Low-power-consumption multi-chip cooperative control method and system for intelligent glasses

Abstract

The invention relates to the technical field of intelligent control, and discloses a low-power-consumption multi-chip cooperative control method for intelligent glasses, and the method comprises the steps: monitoring a user input signal in real time in a low-power-consumption state through a special sensor module, and generating a wake-up trigger signal; obtaining a current resource state corresponding to the intelligent glasses, and dynamically selecting a system operation mode based on the user intention information and the current resource state; calling a corresponding hardware module to execute data acquisition, data processing and data transmission tasks according to the system operation mode; a power supply strategy is adjusted according to the system operation mode and the user real-time behavior data, the power supply strategy comprises only core module operation corresponding to a standby mode, partial module closing corresponding to a sleep mode and full-module cooperative operation corresponding to a high-performance mode, and hardware resource allocation is optimized through a task priority scheduling algorithm. The power consumption of the intelligent glasses can be reduced.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology, and in particular to a low-power multi-chip collaborative control method and system for smart glasses. Background Technology

[0002] With the widespread application of smart glasses in consumer electronics and industrial scenarios, the demand for low power consumption, high performance, and multi-module collaborative control is becoming increasingly prominent. Traditional smart glasses systems typically employ a single-chip architecture or a fixed-function module operating mode, making it difficult to balance power optimization with the needs of complex task processing. For example, when users only need to execute simple voice commands, the system still needs to activate all hardware modules (such as image processing, communication, and sensor arrays), resulting in unnecessary energy waste. Furthermore, when performing tasks such as high-precision environmental analysis or real-time translation, the system's response efficiency and energy efficiency ratio are difficult to meet actual requirements due to fixed power supply strategies and static resource allocation.

[0003] Furthermore, existing technologies rely heavily on traditional serial communication protocols for data interaction between sensor arrays and core processing units, resulting in high transmission latency and limited bandwidth, further exacerbating the conflict between system power consumption and performance. Therefore, reducing the power consumption of smart glasses remains a challenge. Summary of the Invention

[0004] To reduce the power consumption of smart glasses, this application provides a low-power multi-chip collaborative control method and system for smart glasses.

[0005] In a first aspect, this application provides a low-power multi-chip collaborative control method and system for smart glasses, employing the following technical solution: A low-power multi-chip collaborative control method for smart glasses includes: A dedicated sensor module monitors user input signals in real time with low power consumption and generates a wake-up trigger signal. The user input signals include voice, gestures, or environmental changes, and the wake-up trigger signal includes user intent information. The system obtains the current resource status corresponding to the smart glasses, and dynamically selects the system operation mode based on the user intent information and the current resource status. The system operation mode includes low power mode or high performance mode. In low power mode, only the functional modules corresponding to the current task are activated. In high performance mode, all functional modules are activated and a dynamic resource allocation strategy is started. According to the system's operating mode, the corresponding hardware modules are invoked to perform data acquisition, data processing, and data transmission tasks, including: an image processing module, used to combine AI algorithms to complete environmental analysis and location data extraction; a communication module, used to dynamically adjust the data transmission strategy according to task priority; and a sensor array, used to collect user posture, ambient light, and wearing status information in real time. The power supply strategy is adjusted based on the system operation mode and real-time user behavior data. The real-time user behavior data includes wearing status and task duration. The power supply strategy includes standby mode (only core modules run), hibernation mode (some modules are shut down), and high-performance mode (all modules run in coordination). Hardware resource allocation is optimized through a task priority scheduling algorithm.

[0006] Optionally, the dedicated sensor module includes a voice wake-up chip, and the method further includes: The voice wake-up chip continuously listens to voice commands with a microampere-level current and sends a wake-up signal to the main control chip via an SPI or I2C interface. The voice wake-up chip extracts user intent data through a natural language processing algorithm, and transmits the user intent data to the main control chip after feature encoding through a semantic embedding model.

[0007] Optionally, during the process of the communication module adjusting the data transmission strategy according to task priority, the method further includes: In low-power mode, a lightweight data compression algorithm is used to transmit map update information or location data, and low-bandwidth communication is performed based on the 2.4GHz band of Wi-Fi 6E. In high-performance mode, video streams or voice data are transmitted based on H.265 dynamic bitrate control technology, and signal interference is reduced by joint frequency band switching of Wi-Fi 6E and Bluetooth 5.3 protocols, with latency controlled within 2 milliseconds; The dynamic bitrate control technology combines real-time network status monitoring of edge computing nodes to dynamically adjust video encoding resolution and frame rate.

[0008] Optionally, the method for adjusting the power supply strategy further includes: When the wearing status detection circuit triggers the lens removal signal, it enters deep sleep mode with a response time of no more than 50 milliseconds; In high-performance mode, the voltage and charging current are dynamically adjusted through the ET9562 power management chip, combined with a battery health assessment mechanism. The dynamic voltage adjustment includes: adjusting the core module power supply voltage in segments from 1.0V to 1.8V according to changes in task load to reduce dynamic power consumption.

[0009] Optionally, the method further includes: The image processing module calls a pre-set ISP engine, which supports small image sensors from 1 / 4 inch to 1 / 5 inch and small aperture optimization. The ISP engine includes an image denoising module based on a convolutional neural network; The image processing module interacts with the communication module through a shared memory mechanism, wherein the shared memory adopts a circular buffer structure.

[0010] Optionally, the method further includes: The physical integration stage uses Chiplet technology to vertically stack the voice wake-up chip, image processing chip and communication module, and achieves inter-chip interconnection through hybrid bonding process. The hybrid bonding process includes nanoscale copper-copper direct bonding and dielectric layer alignment technology, wherein the interconnect density is higher than 10,000 connection points per square millimeter; The frame uses a multi-layer waterproof sealant process, and silicone gaskets are added at the temple joints, passing the IP67 protection level test.

[0011] Secondly, this application provides a low-power multi-chip collaborative control system for smart glasses, which adopts the following technical solution: A low-power multi-chip collaborative control system for smart glasses. The wake-up trigger signal generation module monitors user input signals in real time in a low-power state through a dedicated sensor module to generate a wake-up trigger signal. The user input signals include voice, gestures or environmental changes, and the wake-up trigger signal includes user intent information. The system operation mode selection module obtains the current resource status corresponding to the smart glasses. Based on the user intent information and the current resource status, it dynamically selects the system operation mode, which includes a low-power mode or a high-performance mode. In the low-power mode, only the functional modules corresponding to the current task are activated. In the high-performance mode, all functional modules are activated and a dynamic resource allocation strategy is initiated. The hardware module invocation module, based on the system operating mode, calls the corresponding hardware modules to perform data acquisition, data processing, and data transmission tasks, including: an image processing module, used to combine AI algorithms to complete environmental analysis and location data extraction; a communication module, used to dynamically adjust the data transmission strategy according to task priority; and a sensor array, used to collect user posture, ambient light, and wearing status information in real time. The power supply strategy adjustment module adjusts the power supply strategy based on the system operation mode and real-time user behavior data. The real-time user behavior data includes wearing status and task duration. The power supply strategy includes standby mode (only core modules run), hibernation mode (some modules are shut down), and high-performance mode (all modules run in coordination). Hardware resource allocation is optimized through a task priority scheduling algorithm.

[0012] Thirdly, this application provides a low-power multi-chip collaborative control method for smart glasses, employing the following technical solution: A low-power multi-chip collaborative control system for smart glasses includes a processor, wherein the processor runs a program for the low-power multi-chip collaborative control method for smart glasses as described in any one of the above-mentioned methods.

[0013] Fourthly, this application provides a storage medium, which adopts the following technical solution: A storage medium storing a program for a low-power multi-chip collaborative control method for smart glasses as described in any one of the above.

[0014] In summary, this application includes at least one of the following beneficial technical effects: By extracting user intent through low-power monitoring (microampere-level current) using a dedicated sensor module and natural language processing algorithms, the system activates only necessary functional modules when a valid command is detected, avoiding prolonged operation of high-power hardware. A dynamic operating mode selection mechanism (low power / high performance) combined with a task priority scheduling algorithm ensures that only core modules (such as image processing and communication) are activated under low load scenarios, while under high load tasks, the ET9562 power management chip dynamically adjusts the voltage (1.0V to 1.8V segmented control), significantly reducing dynamic power consumption. Furthermore, a deep sleep mode (response time ≤50ms) and a battery health assessment mechanism (based on internal resistance changes) further extend battery life and reduce redundant power supply waste.

[0015] The system employs Chiplet technology to vertically stack the voice wake-up chip, image processing chip, and communication module. A hybrid bonding process (nanoscale copper-copper direct bonding) reduces signal transmission latency (≤50ps), minimizing additional power consumption due to data interaction. The image processing module optimizes imaging from a small-aperture sensor using an ISP engine and combines it with a convolutional neural network denoising module to maintain image quality even at low resolutions, avoiding high-power image processing steps. The communication module utilizes Wi-Fi 6E and Bluetooth 5.3 band switching technology and H.265 dynamic bitrate control, combined with real-time network monitoring from edge computing nodes, to compress video stream bandwidth requirements while maintaining low latency (≤2ms), ultimately achieving an overall system power consumption reduction of over 30%. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a low-power multi-chip collaborative control method for smart glasses according to an exemplary embodiment.

[0017] Figure 2 This is a block diagram illustrating a low-power multi-chip collaborative control system for smart glasses, according to an exemplary embodiment. Detailed Implementation

[0018] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0019] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0020] This application discloses a low-power multi-chip collaborative control method for smart glasses, referring to... Figure 1 ,include: The S100 uses a dedicated sensor module to monitor user input signals in real time in a low-power state and generate a wake-up trigger signal.

[0021] The dedicated sensor module employs multimodal sensing technology, integrating multiple sensors such as voice, gesture, and environmental change sensors. It continuously monitors user commands with microamp-level current. For example, the voice wake-up chip operates with extremely low power consumption (≤10μA), activating subsequent processing only when a valid wake-up word (such as "start navigation" or "translate") is detected. The gesture recognition sensor captures user actions through capacitive touch or IMU, and combines motion feature extraction algorithms to interpret intent. The ambient light sensor monitors light intensity in real time to help determine whether the user needs to adjust display brightness or switch to night vision mode.

[0022] To improve system response efficiency, the S100 fuses multi-sensor data through a heterogeneous computing architecture and extracts user intent by combining edge AI algorithms (such as lightweight neural networks). For example, after the voice signal is parsed by a natural language processing model, "translate 'Hello'" can be mapped to specific task parameters; gesture signals are identified by frequency domain analysis of IMU data to identify the action type (such as swiping or pinching) and the intent (such as "zoom in on the image" or "switch pages") is determined by combining the current application state.

[0023] In this embodiment, the generated wake-up trigger signal not only includes the task type, but also integrates sensor status information (such as ambient light intensity and gesture position), and is quickly transmitted to the main control chip via SPI / I2C interface or Bluetooth Low Energy (BLE) to trigger the subsequent system operation mode selection (S200).

[0024] To achieve low-power operation, the S100 employs a dynamic power management strategy. In standby mode, only essential circuits (such as the microphone array of the voice wake-up chip) are activated, while other modules enter deep sleep (current drops to the nanoamp level). When valid user input is detected, the system activates other sensors (such as the IMU or ambient light sensor) through an event-driven mechanism, avoiding energy waste caused by continuous sampling. For example, the voice wake-up chip maintains low-power listening when no command is detected, while the gesture sensor only activates after a voice signal is triggered, thus significantly reducing overall power consumption.

[0025] S200 obtains the current resource status corresponding to the smart glasses, and dynamically selects the system operation mode based on user intent information and the current resource status. The system operation mode includes low power mode or high performance mode. In low power mode, only the functional modules corresponding to the current task are activated. In high performance mode, all functional modules are activated and a dynamic resource allocation strategy is started.

[0026] In step S200, the current resource status of the smart glasses needs to be acquired first. This includes, but is not limited to, battery level, processor load, memory usage, and sensor operating status. By monitoring these parameters in real time, the system can gain a comprehensive understanding of the device's current operating status. For example, when the battery is low, the system will prioritize entering a low-power mode to extend usage time; while when the processor load is low and memory is sufficient, it can support higher-performance task processing. Furthermore, sensor operating status (such as whether high-precision environmental monitoring is in progress) also affects the selection of the system's operating mode.

[0027] After acquiring the current resource status, the system dynamically selects the appropriate system operating mode based on user intent information. If the user intent is to perform a simple task, such as checking the time or answering a phone call, the system will automatically switch to low-power mode. In this mode, only functional modules directly related to the current task are activated, such as the display screen and voice processing unit, while other non-essential modules are in a dormant or off state, thus greatly reducing energy consumption. Conversely, if the user intent involves complex data processing tasks, such as high-definition video calls or augmented reality navigation, the system will select high-performance mode. In this case, all functional modules are activated, and a dynamic resource allocation strategy is initiated to ensure that each module receives sufficient resources to meet the task requirements.

[0028] In low-power mode, the operation of smart glasses focuses primarily on maintaining basic functions. This means that, apart from modules related to the current task, all other unnecessary hardware and software will be minimized or completely shut down. For example, in standby mode, the display may be off, only turning on when a new notification arrives or the user performs a specific action. Simultaneously, the communication module will adjust its activity level as needed, such as reducing Wi-Fi scanning frequency or lowering Bluetooth power consumption. In this way, smart glasses can maximize energy savings without affecting the user's basic experience, allowing the device to remain usable for extended periods even with limited battery power.

[0029] Once in high-performance mode, all core components of the smart glasses operate at full capacity to deliver optimal performance. In this mode, the system not only ensures efficient collaboration between functional modules but also implements effective resource management and scheduling strategies. For example, it employs Dynamic Voltage and Frequency Scaling (DVFS) technology to adjust the processor's operating frequency and voltage based on the actual load, thereby achieving energy savings; or it leverages the advantages of multi-core processors to rationally allocate computing tasks and avoid overloading any single core. Furthermore, for data-intensive applications, the system will activate more advanced compression algorithms and efficient communication protocols (such as Wi-Fi 6E or 5G) to accelerate data transmission speeds and reduce latency. These measures work together to ensure that the smart glasses can operate smoothly even under demanding tasks, providing users with a seamless interactive experience.

[0030] The S300 calls the corresponding hardware modules to perform data acquisition, data processing, and data transmission tasks according to the system operating mode. These include: an image processing module, which combines AI algorithms to complete environmental analysis and location data extraction; a communication module, which dynamically adjusts the data transmission strategy according to task priority; and a sensor array, which collects user posture, ambient light, and wearing status information in real time.

[0031] 1. The image processing module is an AI-based environmental analysis and localization data extraction module. After the system operating mode (low power or high performance) is determined, the image processing module will invoke AI algorithms according to task requirements to complete environmental analysis and localization data extraction. The specific implementation is as follows: AI algorithm integration: Combine deep learning models (such as a lightweight version of YOLOv8 or FastVLM) to process real-time images captured by the camera. For example, identify surrounding objects (such as pedestrians, vehicles, and road signs) through object detection algorithms, and distinguish different scenes (such as indoor, outdoor, and traffic intersections) through semantic segmentation technology.

[0032] Environmental analysis: Using computer vision technologies (such as SLAM, simultaneous localization and mapping) to extract environmental feature points and generate a 3D spatial map. For example, in AR navigation, accurate real-time positioning can be achieved by recognizing landmarks and matching them with a pre-stored geographic information database.

[0033] Location data extraction: Combining sensor arrays (such as IMUs and barometers) with image data, the user's position and orientation are calculated using a multimodal fusion algorithm. For example, by analyzing changes in the user's head posture (relying on patented capacitive sensor technology) and visual feature points, positioning drift caused by motion is corrected.

[0034] AI algorithms enhance image processing capabilities, supporting positioning and navigation in complex scenarios (such as indoor environments without GPS); in low-power mode, only basic image processing functions (such as edge detection) are enabled, while in high-performance mode, the full AI model (such as FastVLM) is activated, balancing performance and energy consumption.

[0035] 2. The communication module employs a dynamic data transmission strategy based on task priority. The module dynamically adjusts the data transmission strategy according to the system's operating mode and task priorities to ensure efficient and low-latency communication. Specific implementation includes: Optimizations in low-power mode: Lightweight compression and low-bandwidth communication: Compress map update information or location data using JPEG-XL or H.265 lightweight encoding (e.g., reducing color depth or resolution) to reduce data volume; Transmit via the 2.4GHz band of Wi-Fi 6E, taking advantage of its lower bandwidth requirements and low power consumption to reduce energy consumption and interference.

[0036] Enhancements in high-performance mode: H.265 dynamic bitrate control, based on real-time network status monitoring of edge computing nodes (such as bandwidth fluctuations and latency), dynamically adjusts the resolution of the video stream (such as reducing from 4K to 1080p) and frame rate (such as reducing from 60fps to 30fps) to balance image quality and transmission efficiency; joint frequency band switching, through the collaboration of Wi-Fi 6E's 5GHz / 6GHz high-frequency band (high bandwidth, low latency) and Bluetooth 5.3 protocol, to achieve dynamic frequency band switching (such as avoiding the congested 2.4GHz band), reduce signal interference and ensure that the latency of critical tasks (such as AR navigation voice commands) is ≤2ms.

[0037] Edge computing supports: network status awareness, edge nodes monitor bandwidth, signal strength and task priority in real time (such as ensuring voice channels for video calls), and dynamically allocate resources (such as prioritizing voice data transmission and downgrading non-critical sensor data); protocol and algorithm linkage, combining the efficient compression of H.265 and the OFDMA technology of Wi-Fi 6E to achieve parallel transmission of multiple devices, further reducing latency and improving throughput.

[0038] By dynamically adjusting the transmission strategy, high-priority data for critical tasks (such as real-time translation and AR navigation) is prioritized for transmission; redundant data transmission is reduced in low-power mode, and communication efficiency is maximized and battery life is extended in high-performance mode.

[0039] 3. The sensor array collects user behavior and environmental data in real time. It is responsible for acquiring real-time information on user posture, ambient light, and wearing status, providing data support for system decision-making. Specific implementation includes: User posture monitoring uses IMU (Inertial Measurement Unit) and capacitive sensors (such as the head posture monitoring technology in Google's patent) to capture the user's head movement trajectory and angle changes. For example, it can detect whether the user is looking down at their phone or turning around to observe their surroundings; combined with AI algorithms (such as convolutional neural networks) to analyze posture data, it can identify the user's intentions (such as "turn right" or "move closer to the screen").

[0040] An ambient light sensor monitors light intensity in real time and dynamically adjusts the display brightness (e.g., increasing brightness in strong light to improve visibility); a wearing status detection system uses pressure or capacitive sensors to determine whether the user has removed their glasses. For example, when the system detects that the frame has stopped contacting the face, it triggers a sleep mode to save power.

[0041] Multimodal data fusion: Integrates posture, ambient light, and wearing status data, and generates comprehensive decisions through edge computing algorithms. For example, navigation prompts can be automatically paused after the user removes their glasses to avoid interfering with the user's operation.

[0042] By collecting user behavior and environmental data in real time, the system can dynamically adjust its operating mode (such as switching from high performance to low power consumption); in addition, it can improve user comfort and device usability by adaptively adjusting based on wearing status and ambient light (such as adjusting screen brightness).

[0043] Through the collaborative work of image processing, communication, and sensor arrays, the S300 enables dynamic resource allocation and task priority management for smart glasses in both low-power and high-performance modes. This is not only reflected in energy efficiency optimization (such as reducing standby power consumption by more than 30%), but also in the deep integration of AI algorithms and sensor data, which enhances the intelligence level of the device and the user interaction experience. This lays a solid foundation for the widespread application of smart glasses in scenarios such as AR navigation, health monitoring, and real-time translation.

[0044] The S400 adjusts its power supply strategy based on the system's operating mode and real-time user behavior data. The real-time user behavior data includes wearing status and task duration. The power supply strategy includes standby mode (where only the core module runs), hibernation mode (where some modules are shut down), and high-performance mode (where all modules run in tandem). Hardware resource allocation is optimized through a task priority scheduling algorithm.

[0045] Among them, the core of the S400's dynamic power supply strategy adjustment mechanism lies in dynamically adjusting the power supply strategy based on the system operating mode (low power / high performance) and real-time user behavior data (such as wearing status and task duration). The specific implementation is as follows: Standby mode activates only core modules (such as the voice wake-up chip and sensor array) and shuts down unnecessary modules (such as the display and communication modules) to maintain basic functions with minimal power consumption; hibernation mode further shuts down some modules (such as the image processing unit) and retains only key sensors (such as the ambient light sensor) to maintain extremely low power consumption via Bluetooth Low Energy (BLE); high-performance mode activates all functional modules (such as the display, camera, and communication modules) and activates Dynamic Voltage and Frequency Scaling (DVFS) technology to adjust the processor voltage and frequency in real time according to the task load, balancing performance and power consumption.

[0046] The system detects whether the user has removed their glasses using pressure or capacitance sensors. For example, when the user removes their glasses, it automatically switches to sleep mode with a response time of no more than 50 milliseconds, shutting down the display and communication module, retaining only the ambient light monitoring function. Resources are dynamically allocated based on task duration. For example, short tasks (such as translating "Hello") are executed quickly in high-performance mode and then immediately return to standby mode; while long tasks (such as AR navigation) maintain high-performance mode, but resource allocation is optimized through a task priority scheduling algorithm.

[0047] By dynamically switching power supply modes, the power consumption of non-critical modules is reduced. For example, in standby mode, the system power consumption can be reduced to <10mW, saving more than 80% of power compared to traditional continuous operation solutions. In addition, the power supply strategy is flexibly adjusted according to different user behaviors (such as changes in wearing status and task type) to ensure a balance between long-term wear (such as all-day use) and sudden high load (such as AR games).

[0048] 2. Real-time user behavior data collection and analysis: User behavior data is collected and analyzed in real-time using sensor arrays and AI algorithms to provide a basis for power supply strategy adjustments. Specifically, this includes: Wearing status monitoring: Sensor technology, combining capacitive sensors (detecting contact between the frame and face) and IMU (inertial measurement unit), determines whether the user has removed their glasses. For example, when the IMU detects a sudden change in the frame angle (such as when the glasses are removed), it triggers a sleep mode.

[0049] With the assistance of AI algorithms, sensor data is analyzed through edge neural networks to distinguish between normal wearing (such as adjusting the position of glasses) and taking them off, thereby reducing false triggers.

[0050] Task duration prediction: Based on the user's historical task records (such as average translation time, navigation path length) and current task characteristics (such as voice command complexity, distance to the target location), the machine learning model predicts the task duration. For example, if it is predicted that the user is engaged in long-term navigation, the high-performance mode is activated in advance and extra power is reserved to avoid task interruption due to insufficient power.

[0051] By sensing wear status and predicting task duration, the system avoids consuming power when the user removes the glasses or affecting the user experience due to insufficient power during long tasks. In addition, by combining real-time behavioral data to optimize the power supply strategy, the device provides sufficient performance during active periods and greatly reduces energy consumption during idle periods.

[0052] 3. Resource optimization of task priority scheduling algorithm: Hardware resources are dynamically allocated through the task priority scheduling algorithm to ensure that critical tasks are executed first. Specific implementation includes: Priority division: High-priority tasks, such as real-time voice translation and AR navigation and positioning, require the full module to run collaboratively (high-performance mode); low-priority tasks, such as ambient light data acquisition and sensor status monitoring, can be executed as needed in standby or sleep mode.

[0053] Resource allocation strategy: Dynamic scheduling algorithm, using the Shortest Remaining Time First (SRT) algorithm, prioritizes the execution of tasks with short expected completion times (such as "translating 'Thank you'"), freeing up resources for subsequent tasks; Multi-core processor collaboration, utilizing the multi-core architecture to allocate high-priority tasks to dedicated cores (such as GPU image processing), and low-priority tasks to energy-efficient cores (such as Cortex-M series), improving overall efficiency.

[0054] Energy efficiency optimization mechanisms: DVFS (Dynamic Voltage and Frequency Scaling) dynamically adjusts the processor frequency according to the task load in high-performance mode (e.g., reducing it from 1GHz to 500MHz) to reduce power consumption; Module-level power gating implements gated power control for non-critical modules (such as Wi-Fi chips), completely shutting off power during task breaks to further save energy.

[0055] By using a priority scheduling algorithm, we ensure that critical tasks (such as AR navigation) receive sufficient resources and avoid performance degradation due to resource contention. In addition, by combining DVFS and power gating technology, we can still achieve more than 30% energy efficiency improvement in high-performance mode and extend the device's battery life.

[0056] By dynamically adjusting power supply strategies, leveraging user behavior data, and employing task priority scheduling algorithms, the smart glasses achieve an optimal balance between performance, energy efficiency, and user experience. The technological benefits are not only reflected in a significant extension of battery life (e.g., exceeding 12 hours), but also in improved stability and usability in complex scenarios through intelligent resource management.

[0057] This application addresses the challenge of real-time user input monitoring under low power consumption by employing multimodal sensor fusion and edge AI algorithms. Dedicated sensor modules (voice, gesture, ambient light) continuously monitor user commands with microamplitude current. For example, the voice wake-up chip activates subsequent processes only when a valid command is detected. The gesture sensor interprets intent using IMU data combined with an AI model, avoiding continuous sampling and wasting energy. Simultaneously, the system fuses multi-sensor data through a heterogeneous computing architecture, transmitting task type and sensor status information (such as ambient light intensity) to the main control chip, triggering dynamic operating mode selection (low power / high performance). For instance, when a "translate" command is detected, the system automatically switches to high-performance mode and activates necessary modules (such as the camera and communication module) based on the current battery level and task complexity. Simple tasks (such as checking the time) maintain low-power mode, retaining only core functions, thus achieving a balance between performance and energy efficiency.

[0058] To address resource allocation and battery life bottlenecks under complex tasks, this application employs a multi-chip collaborative architecture and dynamic power management strategy. The image processing module uses a lightweight AI model (such as YOLOv8) to perform basic environmental analysis in low-power mode, while activating a full model (such as FastVLM) in high-performance mode to improve positioning accuracy. The communication module dynamically adjusts its transmission strategy based on task priority (such as H.265 dynamic bitrate and Wi-Fi 6E band switching) to ensure low-latency transmission of critical data (such as AR navigation). Furthermore, the power supply strategy is dynamically adjusted based on user behavior data (wearing status, task duration). For example, a pressure sensor detects the removal of the glasses and triggers a sleep mode, or a task priority scheduling algorithm (such as SRT) is used to allocate resources. Combined with DVFS technology and module-level power gating, this achieves an energy efficiency improvement of over 30% even in high-performance mode. Ultimately, through on-demand activation of hardware modules and AI-driven resource scheduling, the system balances long battery life (over 12 hours) and high performance requirements in scenarios such as AR navigation and real-time translation.

[0059] In this embodiment, the dedicated sensor module includes a voice wake-up chip, and the method further includes: 1. Low-power monitoring of voice wake-up chip: The voice wake-up chip continuously listens for voice signals in the environment (such as "start navigation") with a microampere-level current and sends a wake-up signal to the main control chip via the SPI / I2C interface. This process has been described in S100, and here we only emphasize its low power consumption characteristics (≤10μA) and event-driven triggering mechanism (activating subsequent processes only when a valid wake-up word is detected).

[0060] 2. Natural Language Processing (NLP) for extracting user intent: In speech signal processing, after wake-up, the speech signal undergoes preprocessing (denoising and framing) and feature extraction (such as MFCC and Mel spectrum) to be converted into text.

[0061] Intent recognition involves parsing text using lightweight NLP models (such as rule-based engines or fine-tuned BERT) to identify user intent (e.g., mapping "translate 'Hello'" to task parameters). This process combines intent extraction techniques to extract key information from the text (such as task type and parameters).

[0062] 3. Feature encoding of semantic embedding models: Feature vectorization uses semantic embedding models (such as Word2Vec, GloVe, or ELMo) to encode user intent text (such as “navigate to Beijing”) into low-dimensional dense vectors, capturing semantic relevance (e.g., the vectors for “navigation” and “route planning” are similar).

[0063] The encoded vector is transmitted to the main control chip via the SPI / I2C interface, serving as the basis for subsequent system operation mode selection (S200) and hardware resource scheduling.

[0064] By leveraging the low-power monitoring capabilities of a microampere-level voice wake-up chip and lightweight edge AI models (such as semantic embedding and NLP algorithms), near-zero power consumption command detection is achieved in standby mode. Simultaneously, it accurately parses user intent (such as multi-hop commands like "translate and save"), significantly reducing false wake-up rates. Combined with multimodal sensor fusion and dynamic operating mode switching (low power / high performance), the system can activate necessary hardware modules (such as cameras and communication modules) in real time based on user behavior data (such as wearing status and task duration). Through task priority scheduling algorithms and dynamic power management (DVFS, power gating), resource allocation is optimized, ensuring the performance of complex tasks such as AR navigation and real-time translation while extending device battery life (e.g., to over 12 hours), ultimately achieving a balance between energy efficiency, intelligence, and user experience.

[0065] In this embodiment of the application, the method for adjusting the power supply strategy further includes: The energy efficiency advantage of segmented voltage regulation lies in its ability to match different task requirements in high-performance mode, avoiding over-power supply (such as maintaining 1.8V all the time, which leads to no-load loss). The battery health assessment mechanism utilizes the Battery Management System (BMS) to collect voltage, temperature, and charge / discharge curve data in real time. A machine learning model is used to assess battery health (such as SoH) and dynamically adjust charging strategies (such as limiting the maximum charging current or entering sleep mode early to protect the battery).

[0066] In this embodiment of the application, the method further includes: 1. ISP engine optimization for small sensors and small apertures The image processing module utilizes an ISP engine designed specifically for small image sensors ranging from 1 / 4 inch to 1 / 5 inch. It enhances detail retention in low-light scenes through multi-frame fusion or HDR processing techniques. Addressing depth-of-field blurring caused by small apertures, the ISP engine combines edge detection or deep learning models like U-Net to identify blurred areas and restore sharpness through non-linear sharpening or high-frequency enhancement. Furthermore, given the limited angle of light incidence in small-aperture scenes, the ISP engine optimizes color reproduction and white balance parameters to reduce color casts or vignetting caused by light path deflection.

[0067] 2. Image denoising module based on convolutional neural network The ISP engine's built-in image denoising module employs a lightweight convolutional neural network. It captures noise distribution features through multi-scale convolutional layers and combines batch normalization and ReLU activation functions to improve training efficiency. The model is pre-trained using a synthetic noise dataset, including images with added Gaussian and salt-and-pepper noise. The SSIM and MSE joint loss function optimizes the structural fidelity and detail preservation of the output image. In practical applications, the CNN denoising module dynamically adjusts network parameters, such as kernel size and number of channels, to match scenes with different noise intensities. It suppresses high-frequency noise in low-light conditions and preserves texture details in high dynamic range scenes, resulting in lower computational overhead and higher image quality compared to traditional filtering algorithms.

[0068] 3. Coordinated Implementation of Shared Memory and Circular Buffer The image processing and communication modules achieve high-speed data interaction through a shared memory mechanism. This shared memory employs a circular buffer structure to improve real-time performance and stability. The two modules create a shared memory segment via Linux system calls, pre-allocating a fixed-size memory space to store image data, such as RAW format or JPEG compressed data. The circular buffer manages the data flow through read / write pointers: after the image processing module writes data, the communication module reads and transmits it. The system introduces a semaphore synchronization mechanism: when the buffer is full, the producer blocks waiting for the consumer to read; when the buffer is empty, the consumer blocks waiting for new data to be written. Simultaneously, atomic operations ensure thread safety for read / write pointer updates, and a fixed memory pool avoids frequent memory allocation, significantly reducing system latency and improving throughput.

[0069] In this embodiment of the application, the method further includes: 1. Vertical integration of Chiplet technology and hybrid bonding During the physical integration phase, Chiplet technology is used to vertically stack the voice wake-up chip, image processing chip, and communication module, achieving inter-chip interconnection through a hybrid bonding process. Chiplet technology, through modular design, manufactures different functional chips independently before integration, significantly reducing complexity and cost. The hybrid bonding process employs direct copper-copper bonding and dielectric layer alignment technology, achieving nanometer-level interconnect spacing (e.g., below 1μm) without traditional solder bumps, thereby significantly increasing the interconnect density per unit area (e.g., over 10,000 connection points per square millimeter). This high-density interconnection shortens the signal transmission path between chips, reduces latency and power consumption, and enhances the overall package's electrical performance and reliability.

[0070] 2. Key Technologies and Advantages of Hybrid Bonding Process The core of hybrid bonding technology lies in nanoscale copper-copper direct bonding and high-precision alignment of the dielectric layer. Copper-copper bonding, through atomic-level surface polishing and vacuum pressure heating, allows the copper layers between chips to fuse directly, forming a low-resistance, low-latency connection. Dielectric layer alignment, through wafer-level photolithography and precision mechanical control, ensures seamless adhesion of the insulating layers between chips, avoiding the risk of short circuits. This process supports ultra-fine pitch (e.g., below 9μm) and is compatible with technologies such as TSV, enabling chip stacking of up to 16 layers or more. Compared to traditional bump bonding, hybrid bonding significantly improves interconnect density, thermal management efficiency, and long-term stability, for example, enabling 420-layer NAND stacking in HBM4 or improving heat dissipation performance in SK Hynix HBM3E.

[0071] 3. Waterproof sealing design and IP67 protection achieved The frame employs a multi-layer waterproof sealant process, where industrial-grade sealant is evenly applied to the U-shaped structure using a high-precision dispensing machine to form a continuous sealing layer. Silicone gaskets are added at the temple joints to further prevent moisture penetration. The sealant and silicone gaskets work synergistically, passing the IP67 protection rating test: after a 30-minute water spray test, the sample showed no bubbles generated when inflated to 10kPa in an IPX7.8 leak detector, proving its dustproof and waterproof capabilities. This design meets the requirements of complex scenarios such as high humidity and high temperature, while maintaining both aesthetic integrity and functionality through a residue-free curing process, similar to the testing and verification procedures for outdoor cameras.

[0072] This application discloses a low-power multi-chip collaborative control system for smart glasses, referring to... Figure 2 ,include: The wake-up trigger signal generation module 001 monitors user input signals in real time in a low-power state through a dedicated sensor module to generate a wake-up trigger signal. The user input signal includes voice, gesture or environmental changes, and the wake-up trigger signal includes user intent information. The system operation mode selection module 002 obtains the current resource status corresponding to the smart glasses and dynamically selects the system operation mode based on user intent information and the current resource status. The system operation mode includes low power mode or high performance mode. In low power mode, only the functional modules corresponding to the current task are activated. In high performance mode, all functional modules are activated and a dynamic resource allocation strategy is started. Hardware module retrieval module 003 is used to call the corresponding hardware modules to perform data acquisition, data processing and data transmission tasks according to the system operation mode, including: an image processing module, used to combine AI algorithms to complete environmental analysis and location data extraction; a communication module, used to dynamically adjust the data transmission strategy according to task priority; and a sensor array, used to collect user posture, ambient light and wearing status information in real time. The power supply strategy adjustment module 004 adjusts the power supply strategy based on the system operation mode and real-time user behavior data. The real-time user behavior data includes wearing status and task duration. The power supply strategies include standby mode (only core modules run), hibernation mode (some modules are shut down), and high-performance mode (all modules run in coordination). Hardware resource allocation is optimized through a task priority scheduling algorithm.

[0073] This application also discloses a low-power multi-chip collaborative control system for smart glasses, including a processor, wherein the processor runs a program of any one of the above-described low-power multi-chip collaborative control methods for smart glasses.

[0074] This application also discloses a storage medium storing a program for the low-power multi-chip collaborative control method for smart glasses as described in any one of the above embodiments.

[0075] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A low-power multi-chip collaborative control method for smart glasses, characterized in that, include: A dedicated sensor module monitors user input signals in real time with low power consumption and generates a wake-up trigger signal. The user input signals include voice, gestures, or environmental changes, and the wake-up trigger signal includes user intent information. The system obtains the current resource status corresponding to the smart glasses, and dynamically selects the system operation mode based on the user intent information and the current resource status. The system operation mode includes low power mode or high performance mode. In low power mode, only the functional modules corresponding to the current task are activated. In high performance mode, all functional modules are activated and a dynamic resource allocation strategy is started. According to the system's operating mode, the corresponding hardware modules are invoked to perform data acquisition, data processing, and data transmission tasks, including: an image processing module, used to combine AI algorithms to complete environmental analysis and location data extraction; a communication module, used to dynamically adjust the data transmission strategy according to task priority; and a sensor array, used to collect user posture, ambient light, and wearing status information in real time. The power supply strategy is adjusted based on the system operation mode and real-time user behavior data. The real-time user behavior data includes wearing status and task duration. The power supply strategy includes standby mode (only core modules run), hibernation mode (some modules are shut down), and high-performance mode (all modules run in coordination). Hardware resource allocation is optimized through a task priority scheduling algorithm.

2. The low-power multi-chip collaborative control method for smart glasses according to claim 1, characterized in that, The dedicated sensor module includes a voice wake-up chip, and the method further includes: The voice wake-up chip continuously listens to voice commands with a microampere-level current and sends a wake-up signal to the main control chip via an SPI or I2C interface. The voice wake-up chip extracts user intent data through a natural language processing algorithm, and transmits the user intent data to the main control chip after feature encoding through a semantic embedding model.

3. The low-power multi-chip collaborative control method for smart glasses according to claim 2, characterized in that, In the process of the communication module adjusting the data transmission strategy according to task priority, the method further includes: In low-power mode, a lightweight data compression algorithm is used to transmit map update information or location data, and low-bandwidth communication is performed based on the 2.4GHz band of Wi-Fi 6E. In high-performance mode, video streams or voice data are transmitted based on H.265 dynamic bitrate control technology, and signal interference is reduced by joint frequency band switching of Wi-Fi 6E and Bluetooth 5.3 protocols, with latency controlled within 2 milliseconds; The dynamic bitrate control technology combines real-time network status monitoring of edge computing nodes to dynamically adjust video encoding resolution and frame rate.

4. The low-power multi-chip collaborative control method for smart glasses according to claim 3, characterized in that, The method for adjusting the power supply strategy further includes: When the wearing status detection circuit triggers the lens removal signal, it enters deep sleep mode with a response time of no more than 50 milliseconds; In high-performance mode, the voltage and charging current are dynamically adjusted through the ET9562 power management chip, combined with a battery health assessment mechanism. The dynamic voltage adjustment includes: adjusting the core module power supply voltage in segments from 1.0V to 1.8V according to changes in task load to reduce dynamic power consumption.

5. The low-power multi-chip collaborative control method for smart glasses according to claim 4, characterized in that, The method also includes: The image processing module calls a pre-set ISP engine, which supports small image sensors from 1 / 4 inch to 1 / 5 inch and small aperture optimization. The ISP engine includes an image denoising module based on a convolutional neural network; The image processing module interacts with the communication module through a shared memory mechanism, wherein the shared memory adopts a circular buffer structure.

6. The low-power multi-chip collaborative control method for smart glasses according to claim 5, characterized in that, The method also includes: The physical integration stage uses Chiplet technology to vertically stack the voice wake-up chip, image processing chip and communication module, and achieves inter-chip interconnection through hybrid bonding process. The hybrid bonding process includes nanoscale copper-copper direct bonding and dielectric layer alignment technology, wherein the interconnect density is higher than 10,000 connection points per square millimeter; The frame uses a multi-layer waterproof sealant process, and silicone gaskets are added at the temple joints, passing the IP67 protection level test.

7. A low-power multi-chip collaborative control system for smart glasses, characterized in that, include: The wake-up trigger signal generation module monitors user input signals in real time in a low-power state through a dedicated sensor module to generate a wake-up trigger signal. The user input signals include voice, gestures or environmental changes, and the wake-up trigger signal includes user intent information. The system operation mode selection module obtains the current resource status corresponding to the smart glasses. Based on the user intent information and the current resource status, it dynamically selects the system operation mode, which includes a low-power mode or a high-performance mode. In the low-power mode, only the functional modules corresponding to the current task are activated. In the high-performance mode, all functional modules are activated and a dynamic resource allocation strategy is initiated. The hardware module invocation module, based on the system operating mode, calls the corresponding hardware modules to perform data acquisition, data processing, and data transmission tasks, including: an image processing module, used to combine AI algorithms to complete environmental analysis and location data extraction; a communication module, used to dynamically adjust the data transmission strategy according to task priority; and a sensor array, used to collect user posture, ambient light, and wearing status information in real time. The power supply strategy adjustment module adjusts the power supply strategy based on the system operation mode and real-time user behavior data. The real-time user behavior data includes wearing status and task duration. The power supply strategy includes standby mode (only core modules run), hibernation mode (some modules are shut down), and high-performance mode (all modules run in coordination). Hardware resource allocation is optimized through a task priority scheduling algorithm.

8. A low-power multi-chip collaborative control system for smart glasses, characterized in that, Includes a processor, wherein the processor runs a program for a low-power multi-chip collaborative control method for smart glasses as described in any one of claims 1-6.

9. A storage medium, characterized in that, The device stores a program for a low-power multi-chip collaborative control method for smart glasses as described in any one of claims 1-6.

Images