State fusion method and device based on gaussian mixture model

Abstract

Embodiments of the present application provide a state fusion method and device based on a Gaussian mixture model. In one embodiment, the method comprises: firstly, for any target observation object in a plurality of target observation objects: performing particle distribution sampling on the observed object according to position information of the target observation object to generate a position probability distribution of the observed object observed by the target observation object; and performing Gaussian mixture model fitting optimization processing on the position probability distribution to generate a quasi-Gaussian mixture model; secondly, performing Chernoff fusion on the quasi-Gaussian mixture model corresponding to each target observation object in the plurality of target observation objects to generate a state probability density function of the observed object. Thus, the plurality of unknown and non-correlated states corresponding to the same observed object at the same time can be effectively fused, thereby improving the accuracy of the state description of the observed object.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data processing, and particularly relates to a state fusion method and device based on a Gaussian mixture model. BACKGROUND

[0002] The state estimation and fusion process under a distributed architecture will be disturbed by multiple sources of correlation coupling. Ignoring this correlation will produce overconfident estimation results or even divergence. In a large sensor network, it is very difficult to accurately estimate these correlations. In some applications, such as mapping or weather prediction, the fusion process may handle hundreds or even thousands of estimates. It is very expensive to maintain and focus on the correlation between these states, and the magnitude grows exponentially with the number of estimates to be fused. Therefore, many studies have proposed suboptimal fusion strategies that do not track the correlation between multiple state estimates when fusing them. Current data fusion techniques for unknown correlations mainly include data decoupling, correlation estimation and ellipsoid fusion.

[0003] Data decoupling: a common problem in distributed architecture occurs when the same data follows different or circular paths to the fusion node, resulting in repeated calculations. An effective way to avoid this problem is to keep a record of the update of the estimate. A method of directly eliminating repeated data counting is proposed, which stores the received measurements and other sensor node state estimates and uses them to update its own state estimate. This can make repeated data counting be deleted before data fusion. This method assumes a specific network topology to avoid the correlation of state estimates due to repeated counting. In addition, as the number of sensors increases, these methods become very inefficient and impractical.

[0004] Correlation estimation: although it is difficult to obtain the exact correlation between local estimates in a distributed architecture, the properties of the joint covariance matrix have certain limitations on the possible correlation. In addition, certain applications may provide prior knowledge and constraints on the degree of correlation, so that we can infer the strength of the correlation between two local estimates. In fact, the estimates provided by multiple sensors are neither independent nor completely dependent, which means that the cross-correlation is not completely unknown. Therefore, the information and prior knowledge of unknown correlation can be used to improve the accuracy of unknown correlation fusion. The correlation estimation method relies on prior information and empirical knowledge, and is suitable for multi-sensor fusion in specific scenarios, and most of them are fixed sensors.

[0005] Ellipsoidal fusion: The two Gaussians to be fused can be described as ellipsoids ε1, ε2 in space, where the lengths of the ellipsoid axes correspond to the eigenvalues, and the covariance matrix and eigenvectors define its orientation. The covariance terms that can exist between data sources are bounded, thus limiting the possible outcomes of the fused covariance to a bounded set. This is consistent with the correlation modeling approach above, except that ellipsoidal fusion considers the geometric properties of Gaussians. For different choices of the covariance terms P 12 , the fused covariance P f will lie within the intersection of the data sources. The goal of the ellipsoidal method (EM) is to find a bounding covariance P EM such that for any possible P 12 , P EM ≥ P f (P 12 ).

[0006] The EM approach attempts to obtain the fused data by approximating the intersection of each distribution, which is represented by the covariance intersection (CI) method. The fusion criteria of the CI method guarantees consistent convergence but is overly conservative and is limited to Gaussian distributions. The CI method can only be applied to Gaussian distributions, while actual measurement models often do not satisfy this requirement. SUMMARY

[0007] In view of the above problems existing in the prior art, embodiments of the present application provide a state fusion method and device based on a Gaussian mixture model; the method can effectively fuse a plurality of unknown and non-correlated states of a same observed object at a same time, thereby improving the accuracy of the state description of the observed object.

[0008] According to a first aspect of embodiments of the present application, a state fusion method based on a Gaussian mixture model is provided, comprising: for any target observed object in a plurality of target observed objects: performing particle distribution sampling on the observed object according to position information of the target observed object, to generate a position probability distribution of the observed object observed by the target observed object; performing Gaussian mixture model fitting optimization processing on the position probability distribution, to generate a quasi-Gaussian mixture model; and performing Chernoff fusion on the quasi-Gaussian mixture model corresponding to each target observed object in the plurality of target observed objects, to generate a state probability density function of the observed object.

[0009] Optionally, the particle distribution sampling of the observed object according to the position information of the target observation object to generate the position probability distribution of the observed object comprises: performing Gaussian distribution random sampling on the observed object by an angle sensor to obtain an angle particle set of the observed object; wherein the Gaussian distribution is used to indicate the measurement error of the angle sensor at a preset angle; performing distance distribution random sampling on the observed object to generate a distance particle set of the observed object; for any angle particle in the angle particle set: selecting a distance particle having the same order as the angle particle from the distance particle set; determining the position information corresponding to the angle particle based on the selected distance particle, the angle particle, and the position information of the target observation object; and determining the position probability distribution of the observed object based on the position information corresponding to each angle particle in the angle particle set.

[0010] Optionally, the distance distribution random sampling on the observed object to generate the distance particle set of the observed object comprises: performing uniform distribution random sampling on the observed object to obtain a sampling particle set; for any sampling particle in the sampling particle set: performing distance conversion on the sampling particle by a sampling inverse transformation method to generate the spatial distance of the sampling particle based on the sampling parameter of the sampling particle, the minimum spatial distance from the target observation object to the sampling particle, and the maximum spatial distance from the target observation object to the sampling particle; and determining the distance particle set of the observed object based on the spatial distance corresponding to each sampling particle in the sampling particle set.

[0011] Optionally, the Gaussian mixture model fitting optimization processing on the position probability distribution to generate a quasi-Gaussian mixture model comprises: configuring the number of Gaussian models in the Gaussian mixture model; performing model parameter estimation on the Gaussian mixture model by an EM algorithm according to the position probability distribution to generate candidate model parameters; wherein the candidate model parameters comprise mean and variance; and performing iterative optimization on the Gaussian mixture model based on the candidate model parameters and the position probability distribution to generate the quasi-Gaussian mixture model.

[0012] Optionally, the iterative optimization on the Gaussian mixture model based on the candidate model parameters and the position probability distribution to generate the quasi-Gaussian mixture model comprises: generating a virtual sampling data set conforming to the Gaussian distribution based on the candidate Gaussian mixture model corresponding to the candidate model parameters; performing optimization processing on the candidate Gaussian mixture model by an EM algorithm according to the virtual sampling data set and a real data set corresponding to the position probability distribution to generate optimized model parameters; generating virtual data conforming to the Gaussian distribution again based on the Gaussian mixture model corresponding to the optimized model parameters, and performing optimization processing again until the mean variation reaches a preset threshold, and then ending the optimization processing to generate quasi-model parameters; and performing updating processing on the Gaussian mixture model based on the quasi-model parameters to generate the quasi-Gaussian mixture model.

[0013] Optionally, the step of performing Chernov fusion on the quasi-Gaussian mixture model corresponding to each of the plurality of target observation objects to generate the state probability density function of the observed object includes: generating a sequence of quasi-Gaussian mixture models based on the quasi-Gaussian mixture model corresponding to each of the plurality of target observation objects; for any quasi-Gaussian mixture model in the sequence of quasi-Gaussian mixture models: constructing a weighted non-negative least squares problem based on the quasi-Gaussian mixture model using the σ-point method; solving the weighted non-negative least squares problem under the condition that the quasi-Gaussian mixture model is converted from a fractional power form to a power form to generate mixture weights; updating the quasi-Gaussian mixture model based on the mixture weights to generate a power form; generating a power form sequence based on the power form corresponding to each quasi-Gaussian mixture model in the sequence of quasi-Gaussian mixture models; and performing Chernov fusion on the power form sequence of power forms in sequence to generate the state probability density function of the observed object.

[0014] Optionally, the step of sequentially performing Chernov fusion on the power model sequence according to the power model sequence to generate the state probability density function of the observed object includes: for any current power model in the power model sequence: obtaining the previous power model located before and adjacent to the current power model; performing fusion parameter optimization processing based on the previous power model and the current power model to generate current quasi-fusion parameters; performing Chernov fusion processing on the previous power model and the current power model based on the current quasi-fusion parameters to generate the current fused model; using the current fused model as the previous power model, continuing to perform the next fusion parameter optimization processing, until the last power model in the power model sequence is fused by Chernov to generate the state probability density function of the observed object.

[0015] According to a second aspect of the present invention, a state fusion apparatus based on a Gaussian mixture model is also provided, comprising: a generation module, configured to, for any one of a plurality of target observation objects, perform particle distributed sampling on the observed object based on the position information of the target observation object to generate a probability distribution of the observed object's position observed by the target observation object; and perform Gaussian mixture model fitting optimization processing on the position probability distribution to generate a quasi-Gaussian mixture model; and a fusion module, configured to perform Chernov fusion on the quasi-Gaussian mixture model corresponding to each of the plurality of target observation objects to generate a state probability density function of the observed object.

[0016] According to a third aspect of the embodiments of the present application, an electronic device is provided, which comprises a processor, a memory for storing executable instructions for the processor, and the processor is configured to read the executable instructions from the memory and execute the instructions to implement the method according to the first aspect.

[0017] According to a fourth aspect of the embodiments of the present application, a computer readable medium is provided, which stores a computer program, and the program is configured to be executed by a processor to implement the method according to the first aspect.

[0018] The embodiments of the present application provide a state fusion method and device based on Gaussian mixture model, which comprises the following steps: firstly, for any target observation object in a plurality of target observation objects: performing particle distribution sampling on the observed object according to position information of the target observation object to generate a position probability distribution of the observed object observed by the target observation object; and performing Gaussian mixture model fitting optimization processing on the position probability distribution to generate a quasi-Gaussian mixture model; secondly, performing Chernoff fusion on the quasi-Gaussian mixture model corresponding to each target observation object in the plurality of target observation objects to generate a state probability density function of the observed object. Thus, a plurality of unknown and non-correlated states corresponding to the same observed object at the same time can be effectively fused, so that the accuracy of the state description of the observed object is improved. BRIEF DESCRIPTION OF DRAWINGS

[0019] Some specific embodiments of the present application will be described in detail hereinafter with reference to the accompanying drawings, which are presented by way of example and are not limiting. Like reference numbers in the figures indicate corresponding or analogous elements or parts. It should be understood by those skilled in the art that the figures are not necessarily drawn to scale. In the figures:

[0020] Figure 1 A flowchart of the state fusion method based on Gaussian mixture model provided by an embodiment of the present application is shown in the figure;

[0021] Figure 2 A flowchart of generating a quasi-Gaussian mixture model in an embodiment of the present application is shown in the figure;

[0022] Figure 3 A flowchart of generating a state probability density function of an observed object in an embodiment of the present application is shown in the figure;

[0023] Figure 4 A structure diagram of the state fusion device based on Gaussian mixture model provided by an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0024] In order to make the objects, characteristics and advantages of the present application more apparent and easy to understand, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0025] As shown in FIG. 1, a flowchart of a state fusion method based on a Gaussian mixture model provided by an embodiment of the present application is shown. Figure 1

[0026] The state fusion method based on the Gaussian mixture model at least includes the following steps:

[0027] S101, for any target observation object in a plurality of target observation objects: according to the position information of the target observation object, performing particle distribution sampling on the observed object to generate a position probability distribution of the observed object observed by the target observation object; performing Gaussian mixture model fitting optimization processing on the position probability distribution to generate a quasi-Gaussian mixture model;

[0028] S102, performing Chernoff fusion on the quasi-Gaussian mixture model corresponding to each target observation object in the plurality of target observation objects to generate a state probability density function of the observed object.

[0029] In S101, here, the method of particle distribution sampling on the observed object is not limited.

[0030] For example, the observed object is randomly sampled by a angle sensor according to a Gaussian distribution to obtain an angle particle set of the observed object; wherein the Gaussian distribution is used to indicate the measurement error of the angle sensor at a preset angle; the observed object is randomly sampled according to a distance distribution to generate a distance particle set of the observed object; for any angle particle in the angle particle set: selecting a distance particle having the same order as the angle particle from the distance particle set; determining the position information corresponding to the angle particle based on the selected distance particle, the angle particle and the position information of the target observation object; determining the position probability distribution of the observed object based on the position information corresponding to each angle particle in the angle particle set.

[0031] ​More specifically, the distance distribution random sampling of the observed object generates a distance particle set of the observed object; including: uniformly distributed random sampling of the observed object to obtain a sampling particle set; for any sampling particle in the sampling particle set: based on the sampling parameter of the sampling particle, the minimum spatial distance of the target observed object to the sampling particle, and the maximum spatial distance of the target observed object to the sampling particle, the sampling inverse transformation method is used to convert the distance of the sampling particle to generate the spatial distance of the sampling particle; based on the spatial distance corresponding to each sampling particle in the sampling particle set, the distance particle set of the observed object is determined.

[0032] For example: in the Gaussian distribution N(r θ ,σ r ), the observed object is randomly sampled N times to obtain an angle particle set S1={r θ,i ,i∈N}; in the uniform distribution U(0,1), the observed object is randomly sampled N times to obtain a sampling particle set S2={l i ,i∈N}; and S2 is converted into a distance particle set S3 by the sampling inverse transformation method.

[0033] The specific expression of S3 is as follows:

[0034]

[0035] Where d i is the spatial distance of the sampling particle; l i is the sampling parameter; R max is the maximum spatial distance of the target observed object to the sampling particle; and R min is the minimum spatial distance of the target observed object to the sampling particle.

[0036] Based on the position information (x0, y0, θ0) of the target observed object, S1, and S3, a position particle set is generated to describe the position probability distribution of the observed object.

[0037] The expression of the position probability distribution S4 of the observed object is as follows:

[0038]

[0039] Where (x i , y i , r θ,i ) represents the position information corresponding to the i-th angle particle.

[0040] The position probability distribution is fitted and optimized by a Gaussian mixture model based on an Expectation-Maximization (EM) algorithm. The EM algorithm is an iterative optimization strategy. In the calculation method of the EM algorithm, each iteration is divided into two steps: an expectation step (E step) and a maximization step (M step).

[0041] It should be noted that, for any sensor, the method of the embodiment can obtain the position probability distribution of the observed object based on the measurement characteristics and the distribution sampling of the measurement error.

[0042] In S102, since the state probability density function of the observed object in the embodiment cannot be constructed like a Gaussian distribution, when a plurality of non-Gaussian distributions of the state of the observed object are obtained, the consistency fusion of the non-Gaussian distributions with unknown correlations can be realized based on Chernoff fusion, so as to generalize the variables to be fused to any probability density function.

[0043] It should be noted that the Gaussian mixture model is used to indicate the probability density function with complex properties. The state probability density function of the embodiment is used to indicate the accumulation of a plurality of quasi-Gaussian mixture models.

[0044] The embodiment combines the sampling method of the probability distribution and the fusion processing method of the Gaussian mixture model of the non-Gaussian distribution, so that a plurality of unknown non-correlated states of the same observed object at the same time can be effectively fused, thereby improving the accuracy of the state description of the observed object.

[0045] As shown in Figure 2 FIG. 1 is a flowchart of generating a quasi-Gaussian mixture model in an embodiment of the application.

[0046] The method for generating a quasi-Gaussian mixture model comprises at least the following steps:

[0047] S201, configuring the number of Gaussian models in the Gaussian mixture model;

[0048] S202, performing model parameter estimation on the Gaussian mixture model by using an EM algorithm according to the position probability distribution, to generate candidate model parameters; wherein the candidate model parameters include mean and variance;

[0049] S203, performing iterative optimization on the Gaussian mixture model based on the candidate model parameters and the position probability distribution, to generate a quasi-Gaussian mixture model.

[0050] For example, the iterative optimization on the Gaussian mixture model based on the candidate model parameters and the position probability distribution, to generate a quasi-Gaussian mixture model; comprises:

[0051] S1, generating a virtual sampling data set conforming to a Gaussian distribution based on a candidate Gaussian mixture model corresponding to the candidate model parameter;

[0052] S2, performing optimization processing on the candidate Gaussian mixture model according to the virtual sampling data set and a real data set corresponding to the position probability distribution, to generate an optimized model parameter;

[0053] S3, repeatedly iterating S1 and S2 until the candidate Gaussian mixture model converges, and outputting a quasi-model parameter; a convergence condition is generally set as a mean value variation reaching a set convergence precision or an iteration number reaching a maximum iteration number. Specifically, based on a Gaussian mixture model corresponding to the optimized model parameter, virtual data conforming to a Gaussian distribution is generated again, and optimization processing is performed until the mean value variation reaches a preset threshold, and then the optimization processing is ended, and the quasi-model parameter is generated;

[0054] S4, performing updating processing on the Gaussian mixture model based on the quasi-model parameter, to generate a quasi-Gaussian mixture model. Here, the mean value variation is used to indicate a mean value variation in the candidate model parameter.

[0055] An expression of the quasi-Gaussian mixture model p(x) is as follows.

[0056]

[0057] wherein, N i represents an i-th Gaussian model, β i is a weight thereof, P i represents a covariance matrix of the i-th Gaussian model.

[0058] Here, the smaller the number of Gaussian models in the Gaussian mixture model, the lower the fitting precision of the Gaussian mixture model, and the faster the convergence speed; the larger the number of Gaussian models in the Gaussian mixture model, the greater the fitting precision of the Gaussian mixture model, and the slower the convergence speed.

[0059] As shown in FIG. 1, it is a flowchart for generating a state probability density function of an observed object in an embodiment of the present application. Figure 3

[0060] The method for generating the state probability density function of the observed object comprises the following steps:

[0061] S301, generating a quasi-Gaussian mixture model sequence based on a quasi-Gaussian mixture model corresponding to each target observed object in a plurality of target observed objects;

[0062] ​S302, for any quasi-Gaussian mixture model in the quasi-Gaussian mixture model sequence: based on the quasi-Gaussian mixture model, a weighted non-negative least squares (WNNLS) problem is constructed using the sigma point method; under the condition of converting the quasi-Gaussian mixture model from a small power form to a first power form, the weighted non-negative least squares problem is solved to generate a mixing weight; the quasi-Gaussian mixture model is updated based on the mixing weight to generate a first power model;

[0063] S303, based on the first power model corresponding to each quasi-Gaussian mixture model in the quasi-Gaussian mixture model sequence, a first power model sequence is generated.

[0064] S304, according to the first power model sequence, sequentially performing Chernoff fusion on the first power model sequence to generate a state probability density function of the observed object.

[0065] Specifically, the Chernoff fusion is performed on the first power model sequence according to the first power model sequence to generate a state probability density function of the observed object, including: for any current first power model in the first power model sequence: obtaining a last first power model located before and adjacent to the current first power model; based on the last first power model and the current first power model, performing fusion parameter optimization processing to generate a current quasi-fusion parameter; based on the current quasi-fusion parameter, performing Chernoff fusion processing on the last first power model and the current first power model to generate a current fused model; taking the current fused model as the last first power model, continuing the next fusion parameter optimization processing until the last first power model in the first power model sequence is processed by Chernoff fusion to generate a state probability density function of the observed object.

[0066] For example: the current fusion parameter ω calculation process is as follows.

[0067]

[0068] The state probability density function p pcf of the observed object is as follows.

[0069]

[0070] Wherein, μ i (w) represents the weight coefficient of the i-th Gaussian distribution, v j (w) represents the weight coefficient of the j-th Gaussian distribution; π ij (w) represents the local fusion weight coefficient of the i-th and j-th Gaussian distributions. and represent the mean and variance of the local fusion of the i-th and j-th Gaussian distributions; Trace represents the trace of a matrix. is a global fusion state estimate value.

[0071] Therefore, the embodiment uses the method of sigma point approximation to approximate the small power of Gaussian mixture model to one, and gives the closed form of the final fusion result. Therefore, not only the process of integral is avoided, but also the sufficient approximation accuracy is ensured, so as to effectively fuse several unknown and uncorrelated states of the observed object, and further improve the accuracy of the state description of the observed object.

[0072] The state fusion method based on the Gaussian mixture model provided by the embodiment will be described in detail in combination with a specific application scenario.

[0073] The state fusion method based on the Gaussian mixture model comprises at least the following steps:

[0074] S1, for any target observed object in several target observed objects: the angle sensor is used to randomly sample the observed object in a Gaussian distribution to obtain an angle particle set of the observed object; wherein the Gaussian distribution is used to indicate the measurement error of the angle sensor at a preset angle; the observed object is randomly sampled in a uniform distribution to obtain a sampling particle set; for any sampling particle in the sampling particle set: based on the sampling parameter of the sampling particle, the minimum spatial distance from the target observed object to the sampling particle, and the maximum spatial distance from the target observed object to the sampling particle, the sampling particle is distance converted by using the sampling inverse transformation method to generate the spatial distance of the sampling particle; based on the spatial distance corresponding to each sampling particle in the sampling particle set, a distance particle set of the observed object is determined. For any angle particle in the angle particle set: the distance particle with the same order as the angle particle is selected from the distance particle set; based on the selected distance particle, the angle particle, and the position information of the target observed object, the position information corresponding to the angle particle is determined; based on the position information corresponding to each angle particle in the angle particle set, the position probability distribution of the observed object is determined.

[0075] S2, the number of Gaussian models in the Gaussian mixture model is configured; the EM algorithm is used to estimate the model parameters of the Gaussian mixture model according to the position probability distribution to generate candidate model parameters; wherein the candidate model parameters include mean and variance.

[0076] S3, based on the candidate Gaussian mixture model corresponding to the candidate model parameter, generate a virtual sampling data set conforming to the Gaussian distribution; based on the virtual sampling data set and the real data set corresponding to the position probability distribution, use the EM algorithm to optimize the candidate Gaussian mixture model to generate the optimized model parameter; based on the Gaussian mixture model corresponding to the optimized model parameter, generate the virtual data conforming to the Gaussian distribution again and perform the optimization processing until the mean variation reaches the preset threshold, and then end the optimization processing to generate the quasi-model parameter; based on the quasi-model parameter, update the Gaussian mixture model to generate the quasi-Gaussian mixture model.

[0077] S4, based on the quasi-Gaussian mixture model corresponding to each target observation object in the plurality of target observation objects, generate a quasi-Gaussian mixture model sequence; for any quasi-Gaussian mixture model in the quasi-Gaussian mixture model sequence: based on the quasi-Gaussian mixture model, use the sigma point method to construct a weighted non-negative least square problem; under the condition that the quasi-Gaussian mixture model is converted from a small number power form to a first power form, solve the weighted non-negative least square problem to generate a mixing weight; based on the mixing weight, update the quasi-Gaussian mixture model to generate a first power model; based on the first power model corresponding to each quasi-Gaussian mixture model in the quasi-Gaussian mixture model sequence, generate a first power model sequence.

[0078] S5, for any current first power model in the first power model sequence: obtain a previous first power model located before and adjacent to the current first power model; based on the previous first power model and the current first power model, perform fusion parameter optimization processing to generate a current quasi-fusion parameter; based on the current quasi-fusion parameter, perform Chernoff fusion processing on the previous first power model and the current first power model to generate a current fused model.

[0079] S6, take the current fused model as the previous first power model and continue the next fusion parameter optimization processing until the last first power model in the first power model sequence is subjected to Chernoff fusion processing to generate a state probability density function of the observed object.

[0080] The embodiment combines the sampling method of the probability distribution and the fusion processing method of the Gaussian mixture model of the non-Gaussian distribution, so that the plurality of unknown non-correlation states corresponding to the same observed object at the same time can be effectively fused, thereby improving the accuracy of the state description of the observed object.

[0081] As shown in Figure 4 Fig. 1 is a structural schematic diagram of a state fusion device based on a Gaussian mixture model provided by an embodiment of the present application.

[0082] The state fusion device based on the Gaussian mixture model comprises: a generation module 401 configured to, for any target observation object in a plurality of target observation objects: perform particle distribution sampling on the observed object according to position information of the target observation object to generate a position probability distribution of the observed object observed by the target observation object; and perform Gaussian mixture model fitting optimization processing on the position probability distribution to generate a quasi-Gaussian mixture model; and a fusion module 402 configured to perform Chernoff fusion on the quasi-Gaussian mixture model corresponding to each target observation object in the plurality of target observation objects to generate a state probability density function of the observed object.

[0083] In the preferred implementation of the embodiment, the generation module comprises: a first sampling unit configured to perform Gaussian distribution random sampling on the observed object by using the angle sensor to obtain an angle particle set of the observed object; wherein the Gaussian distribution is used to indicate a measurement error of the angle sensor at a preset angle; a second sampling unit configured to perform distance distribution random sampling on the observed object to generate a distance particle set of the observed object; a first determination unit configured to, for any angle particle in the angle particle set: select a distance particle having the same order as the angle particle from the distance particle set; and determine position information corresponding to the angle particle based on the selected distance particle, the angle particle, and the position information of the target observation object; and a second determination unit configured to determine the position probability distribution of the observed object based on the position information corresponding to each angle particle in the angle particle set.

[0084] In the preferred implementation of the embodiment, the second sampling unit comprises: a random sampling subunit configured to perform uniform distribution random sampling on the observed object to obtain a sampling particle set; a distance conversion subunit configured to, for any sampling particle in the sampling particle set: perform distance conversion on the sampling particle by using a sampling inverse transformation method based on a sampling parameter of the sampling particle, a minimum spatial distance from the target observation object to the sampling particle, and a maximum spatial distance from the target observation object to the sampling particle to generate a spatial distance of the sampling particle; and a determination subunit configured to determine the distance particle set of the observed object based on the spatial distance corresponding to each sampling particle in the sampling particle set.

[0085] In the preferred implementation of the embodiment, the generation module further comprises: a configuration unit configured to configure a number of Gaussian models in the Gaussian mixture model; a model parameter estimation unit configured to perform model parameter estimation on the Gaussian mixture model by using an EM algorithm according to the position probability distribution to generate candidate model parameters; wherein the candidate model parameters comprise a mean value and a variance; and an iterative optimization unit configured to perform iterative optimization on the Gaussian mixture model based on the candidate model parameters and the position probability distribution to generate the quasi-Gaussian mixture model.

[0086] In the preferred implementation of the embodiment, the iterative optimization unit comprises: a generating subunit configured to generate a virtual sampling data set conforming to a Gaussian distribution based on a candidate Gaussian mixture model corresponding to the candidate model parameter; an optimization processing subunit configured to perform optimization processing on the candidate Gaussian mixture model based on the virtual sampling data set and a real data set corresponding to the position probability distribution, to generate an optimized model parameter; an iteration subunit configured to generate virtual data conforming to a Gaussian distribution based on a Gaussian mixture model corresponding to the optimized model parameter, and perform optimization processing again until a mean variation reaches a preset threshold, and then end the optimization processing to generate a quasi-model parameter; and an updating subunit configured to perform updating processing on the Gaussian mixture model based on the quasi-model parameter, to generate a quasi-Gaussian mixture model.

[0087] In the preferred implementation of the embodiment, the fusion module comprises: a first generating unit configured to generate a quasi-Gaussian mixture model sequence based on a quasi-Gaussian mixture model corresponding to each target observation object in a plurality of target observation objects; an updating unit configured to, for any quasi-Gaussian mixture model in the quasi-Gaussian mixture model sequence: construct a weighted non-negative least square problem based on the quasi-Gaussian mixture model by using a sigma point method; solve the weighted non-negative least square problem under the condition that the quasi-Gaussian mixture model is converted from a fractional power form to a linear power form, to generate a mixing weight; update the quasi-Gaussian mixture model based on the mixing weight, to generate a linear power model; and a second generating unit configured to generate a linear power model sequence based on a linear power model corresponding to each quasi-Gaussian mixture model in the quasi-Gaussian mixture model sequence; and a fusion unit configured to sequentially perform Chernoff fusion on the linear power model sequence according to the linear power model sequence, to generate a state probability density function of the observed object.

[0088] In the preferred implementation of the embodiment, the fusion unit comprises: a generating subunit configured to, for any current linear power model in the linear power model sequence: acquire a previous linear power model located before and adjacent to the current linear power model; perform fusion parameter optimization processing based on the previous linear power model and the current linear power model, to generate a current quasi-fusion parameter; perform Chernoff fusion processing on the previous linear power model and the current linear power model based on the current quasi-fusion parameter, to generate a current fused model; and a fusion processing subunit configured to continue to perform next fusion parameter optimization processing by taking the current fused model as a previous linear power model, until a last linear power model in the linear power model sequence is subjected to Chernoff fusion processing, to generate a state probability density function of the observed object.

[0089] The device can execute the state fusion method based on the Gaussian mixture model provided by the embodiment of the application, and has the corresponding function modules and beneficial effects of executing the state fusion method based on the Gaussian mixture model. Technical details not described in detail in the embodiment can be referred to the state fusion method based on the Gaussian mixture model provided by the embodiment of the application.

[0090] The application further provides an electronic device, comprising: a processor; a memory for storing executable instructions of the processor; and the processor is configured to read the executable instructions from the memory and execute the instructions to implement the state fusion method based on the Gaussian mixture model.

[0091] In addition to the above method and device, the embodiment of the application can also be a computer program product, which comprises computer program instructions, and the computer program instructions enable the processor to execute the steps of the method according to various embodiments of the application described in the above “Exemplary Method” section of the specification when the computer program instructions are run by the processor.

[0092] The computer program product can be written in any combination of one or more programming languages, including an object-oriented programming language such as Java, C++, and conventional procedural programming languages such as the “C” language or similar programming languages. The program code can be executed entirely on a user computing device, partially on a user device, as an independent software package, partially on a user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0093] In addition, the embodiment of the application can also be a computer readable storage medium, which stores computer program instructions, and the computer program instructions enable the processor to execute the steps of the method according to various embodiments of the application described in the above “Exemplary Method” section of the specification when the computer program instructions are run by the processor.

[0094] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium may, for example, include but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0095] The above describes the basic principles of the present application in conjunction with specific embodiments, but it should be noted that the advantages, benefits, effects and the like mentioned in the present application are only examples and are not limiting, and these advantages, benefits, effects and the like cannot be considered as necessary for each embodiment of the present application. In addition, the above specific details disclosed are only for the purpose of example and understanding, and are not limiting, and the above details do not limit the present application to be necessarily implemented with the above specific details.

[0096] The block diagrams of the devices, apparatuses, equipment, systems involved in the present application are only illustrative examples and are not intended to require or imply the connection, arrangement, configuration as shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, equipment, systems can be connected, arranged, configured in any manner. Words such as "include", "contain", "have" and the like are open-ended words, mean "including but not limited to", and can be used interchangeably. The words "or" and "and" used herein mean the word "and / or", and can be used interchangeably unless the context clearly indicates otherwise. The word "such as" used herein means the phrase "such as but not limited to", and can be used interchangeably.

[0097] It should also be noted that in the devices, apparatuses and methods of the present application, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions of the present application.

[0098] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other aspects without departing from the scope of the present application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0099] The above description has been given for the purpose of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

[0100] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. Also, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the specification and the features of the different embodiments or examples without contradiction.

[0101] In addition, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise specifically limited.

[0102] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A state fusion method based on Gaussian mixture model, characterized in that, The method comprises the following steps: For any one of a plurality of target observation objects: Gaussian distribution random sampling is performed on the observed object by an angle sensor to obtain an angle particle set of the observed object; wherein the Gaussian distribution is used to indicate the measurement error of the angle sensor at a preset angle; distance distribution random sampling is performed on the observed object to generate a distance particle set of the observed object; for any angle particle in the angle particle set: a distance particle with the same order as the angle particle is selected from the distance particle set; based on the selected distance particle, the angle particle, and the position information of the target observation object, the position information corresponding to the angle particle is determined; based on the position information corresponding to each angle particle in the angle particle set, a position probability distribution of the observed object is determined; Gaussian mixture model fitting optimization processing is performed on the position probability distribution to generate a quasi-Gaussian mixture model; The quasi-Gaussian mixture models corresponding to each of the plurality of target observation objects are Chernoff fused to generate a state probability density function of the observed object.

2. The method of claim 1, wherein, The distance distribution random sampling is performed on the observed object to generate a distance particle set of the observed object; The method comprises the following steps: Uniform distribution random sampling is performed on the observed object to obtain a sampling particle set; For any sampling particle in the sampling particle set: distance conversion is performed on the sampling particle by a sampling inverse transformation method based on the sampling parameter of the sampling particle, the minimum spatial distance from the target observation object to the sampling particle, and the maximum spatial distance from the target observation object to the sampling particle to generate the spatial distance of the sampling particle; Based on the spatial distance corresponding to each sampling particle in the sampling particle set, a distance particle set of the observed object is determined.

3. The method of claim 1, wherein, The Gaussian mixture model fitting optimization processing on the position probability distribution to generate a quasi-Gaussian mixture model comprises the following steps: The number of Gaussian models in the Gaussian mixture model is configured; According to the position probability distribution, model parameter estimation is performed on the Gaussian mixture model by an EM algorithm to generate candidate model parameters; wherein the candidate model parameters include mean and variance; Based on the candidate model parameters and the position probability distribution, the Gaussian mixture model is iteratively optimized to generate a quasi-Gaussian mixture model.

4. The method of claim 3, wherein, The Gaussian mixture model fitting optimization processing on the position probability distribution to generate a quasi-Gaussian mixture model comprises the following steps: Based on the candidate Gaussian mixture model corresponding to the candidate model parameters, a virtual sampling data set conforming to the Gaussian distribution is generated; According to the virtual sampling data set and the real data set corresponding to the position probability distribution, the candidate Gaussian mixture model is optimized by an EM algorithm to generate optimized model parameters; Based on the Gaussian mixture model corresponding to the optimized model parameters, virtual data conforming to the Gaussian distribution is generated again and optimized until the mean variation reaches a preset threshold, and the optimization processing is ended to generate quasi-model parameters; Based on the quasi-model parameters, the Gaussian mixture model is updated to generate a quasi-Gaussian mixture model.

5. The method of claim 1, wherein, The Chernoff fusion is performed on the quasi-Gaussian mixture model corresponding to each of the target observation objects in the plurality of target observation objects, and a state probability density function of the observed object is generated. A quasi-Gaussian mixture model sequence is generated based on the quasi-Gaussian mixture model corresponding to each of the target observation objects in the plurality of target observation objects. For any quasi-Gaussian mixture model in the quasi-Gaussian mixture model sequence: a weighted non-negative least square problem is constructed by using a sigma point method based on the quasi-Gaussian mixture model; the weighted non-negative least square problem is solved under the condition that the quasi-Gaussian mixture model is converted from a small number power form to a first power form, and a mixing weight is generated; the quasi-Gaussian mixture model is updated based on the mixing weight, and a first power model is generated. A first power model sequence is generated based on the first power model corresponding to each of the quasi-Gaussian mixture models in the quasi-Gaussian mixture model sequence. The first power model sequence is sequentially subjected to Chernoff fusion based on the first power model sequence, and a state probability density function of the observed object is generated.

6. The method of claim 5, wherein, The first power model sequence is sequentially subjected to Chernoff fusion based on the first power model sequence, and a state probability density function of the observed object is generated. For any current first power model in the first power model sequence: an immediately preceding first power model before the current first power model is obtained; a current quasi-fusion parameter is generated based on the immediately preceding first power model and the current first power model; the immediately preceding first power model and the current first power model are subjected to Chernoff fusion processing based on the current quasi-fusion parameter, and a current fused model is generated. The current fused model is taken as the immediately preceding first power model, and the next fusion parameter optimization processing is continued until the last first power model in the first power model sequence is subjected to Chernoff fusion processing, and a state probability density function of the observed object is generated.

7. A state fusion apparatus based on a Gaussian mixture model, characterized by The method comprises the following steps: The generation module is configured to: for any target observation object in the plurality of target observation objects: perform Gaussian distribution random sampling on the observed object by using an angle sensor to obtain an angle particle set of the observed object; wherein the Gaussian distribution is used to indicate the measurement error of the angle sensor at a preset angle; perform distance distribution random sampling on the observed object to generate a distance particle set of the observed object; for any angle particle in the angle particle set: select a distance particle having the same order as the angle particle from the distance particle set; determine the position information corresponding to the angle particle based on the selected distance particle, the angle particle, and the position information of the target observation object; determine the position probability distribution of the observed object based on the position information corresponding to each angle particle in the angle particle set; and perform Gaussian mixture model fitting optimization processing on the position probability distribution to generate a quasi-Gaussian mixture model. A fusion module is configured to perform Chernoff fusion on the quasi-Gaussian mixture model corresponding to each of the target observation objects to generate a state probability density function of the observed object.

8. A computer readable medium having stored thereon a computer program which, when executed by a processor, implements the method of any one of claims 1-6.

Images