36 results about "Bernoulli filter" patented technology

The invention discloses a multi-Bernoulli video multi-target detection tracking method based on YOLOv3, and belongs to the field of machine vision and intelligent information processing. According tothe method, a YOLOv3 detection technology is introduced under a multi-Bernoulli filtering framework, an anti-interference convolution feature is adopted to describe a target, a detection result and atracking result are interactively fused, and accurate estimation on the multi-target state of the video with unknown number and time varying is realized; in the tracking process, a matched detection frame is combined with a target track and a target template, target newborn judgment and occlusion target re-identification are carried out in real time, identity label information of a detection target and an estimation target is considered at the same time, target identity identification and track tracking are achieved, the tracking precision of an occluded target can be effectively improved, andtrack fragments are reduced. Experiments show that the method has a good tracking effect and robustness, and can widely meet the actual design requirements of systems such as intelligent video monitoring, man-machine interaction and intelligent traffic control.

The invention relates to a multi-target passive tracking method based on a wireless sensor network. The technical characteristics are that the multi-target passive tracking method based on the wireless sensor network comprises the following steps that: (1) a measurement model is established according to the receiving signal intensity of the sensor network; (2) according to the established measurement model from the step (1), the variable multi-target positioning and tracking is realized under the indoor environment through the combination of a multi-target Bernoulli filter algorithm and a particle filter algorithm. The multi-target passive tracking method based on the wireless sensor network provided by the invention is reasonable in design; the established measurement model has high accuracy under the indoor environment, and the model forecast value is approximate to the actual observation value; the target detecting and tracking algorithm has high accuracy and stability, and can detect and track a plurality of targets; and the measurement model and the target algorithm are appropriate in calculation complexity, and thus guaranteeing the running real-time performance of the tracking system.

The invention discloses an expansion target tracking method based on GLMB (Generalized labelled multi-bernoulli) filtering and Gibbs sampling. The expansion target tracking method based on GLMB filtering and Gibbs sampling estimates the target number and the shape of the expansion target, provides a multiple expansion target tracking method under a labelled random finite sets (L-RFS) framework, and mainly includes two aspects: dynamic modeling of multiple expansion targets and tracking estimation of multiple expansion targets. The expansion target tracking method based on GLMB filtering and Gibbs sampling includes the steps: combined with a generalized label multi-bernoulli filter, establishing a measurement limit hybrid model of the expansion targets, by means of Gibbs sampling and Bayesian information criterion, deriving the parameters of the limit hybrid model to learn tracking of the state of the multiple expansion targets, using an equivalent measurement method to replace measurement generated from the expansion targets, and performing ellipse approximating modeling on the shape of the expansion targets to realize estimation of the shape of the expansion targets. The simulation experiment shows that the expansion target tracking method based on GLMB filtering and Gibbs sampling can effectively track the multiple expansion targets, can accurately estimate the state and theshape of the expansion targets, and can obtain the track of the targets.

The invention belongs to the technical field of intelligent information processing, and relates to a multi-target tracking method based on a variational Bayesian label multi-Bernoulli superposition model. The noise covariance of the superposition model is estimated. On the basis of an original superposition model, the covariance of measurement noise is unknown, unknown parameters are estimated based on variational Bayes, the prediction and updating process of the superposition model marked with the multi-Bernoulli filter is achieved, state extraction is conducted, and therefore the tracking problem of the superposition model under unknown measurement noise is solved. The method has the characteristics of wide application range, strong robustness, high estimation precision and the like, caneffectively solve the problem of non-cooperation in an actual superposition model scene, realizes multi-target tracking and parameter estimation in a complex scene, can meet design requirements, andhas a good engineering application value.

The invention discloses a random set theory-based multi-sensor adaptive control method. With the random set theory-based multi-sensor adaptive control method, general labeling multi-target Bernoulli filter-based multi-sensor adaptive control can be realized. According to the method, in a filtering stage, each sensor carries out filtering locally and carries out distributed fusion, and therefore, optimal global performance can be realized. In a control stage, firstly, fused multi-target distribution is sampled; secondly, pseudo prediction is carried out, pseudo prediction distribution is obtained, pseudo updated distribution is obtained through a plurality of steps of filtering iteration, and distributed fusion is carried out on the pseudo updated distribution, so that fused pseudo updated distribution can be obtained; and finally, Cauchy-Schwarz divergence between the pseudo updated distribution and the fused pseudo updated distribution is calculated, and therefore, an optimal sensor control decision can be selected out. With the method adopted, problems in multi-sensor optimal control in practical multi-sensor network application can be solved, and thus, multi-sensor adaptive control can be realized.

The invention discloses a maneuvering weak target Track-before-Detect method of interactive multi-mode Bernoulli filtering. The maneuvering weak target Track-before-Detect method of interactive multi-mode Bernoulli filtering is characterized in that on the basis of Bernoulli particle filtering, an interactive multi-mode method is used to predict the sampling particle of each target state, and thena prediction particle is substituted into a traditional Bernoulli TBD (Track-before-Detect) algorithm to perform update estimation of the existing probability and distribution density of the targets.In the low signal to noise ratio situation, the maneuvering weak target Track-before-Detect method of interactive multi-mode Bernoulli filtering can effectively detect and track the weak maneuveringtarget, and can be more accurate in estimation of the target state.

The invention discloses a random set theory-based multi-target tracking method. The method includes the following steps that: a multi-target state space is expanded, model dimension is increased based on original kinetic information, and therefore, characterization of target model information can be realized; a state transfer function and a likelihood function are expanded based on a jump Markov system, so that the state transfer function and the likelihood function can contain the model information; and the prediction and update process of an expanded multi-model generalized label multi-target Bernoulli filter is realized, and target states are extracted, and a target motion model is estimated. With the method adopted, problems in maneuvering multi-target tracking can be solved. The method has the advantages of high robustness, high adaptability and high estimation accuracy. With the method adopted, the problem of high multi-target maneuverability and inconsistency in practical application can be solved, and maneuvering multi-target tracking and target motion model estimation under complex scenes can be realized.

The invention is suitable for the technical field of target tracking, and provides a multi-target tracking method and system under flicker noises, and the method comprises the steps: predicting a prediction distribution function and a prediction label multi-Bernoulli filtering density of an existing target at a current moment through the distribution function and label multi-Bernoulli filtering density of each target at a previous moment; setting a preset distribution function and a preset label multi-Bernoulli filtering density for the new target; combining the two distribution functions andthe label multi-Bernoulli filtering density to obtain a predicted distribution function and a predicted label multi-Bernoulli filtering density of each target at the current moment; and processing thepredicted distribution function and the predicted label multi-Bernoulli filtering density of each target at the current moment to obtain the distribution function and the label multi-Bernoulli filtering density of each target at the current moment, and taking the distribution function and the label multi-Bernoulli filtering density as the input of a filter at the next moment. According to the invention, the filter can accurately extract the target state of each target at the current moment in the flicker noise environment, and the multi-target tracking precision is improved.

The present invention relates to a uniformly dense clutter sparse method aiming at a finite set tracking filter. In order to overcome the problem that in the traditional algorithms, computing time has exponential growth along with the growth of clutter density, the invention provides the method. The method is operated with the hypothesis testing theory as the norm, with the help of a mixed Gaussian-potential probability density filter and a mixed Gaussian-multi-Bernoulli filter. The hypothesis testing theory is used for verifying a clutter sparse process to overcome the problem that in the traditional algorithms, computing time has exponential growth along with the growth of clutter density. Then, computing efficiency is greatly improved.

The invention discloses a multi-sensor adaptive angle control method based on GLMB filter, which comprises the following steps: S1, using a generalized label multi-Bernoulli filter to perform local filtering; S2, performing distributed fusion based on a generalized covariance cross criterion; S3, sampling multi-target; S4, using the generalized label multi-Bernoulli filter to perform a pseudo prediction; S5, calculating an ideal measurement of each sensor; S6, grouping the sensors into two groups, generating a decision set, and performing distributed fusion processing on the sensor pairs in the decision set; S7, calculating the Cauchy Schwarz divergence under the control decision combination of the pairwise coupling; S8, selecting a locally optimal sensor control decision for each sensor pair; S9, combine all local optimal sensor control decisions into one set. The multi- sensor adaptive angle control method based on GLMB filter can ensure small error loss, and solve the multi-sensor adaptive angle control based on GLMB filter which can be realized with precision and calculation cost.

The invention relates to a vehicle detection and tracking method based on an adaptive model, and belongs to the field of automatic driving. The method comprises the following steps: 1) preprocessing point cloud data at t-1 moment and t moment; 2) converting the point clouds at the t-1 moment into a laser radar coordinate system at the t moment, and constructing a grid map under a polar coordinate system according to the point clouds of two continuous frames; 3) determining possible dynamic objects at the moment t, and carrying out data association to find an associated object of each dynamic object at the moment t-1; 4) fitting a dynamic object and an associated object thereof based on a convex hull model, and screening dynamic vehicle candidates by using a motion evidence criterion; 5) deducing the real size and pose of the vehicle; 6) finding an associated object of the dynamic vehicle at the t + 1 moment through data association, and then screening the dynamic vehicle by using a motion consistency criterion; 7) judging whether the new vehicle and other vehicles have a vehicle following relationship or not; and 8) tracking the vehicle by using a tag multi-Bernoulli filter.

The invention discloses a multi-target video tracking system based on fractal feature estimation. The multi-target video tracking system comprises an image local fractal feature estimation module, anobservation model establishment and likelihood function calculation module and a multi-Bernoulli filter tracking module. The local fractal feature estimation module calculates local fractal features of each frame of image and forms a new Hurst index image. The observation model establishing and likelihood function calculating module is used for establishing an observation model and calculating a likelihood function by utilizing local fractal characteristics and a Hurst index. The multi-Bernoulli filter tracking module is used for tracking a plurality of targets in the Hurst index image. According to the method, under the condition that the sensor is static or moves, a plurality of targets can be correctly tracked, and good robustness is achieved for complex tracking such as target enteringand leaving, target part shielding and target number changing.

The invention provides an underwater multi-station combined multi-target tracking method and system. The method comprises the steps: dividing observation nodes into pairs at a current sampling moment, carrying out observation pair combination, randomly selecting the observation pair combination at the same probability, carrying out the azimuth measurement cross positioning, and obtaining a new target set; generating target information at the current moment, wherein the target information comprises a track of each target; taking the target information at the current moment and the target information at the previous moment as the input of a multi-station multi-Bernoulli filter, carrying out one-step prediction and measurement updating, and outputting the maximum posteriori state estimation of the target at the current moment; and comparing the existence probability of the target with a first threshold, performing counter accumulation on the detection times exceeding the threshold, and finally comparing an accumulation result with a second threshold to control the real-time output of the target track. According to the method, a stable and continuous target track can be output in real time, and meanwhile, a short track formed by a false target is filtered out.

The invention discloses a DOA tracking method based on multi-Bernoulli filtering. The method comprises the following steps: receiving superposed measurement data through a sensor array; obtaining filtering posteriori information obtained by a multi-Bernoulli filter at the k-1 moment, wherein the filtering posteriori information comprises the existence probability of a Bernoulli component and a spatial distribution probability density function of a target; predicting the multi-Bernoulli components according to a multi-Bernoulli filter to obtain multi-Bernoulli posteriori information at the moment k; extracting a target state according to the predicted multi-Bernoulli components; and performing iteration, k is equal to k + 1 until all moments are processed. Measurement information does not need to be processed, and the calculated amount is reduced; and during tracking, the number of signal sources does not need to be known, and the signal sources are tracked in real time by directly utilizing prediction prior and current measurement information. A simulation result shows the effectiveness of the algorithm.

The invention provides a fault tracking and positioning method based on the Internet of Things. The method comprises the following steps: constructing a multi-dynamic sensor network; coupling an anti-swarm distributed collaborative search algorithm and a multi-Bernoulli filtering algorithm to realize fault tracking and acquiring real-time sensor data; establishing a weight clustering analysis model for sensor data of a historical normal sample, and calculating to obtain a control limit of a control quantity; performing fault detection on the acquired real-time sensor data according to the established weight clustering analysis model and the control quantity, screening out a fault data segment, and obtaining a control quantity corresponding to the fault data segment; constructing a random forest classification regression algorithm based on the process variable and the control quantity of the fault data segment; acquiring a variable importance measurement of the process variable through the random forest classification regression algorithm, and determining a fault variable according to the variable importance measurement for fault positioning. The method provided by the invention has high tracking and positioning precision and obvious superiority.

The invention discloses a self-adaptive sensor management method for multi-sensor multi-target tracking. The method comprises the following steps of S100, according to sensor network bandwidth limitation and tracking processing real-time requirements, the number of sensors selected at each moment being stipulated to be P; S200, modeling sensor management as a partially observable Markov decision model; S300, estimating the amount of information which can be acquired by each sensor in the sensor network by using the partially observable Markov decision model; S400, P sensors with the most information amount being selected from all the sensors; S500, sorting the selected P sensors according to the sequence of the information amount from small to large; and S600, a control instruction being sent to the sensors, the selected P sensors being activated, target measurement being acquired, and target number and state estimation being realized by using a generalized label multi-Bernoulli filter.

The invention discloses a radar weak fluctuation target track-before-detect algorithm based on multi-Bernoulli filtering, and the algorithm not only considers amplitude information, but also carries out marginalization processing on a phase in MB-TBD, so as to improve the discrimination between a target and noise. And a square modulus likelihood ratio (SLR) is replaced with complex likelihood ratios (CLR) of three Swerling types. In order to adapt to the condition that undulating target new prior information is unknown, a multi-Bernoulli filter self-adaptive new distribution TBD (LABer-STC-TBD) based on the measurement likelihood ratio is provided by referring to the idea of target successive division, and compared with an existing MB-TBD self-adaptive new distribution algorithm, the new algorithm overcomes the defect that when a target undulates, it is difficult to detect a weak target and a strong target which appear at the same time. The Bernoulli components of the same target are combined by an algorithm (DPM) based on distance and particle weight after the updating of the MB-TBD is finished. And finally, the studied estimation and detection performances under different conditions are compared, and the advantages of the LABer-STC-TBD algorithm under target amplitude fluctuation are displayed.

The present invention discloses a Low-computation complexity Bernoulli filter for single-object tracking. The whole Bayes derivation process is simplified, and the target state transfer function and a target measurement likelihood function are optimized. The target measurement and the state independent clutter statistics characteristics are analyzed to derive an improved sequential Monte Carlo realization process so as to effectively extract the real goal in the clutter environment. The filter can effectively track the maneuvering single goal in the clutter environment and has low computation complexity and high tracking precision.