Positioning method and system based on Bayesian filtering neural network, and electronic equipment

Abstract

The invention relates to the technical field of environmental perception fusion positioning, and provides a positioning method and system based on a Bayesian filtering neural network, and electronic equipment, and the method comprises the steps: obtaining an observation waveform matrix generated after a base station receives a waveform signal transmitted by a target object at each sampling moment; inputting the observation waveform matrix into a preset positioning model to obtain positioning information of the target object; the positioning model uses a recurrent neural network to fit a Bayesian filtering result, the recurrent neural network is trained through a cost function including motion likelihood and observation likelihood, the motion likelihood is generated based on a motion model of a target object sample, and the observation likelihood is generated based on position information of the target object sample determined by adopting a parameter estimation algorithm. The method is used for overcoming the defects that in the prior art, when a Bayesian filtering algorithm is applied to environment perception fusion positioning, a positioning result can be obtained only by depending on an approximation algorithm and complex derivation, and then the positioning speed and the positioning precision are affected.

Description

Technical Field

[0001] The present invention relates to the technical field of environment perception fusion positioning, and in particular to a positioning method, system and electronic equipment based on a Bayesian filtering neural network. Background Art

[0002] At present, when performing environmental perception fusion positioning, the Bayesian filtering algorithm is introduced to combine the user's information at multiple times to overcome the problem that multipath information cannot improve the estimation accuracy of the user's position when there is no prior environmental information.

[0003] The traditional Bayesian filter algorithm and its derivative algorithms are based on the hidden Markov model. When the motion equation and observation equation of the hidden variable state are given, the posterior distribution of the hidden variable state at a certain moment can be iteratively inferred to estimate the hidden variable. When the motion equation and the observation equation are both linear equations, the Bayesian filter is the Kalman filter. However, since the motion equation and the observation equation are high-dimensional nonlinear equations in complex environment fusion positioning scenarios, it is difficult to express the results of the Bayesian filter in a closed form. Therefore, several approximate algorithms have been generated, such as the extended Kalman filter algorithm and the volumetric Kalman filter algorithm. For example: the particle filter algorithm is based on Monte Carlo estimation and uses importance sampling to fit the Bayesian filter.

[0004] Therefore, the existing Bayesian filtering algorithm has the following problems when applied to the task of environmental perception fusion positioning:

[0005] In terms of accuracy, the existing Bayesian filtering algorithm cannot obtain a closed-form solution for this task, so it must rely on approximate algorithms such as extended Kalman filtering and particle filtering, which rely on the adjustment of hyperparameters in the problem of environmental fusion positioning. Hyperparameters have a great impact on the convergence speed and accuracy of the Bayesian filtering algorithm, and rely on precise adjustments by technicians who are proficient in this task. In addition, in terms of the simplicity of the model, since the environmental fusion positioning algorithm needs to deal with complex problems such as the birth, death and matching of multipaths, in order to integrate this complex observation model into the Bayesian filtering algorithm in the actual system, the filtering algorithm needs to be complexly derived, which limits the scalability of the Bayesian filtering algorithm for such problems, that is, affects the speed and accuracy of positioning. Summary of the invention

[0006] The present invention provides a positioning method, system and electronic device based on a Bayesian filtering neural network, so as to solve the defect in the prior art that when applying the Bayesian filtering algorithm in environmental perception fusion positioning, it is necessary to rely on approximate algorithms and complex derivations to obtain positioning results, thereby affecting the positioning speed and positioning accuracy.

[0007] The present invention provides a positioning method based on a Bayesian filtering neural network, comprising:

[0008] Obtaining the observed waveform matrix generated by the base station after receiving the waveform signal transmitted by the target object at each sampling moment;

[0009] Inputting the observed waveform matrix into a preset positioning model to obtain positioning information of the target object output by the positioning model;

[0010] Among them, the positioning model uses a recurrent neural network to fit the Bayesian filtering results, and the recurrent neural network is trained by a cost function including motion likelihood and observation likelihood. The motion likelihood is generated based on the motion model of the target object sample, and the observation likelihood is generated based on the position information of the target object sample determined by a parameter estimation algorithm.

[0011] According to the positioning method based on Bayesian filtering neural network of the present invention, the recurrent neural network is constructed based on a long short-term memory network.

[0012] According to the positioning method based on Bayesian filtering neural network of the present invention, the cost function also includes:

[0013] Prior constraints constructed based on other prior information during the training of the recurrent neural network;

[0014] The other prior information includes prior information generated based on loop detection, initial value constraints, map constraints, and birth and death processes.

[0015] According to the positioning method based on Bayesian filtering neural network of the present invention, the inputting of the observed waveform matrix into a preset positioning model to obtain the positioning information of the target object output by the positioning model includes:

[0016] The observed waveform matrix is ​​input into the result fitting layer of the positioning model, and the result fitting layer uses the recurrent neural network to fit the Bayesian filtering result to obtain the posterior distribution of the target object output by the result fitting layer;

[0017] The posterior distribution is input to the positioning confirmation layer of the positioning model, and the positioning confirmation layer determines the positioning information of the target object based on the posterior distribution, so as to obtain the positioning information of the target object output by the positioning confirmation layer.

[0018] The present invention also provides a positioning system based on a Bayesian filtering neural network, comprising:

[0019] An acquisition module, used to acquire an observation waveform matrix generated by the base station after receiving a waveform signal transmitted by a target object at each sampling moment;

[0020] A positioning module, used for inputting the observed waveform matrix into a preset positioning model to obtain the positioning information of the target object output by the positioning model;

[0021] Among them, the positioning model uses a recurrent neural network to fit the Bayesian filtering results, and the recurrent neural network is trained by a cost function including motion likelihood and observation likelihood. The motion likelihood is generated based on the motion model of the target object sample, and the observation likelihood is generated based on the position information of the target object sample determined by a parameter estimation algorithm.

[0022] According to the positioning system based on Bayesian filtering neural network of the present invention, the recurrent neural network is constructed based on a long short-term memory network.

[0023] According to the positioning system based on Bayesian filtering neural network of the present invention, the cost function also includes:

[0024] Prior constraints constructed based on other prior information during the training of the recurrent neural network;

[0025] The other prior information includes prior information generated based on loop detection, initial value constraints, map constraints, and birth and death processes.

[0026] According to the positioning system based on Bayesian filtering neural network of the present invention, the positioning module is specifically used for:

[0027] The observed waveform matrix is ​​input into the result fitting layer of the positioning model, and the result fitting layer uses the recurrent neural network to fit the Bayesian filtering result to obtain the posterior distribution of the target object output by the result fitting layer;

[0028] The posterior distribution is input to the positioning confirmation layer of the positioning model, and the positioning confirmation layer determines the positioning information of the target object based on the posterior distribution, so as to obtain the positioning information of the target object output by the positioning confirmation layer.

[0029] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the positioning method based on the Bayesian filtering neural network as described above when executing the program.

[0030] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the positioning method based on the Bayesian filtering neural network as described above is implemented.

[0031] The positioning method, system and electronic device based on Bayesian filter neural network provided by the present invention obtain the observation waveform matrix generated by the base station after receiving the waveform signal emitted by the target object at each sampling moment, and then input the observation waveform matrix into the preset positioning model, so as to obtain the positioning information of the target object output by the positioning model, and the positioning model uses a recurrent neural network to fit the Bayesian filter result, and the recurrent neural network is trained by a cost function including motion likelihood and observation likelihood, the motion likelihood is generated based on the motion model of the target object sample, and the observation likelihood is generated based on the position information of the target object sample determined by the parameter estimation algorithm. The recurrent neural network is used to fit the Bayesian filter result, thereby solving the problem of affecting the positioning speed and positioning accuracy when applying the Bayesian filter algorithm in the environment perception fusion positioning, and thus can significantly improve the scalability of the traditional environment perception fusion positioning algorithm, and improve its convergence speed and positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 It is a flowchart of a positioning method based on a Bayesian filtering neural network provided by an embodiment of the present invention;

[0034] Figure 2 is a schematic diagram of the structure of a hidden Markov model provided by an embodiment of the present invention;

[0035] Figure 3 is a schematic diagram of a Bayesian filtering network structure provided by an embodiment of the present invention;

[0036] Figure 4 is a schematic diagram of the structure of a positioning system based on a Bayesian filtering neural network provided by an embodiment of the present invention;

[0037] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] First, in order to facilitate understanding of the positioning method based on the Bayesian filtering neural network provided by the present invention, the technical solution provided by the embodiment of the present invention is described below using an ultra-wideband (UWB) signal as an example.

[0040] It should be noted that the ultra-wideband signal is only an example of the signal type to which the positioning method based on the Bayesian filtering neural network provided in the embodiment of the present invention can be applied. The positioning method based on the Bayesian filtering neural network provided in the embodiment of the present invention can also be applied to other types of signals, such as: 5G NR, WIFI, Bluetooth, geomagnetic sensing, etc., which will not be elaborated here.

[0041] It can be understood that, assuming that a single-base station ultra-wideband positioning system includes a base station with P antennas and a user equipment UE, and the center position of the base station array is known, which is a = {a x , a y} T At time k, the unknown UE position is defined as p k Considering that there are L unknown reflective surfaces in the environment, the image of the base station relative to the lth reflective surface (defined as the virtual base station VA) can be used to describe the parameters of the lth reflective surface relative to the base station. Then the position of the lth VA unknown at time k is expressed as and At the same time, for the simplicity of the model, the 0th VA is considered to be the known base station location, that is, a 0，k =a.

[0042] Because, through the virtual base station a l，k , base station a and UE location, the location p of the virtual tag can be calculated l，k , and based on Fresnel's theorem, all MPC (multipath component) signals received by the base station can be regarded as virtual labels p l，k Therefore, at each time k, p can be estimated by receiving the waveform 0：L，k , dually, the UE position p can be estimated by the waveform k and the locations of all virtual base stations VA 0：L，k .

[0043] Furthermore, after receiving the UWB waveform transmitted by the UE, the base station can record the CIR (Channel Impulse Response) of each received segment as Y k , its signal model is as follows:

[0044]

[0045] Among them, s(τ) is the echo signal, a(θ l ) is the steering vector of the lth MPC, n(τ) is Gaussian white noise, and y(τ) is the echo signal received at the τth sampling time.

[0046] By connecting all the unknown variables at time k, we can express them as x k ,Right now:

[0047]

[0048] At each sampling time t k ,by The sampling frequency can collect T snap Snapshots are accumulated into an observation waveform matrix, namely:

[0049] Y k =[y[t k ],y[t k +t snap ],…,y[t k +T snap t snap ] (3)

[0050] It can be seen that through the parameter estimation algorithm, we can directly get k The ToA (Time of arrival) and AoA (Angle of arrival) parameters h of each MPC are estimated k . Therefore, given h k The estimated UE and VA positions x can be directly calculated through the geometric mapping function h(·) k ,Right now:

[0051] f(x k |h k )=δ(x k -h(h k )) (4)

[0052] Then we have:

[0053]

[0054] It can be seen that by estimating the MPC, the parameter x to be estimated at time k can be obtained. k The posterior distribution f(x k |Y k ), and then obtain the estimated results of the user location and the virtual base station location.

[0055] However, in the absence of prior environmental information, multipath information MPC cannot improve the estimation accuracy of the user's location. Therefore, it is necessary to use the Bayesian filtering algorithm to combine the information at multiple times, that is, to calculate f(x 1：k |Y 1：k ).

[0056] However, the existing Bayesian filtering algorithm has the following problems when applied to the task of environmental perception fusion positioning:

[0057] In terms of accuracy, the existing Bayesian filtering algorithm cannot obtain a closed-form solution for this task, so it must rely on approximate algorithms such as extended Kalman filtering and particle filtering. Approximate algorithms rely on the adjustment of hyperparameters in the problem of environmental fusion positioning. Hyperparameters have a great impact on the convergence speed and accuracy of the Bayesian filtering algorithm, and rely on precise adjustments by technicians who are proficient in this task. In addition, in terms of the simplicity of the model, since the environmental fusion positioning algorithm needs to deal with complex problems such as the birth, death and matching of multipaths, in order to integrate this complex observation model into the Bayesian filtering algorithm in the actual system, the filtering algorithm needs to be complexly derived, which limits the scalability of the Bayesian filtering algorithm for such problems.

[0058] Based on this, an embodiment of the present invention provides a positioning method that uses a recurrent neural network to fit the Bayesian filtering results, thereby solving the convergence problem and scalability problem faced by the current Bayesian filtering algorithm when applied to environment fusion positioning tasks, thereby improving the positioning rate and accuracy.

[0059] Combine the following Figures 1 to 3 The present invention describes a positioning method based on a Bayesian filtering neural network, which can be executed by software and / or hardware in electronic devices such as computers, tablets, and mobile phones. Figure 1 As shown, the method comprises the following steps:

[0060] 101. Obtain an observation waveform matrix generated by the base station after receiving a waveform signal transmitted by a target object at each sampling moment.

[0061] It can be understood that, taking the single-base station ultra-wideband positioning system including a base station with a P antenna and a user equipment UE as an example, the target object refers to the UE and the virtual base station. At time k, after the base station receives the waveform signal transmitted by the target object, the generated observation waveform matrix is ​​Y shown in the above formula (3) k .

[0062] 102. Input the observed waveform matrix into a preset positioning model to obtain positioning information of the target object output by the positioning model;

[0063] Among them, the positioning model uses a recurrent neural network to fit the Bayesian filtering results. The recurrent neural network is trained by a cost function including motion likelihood and observation likelihood. The motion likelihood is generated based on the motion model of the target object sample, and the observation likelihood is generated based on the position information of the target object sample determined by a parameter estimation algorithm.

[0064] In this embodiment, a recurrent neural network is obtained by training a cost function including motion likelihood and observation likelihood, and the motion likelihood is generated based on the motion model of the target object sample, and the observation likelihood is generated based on the position information of the target object sample determined by a parameter estimation algorithm, so that the recurrent neural network can adaptively fit the Bayesian filtering results, thereby solving the convergence problem and scalability problem faced by the existing Bayesian filtering algorithm when it is applied to environmental perception fusion positioning tasks.

[0065] The following specifically describes the network structure and training process of the Bayesian filtering network (BFN) adopted by the positioning model provided in the embodiment of the present invention, so as to prove that based on the network structure and the corresponding cost function described in the embodiment of the present invention, the output result of the recurrent neural network is a maximum likelihood estimate of the output of the Bayesian filtering algorithm.

[0066] First, if Figure 2 As shown, the hidden Markov model is used to analyze f(x 1：k |Y 1：k ) is used for modeling, where m represents that we are discussing in a relatively stable environment m, and the environment does not change dramatically during the movement. Figure 2 The fitting of the Bayesian filtering algorithm using a recurrent neural network under the derivation of the hidden Markov model is shown.

[0067] In a specific embodiment, a recurrent neural network distribution f defined by parameter ρ is used. ρ (x 1：K |Y 1：K ) for f(x 1：K |Y 1：K) is fitted. Then the maximum likelihood estimation parameters of the recurrent neural network parameters ρ are:

[0068]

[0069] Among them, f(x k+1 |x k ) is the preset uniform linear motion model, which assumes that the user maintains uniform linear motion between time k and time k+1. Because in indoor positioning scenarios with high sampling frequency and slow user movement speed, it can be considered that this motion model is approximately satisfied. f(Y k |x k ) is the likelihood function given by the parameter estimation algorithm at a single moment. Therefore, in order to fit the Bayesian filter algorithm under the hidden Markov model, it is necessary to use the distribution f ρ (x 1：K |Y 1：K ) generates a series of samples Then, these samples are subjected to gradient descent using the cost function shown in formula (7). After the training is completed, f ρ (x 1：K |Y 1：K ) can output the result of Bayesian filtering.

[0070]

[0071] Therefore, the recurrent neural network structure ρ can be designed to meet the requirements of x 1：K Efficient learning of spatiotemporal constraints. k There is a motion association, so it is assumed that the user maintains a uniform linear motion between time k and k+1. 1：L，k is the coordinate of the virtual base station VA. It can be considered that in common indoor scenarios, relative to the communication frequency between UE and base station, VA will only undergo unpredictable slow migration as the environment evolves. Therefore, it can be considered that a 1：L，k It obeys the Brownian motion process, that is:

[0072]

[0073] Therefore, based on the content of the above embodiments, the recurrent neural network is constructed based on the long short-term memory network.

[0074] That is, using the Long Short-Term Memory (LSTM) network to model f ρ (x 1：K |Y 1：K ), and use an additional 2L (a base station requires two coordinates in a two-dimensional map) random variables that can be gradient-descended to represent the virtual base station a 1：L，k .

[0075] Furthermore, other additional prior information generated based on loop detection, initial value constraints, map constraints, birth and death conditions, etc. can be added to the cost function to train the recurrent neural network to improve the ability to constrain map information.

[0076] Therefore, based on the content of the above embodiment, the cost function also includes:

[0077] Prior constraints constructed based on other prior information during the RNN training process;

[0078] Other prior information includes prior information based on loop detection, initial value constraints, map constraints, and birth and death processes.

[0079] In summary, the network structure provided by the embodiment of the present invention can be as follows Figure 3 As shown. Figure 3 The network structure shown in the figure, the training process of the positioning model for fitting the Bayesian filtering result using the recurrent neural network provided in the embodiment of the present invention specifically includes:

[0080] First, in BFN, the parameters that can be gradient-decreased include the LSTM parameter ρ 0 and the coordinates of all virtual base stations a 1 , …, a L ,Right now:

[0081]

[0082] like Figure 3 As shown, LSTM takes Y at each moment k and the memory c transmitted at the previous moment k-1 As input, use a Gaussian distribution f ρ (x k |Y 1：k ) approximates f(x k |Y 1：k ). At each moment, LSTM outputs f ρ (x k |Y 1：k ), and resample the distribution using a random variable r that obeys a standard normal distribution to obtain the sampling result Each sampling result can be regarded as a node in the hidden Markov model. The following three constraints are imposed:

[0083] 1. Motion constraints: Combined with motion models to generate motion likelihood

[0084] 2. Observation constraints: Construct observation likelihood based on parameter estimation results

[0085] 3. Prior constraints: Based on other additional prior information such as loop detection, initial value constraints, map constraints, birth and death conditions, additional prior constraints can be imposed on each node.

[0086] After training, the output of LSTM is ρ (p 1：K |Y 1：K ) is the result of the recurrent neural network ρ fitting the Bayesian filter. It can be seen that through Figure 3 The BFN shown in the figure not only stabilizes the movement process of the virtual base station and imposes strong constraints on the map information, but also enhances the temporal correlation of the user's state using a recurrent neural network, thereby enhancing the neural network's learning ability for temporal correlation properties. At the same time, through the modeling of BFN and the nodeization of unknown variables, not only can the Bayesian filtering algorithm be easily completed, but also the complex constraints of the birth and death process of the traditional SLAM algorithm, map constraints, generation of matching information, etc., which require the redesign of the filtering algorithm, can be quickly incorporated into BFN, thereby providing greater potential for the deployment of BFN in more complex scenarios.

[0087] The above method is based on the hidden Markov model, represents the nodes to be estimated in the hidden Markov model using the output of the recurrent neural process, and constructs the likelihood represented by the edges represented by the motion model, observation model, and prior model as a cost function to train the recurrent neural network, thereby achieving the purpose of deriving a Bayesian filtering algorithm for environmental perception fusion positioning based on a recurrent neural network. The Bayesian filtering network provided by the embodiment of the present invention can perform maximum likelihood estimation on the Bayesian filtering result, and the design based on nodes and edges can also make it have strong scalability, suitable for complex multipath birth and death, false alarm, matching and other scenes, and can significantly improve the scalability of traditional environmental perception fusion positioning algorithms, and improve their convergence speed and positioning accuracy.

[0088] Based on the content of the above embodiment, the observed waveform matrix is ​​input into a preset positioning model to obtain the positioning information of the target object output by the positioning model, including:

[0089] The observed waveform matrix is ​​input into the result fitting layer of the positioning model, and the result fitting layer uses a recurrent neural network to fit the Bayesian filtering result to obtain the posterior distribution of the target object output by the result fitting layer;

[0090] The posterior distribution is input to the positioning confirmation layer of the positioning model, and the positioning confirmation layer determines the positioning information of the target object based on the posterior distribution, and obtains the positioning information of the target object output by the positioning confirmation layer.

[0091] In this embodiment, by setting a result fitting layer for fitting the Bayesian filtering result using a recurrent neural network in the positioning model, the positioning speed and accuracy are effectively increased.

[0092] The positioning method based on the Bayesian filtering neural network provided in the above embodiment of the present invention can be used in application scenarios that require positioning and navigation, such as smart factories, emergency search and rescue, and other satellite denial or weak signals. At the same time, the Bayesian filtering network used in the positioning model, as a Bayesian filtering algorithm, can be applied to various occasions such as positioning and navigation, satellite tracking, and remote sensing transmission.

[0093] Based on the same general inventive concept, the present invention also protects a positioning system based on a Bayesian filtering neural network. The positioning system based on a Bayesian filtering neural network provided by the present invention is described below. The positioning system based on a Bayesian filtering neural network described below and the positioning method based on a Bayesian filtering neural network described above can be referenced to each other.

[0094] Figure 4 Schematic diagram of the structure of the positioning system based on the Bayesian filtering neural network provided by the present invention. Figure 4 As shown, it includes: an acquisition module 410 and a positioning module 420; wherein,

[0095] The acquisition module 410 is used to acquire the observed waveform matrix generated by the base station after receiving the waveform signal transmitted by the target object at each sampling moment;

[0096] The positioning module 420 is used to input the observed waveform matrix into a preset positioning model to obtain the positioning information of the target object output by the positioning model;

[0097] Among them, the positioning model uses a recurrent neural network to fit the Bayesian filtering results. The recurrent neural network is trained by a cost function including motion likelihood and observation likelihood. The motion likelihood is generated based on the motion model of the target object sample, and the observation likelihood is generated based on the position information of the target object sample determined by a parameter estimation algorithm.

[0098] The positioning system based on the Bayesian filter neural network provided by the embodiment of the present invention obtains the observation waveform matrix generated by the base station after receiving the waveform signal emitted by the target object at each sampling moment, and then inputs the observation waveform matrix into the preset positioning model, thereby obtaining the positioning information of the target object output by the positioning model, and the positioning model uses a recurrent neural network to fit the Bayesian filter result. The recurrent neural network is trained by a cost function including motion likelihood and observation likelihood. The motion likelihood is generated based on the motion model of the target object sample, and the observation likelihood is generated based on the position information of the target object sample determined by the parameter estimation algorithm. The recurrent neural network is used to fit the Bayesian filter result, thereby solving the problem of affecting the positioning speed and positioning accuracy when applying the Bayesian filter algorithm in the environment-aware fusion positioning, and can significantly improve the scalability of the traditional environment-aware fusion positioning algorithm, and improve its convergence speed and positioning accuracy.

[0099] Based on the contents of the above embodiments, the recurrent neural network is constructed based on the long short-term memory network.

[0100] Based on the content of the above embodiment, the cost function also includes:

[0101] Prior constraints constructed based on other prior information during the RNN training process;

[0102] Other prior information includes prior information based on loop detection, initial value constraints, map constraints, and birth and death processes.

[0103] Based on the content of the above embodiment, the positioning module 420 is specifically used for:

[0104] The observed waveform matrix is ​​input into the result fitting layer of the positioning model, and the result fitting layer uses a recurrent neural network to fit the Bayesian filtering result to obtain the posterior distribution of the target object output by the result fitting layer;

[0105] The posterior distribution is input to the positioning confirmation layer of the positioning model, and the positioning confirmation layer determines the positioning information of the target object based on the posterior distribution, and obtains the positioning information of the target object output by the positioning confirmation layer.

[0106] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute a positioning method based on a Bayesian filter neural network, the method comprising: obtaining an observation waveform matrix generated by the base station after receiving a waveform signal transmitted by the target object at each sampling moment; inputting the observation waveform matrix into a preset positioning model to obtain positioning information of the target object output by the positioning model; wherein the positioning model uses a recurrent neural network to fit the Bayesian filter result, the recurrent neural network is trained by a cost function including motion likelihood and observation likelihood, the motion likelihood is generated based on the motion model of the target object sample, and the observation likelihood is generated based on the position information of the target object sample determined by a parameter estimation algorithm.

[0107] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0108] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the positioning method based on the Bayesian filtering neural network provided by the above-mentioned methods, and the method includes: obtaining an observation waveform matrix generated by the base station after receiving the waveform signal transmitted by the target object at each sampling moment; inputting the observation waveform matrix into a preset positioning model to obtain the positioning information of the target object output by the positioning model; wherein the positioning model uses a recurrent neural network to fit the Bayesian filtering result, and the recurrent neural network is trained by a cost function including motion likelihood and observation likelihood, the motion likelihood is generated based on the motion model of the target object sample, and the observation likelihood is generated based on the position information of the target object sample determined by a parameter estimation algorithm.

[0109] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the positioning method based on the Bayesian filtering neural network provided by the above-mentioned methods, the method comprising: obtaining an observation waveform matrix generated by the base station after receiving the waveform signal transmitted by the target object at each sampling moment; inputting the observation waveform matrix into a preset positioning model to obtain the positioning information of the target object output by the positioning model; wherein the positioning model uses a recurrent neural network to fit the Bayesian filtering result, the recurrent neural network is trained by a cost function including motion likelihood and observation likelihood, the motion likelihood is generated based on the motion model of the target object sample, and the observation likelihood is generated based on the position information of the target object sample determined by a parameter estimation algorithm.

[0110] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0111] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A positioning method based on a Bayesian filtering neural network, characterized in that: include: Obtaining the observed waveform matrix generated by the base station after receiving the waveform signal transmitted by the target object at each sampling moment; Inputting the observed waveform matrix into a preset positioning model to obtain positioning information of the target object output by the positioning model; Among them, the positioning model uses a recurrent neural network to fit the Bayesian filtering results, and the recurrent neural network is trained by a cost function including motion likelihood and observation likelihood. The motion likelihood is generated based on the motion model of the target object sample, and the observation likelihood is generated based on the position information of the target object sample determined by a parameter estimation algorithm.

2. The positioning method based on Bayesian filtering neural network according to claim 1 is characterized in that: The recurrent neural network is constructed based on a long short-term memory network.

3. The positioning method based on Bayesian filtering neural network according to claim 2 is characterized in that: The cost function also includes: Prior constraints constructed based on other prior information during the training of the recurrent neural network; The other prior information includes prior information generated based on loop detection, initial value constraints, map constraints, and birth and death processes.

4. The positioning method based on Bayesian filtering neural network according to any one of claims 1 to 3, characterized in that: The step of inputting the observed waveform matrix into a preset positioning model to obtain positioning information of the target object output by the positioning model includes: The observed waveform matrix is ​​input into the result fitting layer of the positioning model, and the result fitting layer uses the recurrent neural network to fit the Bayesian filtering result to obtain the posterior distribution of the target object output by the result fitting layer; The posterior distribution is input to the positioning confirmation layer of the positioning model, and the positioning confirmation layer determines the positioning information of the target object based on the posterior distribution, so as to obtain the positioning information of the target object output by the positioning confirmation layer.

5. A positioning system based on a Bayesian filtering neural network, characterized in that: include: An acquisition module, used to acquire an observation waveform matrix generated by the base station after receiving a waveform signal transmitted by a target object at each sampling moment; A positioning module, used for inputting the observed waveform matrix into a preset positioning model to obtain the positioning information of the target object output by the positioning model; Among them, the positioning model uses a recurrent neural network to fit the Bayesian filtering results, and the recurrent neural network is trained by a cost function including motion likelihood and observation likelihood. The motion likelihood is generated based on the motion model of the target object sample, and the observation likelihood is generated based on the position information of the target object sample determined by a parameter estimation algorithm.

6. The positioning system based on Bayesian filtering neural network according to claim 5, characterized in that: The recurrent neural network is constructed based on a long short-term memory network.

7. The positioning system based on Bayesian filtering neural network according to claim 6, characterized in that: The cost function also includes: Prior constraints constructed based on other prior information during the training of the recurrent neural network; The other prior information includes prior information generated based on loop detection, initial value constraints, map constraints, and birth and death processes.

8. The positioning system based on Bayesian filtering neural network according to any one of claims 5 to 7, characterized in that: The positioning module is specifically used for: The observed waveform matrix is ​​input into the result fitting layer of the positioning model, and the result fitting layer uses the recurrent neural network to fit the Bayesian filtering result to obtain the posterior distribution of the target object output by the result fitting layer; The posterior distribution is input to the positioning confirmation layer of the positioning model, and the positioning confirmation layer determines the positioning information of the target object based on the posterior distribution, so as to obtain the positioning information of the target object output by the positioning confirmation layer.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the positioning method based on the Bayesian filtering neural network as described in any one of claims 1 to 4 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the positioning method based on the Bayesian filtering neural network as described in any one of claims 1 to 4 is implemented.

Images