Perception model management method and device

Abstract

The invention provides a perceptual model management method and device, and the method comprises the steps: transmitting a first request message to a model reasoning device based on the address information of the model reasoning device, the first request message comprising a model identifier and a model size of a perceptual model, sending a first successful response message to the management device by the model reasoning device when the resource size of the local storage resource is not smaller than the model size, the first successful response message comprising the model identifier; if the first successful response message is received, a model issuing message is sent to the model reasoning equipment, and the model issuing message comprises model data corresponding to the model identifier, so that the model reasoning equipment generates a sensing model corresponding to the model identifier based on the model data, inputs the wireless channel data into the sensing model for reasoning, and sends the wireless channel data to the model reasoning equipment; and obtaining a wireless sensing result of the target object. According to the technical scheme of the invention, the management equipment is responsible for the management of the perception model, and the model management process and the signaling interaction method in the AI auxiliary perception are clear.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a perception model management method and apparatus. Background Technology

[0002] A WLAN (Wireless Local Area Network) can include an Access Controller (AC) and multiple Access Points (APs). When a wireless terminal is within the coverage area of ​​an AP, the wireless terminal establishes a wireless link with that AP. The wireless terminal sends data packets to the AP through this wireless link and receives data packets sent by the AP through the same wireless link.

[0003] During data packet transmission, a large number of wireless signals exist in WLAN. These wireless signals can reflect, penetrate, and bend on the surface of objects during propagation. By processing these wireless signals, it is possible to perceive the surrounding environment, detect obstacles, and detect target movement based on wireless signals.

[0004] The core principle of WLAN sensing is that when a human or object moves within the wireless coverage area, information such as the CSI (Channel State Information) and RSSI (Received Signal Strength Indication) of the wireless signal will change accordingly. Some key features can be extracted from the CSI and RSSI of the wireless signal, and WLAN sensing can be achieved based on these key features.

[0005] However, there is no effective way to extract key features from the CSI and RSSI of wireless signals in related technologies, that is, it is impossible to effectively extract key features from the CSI and RSSI of wireless signals. Summary of the Invention

[0006] In view of this, this application provides a sensing model management method, apparatus and device to effectively extract key features of wireless signals through a sensing model and obtain accurate and reliable wireless sensing results.

[0007] In a first aspect, this application provides a perception model management method applied to a management device. The method includes: sending a first request message to the model inference device based on stored address information of the model inference device, the first request message including a model identifier and a model size of the perception model, such that the model inference device has local storage resources of not less than the model size; sending a first success response message to the management device, the first success response message including the model identifier; if the first success response message is received, sending a model distribution message to the model inference device, the model distribution message including model data corresponding to the model identifier, such that the model inference device generates a perception model corresponding to the model identifier based on the model data; inputting acquired wireless channel data into the perception model for inference, and obtaining the wireless perception result of the target object.

[0008] In a second aspect, this application provides a perception model management device applied to a management device. The device includes: a sending module, configured to send a first request message to the model inference device based on stored address information of the model inference device, the first request message including a model identifier and a model size of the perception model, and to send a first success response message to the management device when the model inference device has local storage resources not less than the model size, the first success response message including the model identifier; a receiving module, configured to receive the first success response message; the sending module is further configured to send a model distribution message to the model inference device, the model distribution message including model data corresponding to the model identifier, so that the model inference device generates a perception model corresponding to the model identifier based on the model data, and inputs acquired wireless channel data into the perception model for inference to obtain the wireless perception result of the target object.

[0009] In a third aspect, this application provides a management device, comprising: a processor and a machine-readable storage medium, the machine-readable storage medium storing machine-executable instructions executable by the processor; the processor is configured to execute the machine-executable instructions to implement the perception model management method of the above example of this application.

[0010] In a fourth aspect, this application provides a computer program product, which includes a computer program that, when executed by a processor, implements the perceptual model management method of the above-described example of this application.

[0011] In a fifth aspect, this application provides a machine-readable storage medium storing machine-executable instructions that can be executed by a processor; wherein the processor is configured to execute the machine-executable instructions to implement the perception model management method of the above example of this application.

[0012] As can be seen from the above technical solutions, in the embodiments of this application, the model inference device can input wireless channel data (such as data related to wireless signals in WLAN) into the perception model for inference to obtain the wireless perception result of the target object. In this way, the wireless perception result of the target object can be analyzed based on the perception model to obtain accurate and reliable wireless perception results.

[0013] The management equipment is responsible for the control and management of the entire sensing system and the management of the sensing models. It clarifies the registration and synchronization of sensing services in AI-assisted sensing, as well as the model management process and signaling interaction methods in AI-assisted sensing, thereby facilitating the practical implementation of AI-assisted sensing services. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating a perception model management method in one embodiment of this application;

[0015] Figure 2 This is a schematic diagram of the structure of a sensing system in one embodiment of this application;

[0016] Figure 3A This is a flowchart illustrating the sensing service registration process in one embodiment of this application;

[0017] Figure 3B This is a flowchart illustrating the perception model distribution process in one embodiment of this application.

[0018] Figure 3C This is a flowchart illustrating the perception model deletion process in one embodiment of this application;

[0019] Figure 4 This is a schematic diagram of the structure of a sensing system in one embodiment of this application;

[0020] Figure 5A This is a flowchart illustrating the perception model distribution process in one embodiment of this application.

[0021] Figure 5B This is a flowchart illustrating the perception model deletion process in one embodiment of this application;

[0022] Figure 6 This is a schematic diagram of the structure of a perception model management device in one embodiment of this application;

[0023] Figure 7 This is a hardware structure diagram of the management device in one embodiment of this application. Detailed Implementation

[0024] This application proposes a perception model management method, which can be applied to device management. See [link to relevant documentation]. Figure 1 The diagram shown is a flowchart of the perception model management method, which may include:

[0025] Step 101: Based on the address information of the stored model inference device, send a first request message to the model inference device. The first request message may include the model identifier and model size of the perception model. When the resource size of the model inference device in local storage is not less than the model size, send a first success response message to the management device. The first success response message includes the model identifier.

[0026] Step 102: If the first success response message is received, a model delivery message is sent to the model inference device. The model delivery message may include model data corresponding to the model identifier, so that the model inference device can generate a perception model corresponding to the model identifier based on the model data, input the acquired wireless channel data into the perception model for inference, and obtain the wireless perception result of the target object.

[0027] In one example, if the perception model is stored in a storage device, then the management device sending a model delivery message to the model inference device may include, but is not limited to: sending a second request message to the storage device based on the address information of the stored storage device. The second request message may include the model identifier of the perception model and the address information of the model inference device, so that the storage device can obtain the perception model corresponding to the model identifier and send a model delivery message to the model inference device based on the address information of the model inference device.

[0028] In one example, the model sends multiple messages. For each model message, the model data included in the message is a portion of the data from the perception model, and the model data included in all model messages constitute the perception model. The model message also includes a model identifier, the number of model data segments, and a model data index. The number of model data segments indicates how many pieces of model data the perception model is divided into, and the model data index indicates which piece of model data from the perception model is carried in the message.

[0029] In one example, after sending a model delivery message to the model inference device, a model deletion message can also be sent to the model inference device. The model deletion message includes a model identifier, so that the model inference device deletes the perception model corresponding to the model identifier and sends a second success response message to the management device. The second success response message includes the model identifier. If the second success response message is received, it is determined that the perception model has been successfully deleted.

[0030] In one example, a first model deletion message, including a model identifier, can be sent to the model inference device to cause the model inference device to delete the perception model corresponding to the model identifier; a second success response message, including the model identifier, can be sent to the management device; a second model deletion message, including the model identifier, can be sent to the storage device to cause the storage device to delete the perception model corresponding to the model identifier; and a third success response message, including the model identifier, can be sent to the management device. If the second success response message sent by the model inference device and the third success response message sent by the storage device are received, it is determined that the perception model has been successfully deleted.

[0031] In one example, before sending a first request message to the model inference device based on the acquired address information of the model inference device, the device may also receive a first registration message sent by the model inference device. The first registration message may include the address information of the model inference device, the number K of data acquisition devices supported by the model inference device, and the storage of the address information and the number K of the model inference device. The management device may trigger a maximum of K data acquisition devices to send wireless channel data to the model inference device.

[0032] In one example, before sending a first request message to the model inference device based on the acquired address information of the model inference device, a second registration message can also be received from the storage device. The second registration message includes the address information of the storage device, and the address information of the storage device.

[0033] In one example, the first registration message includes the following fields: serverIPv4Address, serverIPv6Address, port, maxApNumber, and protocol. The serverIPv4Address field carries the IPv4 address used by the model inference device to receive data, the serverIPv6Address field carries the IPv6 address used by the model inference device to receive data, the port field carries the port number used by the model inference device to receive data, the maxApNumber field carries the number of data acquisition devices supported by the model inference device, and the protocol field carries the protocol type of the data packets supported by the model inference device. The first request message includes the modelID and modelSize fields. The modelID field carries the model identifier of the sensing model, and the modelSize field carries the model size of the sensing model.

[0034] The first successful response message includes a modelID field and an ack field; the modelID field is used to carry the model identifier of the sensing model, and the ack field is used to indicate whether the sensing model can be received.

[0035] The model delivery message includes the modelID field, payloadNo field, payloadIndex field, and modelPayload field; wherein, the modelID field is used to carry the model identifier of the perception model, the payloadNo field is used to carry the number of model data segments, the payloadIndex field is used to carry the model data index, and the modelPayload field is used to carry the model data corresponding to the model identifier.

[0036] The second request message includes the modelID field, the inferIPv4Address field, the inferIPv6Address field, and the port field; the modelID field is used to carry the model identifier of the sensing model, the inferIPv4Address field is used to carry the IPv4 address of the model inference device, the inferIPv6Address field is used to carry the IPv6 address of the model inference device, and the port field is used to carry the port number of the model inference device.

[0037] As can be seen from the above technical solutions, in the embodiments of this application, the model inference device can input wireless channel data (such as data related to wireless signals in WLAN) into the perception model for inference to obtain the wireless perception result of the target object. In this way, the wireless perception result of the target object can be analyzed based on the perception model to obtain accurate and reliable wireless perception results.

[0038] The management equipment is responsible for the control and management of the entire sensing system and the management of the sensing models. It clarifies the registration and synchronization of sensing services in AI-assisted sensing, as well as the model management process and signaling interaction methods in AI-assisted sensing, thereby facilitating the practical implementation of AI-assisted sensing services.

[0039] The technical solutions described above in the embodiments of this application will be explained below in conjunction with specific application scenarios.

[0040] WLAN sensing is a technology that utilizes wireless signals to perform sensing tasks. Based on ubiquitous infrastructure and wireless signals, and because wireless signals can reflect, penetrate, and bend on object surfaces during propagation, signal processing can be performed on these wireless signals to sense the surrounding environment, detect obstacles, and detect target movement. WLAN sensing has been applied in various scenarios, such as gesture control, fall detection, tracking, imaging, activity recognition, and vital sign monitoring.

[0041] The core principle of WLAN sensing lies in the fact that when a person or object moves within the wireless coverage area, information such as the CSI and RSSI of the wireless signal changes accordingly. Key features can be extracted from the CSI and RSSI of the wireless signal, and WLAN sensing can be achieved based on these key features. However, there is currently no effective way to extract key features from the CSI and RSSI of wireless signals in related technologies; that is, it is impossible to effectively extract key features from the CSI and RSSI of wireless signals.

[0042] In response to the above findings, this application proposes a perception model management method that can effectively extract key features from the CSI and RSSI of wireless signals using an AI model (referred to as the perception model in this embodiment, i.e., the AI ​​model used to realize WLAN perception). Based on these key features, WLAN perception can be realized, and information such as the location, posture, and behavior patterns of target objects (such as target users) can be inferred.

[0043] See Figure 2 The diagram shows the structure of a sensing system. In the process of using a sensing model to assist WLAN sensing, the sensing system can include a data acquisition device, a model training device, a model inference device, and a management device. The data acquisition device can collect training data (wireless channel data, such as CSI data and / or RSSI data) and send the training data to the model training device. The model training device then trains the sensing model based on the training data, or updates the sensing model based on the training data.

[0044] The model training device sends the trained or updated perception model to the management device, which then stores the perception model. The management device can also control and manage the model training device.

[0045] The data acquisition device can collect inference data (wireless channel data) and send it to the model inference device. The model inference device then performs inference based on the data to obtain the inference result. The model inference device sends the inference result to the management device, which then receives the result. The management device can also control and manage the model inference device. Furthermore, the management device can also control and manage the data acquisition device, such as through control signaling interaction between the two devices.

[0046] The above process involves message interaction between different devices, such as message interaction between the management device and the model inference device, message interaction between the management device and the model training device, message interaction between the management device and the data acquisition device, message interaction between the data acquisition device and the model training device, and message interaction between the data acquisition device and the model inference device. The following describes these message interaction processes.

[0047] For data acquisition devices, the data acquisition devices are responsible for collecting wireless channel data, such as CSI data and / or RSSI data, and sending the collected wireless channel data to model training devices or model inference devices.

[0048] For model training equipment, the system is responsible for training a perception model based on input data (such as wireless channel data) and sending the perception model to the management equipment. After training the perception model, processes such as model verification and testing may also be involved. If the perception model passes model verification and testing, it is sent to the management equipment. Each perception model has a unique function and a unique model identifier (model ID). The model training equipment can train perception models offline or online. Different perception models have different functions. Multiple perception models can be trained from data collected by one data acquisition device, or a single perception model can be trained from data collected by multiple different data acquisition devices, or a single perception model can be trained from data collected by one data acquisition device.

[0049] For model inference devices, the model inference device is responsible for inference based on input data (such as wireless channel data). For example, the wireless channel data is input into the perception model, and the perception model performs perception inference to obtain the inference result. The model inference device can send the inference result to the management device.

[0050] The management device is responsible for controlling and managing the entire perception acquisition, training, and inference process. For example, it controls and manages data acquisition equipment, including data acquisition, updating, and termination. It controls and manages model training equipment, including training triggering, updating, and termination, and receives and saves the trained or updated perception model. It controls and manages the perception model itself, including its distribution (or updating) and deletion. It controls and manages the perception inference equipment, including triggering, updating, and terminating perception inference. The management device maintains the correspondence between perception models and datasets. As a server, the management device provides perception services to external clients.

[0051] The management device is responsible for the control and management of the entire sensing system. When other devices are started, they register with the management device to facilitate the establishment of subsequent sensing processes. Each device maintains synchronization with the management device during operation. The management device is also responsible for managing the sensing model. This embodiment focuses on control and management in AI-assisted sensing, such as the main processes of sensing service registration, synchronization, and model management, as well as the detailed parameter design of the signaling messages involved in the processes. Specifically, it proposes a model management process and signaling interaction method for AI-assisted sensing.

[0052] In one example, the management device (management entity) can be a management module, the data acquisition device (data acquisition entity) can be a data acquisition module, the model inference device (model inference entity) can be a model inference module, and the model training device (model training entity) can be a model training module. Based on this, different modules can be deployed on the same device or on different devices.

[0053] For example, the data acquisition module and the model inference module are deployed on the same device; the data acquisition module and the model training module are deployed on the same device; the model inference module and the model training module are deployed on the same device; the data acquisition module, the model inference module, and the model training module are deployed on the same device; and the data acquisition module, the model inference module, the model training module, and the management module are deployed on the same device.

[0054] In one example, an AP device and / or a wireless terminal can be used as a data acquisition device; that is, the data acquisition device is an AP device, or the data acquisition device is a wireless terminal, or the data acquisition device includes both an AP device and a wireless terminal. A perception server can be used as a model training device; that is, the model training device is a perception server, which is any server capable of completing model training, and may include a GPU (Graphics Processing Unit) unit or other computing units. An AC device or other platform can be used as a management device; that is, the management device can be an AC device.

[0055] The access point (AP) device and / or wireless terminal can be used as the model inference device; that is, the model inference device is an AP device, or the model inference device is a wireless terminal, or the model inference device includes both an AP device and a wireless terminal. In this case, the model inference device and the data acquisition device can be the same device.

[0056] Alternatively, the perception server can be used as the model inference device, that is, the model inference device is the perception server. In this case, the model inference device and the model training device can be the same device.

[0057] In the above application scenarios, this application proposes a perception model management method, involving processes such as perception service registration, perception service synchronization, perception model distribution, and perception model deletion. All messages involved in these processes can be referred to as perception business messages. A perception business message can include a message header and a perception business TLV (Type Length Value). The message header can be 12 bytes long or other bytes, and the perception business TLV can be N bytes long, where N is variable. See Table 1 for the message header format and field definitions; this is just an example of a message header. See Table 2 for the perception business TLV format and field definitions; this is just an example of a perception business TLV.

[0058] Table 1

[0059]

[0060] Table 2

[0061] type length load 2 Bytes 2 Bytes 0 to N Bytes

[0062] In Table 1, the message type is used to indicate the type of the sensing business message. For example, when the message type takes the first value, it means that the sensing business message is the first request message; when the message type takes the second value, it means that the sensing business message is the model-issued message; when the message type takes the third value, it means that the sensing business message is the second request message; when the message type takes the fourth value, it means that the sensing business message is the model deletion message, and so on. Different types of messages can correspond to different values.

[0063] The message type of the response message to the first request message is the first value, that is, the message type of the response message is the same as that of the first request message. The operation type indicates whether it is the first request message or the response message.

[0064] The message type of the response message to the model message is the second value, that is, the message type of the response message is the same as the message type of the model message. The operation type indicates whether it is a model message or a response message.

[0065] The message type of the response message to the second request message is a third value, meaning that the message type of the response message is the same as that of the second request message, and the operation type indicates whether it is the second request message or the response message.

[0066] The message type of the response message to the model deletion message takes the fourth value, meaning that the message type of the response message is the same as that of the model deletion message. The operation type indicates whether it is a model deletion message or a response message.

[0067] In Table 1, the operation type indicates the type of operation performed on the perceived business message. For example, operation type 1 indicates that the perceived business message is a request message, and this request message does not require a response message. Operation type 2 indicates that the perceived business message is a request message, and this request message requires a response message. Operation type 3 indicates that the perceived business message is a response message (response information).

[0068] In Table 2, the perception service TLV may include a type field, a length field, and a payload field. The content of the perception service TLV is described in subsequent embodiments. In the AI-assisted perception system, the type value in the perception service TLV can be predefined for perception service messages in different processes. For example, the type value of the TLV related to perception registration and synchronization can be 0x0000-0x00FF, meaning that values ​​within this range (0x0000-0x00FF) are used as the type value of the perception service TLV during the perception registration and synchronization process. The type value of the TLV related to perception data acquisition can be 0x0100-0x0FFF, meaning that values ​​within this range (0x0100-0x0FFF) are used as the type value of the perception service TLV during the perception data acquisition process. The type value of the TLV related to perception training can be 0x1000-0x1FFF, meaning that values ​​within this range (0x1000-0x1FFF) are used as the type value of the perception service TLV during the perception training process. The type value of the TLV related to perception reasoning can be 0x2000-0x2FFF, meaning that values ​​within this range are used as the type value for perception services during the perception reasoning process. Furthermore, the type value of the TLV related to perception model management can be 0x3000-0x3FFF, meaning that values ​​within this range are used as the type value for perception services during the perception model management process. The range 0x4000-0xFFFF can be reserved for other purposes.

[0069] First, regarding the registration process for perception services.

[0070] See Figure 3A The diagram shown illustrates the process of registering a perception service, which may include:

[0071] Step 301: The model inference device sends a first registration message to the management device. The first registration message may include the address information of the model inference device and the number K of data acquisition devices supported by the model inference device.

[0072] In one example, after the model inference device sends the first registration message to the management device, it can also start a timer, the timeout period of which is denoted as T. rtThe timeout period can be configured according to actual needs. If a response message for the first registration message is received before the timer expires, the timer is stopped. If no response message for the first registration message is received before the timer expires, the first registration message is resent, and then the timer is reset, and so on. The first registration message can be repeated at most N times. rep Next, such as N rep =3. After sending the first registration message 3 times, if no response message for the first registration message is received before the timer expires, the process is abandoned and the process is tried again after a period of time (e.g., 5 minutes).

[0073] In one example, after startup, the model inference device can send a first registration message to the management device. This first registration message includes the model inference device's address information, such as its IP address and port. The IP address can be an IPv4 address and / or an IPv6 address. This first registration message is also known as the Server2Mp_Registeration_Request message. See Table 3 for an example of a perception service TLV for the first registration message.

[0074] In addition, the first registration message may also include a message header. The format of the message header can be seen in Table 1. That is, the message type in the message header indicates that the perception business message is a registration message.

[0075] Table 3

[0076]

[0077]

[0078] maxApNumber represents the number K of data acquisition devices supported by the model inference device. For example, if K is 4, it means that the management device can trigger a maximum of 4 data acquisition devices to send wireless channel data to the model inference device, and the model inference device performs inference based on the wireless channel data of the 4 data acquisition devices.

[0079] Step 302: The management device receives the first registration message sent by the model inference device.

[0080] Step 303: The management device stores the address information and quantity K of the model inference device. The quantity K indicates that the management device can trigger a maximum of K data acquisition devices to send wireless channel data to the model inference device.

[0081] In one example, the management device can parse the address information of the model inference device, such as the IPv4 address, IPv6 address, and port, from the first registration message and store the address information of the model inference device.

[0082] The management device can parse the number K of data acquisition devices supported by the model inference device from the first registration message and store the number K. Based on this number K, the management device can trigger a maximum of K data acquisition devices to send wireless channel data to the model inference device. That is, during the inference process, the management device can send messages to a maximum of K data acquisition devices to enable these data acquisition devices to send wireless channel data to the model inference device, and the model inference device performs inference based on the wireless channel data of these data acquisition devices.

[0083] Step 304: The management device sends a response message to the model inference device in response to the first registration message.

[0084] In one example, after receiving the first registration message, the management device can set the model inference device's status to idle, indicating that the model inference device is ready to perform inference, and the management device can then assign inference tasks to it. The management device then sends a response message to the first registration message (i.e., a registration response message) to the model inference device. Upon receiving this registration response message, the model inference device is considered to have successfully registered and can prepare for subsequent operations.

[0085] In one example, the registration response message may include the storage device's address information, such as its IP address and port. The IP address can be an IPv4 address and / or an IPv6 address. In application scenarios where no storage device exists, the registration response message does not carry the storage device's address information. In application scenarios where a storage device exists, the registration response message may or may not carry the storage device's address information. This registration response message can also be called an Mp2Server_Registeration_Response message. See Table 4 for an example of a service-aware TLV for this registration response message.

[0086] In addition, the registration response message may also include a message header. The format of the message header can be seen in Table 1. That is, the message type in the message header indicates that the perception business message is a registration response message.

[0087] Table 4

[0088]

[0089]

[0090] This completes the registration process for the model inference device. Similarly, the model training device can also register with the management device, following a similar process. The data acquisition device can also register with the management device, again following a similar process.

[0091] Second, regarding the synchronization process of perception services.

[0092] After successful registration, the model inference device (model training device, data acquisition device) can periodically send synchronization messages, such as Hello messages, to the management device to indicate that the model inference device is working normally. For example, after receiving the registration response message, the model inference device sends a synchronization message to the management device for the first time, and then sends a synchronization message to the management device every M seconds thereafter.

[0093] For the management device, if it does not receive a synchronization message from the model inference device for more than M seconds, it is considered that the model inference device has malfunctioned. The management device will stop sending any messages to the model inference device until the model inference device successfully re-registers and receives a synchronization message from the model inference device again.

[0094] Third, regarding the perception model distribution process. The perception model distribution process includes the management device sending the perception model to the model inference device for the first time, and the management device updating the perception model to the model inference device.

[0095] See Figure 3B The diagram shown illustrates the process of distributing the perception model, which may include:

[0096] Step 311: The management device sends a first request message to the model inference device based on the stored address information of the model inference device. The first request message may include the model identifier and model size of the perception model.

[0097] In one example, after the management device sends a first request message to the model inference device, it can also start a timer. If a response message for the first request message is received before the timer expires, the timer is stopped. If no response message for the first request message is received before the timer expires, the first request message is resent, and the timer is reset, and so on. The first request message can be repeated at most N times. rep Second-rate.

[0098] In one example, during the perception service registration process, the management device has stored the address information of the model inference device, so that a first request message can be sent to the model inference device based on the address information.

[0099] For each perception model, its model identifier is unique and used to represent that perception model. Based on this, the first request message may include the model identifier of the perception model.

[0100] Furthermore, the management device can determine the model size of the sensing model, which indicates how much storage resources the sensing model requires—that is, how much storage resources are needed to store and run the sensing model. Based on this, the first request message can include the model size of the sensing model.

[0101] The first request message can also be called the Mp2Infer_Model_Delivery_Query message. See Table 5 for an example of a perception service TLV for the first request message. The first request message may also include a message header, where the message type indicates that the perception service message is a first request message.

[0102] Table 5

[0103]

[0104]

[0105] Step 312: The model inference device receives the first request message sent by the management device.

[0106] Step 313: The model inference device sends a first response message to the management device in response to the first request message. The first response message may be a first success response message or a first failure response message.

[0107] In one example, the model inference device can obtain the model size of the perception model from the first request message and determine the size of its local storage resources (i.e., the available resource size), which indicates how much available storage resources the model inference device has. If the size of the local storage resources is not less than the model size, it means that the available storage resources of the model inference device can receive the perception model. Therefore, the model inference device sends a first success response message to the management device, preparing to receive the perception model. This first success response message indicates that the available storage resources of the model inference device can receive the perception model. If the size of the local storage resources is less than the model size, it means that the available storage resources of the model inference device cannot receive the perception model. Therefore, the model inference device sends a first failure response message to the management device, indicating that the available storage resources of the model inference device cannot receive the perception model.

[0108] The model inference device can obtain the model identifier of the perception model from the first request message. When the model inference device sends the first success response message or the first failure response message to the management device, the first success response message or the first failure response message may also include the model identifier of the perception model.

[0109] This first response message can also be called the Infer2Mp_Model_Delivery_Response message. See Table 6 for an example of the awareness service TLV for this first response message. This first response message may also include a message header, indicating that the awareness service message is a first response message through the message type in the message header.

[0110] Table 6

[0111]

[0112] If the ack field takes the first value (e.g., 0), the first response message is a first failure response message, indicating that the model inference device cannot receive the perception model. If the ack field takes the second value (e.g., 1), the first response message is a first success response message, indicating that the model inference device can receive the perception model.

[0113] Step 314: The management device receives the first response message in response to the first request message.

[0114] For example, if the management device receives the first failure response message, it terminates the subsequent process and does not send the perception model to the model inference device. If the management device receives the first success response message, it continues the subsequent process and sends the perception model to the model inference device.

[0115] Step 315: If the first success response message is received, the management device sends a model distribution message to the model inference device. The model distribution message may include the model data corresponding to the model identifier.

[0116] In one example, the amount of data that a single model delivery message can carry is limited and may not be able to contain all the model data of the perception model. Therefore, if a single model delivery message can carry all the model data of the perception model, the management device sends one model delivery message to the model inference device, which includes all the model data of the perception model. If a single model delivery message cannot carry all the model data of the perception model, the management device sends multiple model delivery messages to the model inference device. For each model delivery message, the model data in that message is a portion of the perception model's data, and the model data in all the model delivery messages constitutes the perception model. For ease of description, the management device sends multiple model delivery messages to the model inference device, using multiple model delivery messages to carry all the model data of the perception model.

[0117] For example, the perception model can be divided into P model data, that is, the perception model is composed of P model data. In this way, the management device sends a model distribution message a-1 to the model inference device. The model distribution message a-1 includes the model data b-1 of the perception model. The management device sends a model distribution message a-2 to the model inference device. The model distribution message a-2 includes the model data b-2 of the perception model. And so on, the management device sends a model distribution message aP to the model inference device. The model distribution message aP includes the model data bP of the perception model.

[0118] In one example, for each model, a message is sent, which also includes a model identifier, the number of model data segments, and a model data index. The number of model data segments indicates how many model data segments the perception model is divided into. For example, if the number of model data segments is P, it means that the perception model is divided into P model data segments. The model data index indicates which model data segment of the perception model is carried in the message.

[0119] For example, when the management device sends a model delivery message a-1 to the model inference device, the model delivery message a-1 includes the model identifier of the perception model, the number of model data segments P, the model data index 1 (indicating that the model delivery message a-1 carries the first model data of the perception model) and the model data b-1 of the perception model.

[0120] When the management device sends a model delivery message a-2 to the model inference device, the model delivery message a-2 includes the model identifier of the perception model, the number of model data segments P, the model data index 2 (indicating that the model delivery message a-2 carries the second model data of the perception model) and the model data b-2 of the perception model, and so on.

[0121] In one example, for each model-delivered message, taking model-delivered message a-1 as an example, after the management device sends model-delivered message a-1 to the model inference device, it can also start a timer, meaning each model-delivered message has its own timer. If a response message for model-delivered message a-1 is received before the timer expires, the timer is closed. If no response message for model-delivered message a-1 is received before the timer expires, model-delivered message a-1 is resent, and then the timer is reset, and so on.

[0122] In one example, during the perception service registration process, the management device has stored the address information of the model inference device, so that each model delivery message is sent to the model inference device based on the address information.

[0123] The model-delivered message can also be called an Mp2Infer_Model_Delivery message. See Table 7 for an example of a perception service TLV for model-delivered messages. Model-delivered messages can also include a message header, with the message type indicating whether the perception service message is a model-delivered message.

[0124] Table 7

[0125]

[0126] In Table 7, the payloadNo field carries the number of model data segments P, i.e., the number of model load segments, indicating that the perception model is divided into P model data segments (model load). The payloadIndex field carries the model data index, i.e., the index of the segmented model load, indicating which model data segment (model load) the message sent by the model carries. The modelPayload field carries the actual model data (model load) of the perception model, i.e., the load of the trained perception model.

[0127] Step 316: The model inference device receives the model distribution message sent by the management device.

[0128] In one example, the model inference device can receive multiple model delivery messages. Since each model delivery message carries a model identifier, it can determine whether the delivery messages belong to the same perceptual model based on the model identifier, and generate the perceptual model corresponding to that model identifier based on the model data in these delivery messages. For instance, since each model delivery message carries the number P of model data segments and a model data index, the model data in the P delivery messages corresponding to that model identifier can be sorted in ascending order of the model data index. Based on the sorting result, the model data in the P delivery messages are combined to obtain the perceptual model, that is, the perceptual model corresponding to that model identifier is generated based on these model data. Thus, the perceptual model corresponding to that model identifier is obtained and stored in the local storage medium of the model inference device (such as a hard drive or memory).

[0129] For example, the model data index in model message a-1 is 1, the model data index in model message a-2 is 2, and so on. By sequentially arranging and combining the model data in model message a-1, model message a-2, ..., model message aP, a perceptual model is obtained. At this point, the model data can be assembled according to the order of the model data indices.

[0130] In one example, after obtaining the perception model, the model inference device can perform inference based on it. For instance, the model inference device can acquire wireless channel data (such as CSI information and / or RSSI information), which can be collected by a data acquisition device and sent to the model inference device. The model inference device can input the acquired wireless channel data into the perception model for inference to obtain the wireless perception result of the target object. The wireless perception result of the target object can be the inference result of the perception model. In this embodiment, the inference process of the perception model is not limited.

[0131] For example, the wireless sensing result may include, but is not limited to, whether a target object (such as a target user) exists in a specified scene. If a target object exists in the specified scene, the wireless sensing result may include, but is not limited to, at least one of the following: the target object's location, the target object's posture, and the target object's behavior pattern. The target object's location refers to its physical location, such as latitude and longitude coordinates. The target object's posture refers to whether the target object is standing, walking, or sitting. The target object's behavior pattern refers to what behavior the target object is performing. Of course, these are just a few examples of wireless sensing results and are not limited to, such as the number of target objects, breathing, heartbeat, etc.

[0132] Step 317: For each model, after receiving the model-issued message, the model inference device sends a response message to the management device for that model-issued message.

[0133] In one example, the model inference device can send a success response message to the management device, indicating that the model delivery message has been successfully received. The model inference device can also send a failure response message to the management device, indicating that the model delivery message was not successfully received. The model inference device can obtain the model identifier of the perception model from the model delivery message; the success or failure response message includes the model identifier of the perception model when sending it to the management device.

[0134] This response message can also be called the Infer2Mp_Model_Delivery_Ack message. See Table 8 for an example of the awareness service TLV for this response message. This response message may also include a message header, which indicates that the awareness service message is a response message to a model-delivered message, based on the message type in the header.

[0135] Table 8

[0136]

[0137] Step 318: The management device receives a response message for the message sent to the model.

[0138] For example, when sending a message to each model, if the management device receives a success response message for that model's message, it terminates the subsequent process and completes the transmission of that model's message. If the management device receives a failure response message for that model's message, it resends the model's message.

[0139] Fourth, regarding the process of deleting a perception model. When a perception model is no longer in use, or when a new perception model replaces an existing one, the management device can initiate a model deletion process to release storage space in the model inference device and save storage resources. The following explains this perception model deletion process.

[0140] See Figure 3C The diagram shown illustrates the process of deleting a perception model, which may include:

[0141] Step 321: The management device sends a model deletion message to the model inference device based on the stored address information of the model inference device. The model deletion message may include the model identifier.

[0142] In one example, after the management device sends a model deletion message to the model inference device, it can also start a timer. If a response message for the model deletion message is received before the timer expires, the timer is stopped. If no response message for the model deletion message is received before the timer expires, the model deletion message is resent, and then the timer is reset, and so on. The model deletion message can be repeated at most N times. rep Second-rate.

[0143] Model deletion messages can also be called Mp2Infer_Model_Delete_Request messages. See Table 9 for an example of a perceptual service TLV for a model deletion message. Model deletion messages may also include a message header, with the message type in the header indicating that the perceptual service message is a model deletion message.

[0144] Table 9

[0145]

[0146] Step 322: The model inference device receives the model deletion message and deletes the perception model corresponding to the model identifier.

[0147] In one example, the model inference device can obtain the model identifier from the model deletion message and delete the perceptual model corresponding to the model identifier from the local storage medium, no longer using the perceptual model for inference.

[0148] Step 323: The model inference device sends a second response message to the management device in response to the model deletion message. The second response message can be a second success response message or a second failure response message.

[0149] In one example, if the model inference device successfully deletes the perception model, it sends a second success response message to the management device; if it fails to delete the perception model, it sends a second failure response message to the management device. The second success or failure response message includes the model identifier.

[0150] This second response message can also be called the Infer2Mp_Model_Delete_Response message. See Table 10 for an example of a perception service TLV for this second response message. This second response message may also include a message header, indicating that the perception service message is a second response message through the message type in the message header.

[0151] Table 10

[0152]

[0153] Step 324: The management device receives a second response message from the model inference device. If the second response message is a second success response message, the management device determines that the model inference device has successfully deleted the perception model corresponding to the model identifier. If the second response message is a second failure response message, the management device determines that the model inference device has not successfully deleted the perception model corresponding to the model identifier.

[0154] See Figure 4 The diagram shown illustrates the structure of a sensing system, which may include data acquisition equipment, model training equipment, model inference equipment, management equipment, and storage equipment. Figure 2 Compared to the perception system, the additional storage device is required. Figure 2 In the middle, the sensing model is stored by the management device, in Figure 4 In this context, the storage device can store the awareness model, and the management device can store the model identifier of the awareness model.

[0155] In the above application scenarios, this application proposes a perception model management method, which involves the processes of perception service registration, perception service synchronization, perception model distribution, and perception model deletion.

[0156] Fifth, regarding the registration process for perception services.

[0157] The model inference device sends a first registration message to the management device. This first registration message includes the model inference device's address information and the number K of data acquisition devices supported by the model inference device. The management device receives the first registration message from the model inference device. The management device stores the model inference device's address information and the number K, where K represents the maximum number of data acquisition devices the management device can trigger to send wireless channel data to the model inference device. The management device then sends a response message to the model inference device in response to the first registration message.

[0158] The storage device sends a second registration message to the management device, the second registration message including the storage device's address information. The management device receives the second registration message sent by the storage device. The management device stores the address information of the storage device. The management device sends a response message to the storage device regarding the second registration message.

[0159] The process of registering perception services is similar to that in the first point, and will not be described again here.

[0160] Sixth, regarding the synchronization process of perception services.

[0161] After successful registration, the model inference device (model training device, data acquisition device, storage device) can periodically send synchronization messages, such as Hello messages, to the management device.

[0162] The process of synchronizing perception services is similar to that in point two, and will not be repeated here.

[0163] Seventh, regarding the perception model distribution process. The perception model distribution process includes the management device sending the perception model to the model inference device for the first time, and the management device updating the perception model to the model inference device.

[0164] See Figure 5A The diagram shown illustrates the process of distributing the perception model, which may include:

[0165] Step 511: The management device sends a first request message to the model inference device based on the stored address information of the model inference device. The first request message may include the model identifier and model size of the perception model.

[0166] Step 512: The model inference device receives the first request message sent by the management device.

[0167] Step 513: The model inference device sends a first response message to the management device in response to the first request message. The first response message may be a first success response message or a first failure response message.

[0168] For example, if the size of the local storage resource is not less than the model size, the model inference device can send a first success response message to the management device; if the size of the local storage resource is less than the model size, the model inference device can send a first failure response message to the management device.

[0169] Step 514: The management device receives the first response message in response to the first request message.

[0170] For example, if the management device receives the first failure response message, it terminates the subsequent process and does not send the perception model to the model inference device. If the management device receives the first success response message, it continues the subsequent process and sends the perception model to the model inference device.

[0171] In one example, steps 511-514 refer to steps 311-314, and will not be repeated here.

[0172] Step 515: If the first success response message is received, the management device sends a second request message to the storage device based on the address information of the stored storage device. The second request message may include the model identifier of the perception model and the address information of the model inference device, such as IP address and port address information.

[0173] In one example, after the management device sends a second request message to the storage device, it can also start a timer. If a response message for the second request message is received before the timer expires, the timer is stopped. If no response message for the second request message is received when the timer expires, the second request message is resent, and then the timer is reset, and so on. The second request message can be repeated at most N times. rep Second-rate.

[0174] In one example, during the sensing service registration process, the management device has stored the address information of the storage device, so that a second request message can be sent to the storage device based on the address information.

[0175] Since the management device has stored the address information of the model inference device during the perception service registration process, the second request information may include the address information of the model inference device. Furthermore, the second request message may also include the model identifier of the perception model, which is unique.

[0176] The second request message can also be called an Mp2Storage_Model_Deliver_Request message. See Table 11 for an example of a second request message for a service-aware TLV. The second request message may also include a message header, with the message type indicating that the service-aware message is a second request message.

[0177] Table 11

[0178]

[0179] Step 516: The storage device receives the second request message and sends a model delivery message to the model inference device. The model delivery message may include model data corresponding to the model identifier.

[0180] In one example, the storage device can obtain a model identifier from the second request message. If the storage device has a perception model corresponding to the model identifier and the perception model is functioning correctly, then the storage device obtains the address information of the model inference device from the second request message. Based on the address information of the model inference device, the storage device sends a model delivery message to the model inference device. If a single model delivery message can carry all the model data of the perception model, the storage device sends one model delivery message to the model inference device, which includes all the model data of the perception model. If a single model delivery message cannot carry all the model data of the perception model, the storage device sends multiple model delivery messages to the model inference device. For each model delivery message, the model data in that message is a portion of the perception model's data, and the model data in all the model delivery messages constitute the perception model. This example illustrates how the storage device sends multiple model delivery messages to the model inference device, carrying all the model data of the perception model through these multiple messages.

[0181] In one example, for each model, a message is sent, which also includes a model identifier, the number of model data segments, and a model data index. The number of model data segments indicates how many model data segments the perception model is divided into. For example, if the number of model data segments is P, it means that the perception model is divided into P model data segments. The model data index indicates which model data segment of the perception model is carried in the message.

[0182] For each model delivery message, after the storage device sends the message to the model inference device, it can also start a timer; that is, each model delivery message has its own dedicated timer. If a response message for the model delivery message is received before the timer expires, the timer is closed. If no response message is received before the timer expires, the model delivery message is resent, and the timer is reset.

[0183] This model-delivered message can also be called a Storage2Infer_Model_Deliver message. See Table 12 for an example of a perception service TLV for model-delivered messages. Model-delivered messages can also include a message header; the message type in the header indicates that the perception service message is a model-delivered message.

[0184] Table 12

[0185]

[0186]

[0187] In Table 12, the payloadNo field carries the number of model data segments P, i.e., the number of model load segments, indicating that the perception model is divided into P model data segments (model load). The payloadIndex field carries the model data index, i.e., the index of the segmented model load, indicating which model data segment (model load) the message sent by the model carries. The modelPayload field carries the actual model data (model load) of the perception model, i.e., the load of the trained perception model.

[0188] Step 517: The model inference device receives the model delivery message sent by the storage device.

[0189] In one example, the model inference device can receive multiple model delivery messages, identify those belonging to the same sensing model based on model identifiers, and generate a sensing model corresponding to that model identifier based on the model data in these delivery messages. For instance, it can sort the model data in P delivery messages corresponding to that model identifier in ascending order of model data index. Based on the sorting result, it can combine the model data from the P delivery messages to obtain a sensing model. After obtaining this sensing model, the model inference device can perform inference based on it. For example, the model inference device can input acquired wireless channel data into the sensing model for inference to obtain the wireless sensing results for the target object.

[0190] Step 518: For each model-issued message, after receiving the model-issued message, the model inference device sends a response message for the model-issued message to the storage device.

[0191] In one example, the model inference device can send a success response message to the storage device, indicating that the model delivery message has been successfully received. The model inference device can also send a failure response message to the storage device, indicating that the model delivery message was not successfully received. When sending a success or failure response message, the message includes the model identifier of the perceptual model.

[0192] This response message can also be called the Infer2Storage_Model_Deliver_Ack message. See Table 13 for an example of the awareness service TLV for this response message. This response message may also include a message header, which indicates that the awareness service message is a response message to a model-delivered message, based on the message type in the header.

[0193] Table 13

[0194]

[0195]

[0196] Step 519: The storage device receives a response message for the message sent to the model.

[0197] For example, for each model message, if the storage device receives a success response message for that model message, it terminates the subsequent process and completes the transmission of that model message. If the storage device receives a failure response message for that model message, it retransmits the model message.

[0198] Step 520: The storage device sends a response message to the management device.

[0199] In one example, if the storage device receives a success response message for all model delivery messages, indicating that the transmission of all model delivery messages is complete, the storage device sends a success response message to the management device. This success response message indicates that all model delivery messages have been successfully transmitted, and the management device determines that the model delivery task has been completed. If the storage device does not receive a success response message for all model delivery messages, indicating that the transmission of all model delivery messages is not complete, the storage device sends a failure response message to the management device. This failure response message indicates that the transmission of all model delivery messages was not successful, and the management device determines that the model delivery task was not completed, meaning the model delivery process failed. For example, when sending a success or failure response message, the success or failure response message may include the model identifier of the sensed model.

[0200] This response message can also be called a Storage2MP_Model_Deliver_Response message. See Table 14 for an example of a service-aware TLV for this response message. This response message may also include a message header, which indicates that the service-aware message is a response message based on the message type.

[0201] Table 14

[0202]

[0203] Eighth, regarding the process of deleting a perception model. When a perception model is no longer in use, or when a new perception model replaces an existing one, the management device can initiate a model deletion process to release storage space in the model inference device and save storage resources. The following explains this perception model deletion process.

[0204] See Figure 5BThe diagram shown illustrates the process of deleting a perception model, which may include:

[0205] Step 521: The management device sends a first model deletion message to the model inference device based on the stored address information of the model inference device. The first model deletion message may include the model identifier.

[0206] In one example, after the management device sends a first model deletion message to the model inference device, it can also start a timer. If a response message for the first model deletion message is received before the timer expires, the timer can be stopped. If no response message for the first model deletion message is received when the timer expires, the model deletion message is resent, the timer is reset, and so on.

[0207] The first model deletion message is also called the Mp2Infer_Model_Delete_Request message. See Table 9 for an example of a first model deletion message in the awareness service TLV. The first model deletion message may include a message header; the message type in the header indicates that the awareness service message is a first model deletion message.

[0208] Step 522: The model inference device receives the first model deletion message, obtains the model identifier from the first model deletion message, and deletes the perceptual model corresponding to the model identifier from the local storage medium.

[0209] Step 523: The model inference device sends a second response message to the management device in response to the first model deletion message. The second response message can be a second success response message or a second failure response message.

[0210] The second success response message or the second failure response message may include the model identifier. For example, the second response message may also be called the Infer2Mp_Model_Delete_Response message. See Table 10 for an example of the perceived service TLV for this second response message.

[0211] Step 524: The management device receives a second response message from the model inference device. If the second response message is a second success response message, the management device determines that the model inference device has successfully deleted the perception model corresponding to the model identifier. If the second response message is a second failure response message, the management device determines that the model inference device has not successfully deleted the perception model corresponding to the model identifier.

[0212] Step 525: The management device sends a second model deletion message to the storage device based on the stored address information of the storage device. The second model deletion message may include a model identifier.

[0213] In one example, after the management device sends a second model deletion message to the storage device, it can also start a timer. If a response message for the second model deletion message is received before the timer expires, the timer can be stopped. If no response message for the second model deletion message is received when the timer expires, the management device resends the model deletion message, then resets the timer, and so on.

[0214] The second model deletion message is also called the Mp2Storage_Model_Delete_Request message. See Table 15 for an example of a second model deletion message in the awareness service TLV. The second model deletion message includes a message header; the message type in the header indicates that the awareness service message is a second model deletion message.

[0215] Table 15

[0216]

[0217] Step 526: The storage device receives the second model deletion message, obtains the model identifier from the second model deletion message, and deletes the perception model corresponding to the model identifier from the local storage medium.

[0218] Step 527: The storage device sends a third response message to the management device in response to the second model deletion message. The third response message can be a third success response message or a third failure response message.

[0219] In one example, if the storage device successfully deletes the sensing model, it sends a third success response message to the management device; if the storage device fails to delete the sensing model, it sends a third failure response message to the management device. When sending the third success or failure response message, the storage device includes the model identifier.

[0220] This third response message is also known as the Storage2MP_Model_Deliver_Response message. See Table 16 for an example of a perceptual service TLV for this third response message. This third response message may also include a message header, indicating that the perceptual service message is a third response message through the message type in the message header.

[0221] Table 16

[0222]

[0223] Step 528: The management device receives a third response message from the storage device. If the third response message is a third success response message, the management device determines that the storage device has successfully deleted the perception model corresponding to the model identifier. If the third response message is a third failure response message, the management device determines that the storage device has not successfully deleted the perception model corresponding to the model identifier. In summary, if the management device receives both the second success response message from the model inference device and the third success response message from the storage device, it determines that the perception model has been successfully deleted, thus completing the perception model deletion process.

[0224] As can be seen from the above technical solutions, in this embodiment, the management device is responsible for the control and management of the entire perception system, and the management device is responsible for the management of the perception model. It clarifies the registration and synchronization of perception services in AI-assisted perception, the process and signaling interaction method of model management in AI-assisted perception, and the signaling interaction process and specific message structure between various devices (such as the management device and the model inference device) during registration, synchronization and management in AI-assisted perception. This is conducive to the actual implementation of AI-assisted perception services.

[0225] Based on the same concept as the methods described above, this application proposes a perception model management device for managing equipment. See [link to relevant documentation]. Figure 6 The diagram shown is a structural schematic of the device, which includes:

[0226] The sending module 61 is configured to send a first request message to the model inference device based on the stored address information of the model inference device. The first request message includes a model identifier of the perception model and a model size of the perception model. When the model inference device has local storage resources that are not less than the model size, it sends a first success response message to the management device. The first success response message includes the model identifier. The receiving module 62 is configured to receive the first success response message. The sending module 61 is also configured to send a model distribution message to the model inference device. The model distribution message includes model data corresponding to the model identifier. This enables the model inference device to generate a perception model corresponding to the model identifier based on the model data, and input the acquired wireless channel data into the perception model for inference to obtain the wireless perception result of the target object.

[0227] In one example, if the perception model is stored in a storage device, when the sending module 61 sends a model delivery message to the model inference device, it is specifically used to: send a second request message to the storage device based on the address information of the stored storage device. The second request message includes the model identifier of the perception model and the address information of the model inference device, so that the storage device can obtain the perception model corresponding to the model identifier and send the model delivery message to the model inference device based on the address information of the model inference device.

[0228] In one example, the model sends multiple messages. For each model message, the model data included in the message is a portion of the data of the perception model. The model data included in all model messages constitute the perception model. The model message also includes the model identifier, the number of model data segments, and the model data index. The number of model data segments indicates how many pieces of model data the perception model is divided into, and the model data index indicates which piece of model data of the perception model is carried in the model message.

[0229] In one example, the sending module 61 is further configured to send a model deletion message to the model inference device, the model deletion message including the model identifier, so that the model inference device deletes the perception model corresponding to the model identifier, and send a second success response message to the management device, the second success response message including the model identifier; the receiving module 62 is further configured to receive the second success response message; the device further includes: a determining module, configured to determine that the perception model has been successfully deleted.

[0230] In one example, the sending module 61 is further configured to send a first model deletion message to the model inference device, the first model deletion message including the model identifier, so that the model inference device deletes the perception model corresponding to the model identifier; send a second success response message to the management device, the second success response message including the model identifier; send a second model deletion message to the storage device, the second model deletion message including the model identifier, so that the storage device deletes the perception model corresponding to the model identifier; and send a third success response message to the management device, the third success response message including the model identifier; the receiving module 62 is further configured to receive the second success response message sent by the model inference device and the third success response message sent by the storage device; the device further includes: a determining module, configured to determine that the perception model has been successfully deleted.

[0231] In one example, the receiving module 62 is further configured to receive a first registration message sent by the model inference device, the first registration message including the address information of the model inference device and the number K of data acquisition devices supported by the model inference device; and to receive a second registration message sent by the storage device, the second registration message including the address information of the storage device;

[0232] The device further includes: a storage module for storing the address information of the model inference device and the quantity K; wherein the management device can trigger up to K data acquisition devices to send wireless channel data to the model inference device; and storing the address information of the storage device.

[0233] In one example, the first registration message includes the following fields: serverIPv4Address, serverIPv6Address, port, maxApNumber, and protocol. The serverIPv4Address field carries the IPv4 address used by the model inference device to receive data, the serverIPv6Address field carries the IPv6 address used by the model inference device to receive data, the port field carries the port number used by the model inference device to receive data, the maxApNumber field carries the number of data acquisition devices supported by the model inference device, and the protocol field carries the protocol type of the data packets supported by the model inference device. The first request message includes the modelID and modelSize fields. The modelID field carries the model identifier of the sensing model, and the modelSize field carries the model size of the sensing model.

[0234] The first successful response message includes a modelID field and an ack field; the modelID field is used to carry the model identifier of the sensing model, and the ack field is used to indicate whether the sensing model can be received.

[0235] The model delivery message includes the modelID field, payloadNo field, payloadIndex field, and modelPayload field; wherein, the modelID field is used to carry the model identifier of the perception model, the payloadNo field is used to carry the number of model data segments, the payloadIndex field is used to carry the model data index, and the modelPayload field is used to carry the model data corresponding to the model identifier.

[0236] The second request message includes the modelID field, the inferIPv4Address field, the inferIPv6Address field, and the port field; the modelID field is used to carry the model identifier of the sensing model, the inferIPv4Address field is used to carry the IPv4 address of the model inference device, the inferIPv6Address field is used to carry the IPv6 address of the model inference device, and the port field is used to carry the port number of the model inference device.

[0237] Based on the same application concept as the above method, this application proposes a management device, see [link to relevant documentation]. Figure 7 As shown, the management device includes a processor 71 and a machine-readable storage medium 72, the machine-readable storage medium 72 storing machine-executable instructions that can be executed by the processor 71; the processor 71 is used to execute the machine-executable instructions to implement the perception model management method disclosed in the above example of this application.

[0238] Based on the same concept as the methods described above, this application also provides a machine-readable storage medium storing a plurality of computer instructions. When these computer instructions are executed by a processor, they can implement the perception model management method disclosed in the examples above. The machine-readable storage medium can be any electronic, magnetic, optical, or other physical storage device, and can contain or store information such as executable instructions, data, etc. For example, the machine-readable storage medium can be: RAM (Random Access Memory), volatile memory, non-volatile memory, flash memory, storage drive (such as hard disk drive), solid-state drive, any type of storage disk (such as optical disc, DVD, etc.), or similar storage media, or combinations thereof.

[0239] Based on the same concept as the methods described above, this application also provides a computer program product, which may include a computer program. When executed by a processor, the computer program implements the perception model management method disclosed in the examples above.

[0240] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, embodiments of this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0241] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A perception model management method, characterized in that, Applied to the management of equipment, the method includes: Based on the address information of the stored model inference device, a first request message is sent to the model inference device. The first request message includes the model identifier and model size of the perception model. When the resource size of the model inference device in local storage is not less than the model size, a first success response message is sent to the management device. The first success response message includes the model identifier. If the first success response message is received, a model delivery message is sent to the model inference device. The model delivery message includes model data corresponding to the model identifier, so that the model inference device generates a perception model corresponding to the model identifier based on the model data, and inputs the acquired wireless channel data into the perception model for inference to obtain the wireless perception result of the target object.

2. The method according to claim 1, characterized in that, If the perception model is stored in a storage device, sending the model delivery message to the model inference device includes: Based on the stored address information of the storage device, a second request message is sent to the storage device. The second request message includes the model identifier of the perception model and the address information of the model inference device, so that the storage device can obtain the perception model corresponding to the model identifier and send the model delivery message to the model inference device based on the address information of the model inference device.

3. The method according to claim 1, characterized in that, The model sends out multiple messages. For each model message, the model data included in the message is a part of the data of the perception model, and the model data included in all the model messages constitute the perception model. The message sent by the model also includes the model identifier, the number of model data segments, and the model data index; the number of model data segments indicates how many model data segments the perception model is divided into, and the model data index indicates which model data segment of the perception model is carried in the message sent by the model.

4. The method according to claim 1, characterized in that, After sending the model delivery message to the model inference device, the method further includes: Send a model deletion message to the model inference device, the model deletion message including the model identifier, so that the model inference device deletes the perception model corresponding to the model identifier; and send a second success response message to the management device, the second success response message including the model identifier. If the second success response message is received, it is determined that the perception model has been successfully deleted; Alternatively, a first model deletion message is sent to the model inference device, the first model deletion message including the model identifier, so that the model inference device deletes the perception model corresponding to the model identifier, and a second success response message is sent to the management device, the second success response message including the model identifier; Send a second model deletion message to the storage device, the second model deletion message including the model identifier, so that the storage device deletes the perception model corresponding to the model identifier; send a third success response message to the management device, the third success response message including the model identifier; If the second success response message sent by the model inference device and the third success response message sent by the storage device are received, it is determined that the perception model has been successfully deleted.

5. The method according to claim 1, characterized in that, Before sending the first request message to the model inference device based on the acquired address information of the model inference device, the method further includes: Receive a first registration message sent by the model inference device, the first registration message including the address information of the model inference device and the number K of data acquisition devices supported by the model inference device; The address information of the model inference device and the quantity K are stored; wherein, the management device triggers a maximum of K data acquisition devices to send wireless channel data to the model inference device; Receive a second registration message sent by the storage device, the second registration message including the address information of the storage device, and store the address information of the storage device.

6. The method according to any one of claims 1-5, characterized in that, The first registration message includes the following fields: serverIPv4Address, serverIPv6Address, port, maxApNumber, and protocol. The serverIPv4Address field carries the IPv4 address used by the model inference device to receive data; the serverIPv6Address field carries the IPv6 address used by the model inference device to receive data; the port field carries the port number used by the model inference device to receive data; the maxApNumber field carries the number of data acquisition devices supported by the model inference device; and the protocol field carries the protocol type of the data packets supported by the model inference device. The first request message includes a modelID field and a modelSize field; where the modelID field is used to carry the model identifier of the perception model, and the modelSize field is used to carry the model size of the perception model; The first successful response message includes a modelID field and an ack field; the modelID field is used to carry the model identifier of the sensing model, and the ack field is used to indicate whether the sensing model can be received. The model delivery message includes the modelID field, payloadNo field, payloadIndex field, and modelPayload field; wherein, the modelID field is used to carry the model identifier of the perception model, the payloadNo field is used to carry the number of model data segments, the payloadIndex field is used to carry the model data index, and the modelPayload field is used to carry the model data corresponding to the model identifier. The second request message includes the modelID field, the inferIPv4Address field, the inferIPv6Address field, and the port field; the modelID field is used to carry the model identifier of the sensing model, the inferIPv4Address field is used to carry the IPv4 address of the model inference device, the inferIPv6Address field is used to carry the IPv6 address of the model inference device, and the port field is used to carry the port number of the model inference device.

7. A sensing model management device, characterized in that, Applied to management equipment, the device includes: The sending module is configured to send a first request message to the model inference device based on the stored address information of the model inference device. The first request message includes the model identifier and model size of the perception model. When the model inference device stores local resources with a resource size not less than the model size, the module sends a first success response message to the management device. The first success response message includes the model identifier. The receiving module is used to receive the first success response message; The sending module is further configured to send a model delivery message to the model inference device. The model delivery message includes model data corresponding to the model identifier, so that the model inference device generates a perception model corresponding to the model identifier based on the model data, and inputs the acquired wireless channel data into the perception model for inference to obtain the wireless perception result of the target object.

8. The apparatus according to claim 7, characterized in that, If the perception model is stored in a storage device, the sending module is specifically used to send a model delivery message to the model inference device when it sends the model delivery message: Based on the stored address information of the storage device, a second request message is sent to the storage device. The second request message includes the model identifier of the perception model and the address information of the model inference device, so that the storage device can obtain the perception model corresponding to the model identifier and send the model delivery message to the model inference device based on the address information of the model inference device.

9. The apparatus according to claim 7, characterized in that, The model sends out multiple messages. For each model message, the model data included in the message is a part of the data of the perception model, and the model data included in all the model messages constitute the perception model. The message sent by the model also includes the model identifier, the number of model data segments, and the model data index; the number of model data segments indicates how many model data segments the perception model is divided into, and the model data index indicates which model data segment of the perception model is carried in the message sent by the model.

10. The apparatus according to claim 7, characterized in that, The sending module is further configured to send a model deletion message to the model inference device, the model deletion message including the model identifier, so that the model inference device deletes the perception model corresponding to the model identifier, and send a second success response message to the management device, the second success response message including the model identifier; The receiving module is further configured to receive the second success response message; The device further includes: a determination module, used to determine that the perception model has been successfully deleted; Alternatively, the sending module is further configured to send a first model deletion message to the model inference device, the first model deletion message including the model identifier, so that the model inference device deletes the perception model corresponding to the model identifier; send a second success response message to the management device, the second success response message including the model identifier; send a second model deletion message to the storage device, the second model deletion message including the model identifier, so that the storage device deletes the perception model corresponding to the model identifier; and send a third success response message to the management device, the third success response message including the model identifier. The receiving module is further configured to receive the second success response message sent by the model inference device and the third success response message sent by the storage device; The device further includes a determination module for determining that the perception model has been successfully deleted.

11. The apparatus according to claim 7, characterized in that, The receiving module is configured to receive a first registration message sent by the model inference device, wherein the first registration message includes the address information of the model inference device and the number K of data acquisition devices supported by the model inference device; And, receive a second registration message sent by the storage device, the second registration message including the address information of the storage device; The device further includes: a storage module for storing the address information of the model inference device and the quantity K; wherein the management device can trigger up to K data acquisition devices to send wireless channel data to the model inference device; and storing the address information of the storage device.

12. The apparatus according to any one of claims 7-11, characterized in that, The first registration message includes the following fields: serverIPv4Address, serverIPv6Address, port, maxApNumber, and protocol. The serverIPv4Address field carries the IPv4 address used by the model inference device to receive data; the serverIPv6Address field carries the IPv6 address used by the model inference device to receive data; the port field carries the port number used by the model inference device to receive data; the maxApNumber field carries the number of data acquisition devices supported by the model inference device; and the protocol field carries the protocol type of the data packets supported by the model inference device. The first request message includes a modelID field and a modelSize field; where the modelID field is used to carry the model identifier of the perception model, and the modelSize field is used to carry the model size of the perception model; The first successful response message includes a modelID field and an ack field; the modelID field is used to carry the model identifier of the sensing model, and the ack field is used to indicate whether the sensing model can be received. The model delivery message includes the modelID field, payloadNo field, payloadIndex field, and modelPayload field; wherein, the modelID field is used to carry the model identifier of the perception model, the payloadNo field is used to carry the number of model data segments, the payloadIndex field is used to carry the model data index, and the modelPayload field is used to carry the model data corresponding to the model identifier. The second request message includes the modelID field, the inferIPv4Address field, the inferIPv6Address field, and the port field; the modelID field is used to carry the model identifier of the sensing model, the inferIPv4Address field is used to carry the IPv4 address of the model inference device, the inferIPv6Address field is used to carry the IPv6 address of the model inference device, and the port field is used to carry the port number of the model inference device.

Images