62 results about "Linear dynamical system" patented technology

A gesture recognition interface for use in controlling self-service machines and other devices is disclosed. A gesture is defined as motions and kinematic poses generated by humans, animals, or machines. Specific body features are tracked, and static and motion gestures are interpreted. Motion gestures are defined as a family of parametrically delimited oscillatory motions, modeled as a linear-in-parameters dynamic system with added geometric constraints to allow for real-time recognition using a small amount of memory and processing time. A linear least squares method is preferably used to determine the parameters which represent each gesture. Feature position measure is used in conjunction with a bank of predictor bins seeded with the gesture parameters, and the system determines which bin best fits the observed motion. Recognizing static pose gestures is preferably performed by localizing the body / object from the rest of the image, describing that object, and identifying that description. The disclosure details methods for gesture recognition, as well as the overall architecture for using gesture recognition to control of devices, including self-service machines.

The invention relates to a Kalman filter prediction-based robot obstacle avoidance method. In a complex environment, the robot travelling environment changes dynamically; and when an environment of a preplanned mission is determined to have a significant change, the target objective is modified, the target is planned in real time, and the path is modified. In the obstacle avoidance method, a path scheduler sorts a target set according to a digital map, the target set and the state of the robot acquired by a sensor system so as to generate a robot travelling point sequence; the robot travelling point sequence is executed by a servo system; when the sensor system detects a new obstacle, a Kalman filter model is established according to observation data; and parameters are identified and modified by using the observation data and an expectation maximization model identification algorithm of a classical linear dynamic system; and the digital map is updated for the next turn of local re-planning of the path scheduler. The method can realize obstacle avoidance path planning of the robot generated locally and dynamically in an undetermined environment, and has the advantages of simple implementation and good real-time performance.

A method generates a synthetic textured data signal by first acquiring a time-invariant input textured data signal. The input textured data signal is sampled to construct an observation matrix. The observation matrix is eigen-coding and factoring to identify a linear dynamic system modeling the input textured data signal. Then, the linear dynamic system can be run forward from an initial state using a quadratic regulator and a random noise signal to generate the synthetic textured data signal.

A family of novel multi-layer discrete-time neural net controllers is presented for the control of an multi-input multi-output (MIMO) dynamical system. No learning phase is needed. The structure of the neural net (NN) controller is derived using a filtered error / passivity approach. For guaranteed stability, the upper bound on the constant learning rate parameter for the delta rule employed in standard back propagation is shown to decrease with the number of hidden-layer neurons so that learning must slow down. This major drawback is shown to be easily overcome by using a projection algorithm in each layer. The notion of persistency of excitation for multilayer NN is defined and explored. New on-line improved tuning algorithms for discrete-time systems are derived, which are similar to e-modification for the case of continuous-time systems, that include a modification to the learning rate parameter plus a correction term. These algorithms guarantee tracking as well as bounded NN weights. An extension of these novel weight tuning updates to NN with an arbitrary number of hidden layers is discussed. The notions of discrete-time passive NN, dissipative NN, and robust NN are introduced.

Aspects of the present invention include systems and methods for segmentation and recognition of action primitives. In embodiments, a framework, referred to as the Continuous Linear Dynamic System (CLDS), comprises two sets of Linear Dynamic System (LDS) models, one to model the dynamics of individual primitive actions and the other to model the transitions between actions. In embodiments, the inference process estimates the best decomposition of the whole sequence into continuous alternating between the two set of models, using an approximate Viterbi algorithm. In this way, both action type and action boundary may be accurately recognized.

The invention provides an abnormal behavior detection method based on a linear dynamic system and a deep network. An LDS dynamic system model and a TSN deep learning network are combined, the space-time feature information of actions is extracted through the TSN, then action features are connected in series through an LDS to form complete behavior features, and finally the behavior type is judgedthrough a support vector machine (SVM). According to the method, the 3D convolutional network is established on the basis of the residual network, and the 3D convolutional kernel is established in a 2D+1D form, so that the network parameter quantity is reduced, and the problem that the original 3D network cannot preload the weight is solved. A residual 3D network is introduced into the TSN structure, so that the feature extraction capability of the network is improved. The number of network layers is increased, and the fitting capability of the network is improved. According to the invention,the high-precision identification of long-sequence abnormal actions can be realized, and finally the accurate monitoring of abnormal behaviors is realized.

The invention relates to a touch information classified computing and modelling method based on machine learning. The method comprises the following steps: acquiring a touch sequence of a training set sample, modelling by adopting a linear dynamic system model, extracting dynamic characteristics of a sub touch sequence, calculating distance of the dynamic characteristics of the sub touch sequence by adopting Martin distance, clustering a Martin matrix by adopting a K-medoids algorithm, constructing a code book, carrying out characterization on each touch sequence by adopting the code book to obtain a system packet model, putting the system packet model of the training set sample and a training set sample label into an extreme learning machine for training a classifier, and putting the system packet model of a to-be-classified sample into the classifier to obtain a label for type of an object. The touch information classified computing and modelling method has the advantages that the actual demand of a robot on stable and complaisant grasping of a non-cooperative target is met, data foundation is provided for completion of a precise operation task, and other sensing results can be fused and computed, so that the description and recognition capability on different targets is enhanced by virtue of multi-source deep perception, and a technical foundation is laid for implementation of intelligent control.

An adaptive method for estimating the selling price for an item at auction in order to set a reserve. The method calculates the selling price as a function of selling prices for items previously-sold at auction and differential values attributable to feature differences between the item to be sold and comparative items previously sold. Distance metrics are calculated by comparing the item to be sold with each item in the set of comparative items, and a subset of most similar items is selected according to the calculated distance metrics. A weighting function is then calculated for each item in the subset based on its respective distance metric value, and the selling price is estimated as a function of the weighting functions and the differential values. The differential values are modeled as a linear dynamical system and updated using a Kalman filter as a function of an actual sales price for the item to be sold and a current estimate of uncertainty for the differential values.

The invention discloses a method and system for acquiring the three-dimensional human body motion in real time on line, comprising the following steps: (1) marking three-dimensional sign points, constructing a linear dynamic system and a rigid constraint, setting and screening the thresholds and the point matching cost weights of the candidate point sets of the three-dimensional sign points, initializing data according to the marked three-dimensional sign points, and training the linear dynamic system; (2) screening the candidate point sets, calculating the matching cost, and marking the results; (3) reconstructing the human body posture of the current frame; (4) constructing the relative positions of all three-dimensional sign points under the human body posture of the current frame, and recovering the positions of the lost three-dimensional sign points; and (5) updating the linear dynamic system, the rigid constraint, the thresholds and the weights. The method solves the problem that the precision of the human body motion data reconstructed in real time in the prior art is not high. The reconstructed human body motion data has good intuitive visual effects.

A new framework for video compressed sensing models the evolution of the image frames of a video sequence as a linear dynamical system (LDS). This reduces the video recovery problem to first estimating the model parameters of the LDS from compressive measurements, from which the image frames are then reconstructed. We exploit the low-dimensional dynamic parameters (state sequence) and high-dimensional static parameters (observation matrix) of the LDS to devise a novel compressive measurement strategy that measures only the dynamic part of the scene at each instant and accumulates measurements over time to estimate the static parameters. This enables us to lower the compressive measurement rate considerably yet obtain video recovery at a high frame rate that is in fact inversely proportional to the length of the video sequence. This property makes our framework well-suited for high-speed video capture and other applications. We validate our approach with a range of experiments including classification experiments that highlight the purposive nature of our framework.

The invention discloses a fault detection method for Kalman filtering sensor information fusion. The method includes the steps of: 1. according to a Kalman filtering theory, establishing a state equation and an observation equation of a linear dynamic system; 2. according to the observation equation obtained in step 1, making use of a least square process to acquire state estimation, a corresponding mean square error matrix and an innovation sequence; 3. utilizing the known innovation sequence to acquire different channel-normalized new innovation sequences, and composing an innovation matrix of an m channel parallel sensor; 4. acquiring the spectral norms and spectral norm mean value of the innovation matrix according to the innovation matrix obtained in step 3; and 5. carrying out fault detection on the Kalman filtering sensor information fusion. The method provided in the invention adopts mathematical statistics and interval estimation, simplifies complex calculation, and substantially improves the fault detection speed.

The invention discloses a small-interference stability rapid analysis method targeted at a large scale electric power system. According to the method, an electric power system network, network parameters, a dynamic element model and model parameters are firstly obtained to form a differential algebraic equation set DAE of a system. Linearization of the DAE is carried out at a system steady state point (x0, y0) to obtain a corresponding linear dynamic system. A system key oscillation modal characteristic value is computed in parallel. Finally, based on the obtained key oscillation modal characteristic value computing result, a damping ratio of the system key oscillation modal is obtained. Through comparison with a damping ratio critical value zeta0, the stability of the system is determined. If a non-stable oscillation modality exists, participation factors of system variables in the modality can be calculated according to the characteristic value obtained through calculating and a characteristic vector thereof so as to further determine dynamic elements strongly related to the non-stable oscillation modality. The method can carry out rapid analysis on small-interference stability of the large scale electric power system so as to satisfy real-time requirements.

The invention discloses a supervision-based industrial process fault detection method of a linear dynamic system model, and is applied to fault detection under the condition of obtainable quality variables in an industrial process. By use of the linear dynamic system model and the quality variables, the linear dynamic system model with supervision is established, and the dynamics of a process and the randomness of data are taken into consideration. Compared to other conventional methods, the method provided by the invention not only improves the fault detection effect of the industrial process, but also enhances the grasp of process operators for process states, and enables industrial production to be safer and product quality to be more stable; and the reliance of the conventional fault detection method on process knowledge is improved to a quite large degree, and automatic implementation of the industrial process is better facilitated.

The invention discloses a soft measurement modeling method based on a monitored linear dynamic system model. The method achieves the soft measurement modeling of the dynamic process of the industrial production in a noise environment, and the prediction of the quality variable which is hard to predict directly. Based on the monitored linear dynamic system model, the method builds an effective soft measurement model, and overcomes the process dynamic nature and the data collection randomness in the industrial production. Compared with other methods in the prior art, the model built by the method is more accurate, the prediction of the model is more accurate, so the product quality is more stable. Besides, the dependency of the soft measurement modeling on the process knowledge is reduced, and the automatic implementation of the industrial process is benefited.

The invention discloses a time-space local feature extraction method based on a linear dynamic system. The method comprises the following steps: 1, collecting a video data file to be processed; 2, extracting the three-dimensional time-space local features of the video data file to be processed; 3, unfolding each three-dimensional time-space local feature along space dimension to obtain a two-dimensional time-space local feature; 4, learning to obtain the model parameter of the linear dynamic system by taking the two-dimension time-space local feature Y as the output of the linear dynamic system, wherein the model parameter of the linear dynamic system is taken as a describer of the two-dimensional time-space local feature Y. The extracted describer can be used for expressing the static apparent information and motion information of the time-space local feature at the same time, and the method can be widely applied to the businesses of video content retrieval, sensitive video detection and filtering, intelligent video monitoring and the like.

The invention discloses an industrial process fault diagnosis method based on a switching linear dynamic system model. The method comprises the steps of firstly carrying out independent repeated sampling on normal operation data and various known fault data of the industrial process, and then establishing the switching linear dynamic system model through a learning algorithm of the switching linear dynamic system model; and then acquiring a diagnosis result of the current monitoring data by using a Gaussian sum filtering method, that is, judging whether the current data is located at a normal operating condition, and if not, judging which fault the current data is located. Compared with other method at present, the industrial process fault diagnosis method not only improves a fault diagnosis effect of the industrial process, enhances mastering of a process operator for the process state, enables industrial production to be safer, and enables the product quality to be more stable. In addition, the industrial process fault diagnosis method improves the dependency of a fault diagnosis method for process knowledge to a great extent, thereby being more conducive to automation implementation of the industrial process.

A method and apparatus is disclosed for improving analysis and design of physical structures whose dynamic behavior can be modeled as a linear dynamical system. The method and apparatus is operative to decouple the equations of motion of any linear dynamical systems therewith by processing the equations of motion of any linear dynamical system using a decoupling unit.

The invention discloses a video keyframe extraction method based on a linear dynamic system. The method comprises the steps of (1), collecting a video data file to be processed; (2), initializing a video segment, calculating linear dynamic system model parameters of the video segment, and calculating the reconstruction error of the video segment according to the model parameters; (3), increasing the length of the video segment frame by frame, and repeating the step (2) till the reconstruction error is beyond a pre-set threshold value; (4), taking the finally determined middle frame of the video segment as the keyframe of the segment; and (5), initializing the next new video segment after the previous video segment, and repeating the steps (2)-(3) till the video data file is ended. By means of the video keyframe extraction method based on the linear dynamic system disclosed by the invention, the description capability of keyframes in semantic contents can be obviously improved; and thus, the video keyframe extraction method can be applied in the services, such as internet video content retrieval, sensitive video detection and filtering and intelligent video monitoring.

The invention discloses a state perception optimization method for improving fault diagnosability in a noise environment. The method comprises the following steps: firstly obtaining quantitative fault diagnosability performance indexes by considering the influence of process noise and observation noise; subsequently building a genetic algorithm fitness function by taking minimum cost of measuring point arrangement as an optimization target; finally performing state perception selection based on a binary system genetic algorithm, thus finishing the state sensor optimizing process. The method has the advantages of transferring the key emphasis in work for improving the system fault diagnosis capability to a design stage, thereby ensuring the stability and the design accuracy of a random linear dynamic system, greatly reducing the number of state perception simultaneously, reducing the configuration number of sensors in a satellite control system and reducing the design cost of the satellite control system.

An adaptive method for estimating the selling price for an item at auction in order to set a reserve. The method calculates the selling price as a function of selling prices for items previously-sold at auction and differential values attributable to feature differences between the item to be sold and comparative items previously sold. Distance metrics are calculated by comparing the item to be sold with each item in the set of comparative items, and a subset of most similar items is selected according to the calculated distance metrics. A weighting function is then calculated for each item in the subset based on its respective distance metric value, and the selling price is estimated as a function of the weighting functions and the differential values. The differential values are modeled as a linear dynamical system and updated using a Kalman filter as a function of an actual sales price for the item to be sold and a current estimate of uncertainty for the differential values.

The invention discloses a similarity measurement computing method of linear dynamical systems. The method comprises the following steps that time series data to be processed are collected; linear dynamical system model parameters corresponding to the time series data are respectively computed; subspace included angles between the corresponding linear dynamical systems are obtained by computing the linear dynamical system model parameters; frame offset optimization is carried out on the subspace included angles to obtain the similarity measurement of the time series data. The method can be applied to time series data analysis, video content retrieval, intelligent video monitoring and other services.

The invention provides a voltage stability judgment method based on a key direct current of a multi-infeed AC / DC system. The method comprises the following steps: collecting network parameters of the multi-infeed AC / DC system, a dynamic element model and model parameters, and building a differential-algebra equation set of the multi-infeed AC / DC system; executing linear processing on a system steady-state point (X<0> and Y<0>), and obtaining a linear dynamic system model; determining a system feature value and corresponding left feature vector and right feature vector on the basis of a small-disturbance stability analysis method, obtaining the feature values lambda<Ui> (i=1,2, ... ,m) related to the stability in the multi-infeed AC / DC system through the extinction angle correlation ratio rho<gammai> and determining the minimum feature value lambda<Umin>; and judging the small-disturbance voltage stability of the multi-infeed AC / DC system by the lambda<Umin>, and determining the key direct current of affecting the system voltage stability according to a participation factor p<jmin> corresponding to the lambda<Umin>. The method is suitable for computational analysis of a power system, can effectively and quickly search the key direct current of affecting the voltage stability of the multi-infeed AC / DC system, and provides a new technical support for prevention of a voltage collapse accident of a large power grid.

The present invention discloses an industrial process fault diagnosis method based on a switching supervised LDSM (Linear Dynamic System Model), which is used for fault diagnosis on the condition that a key quality variable is obtainable in an industrial process. According to the fault diagnosis method, a supervised LDSM is expanded to a multi-modal form, and a switching supervised LDSM is established, thus dynamic characteristics and random characteristics of process data are considered, and important process operation information included in quality variables is also fully utilized. In comparison with the conventional method, the fault diagnosis method improves the capability of describing industrial process operation states by the model, improves a fault diagnosis effect, shortens delay time of diagnosis, enables fault processing to be more timely and effective, and is more beneficial to automatic enforcement of industrial process.

A new framework for video compressed sensing models the evolution of the image frames of a video sequence as a linear dynamical system (LDS). This reduces the video recovery problem to first estimating the model parameters of the LDS from compressive measurements, from which the image frames are then reconstructed. We exploit the low-dimensional dynamic parameters (state sequence) and high-dimensional static parameters (observation matrix) of the LDS to devise a novel compressive measurement strategy that measures only the dynamic part of the scene at each instant and accumulates measurements over time to estimate the static parameters. This enables us to lower the compressive measurement rate considerably yet obtain video recovery at a high frame rate that is in fact inversely proportional to the length of the video sequence. This property makes our framework well-suited for high-speed video capture and other applications. We validate our approach with a range of experiments including classification experiments that highlight the purposive nature of our framework.

The invention relates to a rock-soil mass parameter two-dimensional space variability characterization method based on multi-surface wave exploration, and belongs to the field of geotechnical engineering investigation. The method includes establishing a two-dimensional space linear dynamic system for representing the spatial variability of the rock and earth mass parameters; constructing a Bayesian equation for identifying parameters of a two-dimensional space linear dynamic system according to a Bayesian theory under the condition that two-dimensional shear wave velocity data of multi-surface wave exploration and an empirical relationship between exploration data in literatures and rock-soil body parameters are given; solving optimal model parameters by adopting a Bayesian updating method based on subset simulation; and obtaining the most possible two-dimensional spatial distribution of the rock-soil body parameters by adopting a forward recursive algorithm and a backward recursive algorithm. According to the invention, the influence of the two-dimensional spatial correlation of the shear wave velocity data on the spatial variability characterization recognition of the rock-soil body parameters is reasonably considered, the two-dimensional spatial exploration data are fully utilized to characterize the unstable spatial variability of the rock-soil body parameters, the limitation of a traditional steady-state random field model is overcome, and an effective way is provided for improving the exploration accuracy of a geotechnical engineering site.

The invention relates to a method for modeling and recognizing a time sequence, and belongs to the field of machine learning. The method comprises the steps: enabling dynamic data formed by the sequential arrangement of data, collected by a sensor, according to the sequence of the collection time to form L time sequences, and randomly selecting N time sequences (60%-80%) of the L time sequences as a training set, wherein the remaining time sequences serve as a test set; modeling each time sequence in the training set through employing a linear dynamic system, and employing the features of each time sequence in the training set for representation; randomly extracting J time sequence as dictionaries from the training seat to form a dictionary set, learning the optimal feature representation of each time sequence in the dictionary set from the obtained feature representation of each training time sequence in the training set, and calculating the coding coefficient of each training time sequence in the dictionary set; training a support vector machine model through the coding coefficients of the training set, and achieving the recognition of the time sequences. The method greatly reduces the complexity of data representation, and remarkably improves the recognition precision.

The invention discloses a Gaussian process model-based predictive control method for a multi-variable nonlinear dynamic system model, belongs to the technical field of predictive control of the multi-variable nonlinear dynamic system model and aims at solving the technical problem of providing improvement of the Gaussian process model-based predictive control method for the multi-variable nonlinear dynamic system model. The technical scheme adopted for solving the technical problem is as follows: the predictive control method comprises the following steps of (1) building an external dynamic PLS framework; (2) predicting output data and obtaining a plurality of single-input and single-output systems in hidden space through decoupling of a dynamic GP-PLS model; (3) carrying out control by using the dynamic GP-PLS model and designing a model predictive controller in each single-input and single-output system; (4) obtaining an optimum control action through minimizing an objective function; and (5) reconstructing the model predictive control result in the hidden space back to original space and controlling the original space. The Gaussian process model-based predictive control method is applied to the multi-variable nonlinear dynamic system model.

Aspects of the present invention include systems and methods for segmentation and recognition of action primitives. In embodiments, a framework, referred to as the Continuous Linear Dynamic System (CLDS), comprises two sets of Linear Dynamic System (LDS) models, one to model the dynamics of individual primitive actions and the other to model the transitions between actions. In embodiments, the inference process estimates the best decomposition of the whole sequence into continuous alternating between the two set of models, using an approximate Viterbi algorithm. In this way, both action type and action boundary may be accurately recognized.

We disclose transconductor-capacitor classical dynamical systems that emulate quantum dynamical systems and quantum-inspired systems by composing them with 1) a real capacitor, whose value exactly emulates the value of the quantum constant h termed a Planck capacitor; 2) a ‘quantum admittance’ element, which has no classical equivalent, but which can be emulated by approximately 18 transistors of a coupled transconductor system; 3) an emulated ‘quantum transadmittance element’ that can couple emulated quantum admittances to each other; and 4) an emulated ‘quantum transadmittance mixer element’ that can couple quantum admittances to each other under the control of an input. These four parts can be composed together to create arbitrary discrete-state, traveling-wave, spectral, or other quantum systems.