Pedestrian positioning method and device

Abstract

The present invention discloses a pedestrian positioning method and device, the pedestrian positioning method comprising: obtaining pedestrian historical trajectory data and environmental data; predicting the next position of the pedestrian according to the pedestrian historical trajectory data and environmental data; selecting a beacon near the next position according to the predicted next position of the pedestrian based on a pre-trained reinforcement learning model; and positioning the pedestrian according to the beacon near the selected next position.

Description

Technical Field

[0001] The present invention relates to the field of positioning technology, and in particular to a pedestrian positioning method and device. Background Art

[0002] Ultra-wideband (UWB) technology is a high-precision positioning technology that is widely used in indoor positioning systems. UWB technology determines positioning by measuring the time difference between the positioning terminal and multiple beacons. This process relies on effective communication between the beacons and the terminal. If the approximate location of the pedestrian can be predicted in advance, the optimal beacon can be selected in advance for communication, thereby improving the accuracy and response speed of positioning.

[0003] Existing UWB positioning technology usually relies on the results of the last positioning to select beacons. This method has limitations, especially in cases where pedestrians move frequently or the environment is complex, which can easily lead to delayed or inaccurate positioning. In addition, although existing technologies have used actual geographic information (such as building layout and impassable areas) to assist positioning, these methods are usually static and lack the ability to respond to dynamic changes in pedestrian behavior in real time.

[0004] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of the present invention is to provide a pedestrian positioning method and device, aiming to solve the above-mentioned technical problems in the prior art.

[0006] To achieve the above object, the present invention provides a pedestrian positioning method, the pedestrian positioning method comprising:

[0007] Obtain pedestrian historical trajectory data and environmental data;

[0008] Predicting the next location of the pedestrian based on the pedestrian's historical trajectory data and environmental data;

[0009] According to the predicted next position of the pedestrian, a beacon near the next position is selected based on a pre-trained reinforcement learning model;

[0010] Pedestrians are located based on the beacons near the selected next location.

[0011] Preferably, in the pedestrian positioning method, predicting the next position of the pedestrian based on the pedestrian historical trajectory data and environmental data includes:

[0012] The key points in the environmental data are used as nodes in the graph, and the passability between nodes is used as edges to construct a graph structure, where the characteristics of the nodes include the environmental information of the location of the nodes, and the characteristics of the edges include the length of the path and the difficulty of passing;

[0013] Through the pre-built graph convolutional neural network, the information of the node and its neighboring nodes is aggregated to obtain the feature representation of the node. The aggregation formula is as follows:

[0014] ;

[0015] in, is the feature representation of node v at layer l;

[0016] W （l） is the weight matrix of the lth layer;

[0017] N(v) is the neighbor set of node v;

[0018] du and dv represent the degrees of nodes u and v, respectively;

[0019] σ is the activation function;

[0020] Based on the historical trajectory data and the corresponding environmental data, the next location of the pedestrian is predicted. The prediction model is as follows:

[0021] ;

[0022] in, is the predicted value of the next position;

[0023] H() is the prediction model;

[0024] is the feature representation corresponding to the historical trajectory data, that is, the feature representation of node v in the last layer;

[0025] is the feature representation of the environmental data corresponding to the historical trajectory data, that is, the feature representation of the neighboring nodes of node v;

[0026] is the feature of the edge connected to node v.

[0027] Preferably, in the pedestrian positioning method, before the step of predicting the next position of the pedestrian based on the historical trajectory data and the corresponding environmental data, the method further includes:

[0028] Obtain pedestrian gait data, behavior data and current environment data;

[0029] Extracting features of the gait data, behavior data, and current environment data respectively;

[0030] The extracted gait data, behavior data, and features of the current environment data are used as inputs to a trained long short-term memory network model, and a prediction of the pedestrian's next behavior intention is output;

[0031] Accordingly, predicting the next position of the pedestrian based on the historical trajectory data and the corresponding environmental data includes:

[0032] Predict the next location of the pedestrian based on historical trajectory data, corresponding environmental data, and the pedestrian's next behavioral intention prediction.

[0033] Preferably, in the pedestrian positioning method, respectively extracting features of the gait data, behavior data, and current environment data includes:

[0034] Using a convolutional network to extract features from the gait data and behavior data;

[0035] A fully connected layer is used to extract features of the current environment data.

[0036] Preferably, in the pedestrian positioning method, after taking the extracted features of the gait data, behavior data, and current environment data as inputs into a trained long short-term memory network model and outputting a prediction of the pedestrian's next behavior intention, the method further comprises:

[0037] Pedestrian trajectories are predicted based on historical trajectory data, corresponding environmental data, and the pedestrian's next behavioral intention prediction.

[0038] Preferably, in the pedestrian positioning method, before the step of selecting a beacon near the predicted next position of the pedestrian based on a pre-trained reinforcement learning model, the method further includes:

[0039] Define states t , action space, where state s t Including the current position p t , the distribution and signal strength of surrounding beacons B t , and environmental dynamic characteristics E t , the action space is A={B1, B2, ..., B n}, B i For the i-th beacon, action a t is the beacon selection made by the system at time t;

[0040] Set the reward function, the formula is:

[0041] ; Among them, r tis the reward, the error represents the deviation between the positioning result and the actual position; the delay represents the system response time, and α and β are the weights for adjusting the reward;

[0042] Build a reinforcement learning model through strategy Select an action and update the strategy based on the reward. The update formula is as follows:

[0043] ;

[0044] η is the learning rate and γ is the discount factor.

[0045] Preferably, in the pedestrian positioning method, the pedestrian positioning method is applied to a server, the server aggregates reinforcement learning model parameters sent by multiple distributed terminal devices, and the reinforcement learning model is established on each distributed terminal device; the method further includes:

[0046] The server aggregates the reinforcement learning model parameters sent by each of the distributed terminal devices to obtain an overall model, wherein the calculation formula of the parameters of the overall model is as follows:

[0047] ;

[0048] in, are the parameters of the overall model;

[0049] w k are the parameters of the reinforcement learning model of the kth distributed terminal device;

[0050] D k is the data set of the kth distributed terminal device;

[0051] After the overall model is completed, the new overall model is sent to each of the distributed terminal devices, so that the distributed terminal devices use the overall model to update their own reinforcement learning model parameters;

[0052] Each of the distributed terminal devices minimizes the loss function on its data, and the loss function is:

[0053] ;

[0054] The data set on each of the distributed terminal devices is , the objective function is:

[0055] ;

[0056] Each distributed terminal device will have model parameters

[0057] is the loss function of distributed terminal k;

[0058] w is the parameter of the shared model;

[0059] x i For input data;

[0060] y i is the corresponding target output;

[0061] is the prediction function.

[0062] In order to achieve the above object, the present invention also provides a pedestrian positioning device, the pedestrian positioning device comprising:

[0063] An acquisition unit, used to acquire pedestrian historical trajectory data and environmental data;

[0064] A prediction unit, used to predict the next position of the pedestrian based on the pedestrian's historical trajectory data and environmental data;

[0065] A selection unit, configured to select a beacon near the predicted next position of the pedestrian based on a pre-trained reinforcement learning model;

[0066] The positioning unit is used to locate the pedestrian based on the beacons near the selected next position.

[0067] In order to achieve the above object, the present invention further provides a computer device, comprising:

[0068] at least one processor; and,

[0069] a memory communicatively connected to the at least one processor; wherein,

[0070] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the above-mentioned pedestrian positioning method.

[0071] In order to achieve the above object, the present invention also provides a computer-readable storage medium storing a computer program, and the computer program implements the above pedestrian positioning method when executed by a processor.

[0072] The present invention has at least the following beneficial effects:

[0073] The pedestrian positioning method provided by the present invention obtains pedestrian historical trajectory data and environmental data; predicts the next position of the pedestrian according to the pedestrian historical trajectory data and environmental data; selects beacons near the next position based on the predicted next position of the pedestrian based on a pre-trained reinforcement learning model; and locates the pedestrian according to the beacons near the selected next position, so as to improve the accuracy and response speed in complex environments and dynamic changes.

[0074] Furthermore, the present invention can filter out beacons near the next location by predicting the next location first, and actively communicate with these beacons to obtain better ranging, thereby achieving more accurate positioning;

[0075] Furthermore, assuming that the next location is unknown, one can only scan blindly in the radio environment, and the radio environment is relatively complex, which will waste a lot of time to screen beacons and measure the distance to the beacons based on the signal strength relationship. For example, in the radio scan, a beacon that may be far away can communicate, and it will take a lot of time to communicate with this beacon, but this distance measurement is meaningless. Since the distance measurement of nearby beacons is usually required, it is more accurate. The present invention predicts the next location first and gives an approximate range. Then, according to the next location, it can assist in determining which beacons to actively communicate with, so as to obtain a distance measurement that is more conducive to accurately locating the location of pedestrians. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A schematic diagram of a first implementation of a pedestrian positioning method provided by the present invention;

[0077] Figure 2 A schematic diagram of a second implementation of the pedestrian positioning method provided by the present invention;

[0078] Figure 3 A schematic diagram of a third implementation of the pedestrian positioning method provided by the present invention;

[0079] Figure 4 A schematic diagram of an embodiment of a pedestrian positioning device provided by the present invention;

[0080] Figure 5 A schematic diagram of a computer device according to the present invention.

[0081] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0082] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0083] In the embodiments of the present invention, the term "and / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent three situations: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0084] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0085] In the embodiments of the present invention, the term "plurality" refers to two or more than two, and other quantifiers are similar.

[0086] In the present invention, unless otherwise specified, the directional words used, such as "up, down, top, bottom", usually refer to the directions shown in the drawings, or to the components themselves in the vertical, perpendicular or gravity directions; similarly, for ease of understanding and description, "inside and outside" refer to the inside and outside relative to the outline of each component itself, but the above-mentioned directional words are not used to limit the present invention.

[0087] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. However, it can be understood by those skilled in the art that in the embodiments of the present invention, many technical details are proposed in order to enable the reader to better understand the present invention. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present invention can also be implemented. The division of the following embodiments is for the convenience of description and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined and referenced with each other under the premise of no contradiction.

[0088] In order to solve the above problems, the present embodiment relates to a pedestrian positioning method, which can be applied to a computer device. The computer device can be a desktop computer, a tablet computer, a notebook, and other electronic devices with data processing capabilities. In other embodiments, it can also be other electronic devices with data processing capabilities, and no specific restrictions are made here.

[0089] The implementation details of the pedestrian positioning method according to the first embodiment of the present invention are described below. The following content is only the implementation details provided for the convenience of understanding and is not necessary for the implementation of this solution.

[0090] The specific process of this implementation is as follows: Figure 1 As shown, specifically including:

[0091] Step S100, obtaining pedestrian historical trajectory data and environmental data;

[0092] It should be understood that the execution subject of the present invention may be, but is not limited to, a terminal device. In other embodiments, when the terminal may include multiple distributed terminal devices, the execution subject of the present invention is a server, through which the reinforcement learning model parameters sent by the multiple distributed terminal devices are aggregated.

[0093] Environmental data can be data such as building layout, roads, and traffic information collected by GIS systems and sensors.

[0094] Step S200, predicting the next location of the pedestrian based on the pedestrian's historical trajectory data and environmental data;

[0095] It should be understood that step S200 includes the following steps:

[0096] The key points in the environmental data are used as nodes in the graph, and the passability between nodes is used as edges to construct a graph structure, where the characteristics of the nodes include the environmental information of the location of the nodes, and the characteristics of the edges include the length of the path and the difficulty of passing;

[0097] It should be understood that a node may be a building, a road section where pedestrians may pass, or a key location. An edge may be the passability between nodes, and may be defined based on features such as road width, distance, and relative position between buildings.

[0098] It can be achieved by obtaining building layout and road information through GIS (Geographic Information System) and sensor data to build a graph structure. For each node v and its adjacent node u, use edge e vu Represents the transit relationship between nodes, and the edge weight w vu Calculations can be made based on distance, obstacles, travel time, etc.

[0099] Through the pre-built graph convolutional neural network, the information of the node and its neighboring nodes is aggregated to obtain the feature representation of the node. The aggregation formula is as follows:

[0100] ;

[0101] in, is the feature representation of node v at layer l;

[0102] W （l） is the weight matrix of the lth layer;

[0103] N(v) is the neighbor set of node v;

[0104] du and dv represent the degrees of nodes u and v, respectively;

[0105] σ is the activation function, and a nonlinear function is usually selected to introduce nonlinear characteristics and enhance the expressiveness of the model.

[0106] It should be understood that the characteristics of a node include environmental information of the location, such as crowd density, scene type (e.g., commercial area, residential area), etc. The characteristics of an edge include the length of a path, difficulty of passing, etc.

[0107] Based on the historical trajectory data and the corresponding environmental data, the next location of the pedestrian is predicted. The prediction model is as follows:

[0108] ;

[0109] in, is the predicted value of the next position;

[0110] H() is a prediction model; in this embodiment, it can be but is not limited to a graph neural network model.

[0111] is the feature representation corresponding to the historical trajectory data, that is, the feature representation of node v in the last layer;

[0112] is the feature representation of the environmental data corresponding to the historical trajectory data, that is, the feature representation of the neighboring nodes of node v;

[0113] is the feature of the edge connected to node v.

[0114] It should be noted that the prediction of the next location can be made using a graph neural network model. By modeling building layouts, road information, etc., and using graph convolutional neural networks to capture pedestrian movement patterns in complex environments. By extracting features from nodes and edges, the GNN model can accurately predict the future location of pedestrians in the current environment.

[0115] Step S300, according to the predicted next position of the pedestrian, based on the pre-trained reinforcement learning model, selecting a beacon near the next position;

[0116] It should be understood that step S300 also includes establishing a reinforcement learning model. By taking pedestrian positions, beacon distribution, etc. as states, a trained reinforcement learning model is used to dynamically select beacons to maximize positioning accuracy and minimize positioning delay. Each beacon selection result will be adjusted based on the positioning error and positioning delay feedback to adjust the subsequent strategy.

[0117] The methods for building reinforcement learning models include:

[0118] Define states t , action space, where state s t Including the current position p t , the distribution and signal strength of surrounding beacons B t , and environmental dynamic characteristics E t , the action space is A={B1, B2, ..., B n}, B i For the i-th beacon, action a t is the beacon selection made by the system at time t;

[0119] Set the reward function, the formula is:

[0120] ; Among them, r t is the reward, the error represents the deviation between the positioning result and the actual position; the delay represents the system response time, and α and β are the weights for adjusting the reward;

[0121] Build a reinforcement learning model through strategy Select an action and update the strategy based on the reward. The update formula is as follows:

[0122] ;

[0123] η is the learning rate and γ is the discount factor.

[0124] It should be noted that when inputting the reinforcement learning model, the current position pt is the next position predicted by the input. This module uses reinforcement learning strategy to optimize the selection of UWB beacons. By defining pedestrian positions, beacon distribution, etc. as states, reinforcement learning algorithms are used to dynamically select beacons to maximize positioning accuracy and minimize positioning delay. Each beacon selection result will adjust the subsequent strategy based on the positioning error and positioning delay feedback.

[0125] Step S400: positioning the pedestrian according to the beacons near the selected next position.

[0126] The pedestrian positioning method provided by the present invention obtains pedestrian historical trajectory data and environmental data; predicts the next position of the pedestrian according to the pedestrian historical trajectory data and environmental data; selects beacons near the next position based on the predicted next position of the pedestrian based on a pre-trained reinforcement learning model; and locates the pedestrian according to the beacons near the selected next position, so as to improve the accuracy and response speed in complex environments and dynamic changes.

[0127] Furthermore, the present invention can filter out beacons near the next location by predicting the next location first, and actively communicate with these beacons to obtain better ranging, thereby achieving more accurate positioning;

[0128] Furthermore, assuming that the next location is unknown, one can only scan blindly in the radio environment, and the radio environment is relatively complex, which will waste a lot of time to screen beacons and measure the distance to the beacons based on the signal strength relationship. For example, in the radio scan, a beacon that may be far away can communicate, and it will take a lot of time to communicate with this beacon, but this distance measurement is meaningless. Since the distance measurement of nearby beacons is usually required, it is more accurate. The present invention predicts the next location first and gives an approximate range. Then, according to the next location, it can assist in determining which beacons to actively communicate with, so as to obtain a distance measurement that is more conducive to accurately locating the location of pedestrians.

[0129] Figure 2 The second implementation of the pedestrian positioning method of the present invention is illustrated. Based on the first implementation, before step S200, the method further includes:

[0130] Step S510, obtaining pedestrian's gait data, behavior data and current environment data;

[0131] It should be understood that the gait data of pedestrians can be collected through cameras, sensors or other wearable devices (such as sports watches), where the gait data includes walking speed, step frequency, etc. Behavior data can include but is not limited to whether to turn, whether to stop, etc. Current environment data includes the type of area, traffic flow, etc. Through gait data, behavior data and current environment data, the short-term behavior intention of pedestrians can be inferred.

[0132] Step S520, extracting features of the gait data, behavior data, and current environment data respectively;

[0133] It should be understood that the feature extraction of gait data and behavior data can be extracted by using a recognition algorithm or obtained by manual annotation.

[0134] Specifically, a convolutional network is used to extract features from the gait data and behavior data; and a fully connected layer is used to extract features from the current environment data.

[0135] Perform feature extraction on gait data to obtain gait features. Perform feature extraction on behavior data to obtain behavior features. Perform feature extraction on environment data to obtain environment features. Environment features include area type and crowd density, etc.

[0136] Step S530, taking the extracted features of the gait data, behavior data, and current environment data as input, inputting the trained long short-term memory network model, and outputting the prediction of the pedestrian's next behavior intention;

[0137] It should be understood that by using LSTM to fuse gait features, behavioral features, and environmental features, the pedestrian's next behavioral intention can be predicted.

[0138] Accordingly, the step S200 includes:

[0139] Step S240, predicting the next location of the pedestrian based on the historical trajectory data, the corresponding environmental data, and the pedestrian's next behavior intention prediction.

[0140] It should be understood that by combining the pedestrian's next behavior intention prediction, historical trajectory data, and corresponding environmental data to further predict the pedestrian's next position, the prediction result is more accurate.

[0141] In other embodiments, the pedestrian trajectory may be predicted based on historical trajectory data, corresponding environmental data, and the pedestrian's next behavior intention prediction. In other embodiments, after step S400, the pedestrian's trajectory may be further predicted based on the selected beacon combined with the pedestrian's next behavior intention prediction.

[0142] like Figure 3 As shown, Figure 3 The third implementation of the pedestrian positioning method of the present invention is illustrated. Based on the first implementation, the pedestrian positioning method is applied to a server, the server aggregates reinforcement learning model parameters sent by multiple distributed terminal devices, and the reinforcement learning model is established on each distributed terminal device; the method also includes:

[0143] Step S610: the server aggregates the reinforcement learning model parameters sent by each of the distributed terminal devices to obtain an overall model, wherein the calculation formula of the parameters of the overall model is as follows:

[0144] ;

[0145] in, are the parameters of the overall model;

[0146] w k are the parameters of the reinforcement learning model of the kth distributed terminal device;

[0147] D k is the data set of the kth distributed terminal device.

[0148] It should be understood that in this way the server can integrate parameters from multiple distributed terminal devices to form a global prediction model without accessing the private data of the devices.

[0149] Step S620, after the overall model is completed, the new overall model is sent to each of the distributed terminal devices, so that the distributed terminal devices use the overall model to update their own reinforcement learning model parameters;

[0150] Each of the distributed terminal devices minimizes the loss function on its data, and the loss function is:

[0151] ;

[0152] The data set on each of the distributed terminal devices is , the objective function is:

[0153] ;

[0154] is the loss function of distributed terminal k;

[0155] w is the parameter of the shared model;

[0156] x i For input data;

[0157] y i is the corresponding target output;

[0158] is the prediction function.

[0159] It should be understood that the present invention performs collaborative learning on trajectory data on multiple distributed terminal devices. Each distributed terminal device collects and processes data locally, and then aggregates the model parameters to the server for aggregation to form a global trajectory prediction model. This method improves the generalization ability and adaptability of the overall model while protecting data privacy.

[0160] In order to achieve the above object, the present invention also provides a pedestrian positioning device, such as Figure 4 As shown, the pedestrian positioning device includes an acquisition unit 710 , a prediction unit 720 , a selection unit 730 , and a positioning unit 740 .

[0161] The acquisition unit 710 is used to acquire pedestrian historical trajectory data and environmental data. The environmental data may be data such as building layout, roads, and traffic information collected by the GIS system and sensors.

[0162] The prediction unit 720 is used to predict the next location of the pedestrian based on the pedestrian's historical trajectory data and the environmental data. The prediction unit 720 is also used to construct a graph structure using key points in the environmental data as nodes in the graph and the passability between nodes as edges, wherein the characteristics of the nodes include the environmental information of the location of the nodes, and the characteristics of the edges include the length of the path and the difficulty of passing;

[0163] It should be understood that a node may be a building, a road section where pedestrians may pass, or a key location. An edge may be the passability between nodes, and may be defined based on features such as road width, distance, and relative position between buildings.

[0164] It can be achieved by obtaining building layout and road information through GIS (Geographic Information System) and sensor data to build a graph structure. For each node v and its adjacent node u, use edge e vu Represents the transit relationship between nodes, and the edge weight w vu Calculations can be made based on distance, obstacles, travel time, etc.

[0165] The prediction unit 720 is also used to aggregate the information of the node and its neighboring nodes through a pre-built graph convolutional neural network to obtain a feature representation of the node. The aggregation formula is as follows:

[0166] ;

[0167] in, is the feature representation of node v at layer l;

[0168] W （l） is the weight matrix of the lth layer;

[0169] N(v) is the neighbor set of node v;

[0170] du and dv represent the degrees of nodes u and v, respectively;

[0171] σ is the activation function, and a nonlinear function is usually selected to introduce nonlinear characteristics and enhance the expressiveness of the model.

[0172] It should be understood that the characteristics of a node include environmental information of the location, such as crowd density, scene type (e.g., commercial area, residential area), etc. The characteristics of an edge include the length of a path, difficulty of passing, etc.

[0173] The prediction unit 720 is also used to predict the next position of the pedestrian based on the historical trajectory data and the corresponding environmental data. The prediction model is as follows:

[0174] ;

[0175] in, is the predicted value of the next position;

[0176] H() is a prediction model; in this embodiment, it can be but is not limited to a graph neural network model.

[0177] is the feature representation corresponding to the historical trajectory data, that is, the feature representation of node v in the last layer;

[0178] is the feature representation of the environmental data corresponding to the historical trajectory data, that is, the feature representation of the neighboring nodes of node v;

[0179] is the feature of the edge connected to node v.

[0180] It should be noted that the prediction of the next location can be made using a graph neural network model. By modeling building layouts, road information, etc., and using graph convolutional neural networks to capture pedestrian movement patterns in complex environments. By extracting features from nodes and edges, the GNN model can accurately predict the future location of pedestrians in the current environment.

[0181] The selection unit 730 is used to select a beacon near the next position of the pedestrian according to the predicted next position based on the pre-trained reinforcement learning model. The method for establishing the reinforcement learning model includes:

[0182] Define states t , action space, where state s t Including the current position p t , the distribution and signal strength of surrounding beacons B t , and environmental dynamic characteristics E t , the action space is A={B1, B2, ..., B n}, B i For the i-th beacon, action a t is the beacon selection made by the system at time t;

[0183] Set the reward function, the formula is:

[0184] ; Among them, r t is the reward, the error represents the deviation between the positioning result and the actual position; the delay represents the system response time, and α and β are the weights for adjusting the reward;

[0185] Build a reinforcement learning model through strategy Select an action and update the strategy based on the reward. The update formula is as follows:

[0186] ;

[0187] η is the learning rate and γ is the discount factor.

[0188] It should be noted that when inputting the reinforcement learning model, the current position p t That is, input the predicted next location. This module uses reinforcement learning strategy to optimize the selection of UWB beacons. By defining pedestrian positions, beacon distribution, etc. as states, reinforcement learning algorithm is used to dynamically select beacons to maximize positioning accuracy and minimize positioning delay. Each beacon selection result will adjust the subsequent strategy based on the positioning error and positioning delay feedback.

[0189] The positioning unit 740 is used to locate the pedestrian according to the beacons near the selected next position.

[0190] It should be understood that by combining the pedestrian's next behavior intention prediction, historical trajectory data, and corresponding environmental data to further predict the pedestrian's next position, the prediction result is more accurate.

[0191] In other embodiments, the pedestrian trajectory may be predicted based on historical trajectory data, corresponding environmental data, and the pedestrian's next behavioral intention prediction.

[0192] In order to achieve the above object, the present invention also provides a computer device, such as Figure 5 As shown, the cleaning base station includes at least one processor 801; and a memory 802 that is communicatively connected to the at least one processor 801; wherein the memory 802 stores instructions that can be executed by the at least one processor 801, and the instructions are executed by the at least one processor 801 so that the at least one processor 801 can execute the above-mentioned pedestrian positioning method.

[0193] The memory 802 and the processor 801 are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors 801 and the memory 802 together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor 801 is transmitted on a wireless medium via an antenna, and further, the antenna also receives data and transmits the data to the processor 801.

[0194] The processor 801 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management and other control functions. The memory 802 can be used to store data used by the processor 801 when performing operations.

[0195] In order to achieve the above object, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program implements the above pedestrian positioning method when executed by a processor.

[0196] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned implementation method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each implementation of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0197] Obviously, the embodiments described above are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, ordinary technicians in this field can make other different forms of changes or modifications without creative work, which should fall within the scope of protection of the present invention.

Claims

1. A pedestrian positioning method, characterized in that: include: Obtain pedestrian historical trajectory data and environmental data; Predicting the next location of the pedestrian based on the pedestrian's historical trajectory data and environmental data; According to the predicted next position of the pedestrian, a beacon near the next position is selected based on a pre-trained reinforcement learning model; Locating pedestrians based on beacons near the selected next location; The step of predicting the next location of the pedestrian based on the pedestrian's historical trajectory data and the environmental data includes: The key points in the environmental data are used as nodes in the graph, and the passability between nodes is used as edges to construct a graph structure, where the characteristics of the nodes include the environmental information of the location of the nodes, and the characteristics of the edges include the length of the path and the difficulty of passing; Through the pre-built graph convolutional neural network, the information of the node and its neighboring nodes is aggregated to obtain the feature representation of the node. The aggregation formula is as follows: ； in, is the feature representation of node v at layer l; W （l） is the weight matrix of the lth layer; N(v) is the neighbor set of node v; du and dv represent the degrees of nodes u and v, respectively; σ is the activation function; Based on the historical trajectory data and the corresponding environmental data, the next location of the pedestrian is predicted. The prediction model is as follows: ; in, is the predicted value of the next position; H() is the prediction model; is the feature representation corresponding to the historical trajectory data, that is, the feature representation of node v in the last layer; is the feature representation of the environmental data corresponding to the historical trajectory data, that is, the feature representation of the neighboring nodes of node v; is the feature of the edge connected to node v.

2. The pedestrian positioning method according to claim 1, characterized in that: Before the step of predicting the next position of the pedestrian according to the historical trajectory data and the corresponding environmental data, the method further includes: Obtain pedestrian gait data, behavior data and current environment data; Extracting features of the gait data, behavior data, and current environment data respectively; The extracted gait data, behavior data, and features of the current environment data are used as inputs to a trained long short-term memory network model, and a prediction of the pedestrian's next behavior intention is output; Accordingly, predicting the next position of the pedestrian based on the historical trajectory data and the corresponding environmental data includes: Predict the next location of the pedestrian based on historical trajectory data, corresponding environmental data, and the pedestrian's next behavioral intention prediction.

3. The pedestrian positioning method according to claim 2, characterized in that: The extracting the features of the gait data, the behavior data, and the current environment data respectively includes: Using a convolutional network to extract features from the gait data and behavior data; A fully connected layer is used to extract features from the current environment data.

4. The pedestrian positioning method according to claim 2, characterized in that: After the extracted features of the gait data, behavior data, and current environment data are used as inputs to the trained long short-term memory network model and the prediction of the pedestrian's next behavior intention is output, the method further includes: Pedestrian trajectories are predicted based on historical trajectory data, corresponding environmental data, and the pedestrian's next behavioral intention prediction.

5. The pedestrian positioning method according to claim 1, characterized in that: Before the step of selecting a beacon near the predicted next position of the pedestrian based on a pre-trained reinforcement learning model, the method further includes: Define states t , action space, where state s t Including the current position p t , the distribution and signal strength of surrounding beacons B t , and environmental dynamic characteristics E t , the action space is A={B1, B2, ..., B n }, B i For the i-th beacon, action a t is the beacon selection made by the system at time t; Set the reward function, the formula is: ; Among them, r t is the reward, the error represents the deviation between the positioning result and the actual position; the delay represents the system response time, and α and β are the weights for adjusting the reward; Build a reinforcement learning model through strategy Select an action and update the strategy based on the reward. The update formula is as follows: ； η is the learning rate and γ is the discount factor.

6. The pedestrian positioning method according to claim 5, characterized in that: The pedestrian positioning method is applied to a server, the server aggregates reinforcement learning model parameters sent by multiple distributed terminal devices, and the reinforcement learning model is established on each distributed terminal device; the method also includes: The server aggregates the reinforcement learning model parameters sent by each of the distributed terminal devices to obtain an overall model, wherein the calculation formula of the parameters of the overall model is as follows: ； in, are the parameters of the overall model; w k are the parameters of the reinforcement learning model of the kth distributed terminal device; D k is the data set of the kth distributed terminal device; After the overall model is completed, the new overall model is sent to each of the distributed terminal devices, so that the distributed terminal devices use the overall model to update their own reinforcement learning model parameters; Each of the distributed terminal devices minimizes the loss function on its data, and the loss function is: ； The data set on each of the distributed terminal devices is , the objective function is: ； Each distributed terminal device will model parameters is the loss function of distributed terminal k; w is the parameter of the shared model; x i For input data; y i is the corresponding target output; is the prediction function.

7. A pedestrian positioning device, characterized in that: include: An acquisition unit, used to acquire pedestrian historical trajectory data and environmental data; A prediction unit, used to predict the next position of the pedestrian based on the pedestrian's historical trajectory data and environmental data; A selection unit, configured to select a beacon near the predicted next position of the pedestrian based on a pre-trained reinforcement learning model; a positioning unit for locating the pedestrian based on the beacons near the selected next position; The prediction unit is specifically used for: The key points in the environmental data are used as nodes in the graph, and the passability between nodes is used as edges to construct a graph structure, where the characteristics of the nodes include the environmental information of the location of the nodes, and the characteristics of the edges include the length of the path and the difficulty of passing; Through the pre-built graph convolutional neural network, the information of the node and its neighboring nodes is aggregated to obtain the feature representation of the node. The aggregation formula is as follows: ； in, is the feature representation of node v at layer l; W （l） is the weight matrix of the lth layer; N(v) is the neighbor set of node v; du and dv represent the degrees of nodes u and v, respectively; σ is the activation function; Based on the historical trajectory data and the corresponding environmental data, the next location of the pedestrian is predicted. The prediction model is as follows: ; in, is the predicted value of the next position; H() is the prediction model; is the feature representation corresponding to the historical trajectory data, that is, the feature representation of node v in the last layer; is the feature representation of the environmental data corresponding to the historical trajectory data, that is, the feature representation of the neighboring nodes of node v; is the feature of the edge connected to node v.

8. A computer device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the pedestrian positioning method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the pedestrian positioning method according to any one of claims 1 to 6 is implemented.

Images