107 results about "Adaptive smoothing" patented technology

The invention belongs to the technical field of trajectory planning, and discloses an urban road trajectory planning and control method and system, and the method comprises the steps: generating an obstacle set and a Frenet projection of a road physical boundary through employing a spatial clustering algorithm; utilizing an extreme value method and a self-adaptive smoothing algorithm to dynamically construct a feasible trajectory region trajectory, and synchronously performing interval constraint expansion on a future prediction trajectory of the dynamic obstacle; constructing a path optimization objective function with interval constraints, and solving an optimal path variable in real time by using a numerical optimization method; an expected speed and acceleration sequence of each sampling point is generated based on a space-time domain dynamic planning method, and a longitudinal motion curve is optimized in real time by using an objective function; and inputting the space path and the longitudinal speed sequence into an MPC system, and calculating an optimal front wheel steering angle and acceleration control instruction to realize vehicle trajectory tracking. The method can sense errors and environment changes in a self-adaptive mode, the response speed and robustness of dynamic traffic flow are improved, and multi-target collaborative optimization is achieved.

The invention discloses a path planning method and system based on an improved A * algorithm and an improved DWA algorithm, and relates to the technical field of path planning, and the method comprises the steps: outputting a dynamically optimized heuristic function weight; generating a complete path of the current navigation task in combination with the heuristic function weight; executing self-adaptive smoothing processing, and outputting a smoothed global path; performing dynamic trajectory prediction according to the dynamic window parameters to generate a plurality of motion trajectories; performing comprehensive evaluation on each motion track, generating an evaluation value of each motion track, and selecting the motion track with the optimal evaluation value as a final track; and inputting the path coefficient recorded by the navigation task to the path coefficient optimization network based on the hierarchical attention mechanism. According to the method, the planning time of the A * algorithm in global path planning is shortened, the search efficiency of the A * algorithm in a large-scale scene is improved by optimizing the algorithm structure, and the problem that the DWA algorithm is prone to falling into local optimum is greatly solved.

The invention discloses a video anti-shake method and system based on multi-scale fusion and adaptive smoothing, relates to the technical field of video image processing, and aims to effectively solve the image quality problem caused by shake in a video shooting process. Gradient histograms and wavelet energy distribution characteristics of video frames are extracted through graying and normalization processing, and the gradient histograms and the wavelet energy distribution characteristics are input into a jitter type recognition network to recognize translation, rotation and Z-axis jitter probabilities. And further extracting motion, frequency domain and edge features, and generating multi-modal coupling features through combination of a dynamic feature interaction network and a dot product attention mechanism. And constructing a motion trajectory by using the features, optimizing the trajectory by using a texture perception double-layer smoothing strategy, introducing an adaptive penalty term into a dynamic planning cost function, and outputting a smooth motion compensation parameter. And finally, processing the boundary region through motion compensation and image extrapolation to generate an anti-shake video frame. Through multi-scale feature fusion and a self-adaptive smoothing strategy, the video anti-shake effect is effectively improved, and the method is suitable for complex scenes.

The invention provides a photovoltaic output scene clustering method. High-frequency noise of an original photovoltaic output time sequence is suppressed through self-adaptive smoothing processing, fluctuation characteristics of the high-frequency noise are extracted, it is ensured that target time sequence data accurately reflect photovoltaic power generation working conditions, and high-quality input is provided for subsequent clustering; the target clustering number is determined based on joint optimization parameters (fusing local dynamic characteristics and global distribution characteristics), and scene division deviation caused by subjective setting of the clustering number is avoided; the clustering center is updated by iteratively optimizing the clustering objective function, and it is ensured that the typical scene center truly reflects the mainstream output mode. Therefore, scene representativeness is improved in the aspects of data preprocessing, clustering number decision making and central optimization of a whole link, and generalization ability of a power grid dispatching strategy is remarkably enhanced.

An apparatus comprising an interface and a processor. The interface may be configured to receive pixel data of an environment. The processor may be configured to process the pixel data arranged as video frames, generate a luminance distribution map of the video frames in response to a low-pass filter operation, determine a plurality of defogging intensity weights for the luminance distribution map, perform adaptive smoothing to each of the plurality of defogging intensity weights, and generate defogged video frames in response to the video frames and the plurality of defogging intensity weights with the adaptive smoothing. The plurality of defogging intensity weights may each correspond to one of a plurality of luminance intervals of the luminance distribution map. The adaptive smoothing may be configured to prevent brightness differences in the defogged video frames.

The invention discloses a hydraulic system lag and leakage analysis system based on a multi-scale depth residual PINN and a corresponding compensation method in real-time control of a hydraulic system. According to the method, the deep neural network containing the residual block structure is constructed, so that deep fusion of physical constraint and data driving is realized. The system combines multi-scale feature engineering and a self-adaptive smooth transition mechanism, and the problem of prediction mutation of a traditional segmented model in different load intervals is eliminated. Through multi-scale feature engineering, original load data is converted into a five-dimensional feature vector containing an original value, a logarithm value, a square root value, a hyperbolic tangent value and a region indicator, and nonlinear characteristics of different load intervals can be effectively captured. Through a dynamic weight scheduling strategy, the weights of the physical constraint and the smooth constraint are gradually enhanced in the training process, so that the prediction precision and stability of the model can be remarkably improved. According to the invention, high-precision leakage prediction of the hydraulic loading system is realized, and a reliable technical support and a new idea are provided for real-time control and prediction maintenance of the hydraulic system.

The invention provides a bridge vibration displacement detection method and system based on image enhancement and space-time tracking, and relates to the technical field of bridge detection and computer vision, and the method comprises the steps: obtaining a video image frame sequence of a bridge structure, carrying out the cooperative processing of image frames through employing an ACAE adaptive image enhancement algorithm, and obtaining an enhanced reconstructed image; inputting the reconstructed image into an AHSD-Net detection model, introducing a depth separable convolution module into a backbone network of the AHSD-Net detection model, performing spatial aggregation of preliminary features in a neck network by combining a dynamic upsampling mechanism, and finally outputting to obtain a target position detection result; based on a cascaded ST-Refine Track strategy, density denoising and position correction of a space dimension, cross-frame association of a time sequence dimension and adaptive smoothing processing of a state dimension are sequentially executed on a target position detection result, and a continuous and stable bridge vibration displacement curve is obtained. According to the invention, the displacement measurement precision and reliability are improved.

The invention discloses an EEG attention signal adaptive smoothing method and a focusing method, and relates to the technical field of data processing, and the method comprises the steps: obtaining a real-time attention signal of a target user; calculating signal noise information according to the real-time attention signal and the historical attention signal of the target user; calculating a smoothing parameter according to the signal noise information and preset maximum noise information; and calculating a smoothed real-time attention signal according to the real-time attention signal, the smoothing parameter and the smoothed previous attention signal. By adopting the method and the device, the stability of the attention signal can be effectively improved and the noise can be reduced by adaptively adjusting the smoothing parameter.

The invention discloses a signal denoising method and device based on MEEMD and northern eagle optimization, and belongs to the technical field of signal processing and gas detection. The method comprises the following steps: acquiring a second harmonic signal in photoacoustic spectrometry gas detection; performing MEEMD decomposition on the signal to obtain a plurality of intrinsic mode function IMF components and residual errors; calculating the energy ratio of each IMF component and screening effective IMF components; reconstructing the effective IMF component and the residual error to obtain a preliminary de-noised signal; carrying out self-adaptive smoothing processing on the preliminarily denoised signal; gaussian filtering is carried out on the smoothed signal; and optimizing an energy threshold value, a self-adaptive smooth window parameter and a Gaussian filtering parameter by using a northern eagle optimization algorithm to obtain an optimal parameter combination. According to the method, MEEMD modal decomposition, Gaussian filtering and NGO parameter optimization are fused, accurate denoising and feature retention of gas concentration signals are achieved, and the precision and reliability of photoacoustic spectrum gas detection are remarkably improved.

The invention provides a high-precision line structured light center extraction method, which belongs to the technical field of three-dimensional measurement, and specifically comprises the following steps: carrying out adaptive threshold segmentation on a line structured light image, and extracting a light strip candidate region; constructing a multi-scale multi-direction filter bank, and screening stable edge points through a direction consistency index; a fusion confidence model is established by combining gradient amplitude, gray level energy and curvature, and false edge points are effectively eliminated; after the reserved points are connected into a curve, a fracture area is repaired through piecewise linearity evaluation and a self-adaptive smoothing strategy, and the curve continuity is improved; and for the final curve, realizing sub-pixel-level center positioning by adopting a combined method of gray gravity center and neighborhood matching, and establishing a center point continuity constraint by utilizing an improved Markov random field to obtain a smooth and stable center line. The method has higher robustness and extraction precision under the conditions of high reflection, low contrast and multi-curvature surfaces, and is suitable for industrial visual inspection and three-dimensional scene reconstruction.

The application discloses a multi-agent collaborative perception feature enhancement method based on context aggregation, relates to the technical field of automatic driving and intelligent networked vehicles, and comprises the following steps: acquiring current frame features and at least one frame of historical features; adaptively aligning the historical features based on a motion prediction network and a deformable convolution offset; generating semantic consistency weights and identifying time sequence discontinuous regions based on a correlation prediction network; fusing the current and historical features according to the weights and carrying out state space selective scanning, pooling, denoising and aggregation; generating multi-scale features, combining scene complexity, quality evaluation and time sequence consistency constraints to determine scale weights and then fusing; and inputting an LSTM to perform time sequence modeling and output enhanced context features. Invalid alignment is avoided through motion compensation and semantic consistency constraints, noise and scale jitter are suppressed through position-level denoising and multi-scale adaptive smoothing, and the collaborative perception robustness and detection accuracy are improved.

The invention relates to the technical field of electric data processing, in particular to a medicine logistics transportation monitoring method and system, and the method comprises the steps: collecting temperature data in a box body on a transport vehicle; the validity of the temperature data at the current moment is calculated, the validity is the bumping strength of the box body at the moment and the optimal matching degree of a temperature change sequence at the current moment in the current state and a standard temperature sequence after the box body is normally opened, and the length of the temperature change sequence at the current moment is consistent with that of the standard temperature sequence; and acquiring a temperature prediction value at the next moment based on a sliding moving average method. According to the method, the size of the moving average window is dynamically adjusted by combining the matching degree of the jolting strength and the temperature change trend, and self-adaptive smoothing processing of the temperature data is achieved. Noise can be suppressed during jolting, sensitivity can be kept during real fluctuation, and therefore prediction accuracy is improved, and the false alarm rate of abnormal detection is reduced.

The invention discloses a power distribution network intelligent mapping method based on dot matrix layout and dynamic weight, and relates to the technical field of distribution network mapping, and the method comprises the steps: constructing a dot matrix layout system corresponding to a power distribution network; the dot matrix layout system is divided into a plurality of uniformly distributed dot matrix units through regional gridding topographic factor correction equipment density feedback three-level calculation by taking geographic coordinates of a region where the power distribution network is located as a reference; the method specifically comprises the following steps: after gridding preliminary division, correcting the side length in combination with topographic factors, and then performing iterative adjustment according to equipment density; the self-adaptive smoothing processing is based on GIS vector topographic data, and a boundary is fitted through a Bezier curve; according to the method, the real-time performance and the accuracy of the mapping in a complex environment and an emergency scene are ensured by combining precision verification, on the whole, the intelligent, high-precision and efficient dynamic updating of the mapping of the power distribution network is realized, and the operation and maintenance efficiency, the emergency response speed and the fault processing capability of the distribution network are remarkably improved.

The invention discloses a non-intervention type electro-hydrogen thermal coupling system regulation potential quantification method and system. The method comprises the steps that historical and current operation data of all devices of a system are acquired; establishing an electro-hydrogen thermal coupling external characteristic model comprising a state space and an action space, wherein the state space is a product of the action space and the sensitivity matrix; based on the historical operation data, constructing an initial sensitivity matrix under different typical working conditions; based on the current operation data, performing recursive updating on the sensitivity matrix under each typical working condition; based on the current working condition parameters, self-adaptive smooth weights corresponding to the typical working conditions are calculated, and weighted fusion is carried out on the updated sensitivity matrixes under the typical working conditions to obtain a sensitivity matrix corresponding to the current working condition; and according to the external characteristic model and the sensitivity matrix corresponding to the current working condition, calculating to obtain an instantaneous adjustment potential, an interval adjustment potential and a comprehensive adjustment potential of the electro-hydrogen thermal coupling system. According to the invention, the regulation potential evaluation precision and reusability of the electro-hydrogen thermal coupling system can be improved.

The invention provides a multispectral image fusion method and system for security monitoring, and the method comprises the steps: carrying out the Retinex decomposition based on an obtained visible light image and an infrared image, and obtaining a visible light reflection component, a visible light illumination component, an infrared reflection component and an infrared illumination component; initial division of semantic regions is carried out based on the visible light reflection component and the infrared reflection component, and a joint semantic graph containing a plurality of semantic regions is obtained; based on the joint semantic graph, respectively carrying out regional adaptive smoothing constraint on the visible light illuminance component and the infrared illuminance component to obtain a visible light target smoothing component and an infrared target smoothing component; and performing illumination field alignment fusion on the visible light target smooth component and the infrared target smooth component based on the joint semantic map to obtain a fused illumination component. According to the method, the problems of edge blurring, artifact interference and key target structure distortion caused by uniform processing and ignoring of physical characteristic differences of semantic regions in an existing method are accurately solved.

The application belongs to the technical field of image processing, and relates to an extruded pipe surface defect detection method and system based on machine vision. The method acquires an original gray image of the extruded pipe surface, extracts texture energy response by using a multi-direction filter set, acquires local maximum, minimum and global average texture energy response values; constructs a ratio operation according to the response values to calculate a local spatial anisotropy index; utilizes the nonlinear attenuation characteristics of an exponential function to dynamically map a basic scale constant based on the local spatial anisotropy index, and calculates a dynamic smoothing scale of each pixel point; uses a Gaussian kernel function corresponding to the dynamic smoothing scale to perform adaptive smoothing processing on the image and constructs a Hessian matrix, solves eigenvalues, and combines a reference background curvature constant to calculate a confidence enhancement response, so as to extract real crack defects. The application realizes noise suppression and weak crack reservation, and improves the accuracy of pipe surface defect detection.

A large language model quantization method based on channel arrangement and activation adaptive smoothing belongs to the technical field of large language model compression and efficient inference. In order to solve the problem that the existing post-training quantization method is difficult to simultaneously consider the difference of different channel activation outliers, the group mode quantization has large dynamic range in the group, and the low bit weight and high bit activation alignment efficiency is low, which leads to the problem of precision decline and inference efficiency limitation, the scheme calibrates the maximum activation amplitude of the input channel by statistical data, divides the significant and non significant activation outlier channels according to the percentile threshold, and respectively uses the logarithmic domain and power function domain smoothing strategy to generate channel smoothing scaling factor, and fuses it into the model parameter to extract the weight multidimensional feature to construct the channel quantization sensitivity index and execute the input channel rearrangement, and further generate the joint quantization scaling factor to realize the integer domain alignment inference. It is suitable for low bit compression deployment and efficient inference scene of large language model in resource limited environment.

The application belongs to the technical field of computer artificial intelligence, and relates to a large model field quantization and adaptive model deployment method for citrus intelligent planting management, comprising the following steps: firstly, a large model calibration data set is constructed and preprocessed, and a Transform model is input to perform forward inference; then, the weight matrix of the model to be quantized is left-multiplied and the activation matrix is right-multiplied and rotated, GPTQ quantization is performed, and the dynamic adaptive smoothing factor of each channel of the activation matrix is calculated, normalization and symmetric quantization are performed according to the hidden dimension grouping, and a quantized Transform model is generated; then, citrus industry text data is acquired to construct a fine-tuning instruction set, and the quantized model is supervised and trained to generate a full-quantization and full-precision model; finally, a training task complexity classifier is used to select the full-quantization or full-precision model as a target deployment model according to the task complexity level, and the application realizes efficient model quantization and intelligent task adaptation.

The invention discloses an air quality prediction method, device and equipment based on multi-modal space-time fusion and a medium, and relates to the technical field of air quality prediction.The method comprises the steps that ground monitoring data and satellite remote sensing data are fused, trend and residual errors are dynamically separated from the fused data through an adaptive smoothing filter, and a prediction result is obtained; according to the method, multi-scale periodic features are captured through multi-window time-frequency analysis, then the multi-scale periodic features are predicted through gating fusion, a mixed geographical / data driven graph neural network, Transform time sequence modeling and error correction, an air quality prediction result is obtained, multi-source heterogeneous data fusion and self-adaption to spectral characteristic differences of different stations are achieved, and the air quality prediction accuracy is improved. And the accuracy and reliability of air quality prediction are improved.

The invention provides an eye movement tracking method and system based on a stacked hourglass network and homography transformation. The method comprises the steps that a real-time eye image is collected; eye features are generated based on a stacked hourglass network model; calculating an instantaneous sight angle according to the eye features; mapping the sight line angle into a screen drop point coordinate by using a pre-trained homography transformation model; oneEuro time filtering is applied to the coordinates of the original screen; performing noise reduction on the smoothed coordinate sequence; and generating a KDE attention heat map. According to the invention, high-precision key point extraction of the eyes is realized through the stacked hourglass network, nonlinear mapping of the sight line and the drop point is realized through homography transformation, adaptive smoothing is realized through OneEuro filtering, and the naked eye tracking precision and robustness are remarkably improved.

The application discloses a load impact current adaptive smoothing suppression method and device, a terminal and a medium, wherein the load impact current adaptive smoothing suppression method comprises the following steps: collecting current signals at compensation points in real time; judging working conditions based on the current signals at the signal compensation points; if it is determined that the APF / SVG system is in a load sudden working condition, extracting impact current characteristic parameters; constructing an impact current buffer model and a parameter adjustment model based on the impact current characteristic parameters; integrating the impact current buffer model and the parameter adjustment model in a digital control loop of the APF / SVG system, and performing smoothing suppression by adjusting dynamic parameters; through the impact current buffer model and the parameter adjustment model, different impact intensities and rates are dynamically adapted, the impact current peak value and the mutation rate are effectively reduced, the compensation current fluctuation is reduced, and the equipment operation stability and the power grid power quality are significantly improved.

The invention provides a sea surface wind speed deviation correction method and system based on statistical learning and physical constraint, and belongs to the field of satellite ocean remote sensing and meteorological data reanalysis. According to the method, the technical problems that the correction result of the existing multi-source satellite wind speed data is unstable in statistics and unreasonable in physics due to inherent system errors of a sensor and sparse samples in a high-wind-speed interval, and an accurate and reliable correction result cannot be generated in a full-wind-speed range (especially under a high-wind-speed condition) can be solved. Specifically, quality control and space-time matching are carried out by acquiring multi-source satellite wind speed data and high-precision reference wind field data, an initial deviation correction lookup table is constructed by adopting a double-reference partitioning strategy based on satellite observation wind speed and reference wind field wind speed, statistical stabilization processing is carried out on the lookup table by utilizing a hierarchical Bayesian contraction estimator, and the initial deviation correction lookup table is obtained. The physical continuity of a wind speed-deviation curve is improved through an adaptive LOESS smoothing algorithm, and an exponential decay function based on a turbulence energy spectrum theory is introduced.

This invention discloses a bolt loosening location method and system based on environmental vibration and multi-scale PNN. The location method includes the following steps: S1: Establishing a refined finite element simulation model of the transmission tower and dividing it into multi-scale hierarchical substructures; S2: Simulating bolt loosening conditions in the finite element simulation model and extracting operational modal analysis results; S3: Calculating the frequency change ratio and modal compliance residual matrix norm, and constructing a multi-source damage-sensitive feature vector after weighted fusion; S4: Constructing and training a probabilistic neural network with an adaptive smoothing factor adjustment mechanism; S5: Acquiring on-site environmental vibration data and identifying modal parameters; S6: Extracting and calculating the measured data of the tower using the random subspace identification method, and inputting it into the trained probabilistic neural network to obtain the location results of the main bolt loosening area. This method achieves accurate bolt loosening location through scientific substructure division and multi-source fusion strategies, providing a reference for the safe operation and maintenance of tower bolts.

The invention discloses a landscape lighting intelligent adjusting system based on tone and motion state fusion analysis, and relates to the technical field of digital information processing. The method is used for solving the problems of multi-modal data logic conflict and control signal step. The method comprises the following steps: firstly, executing frequency domain transformation and differential operation on acquired space-time sequence data to generate normalized tone and motion intensity; then, calculating a cooperative gain coefficient and a dispersion index, and quantifying a scene active level and a data deviation; then, calculating a credibility weight based on the dispersion, and executing fusion iterative operation on a preset parameter and a historical state to generate a target control parameter; and finally, generating a pulse width modulation digital sequence in smooth transition by using a linear interpolation algorithm, and outputting a control instruction. According to the system, high-reliability fusion of multi-source data and adaptive smoothing of lighting effect control are realized.

The application discloses a hydraulic system lag and leakage analysis system based on a multi-scale deep residual PINN and a corresponding compensation method in real-time control of the hydraulic system. The method realizes deep fusion of physical constraints and data driving by constructing a deep neural network containing a residual block structure. The system combines multi-scale feature engineering and an adaptive smooth transition mechanism to eliminate the prediction mutation problem of traditional segmented models in different load intervals. Through multi-scale feature engineering, the original load data is converted into a 5-dimensional feature vector containing original values, logarithmic values, square root values, hyperbolic tangent values and region indicators, which can effectively capture the nonlinear characteristics of different load intervals. Through a dynamic weight scheduling strategy, the weights of physical constraints and smooth constraints are gradually increased during the training process, thereby significantly improving the prediction accuracy and stability of the model. The application realizes high-precision leakage prediction of the hydraulic loading system, and provides reliable technical support and new ideas for real-time control and predictive maintenance of the hydraulic system.

The invention provides an ocean internal wave parameter inversion method and system based on temperature chain buoy data. The method comprises the following steps: preprocessing an acquired multi-depth temperature sequence and then constructing a temperature-time profile; detrending and self-adaptive smoothing are carried out through empirical mode decomposition, and a temperature background field and a pulsating field are separated; extracting a vertical gradient of a temperature background profile by using LOESS local regression, and introducing Tikhonov regularization and threshold gating to guarantee gradient robustness; mapping temperature pulsation into vertical displacement based on isothermal surface conservation, and screening effective data in combination with a credibility index; finally, the amplitude, the period and the wave height of an internal wave are inversed through multi-window wavelet spectrum and Hilbert envelope conjoint analysis, and the reliability of a result is improved through vertical modal consistency verification. According to the method, consistent inversion of ocean internal wave parameters can be completed only by means of a single temperature chain buoy, and the inversion accuracy under the background of weak gradient, weak signal and noise is remarkably improved.

This application provides a rendering optimization system and method for altered mineral 3D models. The method first divides the 3D model surface into uniform alteration intensity regions and alteration intensity transition zones. Based on the local gradient direction of the alteration intensity transition zone, directional compensation is applied to the boundary diffusion parameter, which characterizes the visual blurring properties of the boundary, to obtain the visual compensation coefficient for the alteration intensity transition zone. Based on the alteration intensity difference between adjacent uniform alteration intensity regions in the 3D model and the inherent width of the alteration intensity transition zone, the visual loss of boundary continuity is determined. The anisotropic smoothing constraint weights are determined through the visual compensation coefficients and the visual loss of boundary continuity when performing color interpolation calculations for each alteration intensity transition zone. Then, adaptive smoothing rendering optimization is performed on the transition zones of the 3D model surface based on all anisotropic smoothing constraint weights. This application's solution can achieve adaptive smoothing rendering of the transition zones on the 3D model surface, thereby improving boundary continuity.

The invention relates to the field of strain detection, in particular to a self-adaptive strain detection method and system and a medium. The method comprises the following steps: inputting grid point coordinates, noisy displacement data and a maximum smooth half window; calculating the overall noise level sigma through local linear regression analysis; extracting local strain curvature characteristic quantity through cubic polynomial regression; establishing a noise and curvature balance relation based on the sigma and the curvature characteristic quantity, and calculating an adaptive smooth half window; obtaining the strain estimation value of each grid point through linear polynomial regression, and finally obtaining the strain distribution of the whole field. According to the method, the optimal smooth window can be automatically adjusted without manual intervention, balance is achieved between noise suppression and detail reservation, the problem that a traditional fixed window method is insufficient in precision in a non-uniform deformation area is solved, high-precision full-field strain detection of a noise-containing displacement field is achieved, and the detection precision is improved. The method is suitable for derivative calculation of digital image related systems and various noisy continuous signals.

The application belongs to the technical field of robot motion control, and provides a four-legged robot eye-leg cooperative obstacle crossing method and system for complex terrains, which comprises the following steps: acquiring an original depth image of a four-legged robot; repairing the original depth image, performing smoothing processing on the repaired original depth image by combining an adaptive smoothing model based on gradient constraints, and obtaining a depth information matrix of the four-legged robot; performing image cropping and standardization processing based on the depth information matrix, and obtaining visual depth information of the four-legged robot; constructing a robot body state vector according to the current motion state of the four-legged robot; performing mapping of the visual depth information of the four-legged robot and the robot body state vector in a low-dimensional feature space, generating a terrain latent feature vector; fusing the terrain latent feature vector and the robot body state vector, calculating the joint torque of the four-legged robot, generating a joint motion instruction, and completing the four-legged robot eye-leg cooperative obstacle crossing for complex terrains.