30 results about "Kalman algorithm" patented technology

The invention discloses a dual-band infrared radiation attitude estimation method for deep learning assisted extended Kalman. The method comprises the following steps: analyzing the influence of geometric errors and random errors in dual-band infrared radiation measurement signals on attitude estimation reliability; the method comprises the following steps: constructing a CBGTE model to eliminate geometric errors and random errors in dual-band infrared radiation signals, firstly, collecting a dual-band infrared radiation signal data set by using a semi-physical simulation platform, cleaning the dual-band infrared radiation signal data set, inputting the dual-band infrared radiation signal data set into a BGT neural network for feature learning, and storing an optimal weight; and inputting data obtained by the dual-band infrared radiation sensor into the BGT neural network with the optimal weight to obtain infrared radiation data after compensation of random errors, and inputting the data after filtering of the random errors into an extended Kalman algorithm to filter geometric errors to obtain reliable attitude angle information. The invention provides a high-precision and robust attitude information acquisition and processing method for a spinning unmanned aerial vehicle taking infrared radiation as an attitude measurement method.

The invention relates to the technical field of auxiliary driving, and discloses a deep learning-based road signboard recognition method, which comprises the following steps of: acquiring a road traffic video sequence stream in a vehicle driving process; inputting the road traffic video sequence stream into a deep learning detection framework based on improved YOLOv11 for initial positioning to obtain an initial positioning result of the road signboard; performing cross-frame target tracking on the initial positioning result of the road signboard through an improved Kalman algorithm to obtain a road signboard identification result with a motion track; and displaying the identification result of the road signboard on a vehicle-mounted head-up display of the vehicle. According to the road signboard identification method and system based on deep learning provided by the invention, a scheme based on deep learning and optimized Kalman filtering is adopted, traffic signboards in a complex road traffic environment are identified, and a perfect auxiliary driving system is established, so that key signboard information can be quickly and accurately found in the complex road environment; and a driver is assisted to make correct judgment.

The invention provides a clamping force estimation method of an electronic mechanical braking system and related equipment, and relates to the technical field of braking system control, and the clamping force estimation method comprises the following steps: obtaining position information of a rotor of a driving motor and motor rotating speed information in real time; the corresponding relation between the historical clamping force of the electronic mechanical braking system and the position information of the rotor is obtained, and a first estimated value of the current clamping force of the electronic mechanical braking system is determined by combining the current position information of the rotor and the corresponding relation; determining the current output torque of the driving motor according to the motor rotating speed information; the conversion ratio from the output torque of the driving motor to the clamping force is obtained, and a second estimated value of the clamping force is obtained by combining the conversion ratio and the current output torque; and determining a target estimated value of the clamping force by combining a discrete Kalman algorithm, the first estimated value of the clamping force and the second estimated value of the clamping force. The problems that in the prior art, a clamping force sensor is used for obtaining clamping force, the manufacturing cost is high, and installation is inconvenient are solved.

The application provides a conveyor speed detection method, device and equipment based on a dual-antenna RFID. The method comprises: tracking and processing an RFID tag by using a first antenna, and recording a first phase sequence of the RFID tag; tracking and processing the RFID tag by using a second antenna, and recording a second phase sequence of the RFID tag; determining a phase difference between the first phase sequence and the second phase sequence, and determining a phase difference speed according to the phase difference; determining a Doppler speed according to movement data of the RFID tag; fusing the phase difference speed and the Doppler speed by using a Kalman algorithm, determining a final conveyor belt speed, and outputting the final conveyor belt speed. The phase difference speed is determined according to the phase difference between the phase sequences of the two antennas. The Doppler speed is also determined. The phase difference speed and the Doppler speed are fused by using the Kalman algorithm, and the final conveyor belt speed is determined, so that the conveyor belt speed is accurately detected with high precision.

The invention provides a three-axis coil error parameter identification method and device based on Kalman filtering, a medium and electronic equipment, and belongs to the technical field of motor control. According to the method, an error parameter identification model of a three-axis Helmholtz coil under the condition that a single-group coil offset error and a three-axis non-orthogonal error are superposed is constructed, and interference parameters in the error parameter identification model are identified by adopting a Kalman algorithm; according to identification interference parameters obtained through identification, a central uniform region vector magnetic field of the same three-axis Helmholtz coil under all current conditions is obtained through calculation; the method can effectively solve the problem that a traditional least square method or a neural network method is low in reliability of magnetic field simulation of the three-axis Helmholtz coil, magnetic field calculation errors caused by interference coefficient errors of the three-axis Helmholtz coil are remarkably reduced, and the identification accuracy is greatly improved.

The invention relates to a permanent magnet synchronous motor parameter identification method based on a multi-information least two and multi-information extended Kalman algorithm, and the method specifically comprises the steps: S1, modeling a permanent magnet synchronous motor digital model, and simplifying the digital model through a coordinate transformation method; s2, performing online parameter identification on the digital model by using a multi-information least square method; and S3, expanding the nonlinear function into Taylor series around the filtering value by using an extended Kalman filtering algorithm, and removing second-order and above terms to obtain an approximate linearization model, and finally completing parameter estimation of the permanent magnet synchronous motor. Through the method, the parameter acquisition accuracy of the permanent magnet synchronous motor is improved.

The invention relates to the technical field of medical equipment and signal processing, in particular to a cerebral infarction thrombectomy postoperative intracranial pressure management method and system based on fNIRS-TCD combined monitoring, and the method comprises the steps: firstly, synchronously collecting fNIRS cerebral blood oxygen concentration change data and TCD cerebral blood flow velocity data through a head-mounted device, and calculating a pulsation index; then, on the basis of a Bayesian data assimilation framework, constructing the data into a real observation vector, combining a state vector predicted by an intracranial physiological model and an observation vector predicted by an observation model, calculating a Kalman gain through an iterative integrated Kalman algorithm, and further correcting the state vector in real time; and obtaining state estimation containing precise intracranial pressure. And finally, the system carries out three-level early warning according to the real-time intracranial pressure data in combination with brain blood oxygen and pulsation index changes, and carries out stepped individual management from observation, drug intervention to emergency rescue according to the three-level early warning. According to the invention, the intracranial pressure management accuracy is improved.

The invention provides a system-level power station SOH estimation method and device, a terminal and a storage medium, and the method comprises the steps: inputting the charging and discharging data of a power station system into a trained LSTM model, and obtaining an SOH1 prediction value outputted by the trained LSTM model; calculating an equivalent internal resistance value in real-time operation of the power station system based on a power station voltage equivalent value and a power station current RMS value in the charging and discharging data of the power station system, and calculating an SOH2 predicted value by using the equivalent internal resistance value; and fusing the SOH1 predicted value and the SOH2 predicted value based on a Kalman algorithm to obtain an SOH value of the power station. According to the system-level power station SOH estimation method and device, the terminal and the storage medium, the equivalent value of a power station system can be efficiently and rapidly extracted, the overall calculation amount can be reduced, subsequent rapid analysis and processing of data can be achieved, and the input requirement needed by an LSTM model is met.

The present invention discloses a method for adaptive collaborative estimation of the SOC and SOH of a lithium-ion battery, which belongs to the technical field of lithium-ion battery state estimation. The steps are as follows: first, the terminal voltage and load current data of the lithium-ion battery are measured through a mixed power pulse characteristic experiment; then, a fractional-order equivalent circuit model of the lithium-ion battery is established offline; then, according to the state space of the established model, a dual adaptive square root volume Kalman filter for lithium-ion battery state estimation is constructed; finally, the actual working conditions are simulated through a random walk charge and discharge experiment, and the SOC and SOH of the lithium-ion battery are collaboratively estimated online. The present invention realizes high-precision SOC and SOH estimation while considering the influence of battery aging, solves the problem that the filter covariance matrix is ​​easy to lose semi-positive definiteness in collaborative estimation and the square root volume Kalman algorithm is not applicable, has strong robustness and generalization ability, and reduces the computational cost in a more mechanism-oriented way.

The present invention provides a ship image track extraction method for non-continuous visual feature interference, comprising: acquiring ship images, including images of scenes where the ship is occluded and images of scenes where the ship is not occluded; detecting the ship images based on the DeepSnake model and the YOLOX model to obtain the ship's position in the image sequence; inputting the ship's position and the detection results of the DeepSnake model into the YOLOX model to obtain the ship's external contour; and tracking the ship's trajectory based on the external contour and the ship's position to obtain the ship's motion trajectory in the image sequence. The present invention introduces occlusion state determination and further utilizes the Kalman algorithm to obtain a more accurate ship trajectory, thereby improving the accuracy of extracting ship navigation tracks from shipborne surveillance video data and enhancing robustness in complex sea conditions.

The present invention discloses a method for testing the three-dimensional distribution of pipelines based on the fusion of two attitude angles of laser spots, including setting up a pipeline three-dimensional distribution detection device; controlling the pipeline three-dimensional distribution detection device to move forward at a constant speed in the pipeline, and simultaneously obtaining the data of the three-axis accelerometers and three-axis gyroscopes of two MEMS-IMU inertial modules and the position information of the laser spots collected by a COMOS image sensor; processing the data of the MEMS-IMU inertial modules to obtain the deflection angle of the pipeline to be measured; processing the position information of the laser spots collected by the CMOS image sensor to obtain the deflection angle of the pipeline to be measured; using the Kalman algorithm to fuse the deflection angle of the pipeline to be measured obtained by the CMOS image sensor and the deflection angle of the pipeline to be measured obtained by the MEMS-IMU inertial module to obtain the optimal value of the deflection angle. The present invention reduces the errors of sensors with different principles through data fusion, can adapt to the measurement of different deflection angles of pipelines, and has high efficiency and high detection accuracy.

This invention discloses a deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation method. The method analyzes the impact of geometric and random errors in dual-band infrared radiation measurement signals on attitude estimation reliability. A CBGTE model is constructed to eliminate geometric and random errors in the dual-band infrared radiation signal. First, a dual-band infrared radiation signal dataset is collected and cleaned using a hardware-in-the-loop simulation platform. The dataset is then input into a BGT neural network for feature learning and optimized weights are saved. Data acquired by the dual-band infrared radiation sensor is then input into the BGT neural network with optimized weights to obtain infrared radiation data after random error compensation. This data, after random error filtering, is then input into an extended Kalman algorithm to filter out geometric errors, thereby obtaining reliable attitude angle information. This invention provides a high-precision and robust method for acquiring and processing attitude information for a spinning drone using infrared radiation as its attitude measurement method.

The invention relates to a high-real-time power battery linear simplified modeling and self-adaptive state of charge estimation method, and belongs to the technical field of automobile power lithium batteries. The method comprises the following steps: collecting dynamic data of a power battery, preprocessing the data, and preliminarily calculating the SOC of the power battery; establishing a linear relation between the open-circuit voltage Voc and the SOC by adopting a piecewise linear processing mode; establishing a linear simplified power battery model with high real-time performance and a discrete state space equation of the linear simplified power battery model; adopting a multi-forgetting factor least square algorithm to identify battery parameters on line; and estimating the SOC according to the online identified battery parameters through a noise adaptive Kalman algorithm. According to the method, the problems of complex parameter identification of a conventional RC battery model, high calculation burden of a nonlinear battery model, different multi-parameter change rates and weak anti-noise capability can be solved.

The present invention discloses a method for estimating the key parameters of a lithium battery used for braking energy recovery in an electric-driven dump truck for mining. The estimation method uses a fractional-order equivalent circuit model (FOM) and integrates a dual adaptive unscented Kalman algorithm (DAUKF) to achieve multi-time-scale cyclic estimation of the key characteristic parameters of the lithium battery in the braking energy recovery system of the electric-driven dump truck for mining, providing an accurate data basis for the efficient and safe recovery of braking energy. The present invention improves the defects of existing methods for estimating characteristic parameters of lithium batteries, thereby achieving rapid and accurate estimation of characteristic parameters of lithium batteries under extreme working conditions such as frequent charging and discharging, and large current shocks. The safety and stability of the braking energy recovery system of electric-driven dump trucks for mining are improved. The present invention can solve the technical problem that existing methods for estimating key characteristic parameters of lithium batteries are mainly aimed at lithium batteries in ordinary working environments, but are not suitable for estimating key parameters of lithium batteries for braking energy recovery in electric-driven dump trucks for mining.

The invention relates to the technical field of automobile control, and discloses a four-wheel steering vehicle fault-tolerant control method, which comprises the following steps: constructing a vehicle two-degree-of-freedom dynamic model; real-time data in the vehicle running process are collected, and a vehicle state estimation value is obtained through an unscented Kalman algorithm; constructing a residual reference model to obtain a residual sequence, and determining a fault sensor by using the residual sequence; and the vehicle state estimation value is used as the input of a reinforcement learning model-free control algorithm, and tire rotation angle control in the vehicle operation process is adjusted in real time. According to the method, the vehicle kinetic equation is used as a network regular constraint term, the accuracy and reliability of residual calculation are remarkably improved, rapid and accurate fault positioning is achieved through matching analysis of the residual sequence and the preset fault vector table, meanwhile, rapid self-adaption and control law optimization of a dynamic environment are achieved through reinforcement learning without a model, and the method is suitable for large-scale popularization and application. And the fault-tolerant performance and the control stability of the system are improved.

The invention relates to a power battery adaptive linear modeling and state of charge estimation method considering internal and external information of a battery, and belongs to the field of automobile power lithium battery modeling and SOC estimation. The method comprises the following steps: collecting dynamic data of a power battery, preprocessing the data, and preliminarily calculating the SOC of the power battery; establishing a linear relation between the open-circuit voltage Voc and the SOC by adopting a piecewise linear processing mode; establishing a power battery linear model considering internal and external information of the battery and a discrete state space equation thereof; a self-adaptive forgetting factor least square algorithm is adopted to identify battery parameters on line; and estimating the SOC according to the online identified battery parameters through a noise adaptive Kalman algorithm. The method can solve the problems that an existing lithium battery electrochemical model cannot consider the influence of internal and external factors of the battery on a battery model, the calculation burden of a nonlinear battery model is high, the information redundancy of an estimation algorithm is high, and the anti-noise capability is weak.

This invention belongs to the field of state control and control optimization application technology for complex systems, and particularly relates to a SOC calibration algorithm based on Kalman filter multi-sensor fusion. The algorithm includes constructing a lithium battery circuit model, lithium battery model identification, Kalman filter multi-sensor fusion SOC calibration, and improving the accuracy of multi-sensor fusion Kalman filtering. This invention considers the charging and discharging characteristics of the lithium battery model, the complexity of battery internal parameter identification, and the complexity of the Kalman algorithm. It proposes a Kalman filter multi-sensor fusion SOC calibration that takes into account process noise and measurement noise, ensuring accurate SOC. This can improve user experience, extend battery life, ensure safety, and optimize energy management. This invention utilizes the cloud platform of the BMS to define and analyze data, adjust the parameter matrix of the battery model and the covariance of the sensors, and still maintain SOC accuracy even under sensor aging conditions.

The invention belongs to but not limited to the technical field of computer vision, and discloses a lightweight anti-shielding unmanned aerial vehicle target tracking method and system, and the method comprises the steps: obtaining a video image through an airborne camera; inputting the video into a tracking method SiamRPN + + A; obtaining a tracking result confidence map; calculating a weighted neighborhood index value for the current frame; calculating an average weighted neighborhood index value of the current frame in combination with the adjacent frame; setting a threshold value T; judging whether the t-frame tracking result is reliable or not according to a set threshold T; if the tracking is reliable, the tracker outputs a tracking result as a final tracking result; and if not, predicting a final tracking result by adopting an extended Kalman algorithm. According to the invention, the advanced tracker SiamRPN + + is taken as a base line, and lightweight improvement is carried out, so that the operation requirement of embedded equipment is met, and the speed and precision of the target tracker of the unmanned aerial vehicle are improved; a simple and effective index is also designed to detect shielding, and then target position prediction is realized by using extended Kalman filtering.

The invention discloses a state estimation method and device of a power distribution network, equipment and a medium. The state estimation method of the power distribution network comprises the following steps: establishing a dynamic model of the power distribution network; establishing a sliding-mode observer model of the power distribution network according to the dynamic model; and determining a state estimation value of the power distribution network according to the dynamic model and the sliding-mode observer model. According to the technical scheme disclosed by the invention, the strong anti-noise disturbance capability of the Kalman algorithm and the strong anti-parameter disturbance capability of the sliding mode observer can be integrated at the same time, and the state estimation precision and accuracy of the power distribution network parameters are improved.

The invention provides an improved unscented Kalman lithium battery state joint estimation method based on IPOA optimization. The improved unscented Kalman lithium battery state joint estimation method comprises the following steps: S1, firstly, constructing a fractional-order second-order RC equivalent circuit model number; according to the method, a battery fractional-order second-order equivalent circuit model is constructed, model parameters are identified by adopting an adaptive genetic algorithm (AGA), the influence of historical data is considered on the basis of UKF, a multi-information theory is combined, an adaptive attenuation factor is introduced to overcome the influence of historical measured values on an estimation result, initial value deviation is inhibited, and the estimation accuracy is improved. A noise adaptive link is introduced to carry out adaptive updating on system noise covariance, an improved pelican algorithm (IPOA) is adopted to optimize distribution adjustment parameters of a UKF during UT transformation, the improved pelican algorithm is adopted to optimize a fractional order multi-innovation adaptive unscented Kalman algorithm (IPOA-FOMIAUKF) to estimate SOC, and finally, a multi-time scale theory is combined, so that the system noise covariance is optimized. The SOC of the battery is estimated by adopting the IPOA-FOMIAUKF under the micro-scale, the SOH is estimated through the UKF under the macro-scale, and the state joint estimation based on the IPOA-FOMIAUKF-UKF is realized through iterative updating.

The invention discloses a combined positioning method based on geomagnetism / vision / DVL. A geomagnetic induction sensor, a vision sensor and a DVL sensor are combined to position a high-speed intelligent platform and construct a position information system of the platform. According to the right 1, a visual sensor is used for collecting image information, and displacement differences between reference objects in a scene are settled through a visual model, so that the moving speed of the platform is calculated. According to the geomagnetic induction sensor in the right 1, the vector included angle between the high-speed intelligent platform and the geomagnetic field can be obtained by measuring the geomagnetic azimuth and fitting the fast reconfigurable Kalman algorithm. According to the right 1, the DVL sensor is used for measuring the speed information of the high-speed intelligent platform under a navigation coordinate system, and the height sensor provides the vertical height information from the ground for the mobile platform. And a local geomagnetic declination compensation angle is obtained through a global geomagnetic field model in combination with the obtained longitude and latitude height information, targets in the scene are recognized through a Riemannian manifold algorithm in combination with a visual system and the obtained position information, and reference information of all objects in the scene is obtained. And integrating the position information, and identifying the target in the scene through a Riemannian manifold algorithm to obtain reference information of each object in the scene.

The application provides a crane outrigger dynamic electronic fence construction method and device and electronic equipment, and relates to the technical field of crane safety monitoring. The method comprises the following steps: acquiring historical key point information in the continuous previous N frames of history images of a target crane outrigger and a current frame image; performing key point extraction on the current frame image to obtain real-time key point information of a plurality of key points of the current frame image; if the number of key points corresponding to the real-time key point information of the current frame image is less than the number of preset position points of the target crane outrigger, calculating the missing information of the key points of the current frame image according to the historical key point information by using a Kalman algorithm, and obtaining effective point information of the preset position points by comprehensively combining the missing information of the key points and the real-time key point information; and calculating a minimum convex hull region by using a convex hull algorithm according to the effective point information, and taking the minimum convex hull region as an electronic fence. The application can accurately construct an electronic fence when the key points of the crane outrigger are blocked.

The invention provides a conveyor speed detection method, device and equipment based on double-antenna RFID. The method comprises the following steps: performing tracking processing on an RFID tag by using a first antenna, and recording a first phase sequence of the RFID tag; performing tracking processing on the RFID tag by using a second antenna, and recording a second phase sequence of the RFID tag; determining a phase difference between the first phase sequence and the second phase sequence, and determining a phase differential velocity according to the phase difference; the Doppler velocity is determined according to the movement data of the RFID tag; and fusing the phase differential velocity and the Doppler velocity by using a Kalman algorithm, determining a final conveyor belt velocity, and outputting the final conveyor belt velocity. The accurate phase differential velocity is determined according to the phase difference of the phase sequences of the two antennas; the Doppler velocity is determined, and the phase difference velocity and the Doppler velocity are fused by using a Kalman algorithm, so that the accurate final conveyor belt velocity is determined, and the high-precision accurate detection of the conveyor belt velocity is realized.

The method for grabbing dynamic objects by the mechanical arm based on visual detection and deep reinforcement learning comprises the following steps: step 1, a visual servo control system of the mechanical arm is built in a simulation scene; step 2, in the visual servo control system of the mechanical arm, a target object is identified based on an improved YOLOv8-pose network to obtain position information of object key points in a two-dimensional image; step 3, the position information of the object key points in the two-dimensional image is combined with a three-dimensional coordinate system of the object itself, a PNP algorithm is used to restore a 6D pose of the object, and a Kalman algorithm is used to smooth a motion trajectory of the moving object and estimate a pose at a next moment; and step 4, a deep reinforcement learning DDPG algorithm model is constructed according to the estimation result to drive the mechanical arm to complete tracking and grabbing of the dynamic object. The efficiency and stability of pose estimation are improved, and the tracking and grabbing performance of the mechanical arm on the dynamic target is improved as a whole.

The invention provides a fall detection method, system and device for a high-integration wearable device, and relates to the technical field of fall prediction, and the method comprises the steps: obtaining the acceleration and angular velocity data of a human body; estimating the Euler angle of the human body by using an extended Kalman algorithm according to the acquired acceleration and angular velocity data of the human body; setting a data threshold value of the human body acceleration as a first threshold value; when the acquired human body acceleration exceeds a first threshold value, calculating the variation of an Euler angle; setting the Euler angle variation threshold as a second threshold; and when the absolute value of the change of the Euler angle in a certain time window exceeds a second threshold value, performing fall risk early warning. And a new hardware equipment structure is constructed by using a communication module supporting secondary development. According to the invention, by optimizing the hardware architecture, the equipment size is reduced, the wearing comfort is improved, and by optimizing the tumble detection algorithm, the tumble detection precision is improved.

The present invention discloses a method, system, device and readable storage medium for decoupling slope and vehicle weight of a commercial vehicle AMT transmission, including the following steps: Step 1, obtain the vehicle running data when the vehicle is running normally; Step 2, set the first moment delay value; Step 3, obtain the slope signal through the Kalman algorithm; Step 4, obtain the vehicle weight signal through the least square algorithm using the slope signal in Step 3; Step 5, perform weighted calculation on the vehicle weight signal to obtain the final vehicle weight information. By establishing a combined algorithm based on the Kalman-least square method, the acceleration sensor is completely separated to calculate the road slope and vehicle weight; an algorithm design using enabling trigger and interruption is adopted to strictly limit the conditions for the establishment of the dynamic equation, improving the data reliability and data convergence speed; thus, the slope and vehicle weight information can be obtained without the acceleration sensor, ensuring the vehicle driving safety before the acceleration sensor resumes its function.

The invention discloses a battery circuit monomer short-circuit resistance fault identification method and a terminal, and the method comprises the steps: building an equivalent circuit model containing short-circuit equivalent resistance for each battery monomer of a series lithium battery system; according to circuit parameters of the equivalent circuit model, in each detection period, SOC estimation is carried out on each single battery through a Kalman family algorithm to obtain a first SOC data set, and SOC estimation is carried out through an ampere-hour integral method to obtain a second SOC data set; calculating a short-circuit resistance data set according to the first SOC data set and the second SOC data set; according to the short-circuit resistance data set, calculating the deviation between the short-circuit resistance of each single battery and a short-circuit resistance mean value, and judging whether the single short-circuit resistance has a fault or not; and short circuit identification without complex electrochemical principle modeling is realized.

The invention discloses a system for improving the data acquisition precision of a force sensor, and the system employs a force sensor AD chip CS1237 to collect AD values, stores the collected AD values in a buffer area, stores the collected AD values in a data flow manner, removes possible interference values through an amplitude limiting filtering algorithm, obtains stable data through an average value filtering algorithm, and carries out the data collection through a data flow. Whether a fast Kalman algorithm or a slow Kalman algorithm is used is finally judged through noise judgment of the data, and finally accurate data are obtained; compared with the prior art, the method has the advantages that compared with a simple average filtering method and a first-order filtering method in the prior art, data values become more accurate and stable, the stability and accuracy of data acquisition of the force sensor are improved, regular floating data and burst peak data can be recognized and filtered out, and the accuracy of data acquisition is improved. The problem that data collected by the force sensor is inaccurate due to signal interference is solved.

The present invention discloses a passive optical motion capture method, which relates to the technical field of motion capture and includes: S1, filtering by using an adaptive unscented Kalman algorithm for dynamically adjusting observation noise; S2, performing feature reconstruction as an observable quantity for state estimation of a target to be measured based on a feature reconstruction method for a fixed target and a feature reconstruction method based on a monocular EPnP algorithm; S3, performing information fusion between multiple groups of binocular systems in the form of matrix-weighted unscented Kalman filtering based on a loose coupling scheme. In view of the problem that outliers on a two-dimensional image are not thoroughly removed during target extraction in a complex background, the present invention proposes a motion capture method using prior volume constraint in three-dimensional space, and a motion capture method based on adaptive unscented Kalman filtering and target feature reconstruction. For a multi-camera system, information fusion can be performed based on loose coupling. By using binocular cameras and retroreflective targets, high-precision six-degree-of-freedom pose measurement in a complex background can be achieved.

The application discloses a multi-channel dereverberation method, comprising inputting multi-channel audio data into a diffuse noise field model and calculating an MVDR superdirectivity beam output; taking the superdirectivity beam output as an expected signal of a Kalman algorithm and predicting a dereverberation output signal according to the Kalman algorithm; iteratively updating the dereverberation output signal of a current frame according to a convergence strategy until a preset condition is reached, stopping the iterative update to obtain a dereverberation signal; wherein the convergence strategy comprises a correction coefficient and a convergence number. Since the Kalman algorithm has direction interference suppression under the superdirectivity beam forming addition effect, the mid-high frequency signal-to-noise ratio is improved and low-frequency reverberation is suppressed, so that the multi-channel linear prediction dereverberation output signal is more accurate, thereby improving the final dereverberation sound quality effect; and the convergence strategy is used to update the dereverberation output signal, so that the performance loss caused by complete diagonalization is overcome and the dereverberation sound quality is improved.