain refined non-integer peak delay ; according to and sampling frequency Computation of time-domain refinement period Time domain frequency estimate : ； in, This represents the estimated frequency in the time domain. Indicates the time-domain fine-grained period. This indicates the camera's sampling frequency. This represents the refined non-integer peak delay.

5. The visual rotational speed measurement method based on temporal feature analysis and dual-domain refinement estimation as described in claim 1, characterized in that: In step S4, a discrete Fourier transform and spectral resampling interpolation are performed in the frequency domain to obtain a frequency domain frequency estimate. The specific steps are as follows: Perform a Discrete Fourier Transform on the signal segment to obtain the spectrum, and locate the integer index corresponding to the peak value of the amplitude spectrum. ; exist Define frequency resampling grid within the neighborhood The original discrete Fourier transform spectrum is reconstructed using cubic B-splines: ； in, Indicates the resampling frequency point Interpolated spectral amplitude at the location, Indicates the index of the interpolation node. This indicates the total number of interpolation nodes. Indicates the first Interpolation coefficients, Indicates the first A B-spline basis function at frequency The value at; For interpolated spectrum Extremum detection is performed to obtain frequency estimates in the frequency domain. .

6. The visual rotational speed measurement method based on temporal feature analysis and dual-domain refinement estimation as described in claim 1, characterized in that: In step S5, the time-domain confidence level is first calculated according to the normalized peak salience criterion. and frequency domain confidence Then calculate the adaptive weighting factor. ,according to Time-domain frequency estimates and frequency domain frequency estimates Weighted fusion is performed to obtain refined frequencies. .

7. The visual rotational speed measurement method based on temporal feature analysis and dual-domain refinement estimation as described in claim 1, characterized in that: The specific steps of S6 are as follows: Refine frequency Convert to rotational speed in revolutions per minute. : ； in, The instantaneous rotational speed of rotating machinery is expressed in revolutions per minute (rpm). The refined frequency after dual-domain adaptive fusion is represented in Hertz; 60 is the time unit conversion factor.

Images