Motion recognition method based on sensing and communication integrated multi-device cooperation

Abstract

The invention particularly relates to an action recognition method and system based on sensing and communication integrated multi-device cooperation, a storage medium, a computer program product and electronic equipment. According to the method, a deep reinforcement learning online algorithm is provided to adapt to a time-varying sensing environment and a network state, and sensing and communication resources are dynamically distributed. Perception and communication resource allocation of a plurality of devices in a plurality of time periods is optimized, so that the accuracy of human body action recognition is maximized.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a motion recognition method and system based on sensory integration and multi-device collaboration, a storage medium, a computer program product, and an electronic device. Background Art

[0002] In related technologies, with the rapid development of sensing technology, human motion recognition (HMR) has attracted widespread attention in multiple application scenarios, such as elderly care, search and rescue operations, and human-computer interaction. A variety of methods have been proposed for human motion recognition. Among them, wireless recognition has better privacy and stronger robustness than visual recognition and acoustic recognition, and has become an important research direction in the field of human motion recognition. Wireless recognition is based on the influence of human motion on the propagation path of radio signals (such as reflection, diffraction, scattering, etc.), and identifies specific actions by extracting and analyzing the features in the received signal. However, the data collection and transmission of multiple devices to the fusion center often generates huge communication overhead. The communication process and the perception process compete for limited wireless resources (such as energy) on the device. In order to improve resource utilization efficiency, the ISAC technology is proposed. The ISAC technology realizes perception and communication functions through shared modules in hardware, which improves resource utilization. In addition, from the perspective of waveform design, beamforming and resource allocation, the perception and communication processes are also jointly designed to balance the conflicting needs of the two. On this basis, some researchers have also proposed task-oriented ISAC, which maximizes the completion of downstream tasks by jointly designing perception and communication.

[0003] For the task of human motion recognition with multi-device collaboration, since the target is constantly moving, the target echo will appear randomly in the time and space domain. In addition, the appearance of interference signals is also random. Therefore, it is necessary to combine the perception data in multiple time periods to obtain temporal diversity gain. To this end, the resource allocation of multiple devices in multiple time periods must be carefully designed. Especially for devices with limited resources, how to maximize the accuracy of human motion recognition is particularly critical. However, human motion recognition based on multi-device collaboration with integrated communication faces three major challenges. First, due to the different positions of different devices relative to the target, their contributions to the improvement of human motion recognition performance are also different. Therefore, how to allocate network resources among multiple ISAC devices to maximize the accuracy of human motion recognition is not easy. Second, the resource allocation in a time period determines the number of remaining resources next, which affects the resource allocation in subsequent time periods, which makes the resource allocation of each device across time periods complex and coupled. Finally, the perception environment and network status are dynamically changing. The random movement of the target and interference objects causes the echo to change dynamically in the time and space domain, resulting in dynamic changes in the perception environment. As for the network status, the number of cellular users and wireless channels are also time-varying, which increases the difficulty of dynamic resource allocation and thus affects the final accuracy of human motion recognition.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The present invention provides a method and system for motion recognition based on integrated sensory communication and multi-device collaboration, a storage medium, a computer program product, and an electronic device, which improve the accuracy of human motion recognition by sensing dynamic environmental changes and network status changes, and can overcome the defects in the prior art to a certain extent.

[0006] Other features and advantages of the present invention will become apparent from the following detailed description, or may be learned in part by practice of the present invention.

[0007] According to a first aspect of the present invention, there is provided a method for motion recognition based on sensory integration and multi-device collaboration, the method comprising:

[0008] Obtaining current observation data corresponding to the current time slot of the ISAC device and historical observation data corresponding to the historical time slot; wherein the observation data includes: power configuration information of the ISAC device, the number of cellular users, and motion state information of the target to be analyzed;

[0009] Based on the time correlation of the cellular traffic load between consecutive time slots, the current observation data and the historical observation data are used to perform cellular traffic load prediction processing to obtain the predicted number of cellular users corresponding to the next time slot; and based on the predicted number of cellular users, the available communication bandwidth prediction data of all ISAC devices is determined;

[0010] Performing position prediction processing based on the current motion state of the target to be analyzed to obtain the position prediction result corresponding to the next time slot of the target to be analyzed; and determining the posterior distribution of the feature data collected by each ISAC device according to the position prediction result;

[0011] Based on the power corresponding to the current time slot and the historical time slots, estimate the available power prediction data for the next time slot;

[0012] Determine the resource allocation strategy corresponding to each ISAC device according to the available communication bandwidth prediction data, the available power prediction data, and the prediction result of the position of the target to be analyzed corresponding to the next time slot;

[0013] The characteristic data transmitted by each ISAC device based on the resource allocation strategy is received, and the action recognition result corresponding to the target to be analyzed is determined based on the characteristic data.

[0014] In some exemplary embodiments, the method further comprises:

[0015] A tuple is defined according to the number of ISAC devices, the state space, the set of joint actions, the state transition probability function, the reward function, the set of joint observations, and the observation probability function; the tuple is used to represent a decentralized partially observable Markov decision process for estimating the amount of available power in the next time slot;

[0016] The goal of defining the global reward of the tuple includes maximizing the accumulated global reward by optimizing the resource allocation strategy of each agent within M time slots.

[0017] In some exemplary embodiments, the method further comprises:

[0018] Construct a sample data set based on the historical resource allocation strategy of ISAC devices;

[0019] Build an initial policy optimization model based on a deep deterministic policy gradient algorithm;

[0020] Input the sample data set into the policy optimization initial model, use the actor network of the initial model to output the selected action, and use the critic network of the policy optimization initial model to evaluate the action;

[0021] The parameters of the actor network and critic network are updated through the preset loss function to obtain the trained strategy optimization model.

[0022] In some exemplary embodiments, the performing cellular traffic load prediction processing based on the time correlation of the cellular traffic load between consecutive time slots using the current observation data and the historical observation data to obtain the predicted number of cellular users corresponding to the next time slot includes:

[0023] The observation data corresponding to multiple consecutive time slots are input into the cellular traffic load prediction model based on the RNN network to obtain the predicted number of cellular users corresponding to the next time slot output by the model.

[0024] In some exemplary embodiments, the current motion state includes position information, motion direction, and speed information of the target to be analyzed in the current time slot.

[0025] In some exemplary embodiments, the method further comprises:

[0026] The edge server receives the feature set sent by each ISAC device; wherein each ISAC device shares a channel with the current cellular communication user through orthogonal frequency division multiple access;

[0027] Feature data integration processing is performed based on feature sets corresponding to multiple consecutive time slots to obtain action recognition results corresponding to the target to be identified.

[0028] According to a second aspect of the present invention, there is provided a motion recognition system based on sensory integration and multi-device collaboration, comprising: an edge server, an ISAC device; wherein:

[0029] The edge server is used to obtain the current observation data corresponding to the current time slot of the ISAC device and the historical observation data corresponding to the historical time slot; wherein the observation data includes: the power configuration information of the ISAC device, the number of cellular users, and the motion state information of the target to be analyzed; based on the time correlation of the cellular traffic load between consecutive time slots, the current observation data and the historical observation data are used to perform cellular traffic load prediction processing to obtain the predicted number of cellular users corresponding to the next time slot; and the available communication bandwidth prediction data of all ISAC devices is determined based on the predicted number of cellular users; the position prediction processing is performed based on the current motion state of the target to be analyzed to obtain the position prediction result corresponding to the target to be analyzed in the next time slot; and the posterior distribution of the feature data collected by each ISAC device is determined according to the position prediction result; based on the power corresponding to the current time slot and the historical time slots, the available power prediction data of the next time slot is estimated; the resource allocation strategy corresponding to each ISAC device is determined according to the available communication bandwidth prediction data, the available power prediction data, and the position prediction result corresponding to the target to be analyzed in the next time slot; the feature data transmitted by each ISAC device based on the resource allocation strategy is received, and the action recognition result corresponding to the target to be analyzed is determined based on the feature data;

[0030] The ISAC device is used to send feature data sets to the edge server according to the resource allocation strategy.

[0031] According to a third aspect of the present invention, there is provided a storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the above-mentioned motion recognition method based on multi-device collaboration with integrated sensory communication is implemented.

[0032] According to a fourth aspect of the present invention, there is provided an electronic device, comprising:

[0033] Processor; and

[0034] A memory, configured to store executable instructions of the processor;

[0035] Among them, the processor is configured to implement the above-mentioned action recognition method based on multi-device collaboration based on integrated sensory communication when executing the executable instructions.

[0036] According to a fifth aspect of the present invention, there is provided a computer program product having a computer program stored thereon, which implements the above-mentioned motion recognition method based on multi-device collaboration with integrated sensory communication when the computer program is executed by a processor.

[0037] The action recognition method based on the integrated sensory communication multi-device collaboration provided by the embodiment of the present invention obtains the current observation data corresponding to the current time slot of the ISAC device and the historical observation data corresponding to the historical time slot, and uses the current observation data and the historical observation data to perform cellular traffic load prediction processing to obtain the predicted number of cellular users corresponding to the next time slot; and determines the available communication bandwidth prediction data of all ISAC devices based on the predicted number of cellular users; performs position prediction processing based on the current motion state of the target to be analyzed to obtain the position prediction result corresponding to the target to be analyzed in the next time slot; and determines the posterior distribution of the feature data collected by each ISAC device according to the position prediction result; estimates the available power prediction data of the next time slot based on the power corresponding to the current time slot and the historical time slot; determines the resource allocation strategy corresponding to each ISAC device according to the available communication bandwidth prediction data, the available power prediction data, and the position prediction result corresponding to the target to be analyzed in the next time slot; receives the feature data transmitted by each ISAC device based on the resource allocation strategy, and determines the action recognition result corresponding to the target to be analyzed based on the feature data. By optimizing the perception and communication resource allocation of multiple devices in multiple time periods, the accuracy of human action recognition is maximized.

[0038] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present invention, and together with the specification are used to explain the principles of the present invention. Obviously, the accompanying drawings described below are only some embodiments of the present invention, and for those of ordinary skill in the art, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0040] Figure 1 A schematic diagram schematically illustrates an exemplary embodiment of the present invention, a method for motion recognition based on sensory integration and multi-device collaboration;

[0041] Figure 2 A schematic diagram schematically shows a framework of a method for human motion recognition based on sensory integration and multi-device collaboration according to an exemplary embodiment of the present invention;

[0042] Figure 3 A schematic diagram schematically illustrates a variation trend of human motion recognition accuracy with time domain resource budget according to an exemplary embodiment of the present invention;

[0043] Figure 4 A schematic diagram schematically illustrates a variation trend of human motion recognition accuracy along with airspace resource budget according to an exemplary embodiment of the present invention;

[0044] Figure 5 A schematic diagram schematically illustrates the accuracy of human motion recognition under different bandwidth budgets according to an exemplary embodiment of the present invention;

[0045] Figure 6 A schematic diagram schematically illustrates the accuracy of human motion recognition under different energy budgets according to an exemplary embodiment of the present invention;

[0046] Figure 7 A schematic diagram schematically shows an exemplary embodiment of the present invention, a motion recognition system based on sensory integration and multi-device collaboration;

[0047] Figure 8 The figure schematically shows the composition of an electronic device in an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0048] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the present invention will be more comprehensive and complete and fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0049] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0050] In related technologies, with the rapid development of sensing technology, human motion recognition (HMR) has attracted widespread attention in multiple application scenarios, such as elderly care, search and rescue operations, and human-computer interaction. A variety of methods have been proposed for human motion recognition. Among them, wireless recognition has better privacy and stronger robustness than visual recognition and acoustic recognition, and has become an important research direction in the field of human motion recognition. Wireless recognition extracts and analyzes the features in the received signal to identify specific actions by the influence of human motion on the propagation path of radio signals (such as reflection, diffraction, scattering, etc.). In contrast, data collection and transmission of multiple devices to the fusion center usually brings huge communication overhead. The communication process and the perception process compete for limited wireless resources (such as energy) on the device. Therefore, improving resource utilization efficiency is an important issue. In order to solve this problem, the integrated sensing and communication (ISAC) technology came into being. ISAC technology realizes the functional integration of perception and communication through hardware sharing modules, thereby improving resource utilization. In addition, the joint design of ISAC in waveform design, beamforming and resource allocation effectively balances the mutual needs between perception and communication. On this basis, some studies have proposed task-oriented ISAC, which maximizes the completion of downstream tasks by jointly designing perception and communication. For the human action recognition task of multi-device collaboration, since the target is constantly moving, the echo signal will appear randomly in the time and space domain. In addition, the appearance of interference signals is also random. Therefore, in order to improve the accuracy of action recognition, it is necessary to combine the perception data of multiple time periods to obtain temporal diversity gain. In order to achieve this goal, the resource allocation of multiple devices in multiple time periods must be carefully designed. Especially for devices with limited resources, how to maximize the accuracy of action recognition is particularly critical. In practical applications, the multi-device collaborative action recognition method based on integrated perception and communication faces three major challenges. First, the positions of different devices relative to the target are different, so their effects on improving the performance of human action recognition are also different. Therefore, how to reasonably allocate resources to improve recognition accuracy is a complex problem. Secondly, the resource allocation in a certain time period determines the availability of the next resource, which in turn affects the resource allocation in the subsequent time period, forming a cross-time period resource allocation coupling. How to leave resources for the time period with the highest accuracy of human action recognition. Finally, since the perception environment and network status are dynamically changing, the random movement of targets and interfering objects as well as the fluctuations in the number of cellular users and wireless channels further affect the accuracy of human action recognition.

[0051] In view of the shortcomings and deficiencies of the prior art, this example implementation provides a multi-device collaborative action recognition method based on integrated sensory communication. This method optimizes the recognition process of human actions by rationally allocating the perception and communication resources of multiple devices. By jointly designing the resource allocation strategy of the ISAC device, the device can perceive and transmit data more efficiently in each time slot, and finally output accurate action recognition results.

[0052] In this example implementation, reference Figure 1 As shown, the multi-device collaborative action recognition method based on sensory integration may specifically include the following steps:

[0053] Step S11, obtaining current observation data corresponding to the current time slot of the ISAC device and historical observation data corresponding to the historical time slot; wherein the observation data includes: power configuration information of the ISAC device, the number of cellular users, and motion state information of the target to be analyzed;

[0054] Step S12, based on the time correlation of the cellular traffic load between consecutive time slots, the current observation data and the historical observation data are used to perform cellular traffic load prediction processing to obtain the predicted number of cellular users corresponding to the next time slot; and based on the predicted number of cellular users, the available communication bandwidth prediction data of all ISAC devices is determined;

[0055] Step S13, performing position prediction processing based on the current motion state of the target to be analyzed to obtain a position prediction result corresponding to the next time slot of the target to be analyzed; and determining the posterior distribution of the feature data collected by each ISAC device according to the position prediction result;

[0056] Step S14, estimating the available power prediction data of the next time slot based on the power corresponding to the current time slot and the historical time slots;

[0057] Step S15, determining the resource allocation strategy corresponding to each ISAC device according to the available communication bandwidth prediction data, the available power prediction data, and the position prediction result corresponding to the target to be analyzed in the next time slot;

[0058] Step S16, receiving feature data transmitted by each ISAC device based on the resource allocation strategy, and determining a motion recognition result corresponding to the target to be analyzed based on the feature data.

[0059] Below, each step of the multi-device collaborative action recognition method based on sensory integration in this example implementation will be described in more detail with reference to the accompanying drawings and embodiments.

[0060] In step S11, current observation data corresponding to the current time slot of the ISAC device and historical observation data corresponding to the historical time slot are obtained; wherein the observation data includes: power configuration information of the ISAC device, the number of cellular users, and motion state information of the target to be analyzed.

[0061] Exemplarily, the edge server may allocate resources to the ISAC device. For the edge server, when the current time slot is time slot m, in order to allocate resources to each ISAC device in the next time slot, that is, the resource configuration of time slot m+1, the current observation data of the current time slot and the historical observation data of the previous multiple consecutive historical time slots may be used to predict the resource allocation of the next time slot. Among them, the observation data collected for each time slot may include: power configuration information of the ISAC device, the number of cellular users, and the motion state information of the target to be analyzed. Specifically, the power configuration information of the ISAC device may be the available power of the ISAC device, which can affect the effect of the ISAC device in the perception stage and the communication stage. The number of cellular users may be the number of cellular users in the time slot, which can be used to reflect the network load. The target to be analyzed may be an object that needs to be motion recognized, and the motion state information may be the location information of the target; in addition, it may also include motion direction and speed information.

[0062] The observation data of ISAC device k in time slot m can be expressed as:

[0063]

[0064] Among them, m Represents the position of the target to be analyzed; l m Expressed as the number of cellular users; Represents the power configuration of ISAC device k.

[0065] For example, considering the data inference accuracy and data processing effect, the number of historical time slots can be configured to be 1 to 5. One time slot can be configured to be 0.7s.

[0066] In step S12, based on the time correlation of the cellular traffic load between consecutive time slots, the current observation data and the historical observation data are used to perform cellular traffic load prediction processing to obtain the predicted number of cellular users corresponding to the next time slot; and based on the predicted number of cellular users, the available communication bandwidth prediction data of all ISAC devices is determined.

[0067] Exemplarily, a deep reinforcement learning-based online optimization algorithm (Deep Reinforcement Learning Optimization, DRLO) can be provided, which can maximize the accuracy of reasoning in a dynamic environment. The algorithm can predict the environmental state and adaptively adjust to cope with changes in new environmental states to allocate resources of multiple ISAC devices.

[0068] In order to implement the DRLO algorithm, it is necessary to predict the perception environment and network status information of the next time slot. To this end, a prediction model based on a recurrent neural network (RNN) is provided, including a cellular traffic load prediction model and a target location prediction model. Among them, the cellular traffic load prediction model can be used to output the predicted number of cellular users in the next time slot, thereby determining the available communication bandwidth of all ISAC devices. The target location prediction model can be used to output the target location of the next time slot, thereby determining the posterior distribution of the features obtained by each ISAC device.

[0069] Exemplarily, the method of performing cellular traffic load prediction processing based on the time correlation of cellular traffic loads between consecutive time slots using the current observation data and the historical observation data to obtain the predicted number of cellular users corresponding to the next time slot includes:

[0070] The observation data corresponding to multiple consecutive time slots are input into the cellular traffic load prediction model based on the RNN network to obtain the predicted number of cellular users corresponding to the next time slot output by the model.

[0071] Specifically, in the cellular traffic load prediction model, the RNN network is used to predict the number of cellular users in the next time slot. o Taking the observation data of time slots as input, the RNN network can capture the temporal correlation of cellular traffic load between consecutive time slots and provide predictions for the next time slot. Figure 2 As shown, from the first (mM o ) time slot to the (m-1)th time slot, the observation data of each time slot is sequentially input into an RNN network, and the output of the last RNN network is input into the output layer of the entire module.

[0072] Specifically, the RNN network architecture may include an input layer, an RNN layer, a fully connected layer, and an output layer. The process of the RNN network processing the M0 group of observation data usually includes the following steps:

[0073] 1) Data input: Input M0 groups of observation data into the RNN network. Each group of data usually contains time series information such as target location, number of cellular users, and device power.

[0074] The input data can be expressed as [M0, T, N]; where M0 is the number of observation data groups, such as 10-100 groups; T is the time slot length, such as 10 time slots; and N is the number of features contained in each time slot, such as the number of cellular users, device power, etc.

[0075] 2) Time-step propagation: At each time step, the RNN network processes the current observation data together with the hidden state (or memory) of the previous time step.

[0076] The RNN network generates the current output based on the current input and previous memory (through recurrent connections) and updates the internal state of the network.

[0077] The RNN network gradually reads the input data of each time slot and passes the information to the next time step. Each time step updates the current state based on the current input and the previous hidden state (or memory). The specific update formula can include:

[0078]

[0079] Among them, x t is the current input; W and U are weight matrices; b is the bias; and f is the activation function, such as the ReLU activation function.

[0080] 3) Inter-layer transfer: If there are multiple RNN layers, the network will transfer information between layers and continuously extract features in the time series.

[0081] 4) Fully connected layer output: After being processed by the RNN layer, the data will be passed to the fully connected layer for final prediction processing.

[0082] Generate prediction results: Through the fully connected layer, the predicted value of the next time slot is finally output, that is, the number of cellular users. The formula may include:

[0083]

[0084] in, is the predicted number of cellular users, W out , b out are the weights and biases of the output layer.

[0085] Finally, the RNN network outputs the predicted value of the number of cellular users in the next time slot

[0086] In step S13, position prediction processing is performed based on the current motion state of the target to be analyzed to obtain a position prediction result corresponding to the next time slot of the target to be analyzed; and the posterior distribution of the feature data collected by each ISAC device is determined according to the position prediction result.

[0087] For example, assuming that there is no sudden turn in the target's motion trajectory, the next position of the target can be estimated based on the current position and speed. Although this estimation method may introduce some slight deviations, due to the short duration of each time slot, the impact of this deviation on the final action recognition accuracy can be ignored.

[0088] Specifically, let (x m ,y m ) is the position of the target at time slot m, θ m is the current moving direction of the target, v m is the speed of the target. Then at time slot m+1, the position of the target (x m+1 ,y m+1 ) can be estimated by the following formula:

[0089]

[0090] Here, Δm is the duration of one time slot.

[0091] In step S14, based on the power amounts corresponding to the current time slot and the historical time slots, available power amount prediction data for the next time slot is estimated.

[0092] Exemplarily, the amount of power used in the current and historical time slots determines the amount of power available in subsequent time slots. Therefore, the problem can be converted into a decentralized partially observable Markov decision process.

[0093] Exemplarily, the method further includes: defining a tuple according to the number of ISAC devices, the state space, the set of joint actions, the state transition probability function, the reward function, the set of joint observations, and the observation probability function; the tuple is used to represent a decentralized partially observable Markov decision process for estimating the amount of available power in the next time slot;

[0094] The goal of defining the global reward of the tuple includes maximizing the accumulated global reward by optimizing the resource allocation strategy of each agent within M time slots.

[0095] Specifically, a tuple<K,S,A,P,R,O,Z,ψ> To represent the decentralized partially observable Markov decision process, where K is the number of agents (ISAC devices), S is the state space, and A = × k∈K A k is the set of joint actions, P is the state transition probability function, R is the reward function, O = × k∈K O k is the set of joint observations, Z is the observation probability function, and ψ is the discount factor. In time slot m, based on the observed data Agent k follows the strategy Determine its action. The policy is a set of actions When all agents perform actions according to their respective strategies, a global reward R is generated. m =R(s m , a m ), and the state changes to s m+1 The goal is to maximize the cumulative global reward by optimizing the policy of each agent within M time slots.

[0096] Specifically, the state space can be defined according to the problem, including the location of the target, the number of cellular users, and the amount of power available for each ISAC device. The state changes with the time slot. The observation data at time slot m is expressed as where ζ m , l m and They represent the location of the target, the number of cellular users, and the available power of ISAC device k at time slot m, respectively.

[0097] The joint action includes the resources that need to be allocated, namely, quantized gain, sensing power, transmission power, and communication bandwidth. It is necessary to determine the action for each ISAC device in each time slot, that is, to allocate resources. The action of ISAC device k in time slot m is expressed as

[0098] For the above rewards, it can be expressed as For the task-oriented multi-device ISAC system, its goal is to maximize the action recognition performance, which is equivalent to maximizing the discriminant gain and is represented by the objective function. Therefore, the discriminant gain is used in the reward function. The cumulative global reward function at time slot m is:

[0099]

[0100] Where Gd,d′(·) represents the discrimination gain between category d and category d′. m Next, when the time slot is m, take action When , the nth restored feature is obtained as

[0101] In addition, the discount factor determines the importance of future rewards in the cumulative rewards. It is a parameter used to discount future rewards in reinforcement learning, usually 0≤ψ≤1. It reflects the importance of the agent to future rewards. A larger discount factor (close to 1) means that the agent attaches more importance to long-term rewards, while a smaller discount factor means that the agent attaches more importance to short-term rewards.

[0102] The state transition function P is used to describe the probability of one state to another state. In this problem, P represents the probability of transitioning from one state to another given the current state s. m and current action (i.e., the combined action set of all devices), how does the system transfer to the next state s? m+1 .

[0103] Specifically, the ISAC device adjusts the device resource configuration (such as power allocation, bandwidth allocation, etc.) through actions, thereby affecting the system state. m+1 |s m , a m ) gives the current state s m Next, take action Then transfer to s m +1 probability.

[0104] The observation probability function Z describes the observation probability given a hidden state s m and actions In the case of a certain observation data that the ISAC device can observe Each ISAC device has the possibility to obtain a partial observation based on its current state and action in each time slot.

[0105] Observation probability function Z(s m+1 |s m , a m ) describes the observation data of ISAC device k in time slot m In the current state m and actions The probability of the following.

[0106] In step S15, the resource allocation strategy corresponding to each ISAC device is determined according to the available communication bandwidth prediction data, the available power prediction data, and the prediction result of the position of the target to be analyzed corresponding to the next time slot.

[0107] Exemplarily, the method further includes:

[0108] Step S21, constructing a sample data set based on the historical resource allocation strategy of the ISAC device;

[0109] Step S22, constructing an initial policy optimization model based on a deep deterministic policy gradient algorithm;

[0110] Step S23, inputting the sample data set into the strategy optimization initial model, using the actor network of the initial model to output the selected action, and using the critic network of the strategy optimization initial model to evaluate the action;

[0111] Step S24, updating the parameters of the actor network and the critic network through a preset loss function to obtain a trained strategy optimization model.

[0112] Specifically, in order to maximize the cumulative global reward function, we use a deep reinforcement learning algorithm to optimize the strategy of each agent. At the same time, we use random mini-batch sample training, that is, a small number of training samples are used in each iteration to reduce the amount of computation and speed up training. i Represents the index of each sample in the mini-batch. In particular, we use the Deep Deterministic Policy Gradient (DDPG) algorithm to obtain the optimal policy for each agent.

[0113] Using Observation Sets As input, with parameters θ actor The actor network outputs the selected action Based on the observation set and the selected action, with parameters θ critic The critic network evaluates these actions.

[0114] During the training process, the parameters of the critic network are adjusted by minimizing the loss function Update. The gradient of the loss function is expressed as:

[0115]

[0116] in, and are the weights of the critic network and the actor network, respectively. Similar to the Deep Q Network (DQN), these two target networks are updated by partially copying weights from the main network.

[0117] To train the parameters of the actor network, the loss function The gradient of is expressed as:

[0118]

[0119] In step S16, feature data transmitted by each ISAC device based on the resource allocation strategy is received, and a motion recognition result corresponding to the target to be analyzed is determined based on the feature data.

[0120] Exemplarily, after the edge server determines the resource allocation strategy of each ISAC device in the next time slot based on the observation data corresponding to the current time slot and multiple historical time slots, the updated resource allocation strategy can be allocated to the corresponding ISAC device. Thus, each ISAC device can determine the specific action in the next time slot according to the resource allocation strategy, such as sensing power, transmission power and communication bandwidth. And send the feature data corresponding to the collected sensing data to the edge server. The ISAC device can compress the sensing data and send the compressed feature data to the edge server.

[0121] In some exemplary embodiments, the method further includes: an edge server receiving a quantitative feature set sent by each ISAC device; wherein each ISAC device shares a channel with a current cellular communication user via orthogonal frequency division multiple access; and performing feature data integration processing based on feature sets corresponding to a plurality of consecutive time slots to obtain an action recognition result corresponding to a target to be identified.

[0122] For example, the edge server can integrate the feature data of the features of multiple consecutive time slots, use the integrated feature data to perform action recognition, and output the corresponding action recognition result. For example, the integrated feature data can be input into a trained action recognition model, and the action recognition result output by the model can be obtained.

[0123] The present invention proposes a human action recognition problem, and optimizes the perception and communication resource allocation of multiple devices in multiple time periods to maximize the accuracy of human action recognition. Specifically, the accuracy of human action recognition uses the discrimination gain as a measurement indicator, that is, the greater the discrimination gain, the higher the accuracy of human action recognition.

[0124] For example, consider the human action recognition scenario, configure 5 ISAC devices to extract the features of the target action from the received echo, and then transmit the features to the edge server for action recognition. The action candidates considered include children walking, children running, adults walking, and adults running. In the simulation, we use a wireless sensing simulator to simulate high-fidelity target actions. The target moves along a certain trajectory at a specific speed to generate action data. Assume that the height H of children and adults is uniformly distributed in the range of [0.9m, 1.2m] and [1.6m, 1.9m], respectively. The speeds of standing, walking, and running are set to 0m / s, 0.25H m / s, and 0.5H m / s, respectively. The target's movement direction is uniformly distributed in the range of [-180°, +180°]. ISAC devices are distributed in a circular area with a radius of 50 meters, and the channel from the device to the edge server follows an independent and identically distributed (iid) Rayleigh distribution with zero mean and variance Configurable The power of additive white Gaussian noise (AWGN) is N0 =-120dB.

[0125] In order to represent the performance gain of this scheme, the following three baseline algorithms are considered for comparison, as follows: 1) Sensing Mode Allocation (SMA): Only the parameters related to the sensing process, namely, sensing power and quantization gain, are optimized. Communication resources (including the transmission power and communication bandwidth of the ISAC device in each time slot) are evenly distributed. 2) Communication Mode Allocation (CMA): Only communication resources are optimized, including the transmission power and communication bandwidth of the ISAC device in each time slot. The sensing power is evenly distributed among all time slots, and the quantization gain is set to a specific value. 3) Single Slot Sensing Allocation (SSA): The energy budget is first evenly distributed to each time slot. Then, resources are allocated based on the amount of allocated energy and available bandwidth to maximize the performance of each time slot. The comparison results are shown in Figure 2. Figure 3-Figure 6 Extensive numerical results show that the DRLO algorithm proposed in this method outperforms other comparison algorithms in terms of overall performance of human action recognition.

[0126] The method provided by the embodiment of the present invention provides an online optimization algorithm based on deep reinforcement learning, which maximizes the accuracy of reasoning in a dynamic environment. The algorithm can predict the environmental state and adaptively adjust to cope with the changes in the new environmental state to allocate the resources of multiple ISAC devices. At the same time, the DRLO algorithm is applied to a multi-device collaborative human motion recognition system. The system has a dynamic interference and time-varying network state, and is composed of multiple devices and edge servers with limited energy and bandwidth. These devices obtain perception data from the received signals in a continuous time period according to the resources allocated by the DRLO algorithm, and transmit it to the edge server to complete the motion recognition task of the mobile target.

[0127] The motion recognition method based on integrated sensory communication and multi-device collaboration provided by the present invention dynamically allocates communication resources to multiple ISAC devices at different perspectives according to changes in the dynamic perception environment and changes in the network status, thereby achieving reasonable allocation of resources, and enabling the edge server to obtain the perception data of each ISAC device in a timely and accurate manner, thereby improving the accuracy of human motion recognition.

[0128] It should be noted that the above figures are only schematic illustrations of the processes included in the method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It is easy to understand that the processes shown in the above figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0129] For further reference, Figure 7As shown, the implementation of this example also provides a motion recognition system 70 based on multi-device collaboration with integrated sensory communication, including: an edge server 702 and an ISAC device 701.

[0130] The edge server 702 is used to obtain the current observation data corresponding to the current time slot of the ISAC device and the historical observation data corresponding to the historical time slot; wherein the observation data includes: power configuration information of the ISAC device, the number of cellular users, and the motion state information of the target to be analyzed; based on the time correlation of the cellular traffic load between consecutive time slots, the current observation data and the historical observation data are used to perform cellular traffic load prediction processing to obtain the predicted number of cellular users corresponding to the next time slot; and the available communication bandwidth prediction data of all ISAC devices are determined based on the predicted number of cellular users; the position prediction processing is performed based on the current motion state of the target to be analyzed to obtain the position prediction result corresponding to the target to be analyzed in the next time slot; and the posterior distribution of the feature data collected by each ISAC device is determined according to the position prediction result; based on the power corresponding to the current time slot and the historical time slots, the available power prediction data of the next time slot is estimated; the resource allocation strategy corresponding to each ISAC device is determined according to the available communication bandwidth prediction data, the available power prediction data, and the position prediction result corresponding to the target to be analyzed in the next time slot; the feature data set transmitted by each ISAC device based on the resource allocation strategy is received, and the action recognition result corresponding to the target to be analyzed is determined based on the feature data set.

[0131] The ISAC device 701 is used to send a feature data set to an edge server according to a resource allocation strategy.

[0132] Exemplarily, the system may include multiple ISAC devices 701 .

[0133] It should be noted that, although several modules or units of the equipment for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into being embodied by multiple modules or units.

[0134] Figure 8 A schematic diagram of an electronic device suitable for implementing an embodiment of the present invention is shown.

[0135] It should be noted that Figure 8 The electronic device 1000 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0136] like Figure 8As shown, the electronic device 1000 includes a central processing unit (CPU) 1001, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 1002 or the program loaded from the storage part 1008 to the random access memory (RAM) 1003. In the RAM 1003, various programs and data required for system operation are also stored. The CPU 1001, ROM 1002 and RAM 1003 are connected to each other through a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0137] The following components are connected to the I / O interface 1005: an input section 1006 including a keyboard, a mouse, etc.; an output section 1007 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed, so that a computer program read therefrom is installed into the storage section 1008 as needed.

[0138] In particular, according to an embodiment of the present invention, the process described below with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a storage medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part 1009, and / or installed from a removable medium 1011. When the computer program is executed by a central processing unit (CPU) 1001, various functions defined in the system of the present application are executed.

[0139] It should be noted that the storage medium shown in the embodiment of the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any storage medium other than computer-readable storage media, which may send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the storage medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0140] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0141] The units involved in the embodiments of the present invention may be implemented by software or hardware, and the units described may also be arranged in a processor. The names of these units do not, in some cases, limit the units themselves.

[0142] It should be noted that, as another aspect, the present application also provides a storage medium, which can be included in an electronic device; or it can exist independently without being installed in the electronic device. The above storage medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method described in the following embodiments. For example, the electronic device can implement the following Figure 1 The individual steps of the method are shown.

[0143] In one embodiment, the present application provides a computer program product, including a computer program, which implements the steps in the above-mentioned method embodiments when executed by a processor.

[0144] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0145] Other embodiments of the invention will readily occur to those skilled in the art after considering the specification and practicing the invention herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims.

[0146] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A motion recognition method based on multi-device collaboration with integrated sensory communication, characterized in that: The method comprises: Obtaining current observation data corresponding to the current time slot of the ISAC device and historical observation data corresponding to the historical time slot; wherein the observation data includes: power configuration information of the ISAC device, the number of cellular users, and motion state information of the target to be analyzed; Based on the time correlation of the cellular traffic load between consecutive time slots, the current observation data and the historical observation data are used to perform cellular traffic load prediction processing to obtain the predicted number of cellular users corresponding to the next time slot; and based on the predicted number of cellular users, the available communication bandwidth prediction data of all ISAC devices is determined; Performing position prediction processing based on the current motion state of the target to be analyzed to obtain the position prediction result corresponding to the next time slot of the target to be analyzed; and determining the posterior distribution of the feature data collected by each ISAC device according to the position prediction result; Based on the power corresponding to the current time slot and the historical time slots, estimate the available power prediction data for the next time slot; Determine the resource allocation strategy corresponding to each ISAC device according to the available communication bandwidth prediction data, the available power prediction data, and the prediction result of the position of the target to be analyzed corresponding to the next time slot; The characteristic data transmitted by each ISAC device based on the resource allocation strategy is received, and the action recognition result corresponding to the target to be analyzed is determined based on the characteristic data.

2. The method according to claim 1, characterized in that The method further comprises: A tuple is defined according to the number of ISAC devices, the state space, the set of joint actions, the state transition probability function, the reward function, the set of joint observations, and the observation probability function; the tuple is used to represent a decentralized partially observable Markov decision process for estimating the amount of available power in the next time slot; The goal of defining the global reward of the tuple includes maximizing the accumulated global reward by optimizing the resource allocation strategy of each agent within M time slots.

3. The method according to claim 2, characterized in that The method further comprises: Construct a sample data set based on the historical resource allocation strategy of ISAC devices; Build an initial policy optimization model based on the deep deterministic policy gradient algorithm; Input the sample data set into the policy optimization initial model, use the actor network of the initial model to output the selected action, and use the critic network of the policy optimization initial model to evaluate the action; The parameters of the actor network and critic network are updated through the preset loss function to obtain the trained strategy optimization model.

4. The method according to claim 1, characterized in that: The method of performing cellular traffic load prediction processing based on the time correlation of cellular traffic loads between consecutive time slots using the current observation data and the historical observation data to obtain a predicted number of cellular users corresponding to the next time slot includes: The observation data corresponding to multiple consecutive time slots are input into the cellular traffic load prediction model based on the RNN network to obtain the predicted number of cellular users corresponding to the next time slot output by the model.

5. The method according to claim 1, characterized in that The current motion state includes the position information, motion direction, and speed information of the target to be analyzed in the current time slot.

6. The method according to claim 1, characterized in that The method further comprises: The edge server receives the quantized feature set sent by each ISAC device; wherein each ISAC device shares a channel with a current cellular communication user through orthogonal frequency division multiple access; Feature data integration processing is performed according to feature sets corresponding to multiple consecutive time slots to obtain action recognition results corresponding to the target to be identified.

7. A motion recognition system based on sensory integration and multi-device collaboration, characterized in that: The system includes: an edge server and an ISAC device; wherein, The edge server is used to obtain the current observation data corresponding to the current time slot of the ISAC device and the historical observation data corresponding to the historical time slot; wherein the observation data includes: the power configuration information of the ISAC device, the number of cellular users, and the motion state information of the target to be analyzed; based on the time correlation of the cellular traffic load between consecutive time slots, the current observation data and the historical observation data are used to perform cellular traffic load prediction processing to obtain the predicted number of cellular users corresponding to the next time slot; and the available communication bandwidth prediction data of all ISAC devices are determined based on the predicted number of cellular users; the position prediction processing is performed based on the current motion state of the target to be analyzed to obtain the position prediction result corresponding to the target to be analyzed in the next time slot; and the posterior distribution of the feature data collected by each ISAC device is determined according to the position prediction result; based on the power corresponding to the current time slot and the historical time slots, the available power prediction data of the next time slot is estimated; the resource allocation strategy corresponding to each ISAC device is determined according to the available communication bandwidth prediction data, the available power prediction data, and the position prediction result corresponding to the target to be analyzed in the next time slot; the feature data set transmitted by each ISAC device based on the resource allocation strategy is received, and the action recognition result corresponding to the target to be analyzed is determined based on the feature data set; The ISAC device is used to send feature data sets to the edge server according to the resource allocation strategy.

8. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the motion recognition method based on sensory integration and multi-device collaboration as described in any one of claims 1 to 6 is implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the motion recognition method based on sensory integration and multi-device collaboration described in any one of claims 1 to 6 is implemented.

10. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; Wherein, the processor is configured to execute the action recognition method based on sensory integration and multi-device collaboration as described in any one of claims 1 to 6 by executing the executable instructions.

Images