An AI glasses collaborative control system and method based on an offline AI small host

Abstract

The application relates to an AI glasses cooperative control system and method based on an offline AI small host computer, the system comprising an AI glasses body and an offline AI small host computer integrated in the inside of a glasses box; the AI glasses body comprises a main control module, and a scene sensing module, a function execution module, a temporary storage unit and a power supply module which are electrically connected with the main control module respectively; the offline AI small host computer comprises an AI computing power chip, an offline storage unit, an offline model training unit, a power management unit and a dual-mode communication unit which are integrated on a PCB in the inside of the glasses box. The application realizes the division of labor and cooperation of "the offline AI small host computer is responsible for core computing power processing, and the AI glasses body is responsible for scene sensing and function execution", realizes automatic identification of multiple scenes such as meeting rooms, outdoors and driving, and automatic switching of corresponding functions without feeling, and simultaneously solves the technical short board related to existing computing power.

Description

Technical Field

[0001] This invention relates to an AI glasses collaborative control system and method based on an offline AI mini-host, belonging to the technical field. Background Technology

[0002] With the deep integration of AI technology, sensor technology and wearable devices, AI glasses have become a popular product in the consumer electronics field. They integrate multiple functions such as recording and transcription, navigation, HUD display, and voice interaction, and are widely used in various daily scenarios such as conference room office, outdoor travel, and driving, becoming users' personal intelligent assistant.

[0003] Currently, mainstream AI glasses and related technologies on the market still have many core technical shortcomings that fail to meet users' actual needs. Specifically: existing AI glasses mostly rely on cloud computing power or the computing power of the AI ​​glasses themselves for scene recognition and computing power processing. If relying on cloud computing power, the usage scenarios are limited in the absence of network or weak network scenarios (such as underground parking garages, remote outdoor areas, and conference rooms without network). At the same time, users' private data such as meeting recordings and location information need to be uploaded to the cloud, which poses a risk of data leakage. If relying on the computing power of the AI ​​glasses themselves, the small size of the glasses and the power consumption control requirements prevent the integration of high-performance computing chips, resulting in low scene recognition accuracy and slow response. In addition, continuous local computing power will consume a lot of power, shorten the battery life, and affect the user experience.

[0004] Furthermore, in existing technologies, the glasses case is merely a storage and charging tool for AI glasses, with limited functionality. It is not endowed with core functions such as computing power processing, data storage, and model training, and cannot realize its value as a carrier that "interacts frequently with the glasses and has AI host computing capabilities." This results in low resource utilization and fails to support the functional optimization of AI glasses, leading to severe homogenization and competition in the industry.

[0005] The following schemes were disclosed in the relevant existing patent literature after searching: 1. The invention patent application with publication number CN120616237A discloses an intelligent glasses box based on artificial intelligence and its wireless charging control method. Its core focuses only on wireless charging control, and optimizes charging efficiency through time-frequency-space three-dimensional resource allocation. It has no offline computing power processing, scene recognition, or model training functions, and only stays at the charging assistance level. 2. The invention patent application with publication number CN119453640A discloses an eyeglass storage case and eyeglass set, which only optimizes the portability of the storage case and uses the charging cable as a handle, without any other smart functions other than charging.

[0006] 3. The invention patent application with publication number CN117100043A discloses a smart glasses box and its usage method. The core is only to optimize the connection method of the PCB board inside the glasses box to ensure the stability of the electrical connection between the display screen and the smart glasses. It does not involve computing power or scene recognition related designs, but only optimizes the hardware connection and does not break through the positioning of the glasses box as an auxiliary tool.

[0007] 4. The invention patent application with publication number CN121477508A discloses a smart box, smart glasses and split smart glasses. Although it proposes to move the computing center to the smart box to reduce the weight of the glasses, it does not clearly define the "offline computing power closed loop" design, does not integrate an offline model training unit, and cannot realize the full-process offline operation of scene recognition and function control.

[0008] 5. The invention patent application with publication number CN121397131A discloses an AI glasses and smartphone data interaction system under intelligent scene perception. Its core is to realize the data interaction between AI glasses and smartphone, which relies on the computing power and network of smartphone, but does not involve the computing power empowerment of the glasses box. 6. The invention patent with announcement number CN120085760B discloses a method, device and equipment for multimodal AI glasses vision-brain-electrode collaborative control. It only focuses on the vision-brain-electrode collaborative control of the AI ​​glasses themselves, relies on the computing power of the glasses themselves, and cannot solve the problem of excessive power consumption of the glasses themselves.

[0009] In summary, existing technologies can only achieve function switching of a single parameter, manual activation of a single function, or only involve data interaction for scene perception, or only use the glasses case as a storage and charging tool; none of them can solve the technical pain points of existing AI glasses, such as cumbersome scene switching, the need for manual operation, reliance on cloud computing power, insufficient local computing power, excessive power consumption, and easy leakage of privacy.

[0010] Therefore, there is an urgent need to design an AI glasses collaborative control system and method based on an offline AI mini-host. Summary of the Invention

[0011] The purpose of this invention is to provide an AI glasses collaborative control system and method based on an offline AI mini-host, in order to solve the technical pain points of existing AI glasses, such as limited scenarios, excessive power consumption, insufficient battery life, single function of the glasses box which cannot realize the value of the carrier, and easy leakage of user privacy data.

[0012] To solve the above problems, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides an AI glasses collaborative control system based on an offline AI mini-host, including an AI glasses body, wherein the special feature is that it also includes an offline AI mini-host integrated inside the glasses case, and the offline AI mini-host and the AI ​​glasses body establish bidirectional linkage through BLE+Wi-Fi dual-mode communication. The AI ​​glasses body includes a main control module, and a scene perception module, a function execution module, a temporary storage unit and a power supply module that are electrically connected to it respectively; The offline AI mini-host includes an AI computing chip, an offline storage unit, an offline model training unit, a power management unit, and a dual-mode communication unit integrated on the PCB board inside the glasses case.

[0013] Furthermore, in the AI ​​glasses body: The scene perception module includes a MEMS dual microphone, an ambient light sensor, a GPS+UWB dual-mode positioning unit, and an inertial measurement unit (IMU), which are used to collect ambient sound, light intensity, location positioning, and motion state data, respectively, to provide multi-dimensional data support for scene recognition. The main control module uses a low-power MCU to receive scene perception data, forward control commands from the offline AI mini-host, and control the operation of the function execution module. It does not need to undertake core computing power tasks, thus reducing power consumption. The function execution module includes a recording transcription unit, a navigation reminder unit, and a HUD display unit, which correspond to the core functions of preset scenarios such as conference room, outdoor, and driving, respectively, to achieve automatic adaptation between scenarios and functions. The capacity of the temporary storage unit is customized according to the needs. It is used to temporarily store scene feature data when the AI ​​glasses are removed from the glasses case. The data is automatically deleted after being synchronized to the offline AI host, thereby reducing the power consumption of the AI ​​glasses. The power supply module uses a 250mAh lithium battery, supports magnetic wireless charging, and has a power consumption of ≤8mW in low-power sleep mode. It can work continuously for 4-6 hours in recording transcription, navigation, and HUD modes, solving the pain point of insufficient battery life of existing AI glasses.

[0014] Furthermore, in the offline AI mini-host: The AI ​​computing chip uses Rockchip's lightweight edge computing chip, which supports offline floating-point operations, has a computing power of ≥1 TOPS, a power consumption of ≤500mW, and a built-in encryption module. It can run scene recognition models independently without relying on the computing power of the cloud or the AI ​​glasses themselves. The offline storage unit is an eMMC 5.1 interface storage chip with a capacity of ≥64GB. It supports encrypted data storage and is used to store scene feature data, scene recognition models, scene-function mapping relationship library, audio transcription files and model self-learning samples to avoid leakage of user privacy data. The offline model training unit optimizes scene recognition model parameters in real time based on user data, improves scene recognition accuracy, shortens the model optimization cycle, and does not require cloud training platform support. The power management unit uses a power management chip that supports wide voltage input (3.3V-5V) and has overvoltage, overcurrent and overheat protection functions. It is laid out separately from the glasses case charging module and uses an independent power supply circuit to avoid electromagnetic interference generated during charging from affecting the operation of the computing module. The dual-mode communication unit uses Espressif chips, supports BLE low-power communication and Wi-Fi high-speed communication, with a BLE communication distance of ≤10 meters and a Wi-Fi transmission speed of ≥150Mbps. It also supports a data synchronization priority mechanism to ensure the timeliness and stability of data transmission.

[0015] Secondly, the present invention provides a collaborative control method for AI glasses based on an offline AI mini-host, which is characterized by including the following steps: S1, System Initialization S2. The AI ​​glasses themselves collect scene feature data of the current environment in real time through the scene perception module and transmit it to the main control module in real time. S3, the main control module preprocesses the scene feature data, and then the data is synchronously stored to the offline AI mini-host; S4, the offline AI mini-host uses a built-in scene recognition model to perform offline analysis and processing of scene feature data to identify the target scene where the user is currently located; S5. The offline AI mini-host calls the scenario-function mapping relationship library, matches the preset function corresponding to the target scenario, generates function control instructions, and sends them to the main control module of the AI ​​glasses through the dual-mode communication module. S6. The main control module receives control commands and forwards them to the function execution module. The function execution module starts the corresponding preset function, completes the automatic switching between scene and function, and synchronizes the function execution status to the offline AI host. The S7 offline AI mini-host records scene feature data, function execution status, and user manual intervention records in real time. It optimizes scene recognition model parameters through the built-in offline model training unit, enabling model self-learning without cloud training. S8. When the scene feature data collected by the scene perception module continues to change and meets the feature threshold of another target scene, the main control module determines that the scene is switching and repeats steps S3-S6 to complete the automatic function switching.

[0016] Furthermore, the system initialization process in step S includes: The offline AI mini-host inside the glasses case and all modules inside the AI ​​glasses are powered on and started. The offline AI mini-host loads the built-in scene recognition model and scene-function mapping relationship library; The dual-mode communication unit establishes a connection between the AI ​​glasses and the offline AI host and user terminal device, receives user-preset scene thresholds, function preferences and navigation parameters, and stores them in the offline storage unit of the offline AI host.

[0017] Furthermore, the scene feature data in step S2 includes ambient sound data, light intensity data, location data, and motion state data.

[0018] Furthermore, the preprocessing of scene feature data in step S3 includes, but is not limited to, noise filtering (filtering invalid clutter and environmental interference signals), data format unification (unifying data length and data type), data normalization / standardization, outlier removal (data that clearly exceeds the reasonable range), and data compression (reducing transmission volume).

[0019] Furthermore, in step S3, if the AI ​​glasses are outside the glasses case, the scene feature data preprocessed by the main control module is first stored in the temporary storage unit of the AI ​​glasses, and automatically synchronized to the offline AI host after the glasses are put back into the glasses case; if the AI ​​glasses are inside the glasses case, the preprocessed scene feature data is directly synchronized to the offline AI host.

[0020] This invention discloses an AI glasses collaborative control system and method based on an offline AI mini-host. The offline AI mini-host is integrated inside the glasses case, upgrading the case from a simple storage and charging tool into a core computing power carrier. Through BLE + Wi-Fi dual-mode communication, a two-way linkage is established between the offline AI mini-host and the AI ​​glasses themselves. This achieves a division of labor where the offline AI mini-host handles core computing power processing, while the AI ​​glasses handle scene perception and function execution. All core high-power computing tasks, such as scene recognition and data processing, are transferred to the glasses case, significantly reducing the computing load on the AI ​​glasses and extending their battery life. This addresses the core pain point of insufficient battery life in existing AI glasses, ensuring stable long-term use in common modes such as recording and transcription, navigation, and HUD display. Simultaneously, relying on the AI ​​computing chip, offline storage unit, and offline model training unit built into the offline AI mini-host, the analysis and processing of scene feature data, scene recognition, and model self-learning can be completed entirely offline, without relying on cloud computing power, completely eliminating the limitations of use in scenarios with no or weak network coverage. Among them, user scene feature data, audio transcription files and other private data are encrypted and stored in the encrypted storage unit of the offline AI host. There is no need to upload to the cloud at all. The entire process of data storage and transmission blocks the risk of privacy leakage and effectively protects user privacy and security. In addition, through the learning and optimization of user manual intervention records by the offline model training unit, the scene recognition accuracy can be continuously improved, making the function control more in line with the user's personalized usage habits, and further improving the user experience and adaptability of AI glasses. Attached Figure Description

[0021] Figure 1 A flowchart of collaborative control for AI glasses based on an offline AI mini-host; Detailed Implementation

[0022] The invention will now be described in detail with reference to the accompanying drawings.

[0023] Example 1 An AI glasses collaborative control system based on an offline AI mini-host includes an AI glasses body and an offline AI mini-host. The offline AI mini-host is integrated inside the glasses case and adopts a layered integration process. It is separately laid out and independently powered from the charging module built into the glasses case.

[0024] The AI ​​glasses body includes a scene perception module, a main control module, a function execution module, a temporary storage unit, and a power supply module; wherein... The scene perception module includes a MEMS dual microphone, an ambient light sensor, a GPS+UWB dual-mode positioning unit, and an inertial measurement unit (IMU), which are used to collect current ambient sound, light intensity, location positioning, and motion state data, respectively, to provide multi-dimensional data support for scene recognition. The main control module uses a low-power MCU to receive scene perception data, forward control commands from the offline AI mini-host, and control the operation of the function execution module. It does not need to undertake core computing power tasks, thus reducing power consumption. The function execution module includes a recording transcription unit, a navigation reminder unit, and a HUD display unit, which correspond to the core functions of three preset scenarios: conference room, outdoor, and driving, respectively, to achieve automatic adaptation between scenarios and functions. The capacity of the temporary storage unit is customized according to the needs. It is used to temporarily store scene feature data when the AI ​​glasses are removed from the glasses case. The data is automatically deleted after being synchronized to the offline AI host, thereby reducing the power consumption of the AI ​​glasses. The power supply module uses a 250mAh lithium battery, supports magnetic wireless charging, and has a power consumption of ≤8mW in low-power sleep mode. It can work continuously for 4-6 hours in recording transcription, navigation, and HUD modes, solving the pain point of insufficient battery life of existing AI glasses.

[0025] The offline AI mini-host includes an AI computing chip, an offline storage unit, an offline model training unit, a power management unit, and a dual-mode communication unit integrated on a dedicated PCB board inside the glasses case. The PCB board is made of insulating and heat-dissipating material and is fixed inside the glasses case on the side away from the storage cavity. The AI ​​computing chip uses Rockchip's lightweight edge computing chip, which supports offline floating-point operations, has a computing power of ≥1 TOPS, a power consumption of ≤500mW, and a built-in encryption module. It can run scene recognition models independently without relying on the computing power of the cloud or the AI ​​glasses themselves. The offline storage unit is an eMMC 5.1 interface storage chip with a capacity of ≥64GB. It supports encrypted data storage and is used to store scene feature data, scene recognition models, scene-function mapping relationship library, audio transcription files and model self-learning samples to avoid leakage of user privacy data. The offline model training unit optimizes scene recognition model parameters in real time based on user data, improves scene recognition accuracy, shortens the model optimization cycle, and does not require cloud training platform support. The power management unit uses a power management chip that supports wide voltage input (3.3V-5V) and has overvoltage, overcurrent and overheat protection functions. It is independently connected to the glasses case charging module and uses an independent power supply circuit to avoid electromagnetic interference generated during charging from affecting the operation of the computing module. The dual-mode communication unit uses Espressif chips, supports BLE low-power communication and Wi-Fi high-speed communication, with a BLE communication distance of ≤10 meters and a Wi-Fi transmission speed of ≥150Mbps. It also supports a data synchronization priority mechanism to ensure the timeliness and stability of data transmission.

[0026] Example 2 This embodiment relates to an AI glasses collaborative control method based on an offline AI mini-host, including the following specific steps: S1, System Initialization When the offline AI mini-host inside the glasses case and each module inside the AI ​​glasses are powered on, the offline AI mini-host loads the built-in scene recognition model and scene-function mapping relationship library. The communication module establishes a connection between the AI ​​glasses and the offline AI mini-host and the user terminal device, receives the user's preset scene thresholds, function preferences and navigation parameters, and stores them in the offline storage unit of the offline AI mini-host. S2. The scene perception module collects scene feature data of the user's environment in real time. The scene feature data includes environmental sound data, light intensity data, location data and motion state data. The collected data is transmitted to the main control module of the AI ​​glasses in real time. S3. The main control module preprocesses the collected scene feature data and filters out invalid noise. If the AI ​​glasses are outside the glasses case, the preprocessed data is stored in the temporary storage unit of the glasses and automatically synchronized to the offline AI host after being put back into the glasses case. If the glasses are inside the glasses case, the data is directly synchronized to the offline AI host. The S4 offline AI mini-host uses a built-in scene recognition model to perform offline analysis and processing of scene feature data, identifying the target scene where the user is currently located, without relying on cloud computing power. S5. The offline AI mini-host calls the scenario-function mapping relationship library, matches the preset function corresponding to the target scenario, generates function control instructions, and sends them to the main control module of the AI ​​glasses through the dual-mode communication module. S6. The main control module receives control commands and forwards them to the function execution module. The function execution module starts the corresponding preset function, completes the automatic switching between scene and function, and synchronizes the function execution status to the offline AI host. S7, an offline AI mini-host, records scene feature data, function execution status, and user manual intervention records in real time. It optimizes scene recognition model parameters through a built-in offline model training unit, enabling model self-learning without cloud-based training. The user manual intervention records specifically include, but are not limited to: 1) Users manually correct scene recognition results. For example, if the system recognizes it as [meeting room], the user manually changes it to [driving]; if the system recognizes it as [outdoor], the user manually changes it to [meeting room]. This is the most typical intervention. 2) Users manually enable / disable a function; 3) Users can manually adjust scene thresholds / function preferences, or adjust sensitivity, reminder methods, etc., or manually cancel the behavior records generated by the system's automatic control, including intervention time, intervention content, scene status before and after intervention, and function status.

[0027] S8. When the scene feature data collected by the scene perception module continues to change and meets the feature threshold of another target scene, the main control module determines that the scene is switched and repeats steps S3-S6 to complete the automatic function switch. When there is no valid scene data input on the AI ​​glasses, the offline AI host controls the AI ​​glasses to enter a low-power sleep state.

[0028] This invention discloses a collaborative control method for AI glasses based on an offline AI mini-host. Through a technical logic of "integrating an offline AI mini-host into the glasses case + multi-dimensional scene feature acquisition → accurate multi-scene recognition → collaborative control between AI glasses and the glasses case → offline model self-learning," it significantly improves product battery life and enhances privacy protection while also considering practicality and user experience. It enables automatic recognition of multiple scenarios such as meeting rooms, outdoor environments, and driving, as well as seamless automatic switching of corresponding functions, while addressing existing limitations related to computing power.

[0029] This invention is not limited to the embodiments discussed above. The above description of specific embodiments is intended to describe and illustrate the technical solutions involved in this invention. Obvious modifications, substitutions, or combinations based on the teachings of this invention should also be considered to fall within the protection scope of this invention. The above specific embodiments are used to disclose the best implementation methods of this invention, so that those skilled in the art can apply various embodiments and alternatives of this invention to achieve the objectives of this invention.

Claims

1. A collaborative control system for AI glasses based on an offline AI mini-host, comprising AI glasses body, characterized in that, It also includes an offline AI mini-host integrated inside the glasses case, which establishes a two-way linkage with the AI ​​glasses through BLE+Wi-Fi dual-mode communication; The AI ​​glasses body includes a main control module, and a scene perception module, a function execution module, a temporary storage unit and a power supply module that are electrically connected to it respectively; The offline AI mini-host includes an AI computing chip, an offline storage unit, an offline model training unit, a power management unit, and a dual-mode communication unit integrated on the PCB board inside the glasses case.

2. The AI ​​glasses collaborative control system based on an offline AI mini-host as described in claim 1, characterized in that, In the AI ​​glasses body: The scene perception module includes a MEMS dual microphone, an ambient light sensor, a GPS+UWB dual-mode positioning unit, and an inertial measurement unit, which are used to collect ambient sound, light intensity, location positioning, and motion state data, respectively, to provide multi-dimensional data support for scene recognition. The main control module uses a low-power MCU to receive scene perception data, forward control commands from the offline AI mini-host, and control the operation of the function execution module. The function execution module includes a recording transcription unit, a navigation reminder unit, and a HUD display unit, which correspond to the core functions of different preset scenarios, realizing automatic adaptation between scenarios and functions. The capacity of the temporary storage unit is customized according to the requirements. It is used to temporarily store scene feature data when the AI ​​glasses are removed from the glasses case. The data is automatically deleted after being synchronized to the offline AI host, thereby reducing the power consumption of the AI ​​glasses. The power supply module uses a 250mAh lithium battery, supports magnetic wireless charging, has a power consumption of ≤8mW in low-power sleep mode, and can work continuously for 4-6 hours in recording transcription, navigation, and HUD modes.

3. The AI ​​glasses collaborative control system based on an offline AI mini-host as described in claim 1, characterized in that, In the offline AI mini-host: The AI ​​computing chip uses Rockchip's lightweight edge computing chip, which supports offline floating-point operations, has a computing power of ≥1 TOPS, a power consumption of ≤500mW, and a built-in encryption module, enabling it to run scene recognition models independently. The offline storage unit is an eMMC 5.1 interface storage chip with a capacity of ≥64GB. It supports encrypted data storage and is used to store scene feature data, scene recognition models, scene-function mapping relationship library, audio transcription files and model self-learning samples to avoid leakage of user privacy data. The offline model training unit optimizes scene recognition model parameters in real time based on user data, improves scene recognition accuracy, shortens the model optimization cycle, and does not require cloud training platform support. The power management unit uses a power management chip that supports a wide voltage range of 3.3V-5V input and has overvoltage, overcurrent, and overheat protection functions. It is separated from the glasses case charging module and uses an independent power supply circuit to avoid electromagnetic interference generated during charging from affecting the operation of the computing module. The dual-mode communication unit uses Espressif chips, supports BLE low-power communication and Wi-Fi high-speed communication, with a BLE communication distance of ≤10 meters and a Wi-Fi transmission speed of ≥150Mbps. It also supports a data synchronization priority mechanism to ensure the timeliness and stability of data transmission.

4. A control method for the AI ​​glasses collaborative control system according to any one of claims 1-3, characterized in that, Includes the following steps: S1, System Initialization S2. The AI ​​glasses themselves collect scene feature data of the current environment in real time through the scene perception module and transmit it to the main control module in real time. S3, the main control module preprocesses the scene feature data, and then the data is synchronously stored to the offline AI mini-host; S4, the offline AI mini-host uses a built-in scene recognition model to perform offline analysis and processing of scene feature data to identify the target scene where the user is currently located; S5. The offline AI mini-host calls the scenario-function mapping relationship library, matches the preset function corresponding to the target scenario, generates function control instructions, and sends them to the main control module of the AI ​​glasses through the dual-mode communication module. S6. The main control module receives control commands and forwards them to the function execution module. The function execution module starts the corresponding preset function, completes the automatic switching between scene and function, and synchronizes the function execution status to the offline AI host. The S7 offline AI mini-host records scene feature data, function execution status, and user manual intervention records in real time. It optimizes scene recognition model parameters through the built-in offline model training unit, enabling model self-learning without cloud training. S8. When the scene feature data collected by the scene perception module continues to change and meets the feature threshold of another target scene, the main control module determines that the scene is switching and repeats steps S3-S6 to complete the automatic function switching.

5. The control method as described in claim 4, characterized in that, The system initialization process in step S1 includes: The offline AI mini-host inside the glasses case and all modules inside the AI ​​glasses are powered on and started. The offline AI mini-host loads the built-in scene recognition model and scene-function mapping relationship library; The dual-mode communication unit establishes a connection between the AI ​​glasses and the offline AI host and user terminal device, receives user-preset scene thresholds, function preferences and navigation parameters, and stores them in the offline storage unit of the offline AI host.

6. The control method as described in claim 4, characterized in that, The scene feature data in step S2 includes ambient sound data, light intensity data, location data, and motion state data.

7. The control method as described in claim 4, characterized in that, The preprocessing of scene feature data in step S3 includes, but is not limited to, noise filtering, data format unification, data normalization / standardization, outlier removal, and data compression.

8. The control method as described in claim 4, characterized in that, In step S3, if the AI ​​glasses are outside the glasses case, the scene feature data preprocessed by the main control module is first stored in the temporary storage unit of the AI ​​glasses, and automatically synchronized to the offline AI host after the glasses are put back into the glasses case; if the AI ​​glasses are inside the glasses case, the preprocessed scene feature data is directly synchronized to the offline AI host.

Images