Positioning method and device, terminal and storage medium

Abstract

The embodiment of the invention provides a positioning method and device, a terminal and a storage medium. The positioning method comprises the steps that a CSI time domain data set of a target dynamic network is acquired, and one CSI time domain data set comprises at least one piece of CSI time domain data; one piece of CSI time domain data is time domain information of a CSI signal sent to a signal receiver by a signal transmitter in the target dynamic network; fast Fourier transform is carried out on the CSI time domain data set to obtain a CSI frequency domain data set of the target dynamic networking, and one CSI frequency domain data set comprises at least one piece of CSI frequency domain data; one piece of CSI frequency domain data is frequency domain information of a CSI signal sent to a signal receiver by a signal transmitter in the target dynamic network; inputting the CSI time domain data set and the CSI frequency domain data set into the trained first model to obtain predicted position information of the target object; therefore, the positioning of the target object in a large-scale space range can be realized, and the accuracy of target positioning estimation can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless positioning, and in particular to a positioning method and device, a terminal and a storage medium. BACKGROUND

[0002] Indoor positioning technologies generally include inertial navigation-based indoor positioning technology, Bluetooth-based indoor positioning technology and mobile hotspot (Wi-Fi) positioning technology, etc. The inertial navigation-based indoor positioning technology requires carrying corresponding devices (such as inertial sensors, etc.), which increases the positioning cost and inconvenience. The Bluetooth-based indoor positioning technology has a small coverage range and relatively low positioning accuracy. The Wi-Fi positioning technology has a wide coverage range, multiple application scenarios and can utilize existing Wi-Fi devices in buildings for positioning.

[0003] However, the current Wi-Fi positioning technology is generally based on the analysis of channel state information (CSI) of a signal receiver, is generally active positioning analysis, generally realizes two-dimensional plane positioning, and generally only analyzes the frequency domain information of the CSI signal, thereby causing the target object to need to carry a signal transmitter, lacking three-dimensional spatial positioning capability and / or low positioning accuracy, etc. SUMMARY

[0004] In view of this, the embodiments of the present application provide a positioning method and device, a terminal and a storage medium to solve the above technical problems.

[0005] The technical scheme of the present application is implemented as follows:

[0006] In a first aspect, the embodiments of the present application provide a positioning method, comprising: acquiring a CSI time domain data set of a target dynamic networking, wherein one CSI time domain data set comprises at least one CSI time domain data; and one CSI time domain data is time domain information of a CSI signal sent by one signal transmitter in the target dynamic networking to a signal receiver;

[0007] performing fast Fourier transform on the CSI time domain data set to obtain a CSI frequency domain data set of the target dynamic networking, wherein one CSI frequency domain data set comprises at least one CSI frequency domain data; and one CSI frequency domain data is frequency domain information of a CSI signal sent by one signal transmitter in the target dynamic networking to the signal receiver;

[0008] inputting the CSI time domain data set and the CSI frequency domain data set into a trained first model to obtain predicted position information of a target object.

[0009] In the scheme, the method further comprises: based on the topology planar graph, the CSI time domain dataset and the CSI frequency domain dataset are endowed with corresponding spatial information to obtain CSI data with spatial information.

[0010] The CSI time domain dataset and the CSI frequency domain dataset are input into the trained first model to obtain the predicted position information of the target object, comprising: the CSI data with spatial information is input into the trained first model to obtain the predicted position information of the target object.

[0011] In the scheme, the method further comprises:

[0012] The CSI sample dataset of the sample dynamic networking is obtained, wherein the CSI sample dataset comprises at least one CSI sample time domain data and CSI sample frequency domain data corresponding to the at least one CSI sample time domain data, or the CSI sample dataset comprises CSI sample data with spatial information obtained by endowing spatial information based on the CSI sample time domain dataset and the CSI sample time domain dataset.

[0013] The actual position information of the sample object in the sample dynamic networking is obtained.

[0014] The CSI sample dataset and the actual position information are input into the first model for training until a convergence condition is met, and the trained first model is obtained.

[0015] In the scheme, the convergence condition is met when the difference between the predicted position information and the actual position information corresponding to the sample is less than or equal to a threshold value; the predicted position information is obtained by inputting the CSI sample dataset into the first model.

[0016] In the scheme, the CSI time domain dataset of the target dynamic networking is obtained, comprising:

[0017] The total power of the first K candidate dynamic networkings is determined in descending order of total power, wherein the total power of any first candidate dynamic networking is the sum of the power of each signal transmitter to signal receiver in the first candidate dynamic networking; K is a positive integer less than or equal to L; L is the number of multiple signal devices;

[0018] In the scheme, the method further comprises:

[0019] The first signal device is determined as the signal receiver from the multiple signal devices, and the other signal devices in the multiple signal devices except the first L signal devices are determined as the signal transmitters;

[0020] The signal transmitters are selected in a predetermined number according to the signal intensity ranking of each signal transmitter to the first signal device in the multiple signal devices;

[0021] determining the lth signaler and a predetermined number of signal transmitters as a second alternative dynamic networking;

[0022] if the second alternative dynamic networking is multiple, determining a topological dispersion of each second alternative dynamic networking, wherein the topological dispersion is a ratio of a volume and a surface area of a convex polyhedron of each signal transmitter in the second alternative dynamic networking;

[0023] from the topological dispersion of the multiple second alternative dynamic networkings, selecting a second alternative dynamic networking corresponding to the maximum topological dispersion as the lth first alternative dynamic networking corresponding to the lth signaler; wherein, l is a positive integer less than or equal to L.

[0024] In the above scheme, the method further comprises: correcting the predicted position information based on historical predicted position information and a dynamic model.

[0025] In a second aspect, an embodiment of the present application provides a positioning device, comprising:

[0026] The acquisition module is configured to acquire a CSI time domain data set of the target dynamic networking, wherein one CSI time domain data set comprises at least one CSI time domain data; and one CSI time domain data is time domain information of a CSI signal transmitted by one signal transmitter in the target dynamic networking to the signal receiver.

[0027] The first processing module is configured to perform fast Fourier transform on the CSI time domain data set to obtain a CSI frequency domain data set of the target dynamic networking, wherein one CSI frequency domain data set comprises at least one CSI frequency domain data; and one CSI frequency domain data is frequency domain information of a CSI signal transmitted by one signal transmitter in the target dynamic networking to the signal receiver.

[0028] The second processing module is configured to input the CSI time domain data set and the CSI frequency domain data set into the trained first model to obtain predicted position information of the target object.

[0029] In a third aspect, an embodiment of the present application provides a terminal, which comprises a processor and a memory for storing a computer program capable of running on the processor; wherein the processor is configured to run the computer program to implement the positioning method of the embodiment of the present application.

[0030] In a fourth aspect, an embodiment of the present application provides a computer storage medium, which has computer executable instructions; the computer executable instructions are executed by a processor to implement the positioning method of the embodiment of the present application.

[0031] In a fifth aspect, an embodiment of the present application provides a computer program product, which comprises a computer program or instructions; the computer program or instructions are executed by a processor to implement the positioning method of the embodiment of the present application.

[0032] In the embodiment of the present application, a CSI time domain data set of the target dynamic networking is obtained, wherein one CSI time domain data set includes at least one CSI time domain data; one CSI time domain data is time domain information of a CSI signal sent by one signal transmitter to a signal receiver in the target dynamic networking; a fast Fourier transform is performed on the CSI time domain data set to obtain a CSI frequency domain data set of the target dynamic networking, wherein one CSI frequency domain data set includes at least one CSI frequency domain data; one CSI frequency domain data is frequency domain information of a CSI signal sent by one signal transmitter to a signal receiver in the target dynamic networking; the CSI time domain data set and the CSI frequency domain data set are input into the trained first model to obtain the predicted position information of the target object; thus, in the embodiment of the present application, adaptive networking is realized in different local positions through the dynamic networking function, so that the positioning of the target object in a large-scale space range is realized; and in the embodiment of the present application, the CSI time domain data set and the CSI frequency domain data set can be input into the trained first model, and the position of the target object is predicted based on the correlation of the CSI signals before and after, so that the accuracy of the positioning estimation of the target object is improved. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 The flowchart of the first positioning method provided by the embodiment of the present application.

[0034] Figure 2 The structure diagram of a first model provided by the embodiment of the present application.

[0035] Figure 3 The flowchart of the second positioning method provided by the embodiment of the present application.

[0036] Figure 4 The flowchart of the third positioning method provided by the embodiment of the present application.

[0037] Figure 5 The schematic diagram of a three-dimensional space human positioning system based on passive Wi-Fi provided by the embodiment of the present application.

[0038] Figure 6 The flowchart of the fourth positioning method provided by the embodiment of the present application.

[0039] Figure 7 The time domain target positioning mechanism diagram provided by the embodiment of the present application.

[0040] Figure 8 The frequency domain target positioning mechanism diagram provided by the embodiment of the present application.

[0041] Figure 9A schematic diagram of a laser sensor observing a human target position is provided for an embodiment of the present application.

[0042] Figure 10 A schematic diagram of a Wi-Fi dynamic networking is provided for an embodiment of the present application.

[0043] Figure 11 A schematic diagram of a Bayesian error compensation model is provided for an embodiment of the present application.

[0044] Figure 12 A structural schematic diagram of a second positioning device is provided for an embodiment of the present application.

[0045] Figure 13 A hardware structural schematic diagram of a terminal is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0046] The present application will be further described by the following figures and embodiments. It should be understood that the specific embodiments described herein are intended to explain the present application and are not intended to limit the present application.

[0047] In the following description, the suffixes such as "module", "part", or "unit" used for an element are merely intended for facilitating explanation of the present application, and have no specific meaning by itself. Thus, "module", "part", or "unit" can be mixedly used. In addition, in the following description, the prefix such as "first" or "second" used for identification of information is merely intended for facilitating explanation of the present application, and has no specific meaning by itself. In addition, in the following description, "at least one" means one or more; "multiple" means two or more. "At least one" means one or more; "multiple" means two or more.

[0048] In some embodiments, indoor positioning technology generally includes inertial navigation-based indoor positioning technology, Bluetooth-based indoor positioning technology, and Wi-Fi-based positioning technology, etc.

[0049] The basic principle of the inertial navigation-based indoor positioning technology is to obtain the orientation, acceleration, and angular acceleration of the target motion through an inertial sensor (IMU), and to calculate the motion distance and motion direction of the target in combination with the position information of the target at the previous time, thereby obtaining the relative position information at the current time. The inertial sensor is usually composed of an accelerometer, a gyroscope, and a magnetometer. In the application of indoor human positioning, people can obtain the acceleration and angular acceleration of human motion by wearing wearable devices with IMU, such as sports bands or electronic watches, thereby realizing position estimation.

[0050] Bluetooth-based positioning technology, such as Bluetooth-based fingerprint positioning method, for example, a user device can be positioned by installing 19 Bluetooth beacons in a 600 square meter space; but due to the limitation of Bluetooth specification and hardware, Bluetooth-based positioning technology is not suitable for location estimation with fine granularity and / or low latency.

[0051] Wi-Fi-based indoor positioning technology is mainly divided into two categories: fingerprint positioning technology and geometric measurement positioning technology.

[0052] Optionally, the fingerprint positioning technology refers to constructing a fingerprint library by mapping the features of collected signals to corresponding physical location points, and matching the signal features of the collected target signal with the fingerprint library, so as to realize positioning. A common matching method is to calculate the Euclidean distance between the target signal features and the signal features in the fingerprint library, and the position corresponding to the signal feature with the smallest distance is the physical position of the measured signal. Common signal features include received signal strength (RSS) and channel state information (CSI). RSS is the superposition of all signal powers, describes the Wi-Fi media access control (MAC) layer information features, and indicates the attenuation of wireless signal propagation between Wi-Fi transceiver devices, but the fluctuation of the attenuated signal causes obvious multipath effect over time. CSI describes the channel features of the radio frequency signal in a more fine-grained manner, which is beneficial to realize more accurate positioning. For example, in an indoor positioning system that applies received signal strength (RSS) for positioning, it is mainly divided into an offline stage and an online stage. In the offline stage, RSS is measured at each reference point to establish a fingerprint library, which maps the physical points in the plane to the RSS in the fingerprint library; in the online stage, the collected RSS information is searched and matched in the fingerprint library to realize target positioning. However, RSS is the superposition of multipath signal power, and the multipath effect is large, the time stability is poor, which leads to large positioning error in complex indoor environment, and the positioning accuracy is usually 2 to 4 meters. For example, channel state information (CSI) is used for indoor positioning, and a wireless signal propagation loss model based on path attenuation index and environmental variables is constructed, which has a sub-meter level positioning accuracy. For example, CSI is applied to fingerprint positioning, a mapping between spatial point positions and collected CSI is established, an offline map-level CSI fingerprint library is constructed, and an online stage uses the maximum likelihood method to match the CSI values of the measured points with the fingerprint library, so as to obtain the predicted position information. For example, the amplitude and phase information of the CSI of a single carrier in the frequency domain can be used as fingerprint features, which provides signal features of each carrier, has relatively more fine-grained information, and improves the positioning accuracy. At the same time, researchers have established fingerprint libraries for the amplitude and phase information of CSI respectively; for example, the amplitude center of the CSI amplitude information of the received signal of the device is clustered to determine the positioning area; and the carrier arrival angle calculated from the phase information of the CSI is used as the position fingerprint library to realize position estimation.

[0053] Optionally, geometric measurement positioning technology is mainly divided into angle of arrival (AoA) based and distance based measurement positioning technology. The angle of arrival (AoA) based positioning technology usually utilizes the multi-antenna array (single AP) of the Wi-Fi device to perform eigenvalue decomposition of the signal subspace and noise subspace of the radio frequency signal, so as to estimate the AoA, and then estimate the target position according to the geometric relationship. At the same time, the angle of arrival AoA of the signal can also be measured by using multiple AP access points combined with the MUSIC algorithm, so as to realize positioning. However, this scheme needs to customize the receiver, or calibrate and modify the commercial network card, and cannot directly use the existing Wi-Fi device and access to the WLAN system. The principle of the distance based positioning technology is to estimate the distance according to the attenuation law of the received power or strength with the change of distance in the signal propagation process, and to realize target positioning based on the three-edge positioning method. In addition, distance measurement can also be realized by measuring the time of arrival (ToA), and the distance of the node is estimated by estimating the time delay of the multi-node transmitted Wi-Fi signal to the receiving device. However, this scheme also needs to modify the bottom protocol of the network card, and has high requirements on the hardware performance of the device.

[0054] The above-mentioned inertial navigation indoor positioning technology, Bluetooth based indoor positioning technology and Wi-Fi based positioning technology have the following characteristics: the inertial navigation indoor positioning technology needs the target to carry the corresponding device, which increases the positioning cost and inconvenience, and the cumulative positioning error gradually enlarges with the extension of the motion path; the Bluetooth based indoor positioning technology has small coverage range and relatively low positioning accuracy; and the Wi-Fi based indoor positioning technology has wide coverage range, multiple application scenarios and can use the existing Wi-Fi device in the building to more accurately position the target.

[0055] In the positioning route, the Wi-Fi based indoor positioning technology in the related art usually analyzes the CSI of the received signal, and is usually active positioning analysis, that is, the target carries a signal transmitting antenna unit, and the fixed receiving signal node AP is used to realize positioning. In this way, the active Wi-Fi based indoor positioning scheme in the related art needs the target to carry the signal transmitting antenna device, which not only has high cost, but also affects the convenience of human motion to a certain extent, and the portability, endurance and difference of the device will also affect the positioning of the human.

[0056] In terms of positioning function, the Wi-Fi indoor positioning technology in the related art only realizes positioning in a two-dimensional plane. The current scheme for realizing three-dimensional space positioning is to realize it by means of the change of the barometer in the altitude, but the accuracy is poor. Thus, the Wi-Fi indoor positioning scheme in the related art only considers the target positioning in a two-dimensional plane, and lacks the three-dimensional space positioning capability; for example, the Wi-Fi indoor positioning scheme only considers the Wi-Fi antenna in the same horizontal plane or the same floor, and positions the target position of the human body in the same plane or the same floor, and the spatial effectiveness is insufficient. However, it is very important to accurately position the three-dimensional space position of the human body in the positioning, especially in buildings with high vertical height.

[0057] In terms of positioning performance, the Wi-Fi indoor positioning technology in the related art usually analyzes the characteristic quantities such as the amplitude and phase of the CSI of the signal, the characteristic quantities are relatively single, and the CSI time domain information and antenna device layout information are missing; the modeling and matching of the fingerprint library are performed through the common machine learning classification model, and the matching accuracy needs to be further improved; in terms of noise processing, the noise of the input signal is usually filtered through filtering or dimension reduction, and the noise of the input signal is directly filtered or dimension reduced, resulting in loss of effective information of the signal. Thus, the Wi-Fi indoor positioning technology in the related art can use the amplitude, phase and other characteristic quantities of the channel state information (CSI) to realize target positioning in the manner of constructing a position fingerprint library. The RSS or CSI of the wireless signal of a plurality of wireless access nodes is received, the amplitude and phase characteristics of the RSS or CSI are counted, and the RSS or CSI corresponding to the spatial target position is calibrated as a Wi-Fi fingerprint, so as to realize the positioning of the target in the grid map. However, with the increase of the indoor scale area, the calibration quantity increases in the square, thereby bringing a large amount of calculation, and in addition, the RSS or CSI brings positioning error due to the influence of multipath effect and environmental factors such as temperature and humidity.

[0058] Moreover, the Wi-Fi positioning technology in the related art considers the specificity of the CSI time domain mode corresponding to the different positions of the human body when positioning the human body target, but due to the influence of the multipath effect and environmental factors (temperature and humidity, building structure complexity), the wireless signal CSI is also affected. In addition, the target positioning estimation of the Wi-Fi wireless signal lacks the continuity prior of the human body motion in the objective world.

[0059] As shown in Figure 1 The embodiment of the present application provides a positioning method, which comprises the following steps:

[0060] Step S11: acquiring a CSI time domain data set of the target dynamic networking, wherein one CSI time domain data set comprises at least one CSI time domain data; one CSI time domain data is the time domain information of the CSI signal sent by one signal transmitter to a signal receiver in the target dynamic networking;

[0061] Step S12: performing a fast Fourier transform on the CSI time-domain dataset to obtain a CSI frequency-domain dataset of the target dynamic networking, wherein one CSI frequency-domain dataset comprises at least one CSI frequency-domain data; and one CSI frequency-domain data is frequency-domain information of a CSI signal transmitted by one signal transmitter to a signal receiver in the target dynamic networking.

[0062] Step S13: inputting the CSI time-domain dataset and the CSI frequency-domain dataset into the trained first model to obtain predicted position information of the target object.

[0063] The positioning method provided by the embodiments of the present application can be executed by a terminal; the terminal can be any kind of mobile terminal or fixed terminal. For example, the terminal can be, but is not limited to, at least one of the following: a mobile communication device, a computer, a server, a tablet computer, a game device, a smart office device, an industrial device, a wearable device, and the like.

[0064] In some embodiments, the target dynamic networking comprises a plurality of signalers; the plurality of signalers comprises one signal receiver and at least one signal transmitter. Here, the CSI signal transmitted by one signal transmitter can directly reach the signal receiver or can reach the signal receiver via the target object.

[0065] Optionally, the CSI time-domain data can be time-domain information of the CSI signal passing through the target object; and the CSI frequency-domain data can be frequency-domain information of the CSI signal passing through the target object.

[0066] Optionally, the time domain is a time domain; for example, the independent variable is time, i.e., the horizontal axis is time and the vertical axis is the change of the signal. The dynamic signal is a function describing the value of the CSI signal at different times.

[0067] Optionally, the frequency domain is a frequency domain; for example, the independent variable is frequency, i.e., the horizontal axis is frequency and the vertical axis is the amplitude of the frequency signal, which is commonly referred to as a frequency spectrum diagram.

[0068] Optionally, the time-domain information and the frequency-domain information can be feature quantities of the time domain and the frequency domain extracted from the CSI information; for example, they can be amplitude, phase, frequency, and / or sub-sequence waveform, etc.

[0069] In some embodiments, the target object can be an object to be positioned; for example, the target object can be a human body, etc.

[0070] In some embodiments, the CSI signal can also be replaced by other signals; for example, it can be any reference signal or positioning signal or measurement signal, etc.; for example, a positioning reference signal (PRS), etc.

[0071] In some embodiments, the fast Fourier transform in step S12 can be any fast Fourier transform as long as the CSI time domain information can be converted into CSI frequency domain information. Of course, in other embodiments, step S12 can be: determining a CSI frequency domain data set based on the CSI time domain data set; here, as long as the CSI time domain information can be converted into the CSI frequency domain information corresponding to the CSI time domain information.

[0072] In some embodiments, the first model can be any positioning network model; the positioning network model is a neural network model. For example, as shown in Figure 2 The first model can be a neural network model for three-dimensional space human body positioning; the first model is divided into two stages: a training stage and a prediction stage. The training stage is to establish a fingerprint library mapping different target dynamic networking CSI signals and target objects, and to construct a mode library of target dynamic networking CSI signals corresponding to different physical space positions of the target object. The prediction stage is to identify the features of the CSI signals in the target dynamic networking, and then match them with the fingerprint library established in the training stage, so as to predict the predicted position information of the target object.

[0073] In some embodiments, the target positioning information is three-dimensional space target positioning information.

[0074] In the embodiments of the present application, dynamic networking function can be used to achieve adaptive networking in different local positions, so as to realize the positioning of target objects in a large-scale space range; and the CSI time domain data set and the CSI frequency domain data set can be input into the trained first model, the position of the target object can be predicted based on the correlation of the CSI signals, so as to improve the accuracy of the positioning estimation of the target object.

[0075] In some embodiments, the method further comprises: based on the topological flat map, assigning corresponding spatial information to the CSI time domain data set and the CSI frequency domain data set to obtain CSI data with spatial information;

[0076] Step S13 comprises: inputting the CSI data with spatial information into the trained first model to obtain the predicted position information of the target object.

[0077] In some embodiments, the topological flat map is a topological map constructed for the cuboid in which the target dynamic networking and the target object are located; for example, the topological flat map can be used to describe the top view, the bottom view, the side view, the front view and / or the back view of the cuboid; the side view includes the left view and / or the right view. For example, as shown in Figure 2 A side view and a bottom view of a topological flat map are disclosed; the topological flat map can be a Wi-Fi topological flat map.

[0078] In some embodiments, the CSI data with spatial information can be CSI time domain information and CSI frequency domain information that retains spatial information of a CSI signal source.

[0079] For example, a storage space can be allocated for the CSI signal each time the CSI time domain information and the CSI frequency domain information are input into the trained first model, and the corresponding CSI time domain data and CSI frequency domain data in the target dynamic network are mapped into the allocated storage space, and the space of other empty nodes is filled with a tensor of 0. In this way, the CSI data with spatial information is obtained.

[0080] For example, the signal transmitter is placed in a cuboid of 5*4*10 blocks, and each block is a square. If there is a corresponding signal transmission CSI signal in a square, the CSI signal has corresponding CSI time domain information and CSI frequency domain information, the square is marked black, and the corresponding CSI time domain information and CSI frequency domain information are input. The other squares are empty. This is equivalent to a large cuboid space formed by dynamic networking and target object construction. Only a few small squares have data in the large cuboid, and other places are 0. In this way, the spatial information of the CSI signal can be retained to predict the position of the target object.

[0081] In some embodiments, the method further comprises: obtaining identification information of the target dynamic network; and step S13 comprises: inputting the CSI data with spatial information and the identification information into the trained first model to obtain predicted position information of the target object. Here, the identification information is used to indicate the target dynamic network, so that it can be marked which dynamic network the CSI signal belongs to.

[0082] In the embodiments of the present application, the CSI data with spatial information can be input into the trained first model for prediction, so as to form a three-dimensional spatial layout, adapt to Wi-Fi dynamic networking in different layouts in buildings (such as indoor), and can include devices in different positions such as ceiling, wall and floor, and devices arranged in different directions vertically and horizontally, form a three-dimensional spatial layout, and determine the three-dimensional spatial position of the target object in the building, so as to facilitate accurate positioning of the target. The embodiments of the present application propose a three-dimensional spatial positioning technology based on Wi-Fi, which dynamically networks the moving target through devices in different spatial distributions in the room, analyzes the channel state information, and obtains the direction information and height information of the target.

[0083] In addition, the spatial positioning of the target object is combined with the Wi-Fi antennas in the vertical and horizontal multi-topology structure in the dynamic network, which increases the redundancy of spatial positioning and avoids large deviation caused by a single Wi-Fi signal with errors.

[0084] For example,Figure 3 In some embodiments, the CSI time-domain dataset of the target dynamic networking is acquired in step S11, including:

[0085] Step S111: respectively determine the total power of the first K candidate dynamic networkings in descending order of total power, wherein the total power of any first candidate dynamic networking is the sum of the power of each signal transmitter to signal receiver in the first candidate dynamic group; K is a positive integer less than or equal to L; L is the number of multiple signalers;

[0086] Step S112: select the first candidate dynamic networking corresponding to the maximum total power from the total power of the first K candidate dynamic networkings as the target dynamic networking.

[0087] In the embodiments of the present application, steps S111 to S112 can include determining multiple first candidate dynamic networkings, and selecting one first candidate dynamic networking from the multiple first candidate dynamic networkings as the target dynamic networking.

[0088] Optionally, the signaler can be any Wi-Fi signal device, as long as the signaler can send and / or receive signals.

[0089] Optionally, the multiple signalers in step S111 can be multiple signalers for positioning a target object. For example, the multiple signalers can be arranged in a building or a room or a predetermined spatial range; the building or the room or the predetermined spatial range has a target object.

[0090] Optionally, the Lth signaler can be any one of the multiple signalers; for example, the Lth signaler can be the 1st, 2nd, or 10th, etc. of the multiple signalers.

[0091] Optionally, the signal receiver in the Lth first candidate dynamic networking can be the Lth signaler, and the signal transmitter can be other signalers in the multiple signalers except the Lth locator.

[0092] Optionally, step S111 can include: determining the 1st to Lth first candidate dynamic networkings from the L signalers; and respectively determining the total power of the first K candidate dynamic networkings in descending order of total power of the first candidate dynamic networkings from the 1st to Lth first candidate dynamic networkings.

[0093] Optionally, step S111 can also be: selecting the first dynamic networking corresponding to the total power satisfying the first predetermined condition from the total power of the 1st to Kth first candidate dynamic networkings as the target networking. Here, the total power satisfying the first predetermined condition can be: any one of the maximum total power, the second largest total power, or the total power greater than or equal to the power threshold, etc.

[0094] Optionally, the total power can be the sum of the transmission power of the CSI signals transmitted by each signal transmitter to the signal receiver in the first alternative networking. In other embodiments, the total power can also be the sum of the power of each signal transmitter in the first alternative networking, or the sum of the power of the CSI signals received by the signal receiver from each signal transmitter.

[0095] Optionally, the CSI time domain data set of the target dynamic networking obtained in step S11 can include: obtaining the first to Lth first alternative dynamic networkings respectively corresponding to the first to Lth signal transmitters; wherein L is a positive integer and less than or equal to the number of the plurality of signal transmitters used for networking; in the case that the number of the obtained first alternative dynamic networkings is greater than or equal to K, determining the total power of the first to Kth first alternative dynamic networkings respectively, wherein the Kth total power is the sum of the power of each signal transmitter to the signal receiver in the Kth first alternative dynamic networking; K is a positive integer and less than or equal to the data amount of the plurality of signal transmitters; selecting the first alternative dynamic networking corresponding to the maximum total power from the total power of the first to Kth first alternative dynamic networkings as the target dynamic networking.

[0096] For example, if the plurality of signal transmitters is 10, the first to 10th first alternative dynamic networkings can be determined. In the first first alternative dynamic networking, the first signal transmitter is the signal receiver and the other 9 signal transmitters can be signal transmitters, and so on. In the 10th first alternative dynamic networking, the 10th signal transmitter is the signal receiver and the other 9 signal transmitters can be signal transmitters. If the plurality of signal transmitters has determined the first alternative dynamic networking for 6 times, i.e., the first to 6th first alternative dynamic networkings are obtained, the target dynamic networking can be determined from the 6 first alternative dynamic networkings. For example, the total power of the first to 6th first alternative dynamic networkings is determined respectively. For example, the total power of the first first alternative dynamic networking is the first total power, and so on. The total power of the 6th first alternative dynamic networking is the sixth total power. If the maximum value of the first to sixth total powers is the third total power, the third alternative dynamic networking corresponding to the third total power is determined as the target dynamic networking.

[0097] In the embodiments of the present application, a plurality of signal transmitters in a building or a room or a predetermined space range are used for dynamic networking to obtain CSI data at different positions, and a deep neural network is used for fusion analysis of the CSI signals at different positions in the dynamic networking to accurately estimate the three-dimensional spatial position of the target in the indoor space.

[0098] Since the embodiment of the present application carries out passive positioning of the target object based on the signalers, the signalers in the space such as the building can be utilized, and the human body does not need to wear the signal transmitting antenna device, thereby reducing the load burden of the target object, reducing the deployment success, and improving the promotion and coverage.

[0099] In addition, in the embodiment of the present application, the first alternative dynamic network constructed by the plurality of signals (for example, K signals) can be compared, so that the plurality of first alternative dynamic networks can be comprehensively considered, and the appropriate first alternative dynamic network can be selected as the final target dynamic network.

[0100] In some embodiments, the method further comprises determining any first alternative dynamic network. Optionally, the terminal determines the first to the Lth first alternative dynamic network. Optionally, the terminal determines the first to the Lth first alternative dynamic network, comprising: the terminal determines the lth first alternative dynamic network.

[0101] Optionally, the terminal acquires the lth first alternative dynamic network, comprising: determining the lth signaler as the signal receiver and determining the other signalers except the Lth signaler in the plurality of signalers as the signal transmitters; sorting the signal transmitters in the plurality of signalers according to the signal strength to the lth signaler, selecting a predetermined number of signal transmitters in front of the signal strength sorting; determining the lth signaler and the predetermined number of signal transmitters as the second alternative dynamic network; if the second alternative dynamic network is multiple, determining the topological dispersion of each second alternative dynamic network, wherein the topological dispersion is the ratio of the volume and the surface area of the convex polyhedron of each signal transmitter in the second alternative dynamic network; selecting the second alternative dynamic network corresponding to the maximum topological dispersion from the topological dispersion of the plurality of second alternative dynamic networks as the lth first alternative dynamic network corresponding to the lth signaler; wherein, l is a positive integer less than or equal to L.

[0102] Optionally, the terminal obtains other any one first candidate dynamic network in a similar manner to that of obtaining the l first candidate dynamic networks. For example, obtaining the first first candidate dynamic network comprises: determining the first signaler as the signal receiver and determining other signalers in the plurality of signalers as the signal transmitters; ranking the signal transmitters in the plurality of signalers according to the signal strength from the signal transmitters to the first signaler, and selecting a predetermined number of signal transmitters in front of the ranking; determining the first signaler and the predetermined number of signal transmitters as the second candidate dynamic network; if there are a plurality of second candidate dynamic networks, determining the topological dispersion of each second candidate dynamic network, wherein the topological dispersion is the ratio of the volume to the surface area of the convex polyhedron of each signal transmitter in the second candidate dynamic network; and selecting the second candidate dynamic network corresponding to the maximum topological dispersion from the topological dispersion of the plurality of second candidate dynamic networks as the first first candidate dynamic network corresponding to the first signaler. In this way, the L-1th first candidate dynamic network corresponding to the L-1th signaler can be obtained. In the same way, obtaining the Lth candidate dynamic network can comprise: determining the Lth signaler as the signal receiver and determining other signalers in the plurality of signalers as the signal transmitters; ranking the signal transmitters in the plurality of signalers according to the signal strength from the signal transmitters to the Lth signaler, and selecting a predetermined number of signal transmitters in front of the ranking; determining the Lth signaler and the predetermined number of signal transmitters as the second candidate dynamic network; if there are a plurality of second candidate dynamic networks, determining the topological dispersion of each second candidate dynamic network, wherein the topological dispersion is the ratio of the volume to the surface area of the convex polyhedron of each signal transmitter in the second candidate dynamic network; and selecting the second candidate dynamic network corresponding to the maximum topological dispersion from the topological dispersion of the plurality of second candidate dynamic networks as the Lth first candidate dynamic network corresponding to the Lth signaler.

[0103] In the embodiment of the application, when each signaler is the signal receiver, a plurality of second candidate dynamic networks can be determined, and one second candidate dynamic network is selected from the plurality of second candidate dynamic networks to be the first candidate dynamic network.

[0104] Optionally, the predetermined number is less than or equal to L.

[0105] Optionally, when the Lth signaler is the signal receiver, one or more second candidate dynamic networks can be obtained. For example, if there is one second candidate dynamic network, the second candidate dynamic network is determined to be the first candidate dynamic network.

[0106] Optionally, selecting the second candidate dynamic network corresponding to the maximum topological dispersion as the Lth first candidate dynamic network corresponding to the Lth signaler can also be: selecting the second candidate dynamic network corresponding to the topological dispersion satisfying the second predetermined condition as the Lth first candidate dynamic network corresponding to the Lth signaler. Here, the topological dispersion satisfying the second predetermined condition can include: any one of the maximum topological dispersion, the second maximum topological dispersion, or the topological dispersion greater than or equal to the topological dispersion threshold.

[0107] For example, a calculation method of the topological dispersion is provided; for example, the number of signal transmitters in the second candidate dynamic network is 4, and the 4 signal transmitters form a 4-cone in the spatial position; and the topological dispersion is the volume-to-surface area ratio of the 4-cone.

[0108] In the embodiment of the present application, when there are multiple second candidate dynamic networks, the second candidate dynamic network corresponding to the topological dispersion satisfying the second predetermined condition (for example, the maximum topological dispersion) in the second dynamic network can be selected as the first candidate dynamic network; in this way, the size of the unit area flux in the topological structure can be considered, so that the topological structure with a larger flux can better locate the observation target, and the occurrence of the extreme case of small angle and long distance is reduced.

[0109] As shown in Figure 4 The embodiment of the present application provides a positioning method, comprising the following steps:

[0110] Step S14: acquiring a CSI sample data set of the sample dynamic network, wherein the CSI sample data set includes at least one CSI sample time domain data and CSI sample frequency domain data corresponding to the at least one CSI sample time domain data, or the CSI sample data set includes CSI sample data with spatial information based on the CSI sample time domain data set and the CSI sample time domain set.

[0111] Step S15: acquiring actual position information of the sample object in the sample dynamic network;

[0112] Step S16: inputting the CSI sample data set and the actual position information into the first model for training until a convergence condition is met, to obtain the trained first model.

[0113] Optionally, step S16 is performed before step S11.

[0114] Optionally, the embodiments of steps S14 to S16 are independent of the embodiments of steps S11 to S13.

[0115] In some embodiments, the sample dynamic networking includes a plurality of signalers; the plurality of signalers includes one signal receiver and at least one signal transmitter. Here, the CSI signal transmitted by one signal transmitter can directly reach the signal receiver or can pass through the sample object to reach the signal receiver.

[0116] Optionally, the CSI sample time domain data can be time domain information of the CSI signal passing through the sample object; and the CSI sample frequency domain data can be frequency domain information of the CSI signal passing through the sample object.

[0117] In some embodiments, the sample object can be a positioned object; for example, the sample object can be a human body, etc.

[0118] In some embodiments, the step S14 can include obtaining actual position information of the sample object in the sample dynamic networking based on a position sampling manner. For example, as shown in FIG. 6, a certain number of human body spatial positions are obtained in a random sampling manner or a certain probability distribution manner; the human body spatial positions are the actual position information. Figure 2

[0119] In some embodiments, the convergence condition is satisfied when a difference between the predicted position information corresponding to the sample and the actual position information is less than or equal to a threshold value; the predicted position information is predicted by inputting the CSI sample data set into the first model.

[0120] In the embodiments of the present application, the first model can be trained based on the CSI data extracted from the time domain, the frequency domain, or the time domain and the frequency domain with idle information, etc., so that a more accurate trained first model can be obtained, which is beneficial to obtaining more accurate predicted position information of the target object in subsequent prediction.

[0121] In some embodiments, the method further includes correcting the predicted position information based on historical predicted position information and a motion model.

[0122] In some embodiments, the historical predicted position information can be the predicted position information at the previous moment; and the predicted position information is corrected based on the historical predicted position information and the motion model, including: inputting the predicted position information at the previous moment into the motion model to correct the predicted position information at the previous moment to obtain the predicted position information at the next moment.

[0123] Optionally, the motion model further includes a correction value, where the correction value is used to represent the influence of external environmental factors on the position of the target object. The predicted position information is corrected based on the historical predicted position information and the motion model, including: inputting the predicted position information at the previous moment into the motion model including the correction value to correct the predicted position information at the previous moment to obtain the predicted position information at the next moment.

[0124] ​Optionally, the predicted position information comprises predicted position information of a previous time point; the method further comprises: inputting the predicted position information of the previous time point into the motion model to obtain the predicted position information of the current time point; or, based on the predicted position information of the previous time point and the correction value, obtaining predicted position information of a next time point, wherein the correction value is used to represent the influence of external environmental factors on the position of the target object.

[0125] Optionally, the motion model can be: wherein, and are position coordinate components of the target object on the X, Y and Z axes at time t respectively; and are velocity components of the target object on the X, Y and Z axes at time t respectively; △t is a predetermined time interval, and △t is the predetermined time interval of the target dynamic networking; [] represents downward taking. Here, t-1 can be a previous time point, and t can be a next time point.

[0126] Optionally, based on the predicted position information of the previous time point and the correction value, obtaining the predicted position information of the next time point can be: based on the predicted position information of the previous time point inputting the motion model and then adding the correction value to obtain the predicted position information of the next time point.

[0127] In the embodiments of the present application, an error compensation method is proposed, which can fuse the predicted position information of the first trained model in time sequence and the motion model based on the Bayesian principle, compensate and reduce the predicted position information of the first trained model with the prior information of the motion model, so as to realize correction of error compensation correction, improve the robustness of position estimation of the target object in space. Compared with the traditional signal filtering method, the error compensation can avoid the loss of signal effectiveness features caused by filtering, ensure good positioning accuracy, ensure positioning precision (decrease of positioning accuracy), and ensure the robustness of the first model in different scenes.

[0128] In order to further explain any embodiment of the present application, a specific embodiment is provided as follows.

[0129] As shown in Figure 5 , a positioning method is provided, which is applied to a passive Wi-Fi three-dimensional space human positioning system, and the positioning method can be executed by the terminal of the above-mentioned embodiment or by the passive Wi-Fi three-dimensional space human positioning system; the system is composed of three modules, that is, a Wi-Fi dynamic networking system, a self-attention mechanism graph structure network and an error compensation model. The difficulties solved by the three modules and the gains obtained are as follows:

[0130] (1) The Wi-Fi dynamic networking system realizes the spatial optimization networking of multiple signals when positioning the target, increases the positioning redundancy through the spatial topology structure, and reduces the disturbance of noise signals of a single device on the positioning accuracy; after the Wi-Fi dynamic networking is completed, the antenna devices in the networking and the CSI of the received signals are used as the information source of the fingerprint database for the subsequent positioning analysis, and the feature quantities in the time domain and the frequency domain of the information source are further extracted.(2) The graph structure network with the self-attention mechanism is used as the positioning network model, which is used to realize the mapping of the feature quantities of the information source and the three-dimensional space physical position coordinates of the target, that is, the establishment of the fingerprint database. In the process of realizing the construction of the fingerprint database, the graph structure network with the self-attention mechanism has three advantages over the traditional deep neural network: 1) the embodiment of the application can realize the processing of CSI signals of Wi-Fi sources with different quantities in the same network, while retaining the spatial topology structure of the signal source; in the related art, the number of CSI signals input into the network is usually fixed, and the spatial distribution information of the signal source cannot be obtained; 2) the positioning network model adopts the model with the self-attention mechanism, which can establish the semantic correlation of the time sequence waveform signal in the time sequence, better extract the high-dimensional features in the time domain and the frequency domain of the signal, and 3) the position sampling method is adopted, which is different from the traditional scheme in which position sampling is required for each point when constructing the fingerprint database. Only partial sampling is required to achieve the same training effect, which reduces the sampling quantity of the sample and the time cost of training.(3) The error compensation model is used to replace the noise filtering link in the traditional scheme, which avoids the damage of noise filtering to the effective information. The main process is to correct and compensate the CSI positioning prediction based on the Markov chain through the motion information of the target.

[0131] Step S21: determining the Wi-Fi dynamic networking. Optionally, the Wi-Fi dynamic networking can be the target dynamic networking in the above embodiment.

[0132] Wi-Fi dynamic networking (Wi-Fi ad hoc networking): refers to a plurality of Wi-Fi signalers in a certain range; the Wi-Fi signaler can include a Wi-Fi receiver or a Wi-Fi transmitter; the Wi-Fi receiver and the Wi-Fi transmitter form a certain common network structure. The goal of the Wi-Fi ad hoc networking is to optimize the network communication efficiency with the minimum number of nodes and communication links, and to maximize the signal fidelity. The Wi-Fi ad hoc networking proposed in the embodiment of the application adopts a simple topology optimization method: based on the signal power received by the Wi-Fi receiver and the spatial complexity of the signal source networking around the target, a predetermined number of stronger CSI signal sources are screened out for networking. Optionally, the Wi-Fi signaler can be the signaler in the previous embodiment; the Wi-Fi receiver and the Wi-Fi transmitter are the signal receiver and the signal transmitter in the previous embodiment.

[0133] After the optical fiber communication is converted into a wireless network, the static network of the router is constituted by the wired backhaul and the wireless backhaul between the Wi-Fi routers; the dynamic networking manner proposed in the embodiment of the application is to network according to the state of the observation target, and the dynamic network formed in each time interval is adaptively optimized according to the state of the target. Optionally, the target can be the target object or the human body in the above embodiment.

[0134] Optionally, in the Wi-Fi dynamic networking, one router node is randomly selected as a signal receiver, and the wireless signals transmitted by other Wi-Fi router nodes (i.e. signal transmitters) are received, the signals are sorted according to the strength of the received signal strength indication (RSSI) of the signal receiver (for example, the RSSI can be the logarithm of the signal power multiplied by a certain coefficient), the signal transmitters of a predetermined number of stronger signals are selected, and the total power in the ad hoc network is calculated. Here, for the case that the signal strengths of multiple signal receivers are the same, a concept of spatial topological divergence is proposed, and the topological structure with higher topological divergence is selected. The spatial topological divergence calculation is the ratio of the volume of the convex polyhedron with the signal transmitters as vertices to the surface area thereof, which reflects the size of the flux per unit area of the topological structure. The topological structure with larger flux can better locate the observation target, avoiding the extreme case of small angle and long distance; at the same time, the three-dimensional characteristics of the spatial networking are considered, which is beneficial to the positioning of the human target in the three-dimensional space. Continue to traverse other nodes in the ad hoc network, compare the total power of all ad hoc networks, and select the ad hoc network with the highest total power as the final ad hoc network (i.e. the target dynamic network in the above embodiment).

[0135] Optionally, as shown in Figure 6 the step S21 can include the following steps:

[0136] Step S2101: acquiring information of all router nodes; optionally, the router nodes can be the signal transmitters in the above embodiment;

[0137] Step S2102: randomly selecting one router node as a signal receiver;

[0138] Step S2103: determining the strength RSSI of the signal receiver receiving other router nodes; optionally, the other router nodes can be the router nodes other than the signal receiver in all router nodes, and the other router nodes can be the signal transmitters in the above embodiment;

[0139] Step S2104: ranking the RSSI of each signal transmitter according to the strength, and screening out a predetermined number of signal transmitters;

[0140] Step S2105: determining whether there are multiple topology schemes; if yes, performing step S2106, and if no, performing step S2108;

[0141] Step S2106: calculating the topology divergence;

[0142] Step S2107: selecting the topology scheme with the largest topology divergence;

[0143] Step S2108: determining the ad hoc network node, and calculating the total power of the ad hoc network node; optionally, the ad hoc network node can be the first alternative dynamic network in the above embodiment;

[0144] Step S2109: selecting other router nodes as signal receivers;

[0145] Step S2110: determining whether the number of ad hoc networks is greater than or equal to K times; if yes, performing step S211, and if no, performing step S2103;

[0146] Step S2111: comparing and selecting the ad hoc network node with the largest total power in the ad hoc network node as the Wi-Fi ad hoc network; optionally, the Wi-Fi ad hoc network can be the target dynamic network in the above embodiment;

[0147] Step S2112: outputting the Wi-Fi ad hoc network.

[0148] An important feature of the dynamic network (Wi-Fi dynamic network or target dynamic network, etc.) is to consider the three-dimensional spatial layout characteristics of the Wi-Fi signal transmitter to the greatest extent while maintaining the effective information of the signal, to accommodate the ceiling, wall, and ground layout of the Wi-Fi signal transmitter through the topology structure, to optimize the network based on the spatial topology divergence, to better predict the position of the human target in the three-dimensional space, and to increase the robustness of positioning and reduce the positioning deviation caused by the Wi-Fi router with signal noise through the certain structure and quantity redundancy of the devices in the dynamic network.

[0149] Step S22: training the positioning network model and predicting the position information of the target object. Optionally, the positioning network model can be the first model in the above embodiment.

[0150] The positioning network model is a neural network model used to realize the positioning of the human body in indoor three-dimensional space. The model is divided into two stages: training stage and prediction stage. The training stage is to establish a fingerprint library mapping different network CSI signals and human body positions (such as the position of the target object in the above embodiment), and construct a pattern library of Wi-Fi network CSI signals corresponding to the human body in different physical space positions. The prediction stage is to identify the characteristics of the network CSI signal, and then match it with the fingerprint library established in the training stage to predict the spatial position information of the human body. The positioning network models in the training stage and the prediction stage are both graph structure networks based on the self-attention mechanism. The main difference is that in the training stage, the physical position information of the human body obtained by the laser sensor is used as a supervisory signal to train the network model, and the model weights are continuously updated until the loss converges; in the prediction stage, the weights trained in the training stage are used for data inference to directly predict the physical space position of the human body.

[0151] The input of the positioning network model utilizes the time domain waveform information of CSI, the CSI frequency domain waveform information obtained by Fourier transforming the time domain waveform, and the identification information in the Wi-Fi ad hoc network. In order to retain the spatial information of the source of the CSI signal, a CSI data structure with spatial information is adopted. The implementation method is to allocate storage space for the CSI corresponding to the identification information of all Wi-Fi ad hoc networks each time input, and at the same time map the identification information and corresponding CSI data in the ad hoc network to the corresponding storage space. The space of other empty nodes is filled with a tensor of 0, and CSI data with spatial information is obtained at this time. The principle of associating CSI time domain information and frequency domain information with the target position is based on the principles of time domain analysis and frequency domain analysis. The time domain analysis principle is a geometric positioning method based on the CSI signal. The target positioning is achieved by analyzing CSI signals of different modes to estimate the distance, yaw angle and pitch angle of the target; such as Figure 7 As shown, θ n and θm is the pitch angle, and The principle of frequency domain analysis is to estimate the position of human targets based on the frequency and phase changes (Doppler shift phenomenon) generated when the CSI signal encounters human targets in different positions and motion states, such as Figure 8 As shown, the frequency of CSI changes after encountering the human body.

[0152] Optionally, step S22 may include the following steps:

[0153] Step S2201: Obtain the identification information of the i-th Wi-Fi dynamic network within a period of time Δt and the corresponding CSI time domain waveform data tensor χ i .

[0154] Step S2202: Fast Fourier transform is performed on the CSI time domain data to obtain the CSI frequency domain waveform data of the i-th Wi-Fi dynamic networking: phase i .

[0155] Step S2203: According to the Wi-Fi topology planar diagram, the CSI time domain data and the frequency domain data of the nodes in the networking are assigned to the data of the input model network to obtain the CSI data with spatial information.

[0156] Step S2204: In the training stage, a position sampling method is used, such as Figure 2 The black points in the middle grid represent the sampling positions, that is, a random sampling (equal opportunity) or a sampling method with a certain probability distribution is used to plan the relative three-dimensional spatial positions x i ', y i ', z i of the human body, and then the absolute position Loc(x i ,y i ,z i ) of the target object (i.e. the human body target) reaching the i-th sampling point is known by using a laser sensor, as shown in Figure 9 . In addition, the coordinate origin can be determined before coordinate conversion. In the training stage, when the target object is at different sampling position points in the three-dimensional space, steps S2201, S2202 and S2203 are repeated to obtain multiple sets of CSI data with spatial information.

[0157] Step S2205: In the training stage, the multiple sets of time domain and frequency domain CSI data with spatial information obtained in step S2204 are divided into small batches (to realize parallel calculation in different graphics cards) and input into a self-attention-based network with randomly initialized weights, and the network outputs the three-dimensional position prediction value F W (e1χ i ,e2φ i ) of the target. The specific positioning network model proposed in the embodiment of the application is a self-attention-based graph structure network, and the core unit module is the network module in the green box in the graph, that is, the CSI data is respectively subjected to three convolution layers, the matrixes output by two of the convolution layers are multiplied to obtain a correlation matrix of the CSI data features, and after Softmax processing, the correlation matrix is used as a feature weight and multiplied with a CSI feature matrix extracted by convolution to obtain a feature matrix with learning weights. In order to facilitate the convergence of the model training, a residual connection is added. Each core unit module can obtain a CSI feature mode, and each layer of network contains M unit modules, so M kinds of CSI features can be obtained. The entire model network adopts N layers of network, and the number of layers N can be set according to the hardware resource conditions. The network can extract multiple features of sequence data, increase the richness of the features, and well preserve the correlation between the sequence data.

[0158] Step S2206: During the training phase, calculate the three-dimensional position prediction value F of the target predicted by the model in step S2205 W (e1χ i ,e2φ i ) and the real physical location Loc(x i ,y i ,z i ) in the metric space Ω W , D W The representation is as follows: or Here, e1 and e2 can be the residuals of the positioning network model or the parameters used for training the positioning network model; W refers to the weight of the positioning network model. The goal of network optimization is to make the network predicted position as close as possible to the actual position, and the distance in the measurement space can be measured using Euclidean distance or cosine similarity. The above distance D W The formula uses the Euclidean distance for measurement, and the distance D in the metric space is W As the model's loss function, during the training phase, batches of data corresponding to the position sampling points are fed into the model for training, so that the model's loss function converges. The model with the minimum loss function value is the trained model. Here, the convergence of the loss function refers to satisfying the convergence condition in the above embodiment.

[0159] Step S2207: In the prediction phase, steps S2201, S2202, and S2203 are used to obtain a CSI signal with spatial information within the Wi-Fi ad hoc network for a human body in the three-dimensional space. The prediction model uses the fixed-weight model trained and converged in step S2206. The fixed-weight model infers the SCI data within the Wi-Fi ad hoc network and predicts the planar area location and floor number of the target.

[0160] Step S23: Perform free supplementation based on the Bayesian error compensation model.

[0161] Based on the CSI signal, the target positioning is obtained through the neural network model in the time domain and the frequency domain, but the electromagnetic wave signal is affected by the change of the external environment (environment temperature and humidity, signal attenuation of building wall, and electromagnetic signal interference), so that the CSI signal has noise. In addition, due to the existence of multipath effect and the limitation of learning and generalization ability of the neural network model, there is a certain deviation between the position prediction value of the neural network based on the CSI signal and the real position of the target; in order to reduce the positioning error and more accurately estimate the position, the embodiment of the present application proposes a Bayesian compensation model, which corrects or compensates the prediction of the neural network based on the motion model estimator of the Markov process. The compensation process can also be regarded as weighted fusion of the motion model estimator and the neural network prediction.

[0162] The Bayesian compensation model proposed in the embodiment of the present application is based on the state variable of Gaussian distribution, as shown in the following formula (1) : Figure 10 The target position predicted by the neural network model at the historical time (for example, at the time t-1 or the last time) is combined with the motion trajectory model of the target to update the fused target position information (which can be the predicted position information in the above embodiment) at the current time (for example, at the time t or the next time); the prediction value of the network model at the current time is corrected, and the iteration is performed in the Markov process (the state at the current time is only determined by the state at the last time), so that the prediction error of the positioning network model and the estimation error of the motion model are reduced. Alternatively, the Bayesian compensation model is as shown in the following formula (2) : Figure 11

[0163] Firstly, the target motion state at the current time is estimated according to the target state at the historical time. The target is modeled and analyzed as a dynamic system, a spatial rectangular coordinate reference system (the third vertical coordinate is the floor height) with the floor of the building as the coordinate plane is established, and the speed of the target can be regarded as a constant in a very short time δt, so that the dynamic model is established as follows: t Among them, and are the position coordinate components of the human target (which can be the target object in the above embodiment) on the X, Y and Z axes at the time t; and are the speed components of the human target on the X, Y and Z axes at the time t; △t is a predetermined time interval, which corresponds to the predetermined time interval of the Wi-Fi ad hoc network; [] represents the downward value.

[0164] Considering the influence of the external environment, such as the old man being pushed or trying to speed up the pace and falling down by losing the center of gravity, an external control variable is added, and the dynamic equation is as follows: ​​

[0165] Write in matrix form, as follows:

[0166]

[0167] Convert to vector form:

[0168] In the formula, Indicates the motion state of the human target at time t, including the estimated three position components and three velocity components of the human target in the X, Y and Z axes at time t, Indicates the estimated motion state of the human target at time t-1, Is the state transition matrix, and B is the control matrix, Is the control vector, Including two acceleration components of the human target in the X and Y axes at time t, and the acceleration component in the Z axis is 0, which is ignored.

[0169] Due to the influence of possible resistance and other environmental factors, there is a deviation between the estimated position of the target by the motion model and the true state of the target, and the uncertainty variance of the external environmental factors is Q t According to the randomness of the variable, the error variable obeys the Gaussian distribution, and the complete description of the estimated target state sequence by the motion model is:

[0170]

[0171]

[0172]

[0173] Among them, Is the motion model state estimate of the target at time t, Is the covariance matrix of the motion model state estimate of the target. Optionally, Q can be the modified value in the above embodiment.

[0174] In the fusion process, it is necessary to unify the scale space, map the motion model estimate to the neural network model prediction space, and the Gaussian distribution of the state variable is as follows:

[0175]

[0176]

[0177]

[0178] The probability density function of the state variable is:

[0179]

[0180]

[0181] wherein, is the state variable estimated by the neural network model at time t, is the covariance matrix of the neural network model in estimating the state variable x t .

[0182] The intersection of the state variable x t estimated by the neural network model and the state variable estimated by the motion model is taken, and the distribution of the fused state variable x :

[0183]

[0184]

[0185]

[0186] It can be concluded that:

[0187]

[0188]

[0189] wherein, x t is the state variable estimated by the target at time t, is the mean of the fused state variable of the neural network prediction and the motion model state variable at time t, is the covariance of the fused variable, F W (e1χ i ,e2φ i ) is the state variable estimated by the neural network model at time t, is the covariance matrix of the neural network model in estimating the state variable x t , is the state variable estimated by the motion model of the target at time t in the neural network model space, is the covariance matrix of the motion model state estimation variable, H t is the scale conversion matrix at time t.

[0190] In the Markov process, the fused mean is taken as the estimated value of the state x t at the current time t, the next time model prediction value is predicted based on the motion model, and the prediction value of the neural network model based on the CSI signal at the next time t+1 is combined for fusion, so as to perform state estimation iteration and accurately estimate the state of the target.

[0191] The embodiment of the present application effectively corrects the human target position (for example, the predicted position information of the target object) predicted by the neural network model (for example, the positioning network model) based on the Bayesian principle, and reduces the motion model estimation error and the error during the neural network model estimation.

[0192] It should be noted that the method provided by the embodiment of the present application can be executed alone or together with some methods in the embodiment of the present application or some methods in related technologies.

[0193] It should be noted that the following description of the positioning device is similar to the description of the positioning method, and the beneficial effects of the method are not described in detail. For technical details not disclosed in the embodiment of the positioning device of the present application, please refer to the description of the embodiment of the positioning method of the present application.

[0194] As shown in Figure 12 The embodiment of the present application provides a positioning device, which comprises:

[0195] The acquisition module 41 is configured to acquire a CSI time domain data set of the target dynamic networking, wherein one CSI time domain data set comprises at least one CSI time domain data; and one CSI time domain data is time domain information of a CSI signal sent by one signal transmitter to a signal receiver in the target dynamic networking.

[0196] The first processing module 42 is configured to perform fast Fourier transform on the CSI time domain data set to obtain a CSI frequency domain data set of the target dynamic networking, wherein one CSI frequency domain data set comprises at least one CSI frequency domain data; and one CSI frequency domain data is frequency domain information of a CSI signal sent by one signal transmitter to a signal receiver in the target dynamic networking.

[0197] The second processing module 43 is configured to input the CSI time domain data set and the CSI frequency domain data set into the trained first model to obtain the predicted position information of the target object.

[0198] The positioning device provided by the embodiment of the present application can be applied in a terminal.

[0199] In some embodiments, the first processing module 42 is configured to assign corresponding spatial information to the CSI time domain data set and the CSI frequency domain data set based on a topological flat map to obtain CSI data with spatial information.

[0200] The second processing module 43 is configured to input the CSI time domain data set and the CSI frequency domain data set into the trained first model to obtain the predicted position information of the target object, comprising: inputting the CSI data with spatial information into the trained first model to obtain the predicted position information of the target object.

[0201] In some embodiments, the obtaining module 41 is configured to obtain a CSI sample dataset of the sample dynamic networking, where the CSI sample dataset comprises at least one CSI sample time domain data and CSI sample frequency domain data corresponding to the at least one CSI sample time domain data, or the CSI sample dataset comprises a CSI sample dataset with spatial information based on the CSI sample time domain dataset and the CSI sample time domain dataset with spatial information.

[0202] The obtaining module 41 is further configured to obtain actual position information of the sample object in the sample dynamic networking.

[0203] The second processing module 43 is configured to input the CSI sample dataset and the actual position information into the first model for training until a convergence condition is met, to obtain the trained first model.

[0204] In some embodiments, the convergence condition is met when a difference between the predicted position information and the actual position information corresponding to the sample is less than or equal to a threshold value, and the predicted position information is predicted by inputting the CSI sample dataset into the first model.

[0205] In some embodiments, the obtaining module 41 is configured to determine total power of the first K candidate dynamic networkings in descending order of total power, where the total power of any first candidate dynamic networking is a sum of power from each signal transmitter to signal receiver in the first candidate dynamic networking, K is a positive integer less than or equal to L, L is the number of the plurality of signal devices, and the first candidate dynamic networking corresponding to the maximum total power is selected as the target dynamic networking from the total power of the first K candidate dynamic networkings.

[0206] In some embodiments, the obtaining module 41 is configured to obtain the first L candidate dynamic networkings corresponding to the first L signal devices respectively, where L is a positive integer less than or equal to the number of the plurality of signal devices used for networking, and the total power of the first K candidate dynamic networkings is determined when the number of the obtained first candidate dynamic networkings is greater than or equal to K, where the total power of the Kth candidate dynamic networking is a sum of power from each signal transmitter to signal receiver in the Kth candidate dynamic networking, K is a positive integer less than or equal to the number of the plurality of signal devices, and the first candidate dynamic networking corresponding to the maximum total power is selected as the target dynamic networking from the total power of the first K candidate dynamic networkings.

[0207] In some embodiments, the obtaining module 41 is configured to obtain any first candidate dynamic networking. Optionally, obtaining any first candidate dynamic networking comprises obtaining the first L candidate dynamic networkings.

[0208] In some embodiments, the acquisition module 41 is used to determine the lth signal transmitter from multiple signal transmitters as a signal receiver and determine other signal transmitters other than the Lth signal transmitter among the multiple signal transmitters as signal transmitters; sort the signal strengths of each signal transmitter in the multiple signal transmitters to the lth signal transmitter, and select a predetermined number of signal transmitters before the signal strength sorting; determine the lth signal transmitter and the predetermined number of signal transmitters as the second alternative dynamic network; if there are multiple second alternative dynamic networks, determine the topological divergence of each second alternative dynamic network, wherein the topological divergence is the ratio of the volume to the surface area of ​​the convex polyhedron of each signal transmitter in the second alternative dynamic network; from the topological divergences of multiple second alternative dynamic networks, select the second alternative dynamic network corresponding to the maximum topological divergence as the lth first alternative dynamic network corresponding to the lth signal transmitter; wherein l is a positive integer less than or equal to L.

[0209] In some embodiments, the second processing module 42 is configured to modify the predicted position information based on historical predicted position information and a motion model.

[0210] In some embodiments, the predicted position information includes the predicted position information at the previous moment; the second processing module 42 is used to input the predicted position information at the previous moment into the motion model to obtain the predicted position information at a moment; or, the second processing module 42 is used to obtain the predicted position information at the next moment based on the target position and correction value at the previous moment, wherein the correction value is used to represent the influence of external environmental factors on the position of the target object.

[0211] like Figure 13 As shown, an embodiment of the present invention further provides a terminal, which includes a processor 51 and a memory 52 for storing a computer program that can be run on the processor 52; wherein, when the processor 51 is used to run the computer program, it implements the positioning method of any embodiment of the present invention.

[0212] In some embodiments, the memory in the embodiments of the present application can be volatile memory or nonvolatile memory, or can include both volatile and nonvolatile memory. Among them, the nonvolatile memory can be Read-Only Memory (ROM), Programmable ROM (PROM), Erasable PROM (EPROM), Electrically EPROM (EEPROM) or flash memory. The volatile memory can be Random Access Memory (RAM) used as an external cache. By way of example, and not limitation, many forms of RAM can be used, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM) and Direct Rambus RAM (DRRAM). The memory of the system and method described herein is intended to include, but not be limited to, these and any other suitable types of memory.

[0213] The processor can be a chip that has a signal processing capability. In implementation process, each step of the above method can be completed by integrated logic circuit of hardware in the processor or instruction in the form of software. The processor mentioned above can be a general processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. Each method, step and logic block disclosed in the embodiment of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can be any conventional processor. The steps of the method disclosed in combination with the embodiment of the present application can be directly embodied as hardware code processor for execution, or executed by combination of hardware and software modules in the code processor. The software module can be located in a random access memory, a flash memory, a read only memory, a programmable read only memory or an electrically erasable programmable memory, a register or other mature storage medium in the art. The storage medium is located in the storage, and the processor reads the information in the storage and combines the hardware to complete the steps of the above method.

[0214] In some embodiments, the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or any combination thereof. For a hardware implementation, the processing units can be implemented within one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSP Devices), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, micro-controllers, microprocessors, other electronic units designed to perform the functions described herein, or a combination thereof.

[0215] For a software implementation, the techniques described herein can be implemented with modules (e.g., procedures, functions, and so on) that perform the functions described herein. The software codes can be stored in memory and executed by processors. The memory can be implemented within the processor or external to the processor.

[0216] The embodiment of the present application provides a computer storage medium, the computer readable storage medium stores an executable program, and the executable program is executed by a processor to realize the steps of the positioning method of any embodiment of the present application.

[0217] The embodiment of the present application provides a computer program product, the computer program product comprises a computer program or instruction, and the computer program or instruction is executed by a processor to realize the steps of the positioning method of any embodiment of the present application.

[0218] In some embodiments, the computer storage medium can comprise a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes.

[0219] It should be noted that the technical solutions disclosed in the embodiments of the present application can be combined arbitrarily without conflict.

[0220] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited to this, 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 in 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 positioning method, characterized by, The method comprises: obtaining a channel state information (CSI) time domain data set of a target dynamic networking, wherein one CSI time domain data set comprises at least one CSI time domain data, and one CSI time domain data is time domain information of a CSI signal sent by one signal transmitter in the target dynamic networking to a signal receiver; performing fast Fourier transform on the CSI time domain data set to obtain a CSI frequency domain data set of the target dynamic networking, wherein one CSI frequency domain data set comprises at least one CSI frequency domain data, and one CSI frequency domain data is frequency domain information of the CSI signal sent by one signal transmitter in the target dynamic networking to the signal receiver; inputting the CSI time domain data set and the CSI frequency domain data set into a trained first model to obtain predicted position information of a target object.

2. The method of claim 1, wherein: the method further comprises: based on a topology planar graph, assigning corresponding spatial information to the CSI time domain data set and the CSI frequency domain data set to obtain CSI data with spatial information; the inputting the CSI time domain data set and the CSI frequency domain data set into the trained first model to obtain the predicted position information of the target object comprises: inputting the CSI data with spatial information into the trained first model to obtain the predicted position information of the target object.

3. The method according to claim 1 or 2, characterized in that, The method further comprises: obtaining a CSI sample data set of a sample dynamic networking, wherein the CSI sample data set comprises at least one CSI sample time domain data and CSI sample frequency domain data corresponding to the at least one CSI sample time domain data, or the CSI sample data set comprises CSI sample data with spatial information obtained by assigning spatial information to the CSI sample time domain data set and the CSI sample time domain set; obtaining actual position information of a sample object in the sample dynamic networking; inputting the CSI sample data set and the actual position information into the first model for training until a convergence condition is met to obtain the trained first model.

4. The method of claim 3, wherein, The convergence condition is that a difference between the predicted position information corresponding to the sample and the actual position information is less than or equal to a threshold value, and the predicted position information is predicted by inputting the CSI sample data set into the first model.

5. The method according to claim 1 or 2, characterized in that, The obtaining of the CSI time domain data set of the target dynamic networking comprises: determining total power of a first K candidate dynamic networking in descending order of total power, wherein the total power of any first candidate dynamic networking is a sum of power of each signal transmitter to the signal receiver in the first candidate dynamic networking; K is a positive integer less than or equal to L; and L is a number of the plurality of signal transmitters; selecting the first candidate dynamic networking corresponding to the maximum total power from the total power of the first K candidate dynamic networking as the target dynamic networking.

6. The method of claim 5, wherein, The method further comprises: Determining a first signaling device from a plurality of signaling devices as the signal receiver and determining other signaling devices other than the first signaling device from the plurality of signaling devices as the signal transmitters; According to the signal strength ranking of each of the signal transmitters in the plurality of signal transmitters up to the first signal transmitter, a predetermined number of the signal transmitters before the ranking of signal strength are selected; Determining the first signaler and a predetermined number of signal transmitters as a second candidate dynamic network; If there are multiple second candidate dynamic networks, determine the topological divergence of each second candidate dynamic network, where the topological divergence is a ratio of the volume to the surface area of ​​the convex polyhedron of each signal transmitter in the second candidate dynamic network; From the topological divergences of multiple second candidate dynamic networks, select the second candidate dynamic network corresponding to the largest topological divergence as the lth first candidate dynamic network corresponding to the lth signaler; wherein l is a positive integer less than or equal to L.

7. The method according to claim 1 or 2, characterized in that, The method further comprises: Based on the historical predicted position information and the motion model, the predicted position information is modified.

8. A positioning device, characterized in that include: An acquisition module is configured to acquire a channel state information (CSI) time domain data set of a target dynamic network, wherein one CSI time domain data set includes at least one CSI time domain data; one CSI time domain data is time domain information of a CSI signal sent by a signal transmitter to a signal receiver in the target dynamic network; a first processing module, configured to perform a fast Fourier transform on the CSI time domain data set to obtain a CSI frequency domain data set of the target dynamic network, wherein one CSI frequency domain data set includes at least one CSI frequency domain data; one CSI frequency domain data is frequency domain information of the CSI signal sent by one of the signal transmitters to the signal receiver in the target dynamic network; The second processing module is configured to input the CSI time domain dataset and the CSI frequency domain dataset into the trained first model to obtain predicted position information of the target object.

9. A terminal, characterized by comprising: The terminal includes a processor and a memory for storing a computer program that can be run on the processor; wherein, when the processor is used to run the computer program, it implements the positioning method according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The computer storage medium contains computer executable instructions, wherein the computer executable instructions are executed by a processor to implement the positioning method according to any one of claims 1 to 7.

11. A computer program product comprising computer programs or instructions, characterized in that, When the computer program or instruction is executed by a processor, the positioning method according to any one of claims 1 to 7 is implemented.

Images