Vision laser displacement detection system and method based on multi-mode fusion and intelligent calibration

Abstract

The application discloses a visual laser displacement detection system and method based on multi-mode fusion and intelligent calibration, and relates to the technical field of non-contact displacement detection.The system comprises an optical geometric constraint calibration module, a multi-mode detection module, a frequency domain analysis module, a network parameter mapping module, a topology analysis module and a fusion output module.The system maintains imaging stability through optical geometric constraints; signals are synchronously collected by using a vision sensor and a laser sensor; dynamic response differences are extracted through frequency domain analysis; the differences are mapped into a network parameter group and a transfer function model is established; finally, the model is used to fuse and compensate double-channel signals, and a displacement detection result compensated for high-frequency distortion is generated.The application effectively solves the fusion error problem caused by the dynamic characteristic mismatch of the vision sensor and the laser sensor, and realizes the complementary advantages of a large range and high precision and high dynamic response.

Description

Technical Field

[0001] This invention relates to the field of non-contact displacement detection technology, and in particular to a visual laser displacement detection system and method based on multi-mode fusion and intelligent calibration. Background Technology

[0002] In the fields of modern precision manufacturing, automated inspection, and metrology, non-contact displacement detection technology plays a crucial role. Among them, visual measurement methods and laser measurement methods are widely used due to their unique advantages.

[0003] Visual measurement methods, such as imaging systems based on high-resolution CMOS or CCD sensors, have the advantages of a large measurement range and the ability to acquire two-dimensional and even three-dimensional spatial information. However, the dynamic response speed of these methods is usually limited by the exposure and readout time of the image sensor, resulting in significant phase lag in high-speed or dynamic measurement scenarios. They are also susceptible to ambient lighting interference, and there is a bottleneck in improving absolute accuracy.

[0004] Laser measurement methods, such as those based on position-sensitive detectors (PSDs) or laser interferometers, excel in their extremely high resolution, fast response speed, and excellent resistance to ambient light interference. PSDs, in particular, can provide continuous, analog position signals with a response frequency far exceeding that of typical image sensors. However, these methods suffer from drawbacks such as a typically small effective measurement range and the high cost of large-size, high-precision two-dimensional detectors, limiting their widespread adoption in large-scale scanning applications.

[0005] To balance the demands for wide coverage and high precision, existing technologies have attempted to combine visual and laser sensors. However, such multi-sensor systems face a fundamental technical challenge: due to their vastly different physical principles and structures, visual and laser sensors exhibit significant differences in their inherent dynamic response characteristics. Specifically, their responses to the same dynamic displacement excitation are inconsistent in both time (phase) and amplitude, especially at high frequencies. This dynamic response mismatch leads to distortion in the fused signal, introducing additional measurement errors and making it difficult for simple data stitching or weighted averaging methods to realize the expected advantages of multi-sensor fusion.

[0006] Furthermore, existing fusion technologies mostly focus on data calibration under static or quasi-static conditions, lacking online modeling and compensation mechanisms for the dynamic characteristics of sensors. During long-term system operation, due to factors such as temperature drift, mechanical stress relaxation, or optical component aging, the characteristics of the sensors may slowly change, further exacerbating the dynamic response mismatch problem and causing the system accuracy to gradually deteriorate over time.

[0007] Therefore, there is a need in the field for a displacement detection technology and system that can fundamentally understand and compensate for the dynamic response differences between visual and laser sensors, and maintain long-term stability and high accuracy during system operation. Summary of the Invention

[0008] To address the above problems, this invention proposes a visual laser displacement detection system and method based on multi-mode fusion and intelligent calibration.

[0009] The present invention achieves the above objectives through the following technical solutions:

[0010] A visual laser displacement detection system based on multi-mode fusion and intelligent calibration, the system comprising:

[0011] The optical geometry constraint calibration module is used to configure the angle matching relationship between the laser incident optical axis of the laser sensor and the imaging plane of the vision sensor according to the Scheimpflug geometry, so as to maintain the stability of the imaging spot shape during the measurement process.

[0012] The multi-mode detection module is used to acquire visual signals and laser signals from the target surface through a visual sensor and a laser sensor, respectively.

[0013] The frequency domain analysis module is used to perform frequency domain transformation on the sampling sequences of the visual signal and the laser signal, and analyze the frequency domain response differences between the visual channel and the laser channel.

[0014] The network parameter mapping module is used to map the frequency domain analysis processed visual signal and laser signal into a first network parameter group characterizing the transmission characteristics of the visual sensor and a second network parameter group characterizing the transmission characteristics of the laser sensor, respectively, based on the frequency domain response difference.

[0015] The topology analysis module is used to perform network topology analysis on the first network parameter group and the second network parameter group to identify the equivalent relationship of dynamic response speed between the visual sensor and the laser sensor, and to establish a transfer function model from displacement excitation to output signal.

[0016] The fusion output module is used to perform fusion compensation on the output signals of the visual channel and the laser channel using the transfer function model, and generate a multi-mode fusion displacement detection result after high-frequency distortion compensation.

[0017] As a preferred embodiment of the present invention, the multi-mode detection module includes:

[0018] The vision sensor, using a high-resolution CMOS camera, is used to achieve large-range two-dimensional light spot position measurement, with a measurement range of 100mm×100mm and a resolution of not less than 0.05mm.

[0019] The laser sensor employs a two-dimensional PSD based on the lateral photoelectric effect to achieve small-range two-dimensional spot position measurement, with a measurement range no greater than 10 mm and a resolution no less than 6. ;

[0020] The visual sensor and laser sensor respectively collect visual signals and laser signals from the target surface, and the displacement excitation of the light spot is realized by an electronically controlled two-dimensional motion stage;

[0021] The signal synchronization and alignment unit is used to synchronize the acquisition of the visual channel and the laser channel, and to achieve time series alignment of the output signals of the two channels at the sampling point level.

[0022] As a preferred embodiment of the present invention, the frequency domain analysis module includes:

[0023] The signal decoupling unit is used to perform variational mode decomposition on the sampling sequences of the visual signal and the laser signal respectively, and extract the IMF component sets of the visual channel and the laser channel respectively.

[0024] The dominant modality screening unit is used to screen out the visual target IMF component and the laser target IMF component from the IMF component sets of the visual channel and the laser channel respectively, based on the center frequency and energy entropy of each IMF component.

[0025] The cross-power spectrum analysis unit is used to calculate the cross-power spectrum between the visual target IMF component and the laser target IMF component, and to extract the phase difference spectrum and amplitude ratio spectrum between the two channels from the cross-power spectrum.

[0026] The coherence analysis unit is used to calculate the coherence function of the two channels in the effective frequency band based on the cross power spectrum, the autopower spectrum of the visual target IMF component and the autopower spectrum of the laser target IMF component.

[0027] The dominant resonance band identification unit is used to identify the dominant resonance band of the dynamic response coupling between the visual sensor and the laser sensor based on the peak value of the coherence function and in combination with the amplitude peak value of the cross power spectrum and the linearity of the phase difference spectrum.

[0028] The difference feature synthesis unit is used to perform in-band weighted fusion of the phase difference spectrum and the amplitude ratio spectrum within the dominant resonance frequency band to generate a comprehensive frequency domain response difference feature vector.

[0029] As a preferred embodiment of the present invention, the network parameter mapping module includes:

[0030] The state-space modeling unit is used to construct a state-space equation with the dynamic characteristics of the visual channel and the laser channel as internal state variables, using the frequency domain response difference feature vector as the system observation value.

[0031] The physical constraint feature encoding unit is used to encode the normalized frequency domain response difference feature vector into a composite input tensor containing optical transmission delay parameters, system inertial parameters, and dynamic damping parameters, based on the optical transmission models of the visual sensor and the laser sensor.

[0032] The parameterized mapping unit is used to map the composite input tensor into a first network parameter set representing the transmission characteristics of the visual sensor and a second network parameter set representing the transmission characteristics of the laser sensor, based on the state space equation and by introducing a Tikhonov regularized subspace system identification algorithm. Both the first network parameter set and the second network parameter set include a state matrix, an input matrix, an observation matrix, and a direct transfer matrix.

[0033] As a preferred embodiment of the present invention, the network parameter mapping module performs parameterized mapping based on the following state-space equation:

[0034] ;

[0035] In the formula, , The system is respectively in time, The state variables at any given time correspond to the dynamic characteristics of the visual channel and the laser channel; , They are respectively time, The time-based system input corresponds to the composite input tensor. for The system observations at time 1, corresponding to the frequency domain response difference feature vector; These are process noise and observation noise, respectively. These are the state matrix, input matrix, observation matrix, and direct transfer matrix, respectively.

[0036] The frequency domain difference feature correction term is defined as: ,in for The frequency domain response difference feature vector at time step; For adaptive regularization parameters; For the reason The frequency weighting matrix for calculating the dominant resonance band at time t; This is the steady-state mean of the comprehensive frequency domain difference eigenvector.

[0037] As a preferred embodiment of the present invention, the topology analysis module establishes the transfer function model in the following manner:

[0038] Based on the state matrix, input matrix, observation matrix, and direct transfer matrix, a nodal impedance equivalent analysis is performed. The eigenvalues ​​in the state matrix are mapped to the equivalent complex impedances of the network topology nodes. The signal injection relationship from displacement excitation to the equivalent complex impedance network is determined based on the input matrix. The signal acquisition relationship from the equivalent complex impedance network to the output signal is determined based on the observation matrix.

[0039] Based on the equivalent complex impedance, the signal injection relationship, and the signal acquisition relationship, an equivalent RLC network model for the dynamic response speed between the visual sensor and the laser sensor is constructed.

[0040] The equivalent RLC network model is transformed into a transfer function model from displacement excitation to output signal by Laplace transform, wherein the direct transfer matrix represents the feedforward path from input to output in the Laplace transform and directly contributes to the transfer function model.

[0041] As a preferred embodiment of the present invention, the network parameter mapping module further includes:

[0042] The model reduction and verification unit is used to balance the first network parameter set and the second network parameter set and reduce the model order, and to confirm the dominant state by calculating the Hankel singular value, so as to extract the reduced network parameter set that characterizes the core dynamic characteristics of the visual sensor and the laser sensor.

[0043] The meta-learning parameter self-updating unit is used to adaptively fine-tune the parameters of the state-space equation based on real-time collected sample data when the external environment or measurement conditions change, thereby achieving online recalibration under small sample conditions. The parameters of the state-space equation include the state matrix, input matrix, observation matrix, direct transfer matrix, and adaptive regularization parameters.

[0044] At this point, the topology analysis module uses the reduced-order network parameter set instead of the original network parameter set to perform equivalent node impedance analysis.

[0045] As a preferred embodiment of the present invention, the fusion output module includes:

[0046] The channel inverse compensation unit is used to construct inverse models of the visual channel and the laser channel based on the transfer function model, and to compensate for phase lag and high-frequency attenuation of the output signals of the visual channel and the laser channel.

[0047] The frequency domain adaptive fusion unit is used to construct a frequency adaptive weighting function based on the frequency domain response characteristics of the visual channel and the laser channel as represented by the transfer function model within the dominant resonance frequency band, and to perform frequency domain weighted fusion on the two channel signals after inverse compensation. The weight of the visual channel is increased in the frequency band below the crossover frequency, and the weight of the laser channel is increased in the frequency band above the crossover frequency.

[0048] The time-domain reconstruction unit is used to reconstruct the frequency-domain weighted fusion result into a time-domain signal through inverse Fourier transform, generating an initial fused shift signal;

[0049] The high-frequency distortion compensation unit is used to design an adaptive notch filter based on the system resonance characteristics identified by the equivalent RLC network model to selectively filter the resonance distortion component in the initial fused displacement signal, thereby generating a multi-mode fused displacement detection result compensated for high-frequency distortion.

[0050] As a preferred embodiment of the present invention, the system further includes an adaptive fusion feedback module, which is used to receive the fusion residual signal generated by the fusion output module, generate parameter adjustment instructions based on the statistical characteristics and trends of the fusion residual signal, and feed them back to the channel inverse compensation unit and the frequency domain adaptive fusion unit, for dynamically adjusting the parameters of the inverse model and the shape and cross frequency of the frequency adaptive weighting function.

[0051] The fusion residual signal is the difference between the initial fusion displacement signal and the multi-mode fusion displacement detection result.

[0052] A visual laser displacement detection method based on multi-mode fusion and intelligent calibration, the method comprising:

[0053] Based on the Scheimpflug geometry, the angle matching relationship between the laser incident optical axis of the laser sensor and the imaging plane of the vision sensor is configured, and the visual signal and laser signal of the target surface are collected by the vision sensor and the laser sensor respectively.

[0054] The sampling sequences of the visual signal and the laser signal are subjected to frequency domain transformation to analyze the difference in frequency domain response between the visual channel and the laser channel;

[0055] Based on the frequency domain response difference, the visual signal and laser signal after frequency domain analysis and processing are respectively mapped to a first network parameter set characterizing the transmission characteristics of the visual sensor and a second network parameter set characterizing the transmission characteristics of the laser sensor.

[0056] Network topology analysis is performed on the first network parameter group and the second network parameter group to identify the equivalent relationship of dynamic response speed between the visual sensor and the laser sensor, and to establish a transfer function model from displacement excitation to output signal.

[0057] The output signals of the visual channel and the laser channel are fused and compensated using the transfer function model to generate a multi-mode fused displacement detection result compensated for high-frequency distortion.

[0058] The beneficial effects of this invention are as follows: By introducing optical geometric constraint calibration and configuring according to the Scheimpflug principle, the morphology of the imaging spot of the visual sensor is stable and the focus is clear under non-ideal imaging conditions, providing a reliable and consistent signal source for subsequent high-precision fusion processing, and eliminating the systematic error introduced by spot distortion at the physical level. The system can simultaneously acquire large-range, high-resolution information from the visual sensor and small-range, high-speed, high-precision information from the laser sensor, providing a complete data foundation for high-precision displacement measurement over a wide bandwidth. Deep decoupling and analysis of the dual-channel signals can accurately identify the dominant resonant frequency band of the dynamic response coupling between the two, and extract quantitative frequency domain response difference characteristics, transforming the abstract problem of response inconsistency into a calculable and analyzable mathematical model input. The frequency domain difference is mapped to a set of network parameters characterizing the sensor's transmission characteristics, and a transfer function model from displacement excitation to output signal is further established. This model not only reveals the equivalent relationship of dynamic response speed between the visual and laser sensors, but also describes the dynamic behavior of the sensors with a precise mathematical model, laying a theoretical foundation for subsequent accurate compensation. By performing inverse compensation and frequency domain adaptive fusion on the dual-channel signals, phase lag and high-frequency attenuation are effectively corrected, and high-frequency distortion caused by system resonance is specifically suppressed. Finally, a displacement detection result that maintains high accuracy and high stability across a wide frequency band is output. Attached Figure Description

[0059] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0060] Figure 1 This is a schematic diagram of the modular structure of the system of the present invention;

[0061] Figure 2 This is a flowchart of the method in an embodiment of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 described embodiments of the present invention are within the scope of protection of the present invention.

[0063] like Figure 1 As shown, this is an embodiment of the present invention, which provides a visual laser displacement detection system based on multi-mode fusion and intelligent calibration, including an optical geometric constraint calibration module, a multi-mode detection module, a frequency domain analysis module, a network parameter mapping module, a topology analysis module, a fusion output module, and an adaptive fusion correction module.

[0064] The system adopts a modular design: the front end ensures imaging consistency through optical geometric calibration, the middle section establishes a mathematical model through multi-mode acquisition and frequency domain modeling, and the back end achieves system-level accuracy optimization through fusion compensation and adaptive correction. Signal interaction and parameter updates between modules are achieved through host computer software, thereby completing the joint displacement detection of the visual and laser channels.

[0065] The optical geometry constraint calibration module is used to constrain the optical path relationship between the vision sensor and the laser sensor using Scheimpflug geometry conditions during the system installation or startup phase.

[0066] The Scheimpflug condition states that when the laser beam is not parallel to the imaging plane, as long as the object plane, lens plane, and image plane intersect on the same straight line, the sharpness of the light spot on the imaging plane can remain consistent under tilted imaging conditions.

[0067] In this system, the incident optical axis of the laser emitted by the laser sensor is at an angle to the imaging plane of the vision sensor. Without correction, when the target surface is tilted or displaced, the shape of the light spot will be distorted due to the focal plane shift, affecting the accuracy of subsequent image measurements.

[0068] To this end, the optical geometric constraint calibration module adjusts the relative angle between the laser emitter and the camera lens using an adjustable bracket or a rotary fine-tuning mechanism, ensuring that the laser incident optical axis and the imaging plane satisfy the Scheimpflug geometric conditions. The system can automatically calculate the optimal angle matching relationship by photographing a standard planar target and detecting the spot ellipticity. The calibration results are stored in the system parameter library in the form of an angle matrix for subsequent use by the multi-mode detection module.

[0069] The above geometric constraints ensure that the shape of the visual imaging spot remains stable and the focus remains consistent throughout the entire measurement journey, providing a reliable spatial geometric consistency basis for the subsequent fusion of visual and laser signals.

[0070] The multi-mode detection module is used to acquire visual signals and laser displacement signals from the target surface through a visual sensor and a laser sensor, respectively. It includes a visual sensor, a laser sensor, an electronically controlled two-dimensional motion stage, and a signal synchronization and alignment unit.

[0071] The visual sensor employs a high-resolution CMOS camera for large-range two-dimensional spot position measurement, with a measurement range of 100mm × 100mm and a resolution of at least 0.05mm. The laser sensor uses a two-dimensional PSD (Position Sensitive Detector) based on the lateral photoelectric effect for small-range two-dimensional spot position measurement, with a measurement range no greater than 10mm and a resolution of at least 6. The visual sensor and laser sensor respectively collect visual signals and laser displacement signals from the target surface, and the displacement excitation of the light spot is achieved through an electronically controlled two-dimensional motion stage.

[0072] The core idea of ​​this module is to utilize the complementarity of sensors with different measurement principles in terms of accuracy and range.

[0073] The camera extracts the coordinates of the laser spot on the two-dimensional imaging plane based on an image grayscale centroid algorithm or edge recognition algorithm. Its theoretical resolution can be calculated based on pixel size and optical magnification; in this embodiment, a spatial resolution of 0.05mm can be achieved, suitable for large-area scanning. The PSD operates based on the semiconductor lateral photoelectric effect principle. When the laser spot illuminates the PSD surface, charge carriers diffuse laterally along the photosensitive layer, forming a current difference between the two end electrodes proportional to the position of the laser spot. The current between the two electrodes is controlled... The position coordinates of the light spot on the PSD surface are obtained through data acquisition and normalization calculation. : ,in This is the effective length of the PSD. The electrically controlled 2D motion stage is driven by a stepper motor to achieve planar displacement excitation of the measured optical point along the X–Y direction. The motion stage's driving resolution is better than 20. It communicates with the host computer via an RS-485 communication interface to realize the issuance of displacement commands and feedback sampling.

[0074] To ensure time consistency between the visual and laser signals, the signal synchronization and alignment unit uses a unified clock source for trigger sampling. The camera exposure trigger signal and the PSD sampling signal are output from the same timing control card, ensuring complete correspondence between the sampling points of the two channels in each measurement cycle. After acquisition, the system further performs timestamp alignment and interpolation correction to ensure consistency between the signal sequences of the two channels at the sampling point level.

[0075] Through the multi-mode detection module, the system can simultaneously obtain high-precision displacement responses from both the visual and laser channels, providing dual-mode data input for subsequent frequency domain analysis, fusion compensation, and other tasks.

[0076] In multi-mode fusion systems, visual sensors (such as CMOS cameras) and laser sensors (such as PSDs) differ in their dynamic response characteristics:

[0077] The visual channel has a narrow response bandwidth and significant phase lag.

[0078] Laser channels have a fast response speed, but are susceptible to high-frequency noise.

[0079] Direct frequency domain analysis of the original signal introduces interference. Therefore, it is necessary to accurately separate and quantify the dynamic coupling relationship between the two channels by performing frequency domain decomposition, coherence analysis, and resonant band identification.

[0080] To this end, this embodiment sets up a frequency domain analysis module to perform frequency domain transformation on the sampling sequences of visual signals and laser displacement signals, and analyze the frequency domain response differences between the visual channel and the laser channel. It mainly includes a signal decoupling unit, a dominant mode screening unit, a cross power spectrum analysis unit, a coherence analysis unit, a dominant resonance frequency band identification unit, and a difference feature synthesis unit.

[0081] The signal decoupling unit is used to perform variational mode decomposition on the sampling sequences of visual signals and laser signals respectively, and extract the IMF component sets of the visual channel and the laser channel respectively.

[0082] Variational Mode Decomposition (VMD) is a non-recursive time-frequency analysis method that decomposes the input signal into several intrinsic mode functions (IMFs) with finite bandwidth by solving a constrained variational problem. Each IMF represents a stable frequency component in the signal, thereby effectively separating signal components from different dynamic sources.

[0083] In the specific implementation process, the synchronized and aligned sampling sequences of the visual channel and the laser channel are acquired separately, denoted as... and .right The VMD algorithm is executed with a preset number of decomposition modes. (This can be set according to the number of peaks in the signal spectrum), by iteratively solving the variational problem, the IMF component set of the visual channel is finally obtained. Similarly, for The VMD algorithm is executed with a preset number of decomposition modes. (can be combined with) (Same or different), to obtain the IMF component set of the laser channel Each IMF component is accompanied by a definite center frequency, which lays the foundation for subsequent frequency band-based screening.

[0084] However, not all decomposed IMF components are dynamically related to the displacement excitation. Some components may originate from sensor noise, environmental vibration, or other disturbances. Therefore, the goal of the dominant mode selection unit is to intelligently select the IMF components that best represent the sensor's response characteristics to the core displacement excitation from the IMF component set of each channel.

[0085] For each IMF component (whether visual or laser channel), calculate its energy. Then, the proportion of the energy of this component to the total energy of the signal in its channel is calculated, thus obtaining the energy entropy value of this component. The larger the energy entropy, the more significant the response information carried by this component.

[0086] A joint screening criterion is established. For example, the center frequency of the component should fall within the effective frequency band where the electronically controlled 2D motion stage may generate excitation (e.g., between 0Hz and the first-order resonant frequency of the motion stage), and the energy entropy should be higher than a set threshold (e.g., 1.5 times the average energy entropy of all components). Alternatively, in the vision channel, IMF components with low energy entropy (energy concentration) and center frequencies falling within ±20% of the motion stage excitation frequency are selected, while in the laser channel, IMF components with center frequencies slightly higher than the excitation frequency and stable energy distribution are selected. Based on these criteria, visual target IMF components (possibly one or more) that meet the conditions are screened from the IMF component set of the vision channel. Similarly, laser target IMF components are screened from the IMF component set of the laser channel.

[0087] To accurately compare the dynamic relationship between two channels in the frequency domain, it is necessary to analyze their cross-power spectrum. The phase component of the cross-power spectrum directly reflects the phase difference between the two signals, while its amplitude component is related to the amplitude ratio of the two signals.

[0088] The cross-power spectral analysis unit pairs the selected visual target IMF components with the laser target IMF components. Typically, the pair with the closest center frequencies is selected for calculation. If multiple pairs are selected, they can be calculated separately and then averaged, or the pair with the strongest energy can be selected for calculation. For the selected component pairs... and Calculate its cross-power spectrum using Fast Fourier Transform (FFT) Then, the phase difference spectrum is extracted. This refers to the phase angle of the cross-power spectrum, which varies with frequency, forming the phase difference spectrum. Calculate the self-power spectrum of the two components. and (This can be calculated by this unit or provided by the coherence analysis unit), then the amplitude ratio spectrum It can be estimated as .

[0089] To further determine the correlation between the two channels at different frequencies, a coherence function is introduced to quantify the linear correlation between the two signals. This is achieved using the cross-power spectrum analysis unit. and the self-power spectrum of the visual target IMF component. and laser target IMF components The coherence function is calculated by the coherence analysis unit. : , The value range is [0,1]. When it is close to 1, it indicates that the two signals are highly linearly correlated at that frequency.

[0090] The goal of the dominant resonance band identification unit is to find one or more dominant resonance bands. Within these bands, the two sensors not only have strong responses (large amplitude) and high linear correlation (strong coherence), but also stable phase relationships (good linearity). This provides an ideal data foundation for establishing a high-fidelity transfer function model.

[0091] In coherence function On the curve, find all local peak points greater than a set threshold (e.g., 0.8), and simultaneously, in the cross-power spectrum amplitude... We also search for significant local peaks on the curve, and use the frequencies containing these two sets of peaks as candidate frequencies. For each candidate frequency... In a narrow band in its vicinity Inside, examine the phase difference spectrum. The phase variation of a stable, linear system within this narrow band should be approximately linear. The "linearity" can be quantified by linearly fitting the phase-frequency relationship within this narrow band and calculating the fitting residual or correlation coefficient. Only candidate frequency bands with linearity exceeding a threshold are retained. Ultimately, a frequency band that simultaneously satisfies the following conditions is determined as the dominant resonance frequency band:

[0092] coherence function There is a peak value in this frequency band, and the value is relatively high.

[0093] Cross power spectrum amplitude There is a peak value in this frequency band;

[0094] Phase difference spectrum It exhibits good linearity within this frequency band.

[0095] In order to provide a stable and condensed input to the subsequent network parameter mapping module, the difference feature synthesis unit synthesizes the frequency domain feature information in the dominant resonant band into a comprehensive frequency domain response difference feature vector.

[0096] First, determine the dominant resonant frequency band identified by the dominant resonant frequency band identification unit. Within this frequency band, for the existing phase difference spectrum Amplitude Ratio Spectrum Discrete sampling is performed to obtain a series of frequency points. eigenvalues ​​on (Phase difference) and (Amplitude ratio). For each frequency point Assign a weight This weight can be derived from the coherence function value at that frequency point. The weighting (higher coherence, greater weighting) can also be determined by the cross-power spectrum amplitude. Decision. Then, calculate the weighted average phase difference. And amplitude ratio .

[0097] Will , Together with information such as the center frequency and bandwidth of the dominant resonant frequency band, these are combined into a one-dimensional array, which constitutes the frequency domain response difference feature vector. This vector centrally characterizes the key differences between visual and laser sensors within the core dynamic response frequency band.

[0098] The network parameter mapping module transforms the frequency domain response difference feature vector output by the frequency domain analysis module into mathematical model parameters that describe the dynamic characteristics of the visual channel and the laser channel, namely the first network parameter set and the second network parameter set. It mainly includes a state-space modeling unit, a physical constraint feature encoding unit, and a parameterization mapping unit.

[0099] The state-space modeling unit establishes a unified mathematical framework for the system—the state-space equations. These equations are the standard model describing linear dynamic systems and generally consist of two parts:

[0100] State equations: describe the internal states of a system (denoted as...) This refers to how internal variables (such as the transient energy and phase accumulation of sensors, which cannot be directly measured but determine the behavior of the system) evolve over time.

[0101] Observation equations: describe how the system's internal state and external inputs together produce the output signal that we can measure (denoted as...). In this embodiment, it is the frequency domain response difference feature vector.

[0102] Its basic form is:

[0103] ;

[0104] in, For the state variables that need to be estimated, Input for the system; For the response of the visual or laser channel; Noise term; That is, the system matrix to be determined.

[0105] Preferably, this embodiment employs an enhanced state-space equation to perform parameterized mapping, thereby integrating frequency domain analysis results more deeply into the model identification process. The equation is as follows:

[0106] ;

[0107] In the formula, , The system is respectively in time, The state variables at any given time correspond to the dynamic characteristics of the visual channel and the laser channel; , They are respectively time, The time-based system input corresponds to the composite input tensor. for The system observations at time 1, corresponding to the frequency domain response difference feature vector; These are process noise and observation noise, respectively. These are the state matrix, input matrix, observation matrix, and direct transfer matrix, respectively.

[0108] The frequency domain difference feature correction term is defined as: ,in for The frequency domain response difference feature vector at each time step comes directly from the frequency domain analysis module; The adaptive regularization parameter is a variable scaling factor used to control the strength of the correction term; For the reason The frequency weight matrix for calculating the dominant resonant frequency band at time 1 is used to amplify the difference contribution within the resonant frequency band, making the model more focused on the key frequency band. This is the steady-state mean of the comprehensive frequency domain difference eigenvector, used as a calibration benchmark.

[0109] Optionally, The adaptive update rule is: ,in, It is the preset learning rate; It is the model's predicted output. These are actual observed values. , These represent the adaptive regularization parameters before and after the update, respectively. This rule means... It will automatically adjust based on the error in the model's predictions: if the error is large, it will adjust more aggressively. The model is then corrected; if it is already accurate, it remains stable. This forms a built-in optimization loop that allows the model to continuously improve itself.

[0110] To avoid the emergence of physically uninterpretable virtual features in the state-space equations in high-dimensional cases, the physical constraint feature encoding unit normalizes the original frequency domain response difference feature vector to eliminate differences in dimensions and orders of magnitude. Based on the pinhole imaging model of the visual sensor (camera) and the photoelectric conversion model of the laser sensor (PSD), the key physical parameters affecting its dynamic response are determined.

[0111] Both visual sensors and laser sensors possess inherent optical path delay, response inertia, and damping characteristics; these physical quantities can be obtained through calibration experiments. These parameters are used as constraints and encoded into a multidimensional data structure (i.e., a composite input tensor). For example, the composite input tensor includes:

[0112] Optical transmission delay parameters: can be theoretically estimated or initially calibrated based on hardware parameters such as lens focal length and PSD sensor position;

[0113] System inertial parameters: can be correlated with mechanical characteristics such as the mass of the motion platform and the response hysteresis of the motor;

[0114] Dynamic damping parameters: can be correlated with the steady-state velocity of the system under vibration.

[0115] This process is equivalent to injecting physical common sense into the mathematical model, making its solution more consistent with the behavior of real sensors.

[0116] The parameter mapping unit is used to ultimately map the input data into the required set of network parameters. Its implementation includes:

[0117] Subspace system identification algorithms (such as N4SID or MOESP) are employed. These algorithms can robustly estimate the state-space equations directly from the input-output data (i.e., the composite input tensor and the observations of the state-space equations). matrix;

[0118] To prevent numerical instability (i.e., "ill-conditioned problems") in the algorithm when the data is noisy, Tikhonov regularization is introduced in the identification process. Essentially, it adds a penalty term to the algorithm's loss function, which tends to select solutions with smaller and smoother values, thereby effectively suppressing noise interference and obtaining physically more reasonable parameter estimation results.

[0119] The final output consists of two sets of parameters:

[0120] First network parameter set: This characterizes the transmission characteristics of the visual sensor;

[0121] Second network parameter set: This characterizes the transmission characteristics of the laser sensor.

[0122] The topology analysis module is used to perform network topology analysis on the first network parameter group and the second network parameter group to identify the equivalent relationship of dynamic response speed between the visual sensor and the laser sensor, and to establish a transfer function model from displacement excitation to output signal.

[0123] In one specific embodiment, the topology analysis module establishes the transfer function model as follows:

[0124] Nodal impedance equivalent analysis: The state matrices in the first and second network parameter sets are compared. eigenvalues ​​( , These eigenvalues, calculated from , ..., describe the inherent dynamic patterns of the system (such as oscillation frequency and decay rate). Each eigenvalue... Through formula The analogy is mapped to an equivalent complex impedance. Among them, the resistance ,inductance ,capacitance The equivalent value can be determined through mathematical fitting, so that the frequency response characteristics of the impedance are similar to the eigenvalue. The modes they represent match. It is the imaginary unit. Angular frequency is a standard symbol in the fields of electronics and control; this process concretizes the abstract mathematical model into circuit elements; the signal injection relationship from displacement excitation to the equivalent complex impedance network is determined based on the input matrix, and the signal acquisition relationship from the equivalent complex impedance network to the output signal is determined based on the observation matrix;

[0125] Constructing an equivalent RLC network model: This involves combining the various equivalent complex impedances obtained in the previous step. As a basic node, according to its physical meaning and The input-output relationships defined by the matrix are connected in series, parallel, or more complex network topologies to construct a virtual equivalent RLC network model. The dynamic characteristics of this network (impedance variation with frequency) are analogous to the coupling relationship of dynamic response speed between vision and laser sensors;

[0126] Establishing the transfer function model using the Laplace transform: Apply the Laplace transform to the equivalent RLC network model constructed in the previous step. This is a mature mathematical tool that can transform differential equations (circuit equations) in the time domain into algebraic equations in the complex frequency domain (s-domain). These algebraic equations are the transfer function model required by the system. Among them, the direct transfer matrix This transformation directly manifests as a feedforward path, i.e. This represents the portion of the input signal that directly affects the output without going through the system's internal inertia. It is a Laplace variable. It is related to the state matrix Identity matrices of the same dimension.

[0127] Optionally, the network parameter mapping module obtains the network parameter set ( The matrix (often a high-dimensional system) may contain redundant states with minimal impact on system dynamics, hindering real-time processing and subsequent analysis. A model reduction and validation unit can be configured within the network parameter mapping module. The reduction process employs a balance implementation method, a standard algorithm that transforms the system's coordinates to balance the controllability and observability of each state's input and output. Then, the Hankel singular values ​​of the transformed system are calculated; these values ​​directly reflect the importance of each state to the system's input and output behavior. Based on the distribution of Hankel singular values, dominant states with larger values ​​are retained, while minor states with smaller values ​​are truncated. Thus, from the original high-order... From the matrix, a reduced-dimensional network parameter set that retains the core dynamic characteristics is extracted. The reduced-order network parameter set also consists of two sets, which respectively characterize the core dynamic characteristics of the visual sensor and the laser sensor.

[0128] The sensor's characteristics change slowly when ambient temperature changes, lens becomes contaminated, or laser power drifts. The network parameter mapping module also includes a meta-learning parameter self-updating unit. The meta-learning framework can pre-train the parameter mapping model under various simulation conditions, enabling it to quickly adapt to new situations. When the system detects performance drift (such as increased fusion residuals), it automatically collects a small amount of new sample data under the current environment. Utilizing meta-learning capabilities, these new samples are used as input to update the parameters of the state-space equations (including...). as well as Instead of retraining the entire complex model, the system performs rapid fine-tuning through one or more iterations. This allows the system to perform online recalibration with minimal computational resources and time, consistently maintaining high accuracy.

[0129] In this optimization scheme, the topology analysis module will no longer directly use the original network parameter set output (i.e., the first network parameter set and the second network parameter set), but will use the reduced network parameter set refined by the model reduction and validation unit to perform node impedance equivalent analysis, which ensures that the model is both accurate and efficient.

[0130] The fusion output module uses a transfer function model to fuse and compensate the output signals of the visual and laser channels, generating a multi-mode fused displacement detection result compensated for high-frequency distortion. It mainly includes a channel inverse compensation unit, a frequency domain adaptive fusion unit, a time domain reconstruction unit, and a high-frequency distortion compensation unit. The signal processing relationships between these units are as follows:

[0131] In one embodiment, to ensure that the two channel outputs have uniform time response characteristics, the channel inverse compensation unit is based on a transfer function model. Construct inverse models for the visual and laser channels respectively. The role of the inverse model is to "cancele" the dynamic defects of the original sensor. By passing the original output signals of the two channels through their inverse models, the phase lag and high-frequency attenuation can be pre-compensated, resulting in "pre-compensated visual signals" and "pre-compensated laser signals". The compensation method can be implemented using inverse system design based on transfer function models. This inverse modeling process can be achieved through standard control theory algorithms or system identification and control design toolkits in tools such as MATLAB.

[0132] The frequency-domain adaptive fusion unit constructs a frequency-adaptive weighting function based on the frequency domain response characteristics of the two channels as represented by the transfer function model. Its core strategy is to increase the weight of the visual channel signal in frequency bands below a certain crossover frequency, and increase the weight of the laser channel signal in frequency bands above the crossover frequency. This achieves the goal of leveraging the strengths and mitigating the weaknesses of different sensors in their advantageous frequency bands. Frequency-adaptive weighting function The specific form can be dynamically optimized based on real-time signal characteristics. For example, the weights can be calculated based on the signal-to-noise ratio (SNR) and coherence function values ​​of the two channels within the dominant resonant frequency band. First, the spectra of the two channels can be obtained through FFT transformation. Then, the SNR at each frequency point can be estimated using normalized power spectral density. Finally, the peak value of the coherence function within the resonant frequency band can be combined to comprehensively calculate the dynamic fusion weight coefficients that change with frequency. The two pre-compensated signals are then transformed to the frequency domain, and then weighted and summed using the above frequency-adaptive weight function to achieve fusion, forming a "fused spectrum".

[0133] The time-domain reconstruction unit reconstructs the time-domain signal from the frequency-domain fusion result (fused spectrum) using inverse Fourier transform, generating an initial fused displacement signal. This signal is continuous in time and has a consistent dynamic response, reflecting the preliminary measurement results after the fusion of the two channels.

[0134] The initial fused signal may still contain high-frequency distortion caused by sensor resonance or environmental disturbances. The high-frequency distortion compensation unit identifies the system's resonant frequencies based on the pole distribution of an equivalent RLC network model or directly from the transfer function model. An adaptive notch filter bank is designed accordingly, with its center frequency aligned to these resonant frequencies. This filter can be implemented using digital signal processing techniques (such as IIR notch filter design functions). Passing the initial fused displacement signal through this filter bank selectively filters out the (notch) resonant distortion components, ultimately outputting a multi-mode fused displacement detection result compensated for high-frequency distortion.

[0135] To achieve optimal long-term performance, the system adds a closed-loop feedback loop. Its core is to use the difference between the final output and intermediate results to inversely adjust the parameters of the preceding stages—this is the adaptive fusion feedback module. The adaptive fusion feedback module calculates the difference between the initial fused displacement signal and the final multi-mode fused displacement detection result in real time, using this as the fusion residual signal. This residual reflects the error present throughout the entire process from initial fusion to final compensation. Subsequently, statistical analysis is performed on this residual signal (e.g., calculating its mean square value and variance), and its trend over time is identified to comprehensively judge the stability and accuracy level of the current fusion process. Based on the statistical characteristics and trends of the fusion residual signal, parameter adjustment instructions are generated. The parameter adjustment algorithm can employ a classic proportional-integral-derivative (PID) control strategy or an optimization algorithm such as gradient descent. For example, when the residual continues to increase and the mean value of the coherence function output by the frequency domain analysis module decreases, it can be determined that the consistency of the inter-channel response is deteriorating. In this case, an instruction should be generated to prioritize adjusting the parameters of the inverse model in the channel inverse compensation unit. When spectral analysis of the residual signal reveals that its energy is concentrated in a specific frequency band, it can be determined that the crossover frequency setting of the weight function in the frequency domain adaptive fusion unit is improper or the weight shape is unreasonable. In this case, instructions should be generated to adjust the fusion weight function. The above strategy can be implemented using the adaptive control module in tools such as MATLAB / Simulink, without the need for special hardware.

[0136] The parameter adjustment instructions are sent to the channel inverse compensation unit and the frequency domain adaptive fusion unit. The compensation effect is improved by fine-tuning the internal parameters of the inverse model (e.g., adjusting the corresponding matrix elements in its state-space implementation). This is achieved by updating the frequency adaptive weighting function. morphology and crossover frequency This optimizes the fusion strategy. Furthermore, when the system evaluation output confidence level consistently falls below a set threshold, this module can also invoke the meta-learning parameter self-update unit in the network parameter mapping module, triggering a deeper online recalibration process.

[0137] Through the above feedback mechanism, the system forms a complete closed-loop adaptive architecture encompassing data acquisition, model building, signal fusion, result evaluation, and parameter callback. This architecture enables the fusion measurement system to self-correct and operate stably for a long period, effectively addressing challenges such as sensor aging and environmental changes.

[0138] like Figure 2 As shown, another embodiment of the present invention provides a visual laser displacement detection method based on multi-mode fusion and intelligent calibration, comprising the following steps:

[0139] S1: Based on the Scheimpflug geometry, configure the angle matching relationship between the laser incident optical axis of the laser sensor and the imaging plane of the vision sensor, and collect the visual signal and laser signal of the target surface through the vision sensor and the laser sensor respectively.

[0140] S2: Perform frequency domain transformation on the sampling sequences of visual and laser signals to analyze the difference in frequency domain response between the visual and laser channels;

[0141] S3: Based on the difference in frequency domain response, the visual signal and laser signal after frequency domain analysis and processing are mapped to a first network parameter set characterizing the transmission characteristics of the visual sensor and a second network parameter set characterizing the transmission characteristics of the laser sensor, respectively.

[0142] S4: Perform network topology analysis on the first network parameter group and the second network parameter group to identify the equivalent relationship of dynamic response speed between the visual sensor and the laser sensor, and establish a transfer function model from displacement excitation to output signal;

[0143] S5: The output signals of the visual channel and the laser channel are fused and compensated using a transfer function model to generate a multi-mode fused displacement detection result after high-frequency distortion compensation.

[0144] In summary, this invention, through a complete innovation chain from physical optical path calibration to dynamic system modeling and intelligent fusion compensation, successfully solves the inherent mismatch problem in dynamic response between vision and laser sensors, ultimately realizing a displacement detection system that can maintain high resolution over a large range, possess high dynamic performance across the entire frequency band, and operate stably over a long period of time.

[0145] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A visual laser displacement detection system based on multi-mode fusion and intelligent calibration, characterized in that, The system includes: The optical geometry constraint calibration module is used to configure the angle matching relationship between the laser incident optical axis of the laser sensor and the imaging plane of the vision sensor according to the Scheimpflug geometry, so as to maintain the stability of the imaging spot shape during the measurement process. The multi-mode detection module is used to acquire visual signals and laser signals from the target surface through a visual sensor and a laser sensor, respectively. The frequency domain analysis module is used to perform frequency domain transformation on the sampling sequences of the visual signal and the laser signal, and analyze the frequency domain response differences between the visual channel and the laser channel. The network parameter mapping module is used to map the frequency domain analysis processed visual signal and laser signal into a first network parameter group characterizing the transmission characteristics of the visual sensor and a second network parameter group characterizing the transmission characteristics of the laser sensor, respectively, based on the frequency domain response difference. The topology analysis module is used to perform network topology analysis on the first network parameter group and the second network parameter group to identify the equivalent relationship of dynamic response speed between the visual sensor and the laser sensor, and to establish a transfer function model from displacement excitation to output signal. The fusion output module is used to perform fusion compensation on the output signals of the visual channel and the laser channel using the transfer function model, and generate a multi-mode fusion displacement detection result after high-frequency distortion compensation. The frequency domain analysis module includes: The signal decoupling unit is used to perform variational mode decomposition on the sampling sequences of the visual signal and the laser signal respectively, and extract the IMF component sets of the visual channel and the laser channel respectively. The dominant modality screening unit is used to screen out the visual target IMF component and the laser target IMF component from the IMF component sets of the visual channel and the laser channel respectively, based on the center frequency and energy entropy of each IMF component. The cross-power spectrum analysis unit is used to calculate the cross-power spectrum between the visual target IMF component and the laser target IMF component, and to extract the phase difference spectrum and amplitude ratio spectrum between the two channels from the cross-power spectrum. The coherence analysis unit is used to calculate the coherence function of the two channels in the effective frequency band based on the cross power spectrum, the autopower spectrum of the visual target IMF component and the autopower spectrum of the laser target IMF component. The dominant resonance band identification unit is used to identify the dominant resonance band of the dynamic response coupling between the visual sensor and the laser sensor based on the peak value of the coherence function and in combination with the amplitude peak value of the cross power spectrum and the linearity of the phase difference spectrum. The difference feature synthesis unit is used to perform in-band weighted fusion of the phase difference spectrum and the amplitude ratio spectrum within the dominant resonance frequency band to generate a comprehensive frequency domain response difference feature vector.

2. The visual laser displacement detection system based on multi-mode fusion and intelligent calibration according to claim 1, characterized in that, The multi-mode detection module includes: The vision sensor, using a high-resolution CMOS camera, is used to achieve large-range two-dimensional light spot position measurement, with a measurement range of 100mm×100mm and a resolution of not less than 0.05mm. The laser sensor employs a two-dimensional PSD based on the transverse photoelectric effect to achieve small-range two-dimensional spot position measurement, with a measurement range of no more than 10 mm and a resolution of no less than 6. The visual sensor and laser sensor respectively collect visual signals and laser signals from the target surface, and the displacement excitation of the light spot is realized by an electronically controlled two-dimensional motion stage; The signal synchronization and alignment unit is used to synchronize the acquisition of the visual channel and the laser channel, and to achieve time series alignment of the output signals of the two channels at the sampling point level.

3. The visual laser displacement detection system based on multi-mode fusion and intelligent calibration according to claim 1, characterized in that, The network parameter mapping module includes: The state-space modeling unit is used to construct a state-space equation with the dynamic characteristics of the visual channel and the laser channel as internal state variables, using the frequency domain response difference feature vector as the system observation value. The physical constraint feature encoding unit is used to encode the normalized frequency domain response difference feature vector into a composite input tensor containing optical transmission delay parameters, system inertial parameters, and dynamic damping parameters, based on the optical transmission models of the visual sensor and the laser sensor. The parameterized mapping unit is used to map the composite input tensor into a first network parameter set representing the transmission characteristics of the visual sensor and a second network parameter set representing the transmission characteristics of the laser sensor, based on the state space equation and by introducing a Tikhonov regularized subspace system identification algorithm. Both the first network parameter set and the second network parameter set include a state matrix, an input matrix, an observation matrix, and a direct transfer matrix.

4. The visual laser displacement detection system based on multi-mode fusion and intelligent calibration according to claim 3, characterized in that, The network parameter mapping module performs parameter mapping based on the following state-space equation: ； In the formula, , The system is respectively in time, The state variables at any given time correspond to the dynamic characteristics of the visual channel and the laser channel; , They are respectively time, The time-based system input corresponds to the composite input tensor. for The system observations at time 1, corresponding to the frequency domain response difference feature vector; These are process noise and observation noise, respectively. These are the state matrix, input matrix, observation matrix, and direct transfer matrix, respectively. The frequency domain difference feature correction term is defined as: ,in for The frequency domain response difference feature vector at time step; For adaptive regularization parameters; For the reason The frequency weighting matrix for calculating the dominant resonance band at time t; This is the steady-state mean of the comprehensive frequency domain difference eigenvector.

5. The visual laser displacement detection system based on multi-mode fusion and intelligent calibration according to claim 4, characterized in that, The topology analysis module establishes the transfer function model in the following manner: Based on the state matrix, input matrix, observation matrix, and direct transfer matrix, a nodal impedance equivalent analysis is performed. The eigenvalues ​​in the state matrix are mapped to the equivalent complex impedances of the network topology nodes. The signal injection relationship from displacement excitation to the equivalent complex impedance network is determined based on the input matrix. The signal acquisition relationship from the equivalent complex impedance network to the output signal is determined based on the observation matrix. Based on the equivalent complex impedance, the signal injection relationship, and the signal acquisition relationship, an equivalent RLC network model for the dynamic response speed between the visual sensor and the laser sensor is constructed. The equivalent RLC network model is transformed into a transfer function model from displacement excitation to output signal by Laplace transform, wherein the direct transfer matrix represents the feedforward path from input to output in the Laplace transform and directly contributes to the transfer function model.

6. The visual laser displacement detection system based on multi-mode fusion and intelligent calibration according to claim 5, characterized in that, The network parameter mapping module also includes: The model reduction and verification unit is used to balance the first network parameter set and the second network parameter set and reduce the model order, and to confirm the dominant state by calculating the Hankel singular value, so as to extract the reduced network parameter set that characterizes the core dynamic characteristics of the visual sensor and the laser sensor. The meta-learning parameter self-updating unit is used to adaptively fine-tune the parameters of the state space equation based on real-time collected sample data when the external environment or measurement conditions change, so as to realize online recalibration under small sample conditions; the parameters of the state space equation include the state matrix, input matrix, observation matrix, direct transfer matrix and adaptive regularization parameter; At this point, the topology analysis module uses the reduced-order network parameter set instead of the original network parameter set to perform equivalent node impedance analysis.

7. The visual laser displacement detection system based on multi-mode fusion and intelligent calibration according to claim 5, characterized in that, The fusion output module includes: The channel inverse compensation unit is used to construct inverse models of the visual channel and the laser channel based on the transfer function model, and to compensate for phase lag and high-frequency attenuation of the output signals of the visual channel and the laser channel. The frequency domain adaptive fusion unit is used to construct a frequency adaptive weighting function based on the frequency domain response characteristics of the visual channel and the laser channel as represented by the transfer function model within the dominant resonance frequency band, and to perform frequency domain weighted fusion on the two channel signals after inverse compensation. The weight of the visual channel is increased in the frequency band below the crossover frequency, and the weight of the laser channel is increased in the frequency band above the crossover frequency. The time-domain reconstruction unit is used to reconstruct the frequency-domain weighted fusion result into a time-domain signal through inverse Fourier transform, generating an initial fused shift signal; The high-frequency distortion compensation unit is used to design an adaptive notch filter based on the system resonance characteristics identified by the equivalent RLC network model to selectively filter the resonance distortion component in the initial fused displacement signal, thereby generating a multi-mode fused displacement detection result compensated for high-frequency distortion.

8. The visual laser displacement detection system based on multi-mode fusion and intelligent calibration according to claim 7, characterized in that, The system also includes an adaptive fusion feedback module, which receives the fusion residual signal generated by the fusion output module, generates parameter adjustment instructions based on the statistical characteristics and trends of the fusion residual signal, and feeds them back to the channel inverse compensation unit and the frequency domain adaptive fusion unit, for dynamically adjusting the parameters of the inverse model and the shape and cross frequency of the frequency adaptive weighting function. The fusion residual signal is the difference between the initial fusion displacement signal and the multi-mode fusion displacement detection result.

9. The detection method of the visual laser displacement detection system based on multi-mode fusion and intelligent calibration as described in any one of claims 1-8, characterized in that, The method includes: Based on the Scheimpflug geometry, the angle matching relationship between the laser incident optical axis of the laser sensor and the imaging plane of the vision sensor is configured, and the visual signal and laser signal of the target surface are collected by the vision sensor and the laser sensor respectively. The sampling sequences of the visual signal and the laser signal are subjected to frequency domain transformation to analyze the difference in frequency domain response between the visual channel and the laser channel; Based on the frequency domain response difference, the visual signal and laser signal after frequency domain analysis and processing are respectively mapped to a first network parameter set characterizing the transmission characteristics of the visual sensor and a second network parameter set characterizing the transmission characteristics of the laser sensor. Network topology analysis is performed on the first network parameter group and the second network parameter group to identify the equivalent relationship of dynamic response speed between the visual sensor and the laser sensor, and to establish a transfer function model from displacement excitation to output signal. The output signals of the visual channel and the laser channel are fused and compensated using the transfer function model to generate a multi-mode fused displacement detection result compensated for high-frequency distortion.

Images