Commute mode recognition and intervention method and system, readable medium

Abstract

This disclosure provides a commuting pattern recognition and intervention method and system. The method includes: collecting multimodal sensor data of a wearer, including at least inertial sensor data, when a preset commuting trigger condition is met based on a preset time window and changes in environmental beacon signals obtained through passive scanning; segmenting the multimodal sensor data and extracting time-frequency domain features while retaining the amplitude information of the data to obtain a feature data sequence; inputting the feature data sequence into a machine learning model to obtain a traffic state recognition result; performing time-series constraint processing on the traffic state recognition result to generate commuting trajectory slices; determining the walking equivalent based on the commuting trajectory slices, and generating an intervention strategy when the walking equivalent is lower than a preset threshold; and during subsequent commuting, detecting whether the trigger condition corresponding to the intervention strategy is met based on the commuting trajectory slices, and outputting prompt feedback information when the condition is met.

Description

Technical Field

[0001] This disclosure relates to the fields of smart wearable devices and digital health technology, and in particular to a commuting pattern recognition and intervention method and system, and a computer-readable medium. Background Technology

[0002] With the development of digital health technology, the continuous monitoring and analysis of users' daily activities using smart wearable devices has become an important technological means to promote public health management. In the practice of promoting children's and adolescents' health, commuting between home and school, as an important daily life scenario, has high potential for health intervention. By increasing physical activity such as walking during commutes, it is possible to effectively improve the daily exercise levels of children and adolescents, which has a positive effect on preventing health problems such as obesity and myopia.

[0003] How to accurately identify the complex patterns of multiple modes of transportation during commuting between home and school, and provide personalized health intervention suggestions based on the identification results, has become a technical problem that urgently needs to be solved in the field of digital health technology. Summary of the Invention

[0004] This disclosure provides a commuting pattern recognition and intervention method and system, a computer-readable medium, and a computer program product. This disclosure aims to address the following problems in commuting recognition and health intervention technologies in some related technologies: First, reliance on high-frequency GPS positioning leads to high device power consumption and poses a risk of location privacy leakage; second, traditional algorithms struggle to distinguish between signals generated by vehicle bumps and actual walking, failing to accurately filter false step counts and perform refined recognition and segmentation of the "mixed travel and walking" commuting process; third, many related technologies rely primarily on post-event statistics, lacking dynamic analysis and real-time intervention capabilities based on user commuting habits, making it difficult to provide actionable, scenario-based behavioral guidance.

[0005] In a first aspect, embodiments of this disclosure provide a commuting pattern recognition and intervention method, comprising: collecting multimodal sensor data of a wearer when a preset commuting trigger condition is met based on a preset time window and changes in environmental beacon signals obtained through passive scanning, wherein the multimodal sensor data includes at least inertial sensor data; segmenting the multimodal sensor data and extracting time-frequency domain features while retaining the amplitude information of the data to obtain a feature data sequence; inputting the feature data sequence into a machine learning model to obtain a traffic state recognition result; performing time-series constraint processing on the traffic state recognition result to generate commuting trajectory slices; determining the walking equivalent based on the commuting trajectory slices, and generating an intervention strategy when the walking equivalent is lower than a preset threshold; and during subsequent commuting, detecting whether the trigger condition corresponding to the intervention strategy is met based on the commuting trajectory slices, and outputting prompt feedback information when the condition is met.

[0006] In some embodiments, satisfying preset commuting trigger conditions includes at least one of the following situations: First situation: When the strength of a beacon signal corresponding to a first environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within a first commuting time window, it is determined that the first commuting start condition is met, and the collection of the multimodal sensing data begins; When a beacon signal corresponding to a second environmental feature is detected to change from an undetectable state to a detectable state in multiple consecutive scans, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state for a preset duration, it is determined that the first commuting end condition is met. The collection of multimodal sensor data is stopped. Second scenario: When the strength of a beacon signal corresponding to the second environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time falls within the second commuting time window, the second commuting start condition is determined to be met, and the collection of multimodal sensor data begins. When the beacon signal corresponding to the first environmental feature changes from an undetectable state to a detectable state in multiple consecutive scans, and the traffic state identified based on the multimodal sensor data continuously and stably changes from a walking state to a stationary state for a preset duration, the second commuting end condition is determined to be met, and the collection of multimodal sensor data is stopped.

[0007] In some embodiments, inputting the feature data sequence into a machine learning model to obtain traffic state recognition results includes: inputting the feature data sequence into a lightweight machine learning model deployed on the edge; the lightweight machine learning model outputs a transient traffic state probability distribution for each feature vector in the feature data sequence corresponding to a time window; the probability distribution characterizes the confidence level of each traffic state and constitutes the traffic state recognition result; the traffic state includes at least one or more of the following: stationary state, smooth riding state, bumpy riding state, cycling state, and walking state.

[0008] In some embodiments, the traffic state identification results are subjected to temporal constraint processing to generate commuter trajectory slices, including: performing temporal smoothing processing based on state transition constraints on the sequence of traffic state identification results arranged in chronological order to filter out isolated state jumps caused by instantaneous interference; and aggregating continuous time periods with a single traffic semantic based on the smoothed sequence to generate the commuter trajectory slices.

[0009] In some embodiments, a walking equivalent is determined based on the commuting trajectory slices, and an intervention strategy is generated when the walking equivalent is lower than a preset threshold. This includes: constructing a user commuting profile based on historical commuting trajectory slices or current and historical commuting trajectory slices, wherein the user commuting profile includes commuting mode combinations, commuting distance, and walking equivalent; and generating an intervention strategy containing spatial anchor points based on the user commuting profile when the walking equivalent is lower than the preset threshold.

[0010] In some embodiments, generating an intervention strategy that includes spatial anchor points includes: if the commuting distance is less than or equal to a preset threshold, generating a full-journey walking intervention task; if the commuting distance is greater than the preset threshold, then based on historical commuting trajectory slices or the time distribution corresponding to the current and historical commuting trajectory slices, reversely calculating recommended spatial anchor points for early termination of commuting, and generating a phased walking intervention task that switches from commuting to walking accordingly.

[0011] Secondly, embodiments of this disclosure provide a commuting pattern recognition and intervention system, comprising: a wearable device, including: a multimodal sensing module, configured to periodically perform passive scanning at a preset duty cycle to detect environmental beacon signals, and to wake from a dormant state when a preset commuting trigger condition determined based on a preset time window and changes in the environmental beacon signals is met, and to collect multimodal sensing data of the wearer, wherein the multimodal sensing data includes at least inertial sensing data; and a microcontroller, connected to the multimodal sensing module, configured to: segment the multimodal sensing data, extract time-frequency domain features while retaining the data amplitude information, and obtain features. The system comprises: a data sequence; inputting the feature data sequence into a machine learning model to obtain traffic state recognition results; performing temporal constraint processing on the traffic state recognition results to generate commuting trajectory slices; an interactive feedback and communication module for uploading the commuting trajectory slices and receiving intervention strategies, and outputting prompt feedback information when the microcontroller detects that the triggering conditions corresponding to the intervention strategy are met based on the commuting trajectory slices during subsequent commutes; and a server for acquiring the commuting trajectory slices, determining the walking equivalent based on the commuting trajectory slices, generating intervention strategies based on the walking equivalent, and sending the intervention strategies to the wearable device.

[0012] In some embodiments, the system further includes: a user terminal device, which is communicatively connected to the wearable device and the server, respectively, for receiving the commuting trajectory slices and / or the intervention strategy, and displaying commuting information and the intervention strategy to the user.

[0013] In some embodiments, the multimodal sensing module includes: a low-frequency scanning and assisted positioning unit, configured to periodically perform passive scanning at a preset duty cycle to detect environmental beacon signals; an inertial measurement unit, initially in a sleep state; the microcontroller is further configured to: generate a wake-up command and send it to the inertial measurement unit when the strength of a beacon signal corresponding to a first environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within a first commuting time window; generate a sleep command and send it to the inertial measurement unit when a beacon signal corresponding to a second environmental feature changes from an undetectable state to a detectable state in multiple consecutive scans, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state for a preset duration. An inertial measurement unit (IMU) generates a wake-up command and sends it to the IMU when, during multiple consecutive scans, the strength of a beacon signal corresponding to a second environmental feature is detected to be lower than a preset threshold, and the current time falls within a second commuting time window. When, during multiple consecutive scans, the strength of a beacon signal corresponding to a first environmental feature changes from an undetectable state to a detectable state, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state for a preset duration, the IMU generates a sleep command and sends it to the IMU. The IMU, in response to the wake-up command, switches from a sleep state to a working state to initiate the acquisition of the multimodal sensing data, and in response to the sleep command, stops the data acquisition and returns to a sleep state.

[0014] Thirdly, embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method according to any one of the first aspects.

[0015] Fourthly, embodiments of this disclosure provide a computer program product, including a computer program or instructions, which, when executed by a processor, implement the method according to any one of the first aspects.

[0016] This disclosure employs a collaborative triggering mechanism combining passive scanning of environmental beacons with low duty cycle and inertial sensor data acquisition. High-frequency acquisition is only activated when commuting trigger conditions are met, thereby reducing system power consumption. Traffic state recognition and trajectory slice generation are performed at the edge, reducing reliance on continuous positioning data and enhancing privacy protection. Simultaneously, by preserving data amplitude information during feature extraction and combining it with time-frequency domain feature modeling, the differences between walking and vehicle bumpy states are effectively distinguished, and temporal constraint processing improves the stability and accuracy of the recognition results. Based on this, walking equivalents are determined based on commuting trajectory slices to generate intervention strategies. Trigger conditions are detected during subsequent commutes, and feedback information is output, thus achieving a closed loop of recognition, evaluation, decision-making, and feedback, improving the executability of intervention measures. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the hardware configuration of the smart wearable device provided in the embodiments of this disclosure.

[0018] Figure 2 This is a schematic diagram of the hardware architecture of a commuting pattern recognition and intervention system based on multi-sensor fusion according to an embodiment of this disclosure.

[0019] Figure 3 This is a flowchart of a commuting pattern recognition and intervention method according to an embodiment of this disclosure.

[0020] Figure 4 This is a schematic diagram of the workflow of the scenario-based health intervention and recommendation engine according to an embodiment of this disclosure.

[0021] Figure 5 This is a block diagram of a commuting pattern recognition and intervention system according to an embodiment of the present disclosure. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions of this disclosure will be described in detail below with reference to the accompanying drawings.

[0023] Exemplary embodiments will be described more fully below with reference to the accompanying drawings; however, these exemplary embodiments may be embodied in different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will enable those skilled in the art to fully understand the scope of this disclosure.

[0024] Where there is no conflict, the various embodiments of this disclosure and the features thereof in the embodiments may be combined with each other.

[0025] As used herein, the term “and / or” includes any and all combinations of one or more related enumerated entries.

[0026] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. As used herein, the singular forms “a” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “made of” are used in this specification, the presence of the stated feature, integral, step, operation, element, and / or component is specified, but the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded.

[0027] Unless otherwise specified, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and will not be interpreted as having an idealized or overly formal meaning, unless expressly so defined herein.

[0028] Currently, the monitoring technologies for user commuting behavior mainly rely on the following types of solutions: The first type of approach is trajectory analysis based on the Global Positioning System (GPS). This technology typically involves continuously collecting the user's location information through smartphones or wearable devices with positioning capabilities, and determining whether the user is walking, cycling, or traveling by car based on movement speed and trajectory characteristics. However, continuously using GPS positioning results in high energy consumption, which is detrimental to the long-term operation of lightweight wearable devices with limited battery capacity (such as smart glasses, smart bracelets, or smart rings). Furthermore, high-frequency collection of user location trajectories may raise privacy concerns.

[0029] The second type of approach is motion detection based on accelerometers. Common pedometers or smart bracelets use acceleration signals to identify a user's gait and thus count steps or activity levels. However, these methods typically only output simple statistical metrics like total steps and lack the ability to identify specific activity scenarios. In complex commuting environments, such as riding a bumpy bus or using a mix of modes of transportation, these algorithms are prone to misjudgments and struggle to accurately identify changes in mode of transport.

[0030] Furthermore, most health monitoring systems primarily focus on passively recording users' exercise data and performing statistical analysis afterward, lacking the ability to intervene in real-time behaviors tailored to specific life scenarios. Therefore, in the context of children and adolescents commuting between home and school, it is difficult to provide actionable, phased exercise recommendations based on actual commuting patterns.

[0031] Therefore, how to accurately identify the complex patterns of multiple modes of transportation during commuting between home and school without relying on continuous positioning technology, and to provide personalized health intervention suggestions based on the identification results, has become a technical problem that urgently needs to be solved in the field of digital health technology.

[0032] This disclosure aims to solve the following technical problems: In some related technologies, commuting mode recognition technology typically relies on Global Positioning System (GPS) positioning or simple accelerometer motion detection algorithms, which struggle to balance power consumption, recognition accuracy, and privacy protection in lightweight smart wearable devices. Furthermore, some related technologies struggle to accurately identify combined commuting modes such as "driving + walking," and easily misinterpret vibrations from transportation as walking motion. In addition, health monitoring devices mostly remain at the passive recording stage, lacking the ability to dynamically intervene in commuting scenarios.

[0033] Therefore, the technical problem this disclosure aims to solve is how to accurately identify commuting modes between home and school using only low-power sensors without relying on continuous GPS positioning, and how to generate health intervention strategies with scene awareness capabilities based on the identification results.

[0034] Figure 1 This is a schematic diagram of the hardware configuration of a smart wearable device provided in an embodiment of this disclosure. The hardware architecture of this disclosure does not rely on a high-power independent GPS module, or only activates the GPS module at a low frequency when necessary, and is integrated entirely within a lightweight smart wearable device. The smart wearable device can be a wrist-worn device (such as a smart bracelet or smartwatch) or a head-worn device (such as smart glasses or smart headphones), and this disclosure does not impose any special limitations on it.

[0035] like Figure 1 As shown, the smart wearable device includes at least: a processor 101, a memory 102, an inertial measurement unit 103, a low-frequency scanning unit 104, a communication module 105, a vibration module 106, and a power management unit 107.

[0036] The power management unit 107 provides power to all hardware components and works with the processor 101 to implement multi-level sleep and wake-up control. The processor 101 is connected to the memory 102, inertial measurement unit 103, low-frequency scanning unit 104, communication module 105, and vibration module 106, and is used to perform data processing, status recognition, and feedback control. The inertial measurement unit 103 is in sleep mode by default and is woken up by the processor 101 as needed based on the detection results of the low-frequency scanning unit 104 and time window conditions.

[0037] The specific functions of each hardware component are as follows: The processor 101 is a low-power microcontroller, such as an MCU based on the Cortex-M4 architecture or with equivalent computing power, with a built-in DSP extended instruction set and a sufficient on-chip SRAM contiguous memory pool (Arena) for executing edge feature extraction, model inference, timing smoothing and other algorithms to generate commuting trajectory slices.

[0038] The memory 102, including, for example, ROM, RAM, and non-volatile memory (such as flash memory), is used to store program instructions, model parameters, and commuting trajectory slice data. The instructions are used to control the processor 101 to execute the commuting pattern recognition and intervention method described in this embodiment of the disclosure.

[0039] An inertial measurement unit (IMU) 103, such as a high-frequency sensor, employing a 6-axis or 9-axis structure (including an accelerometer, gyroscope, and optional geomagnetic sensor), is used to accurately capture the wearer's three-dimensional motion posture and physical force characteristics under different traffic conditions. This unit is in deep sleep mode by default and is only awakened by the processor 101 when preset commuting trigger conditions are met, acquiring motion data at a preset sampling rate (e.g., 50Hz).

[0040] The low-frequency scanning unit 104, including, for example, a low-power Bluetooth probe or a Wi-Fi probe, performs a passive scan periodically at an extremely low duty cycle (e.g., waking up every few seconds for hundreds of milliseconds) to detect beacon signals (including MAC addresses and RSSI signal strength) in the environment. This unit employs BLE Observer mode or Wi-Fi sniffing mode, receiving only broadcast packets and not sending active scan requests or initiating any pairing or connection handshakes.

[0041] The communication module 105 includes at least a Bluetooth Low Energy unit for asynchronously transmitting structured small-data features generated by edge computing to an external device and receiving intervention tasks. Optionally, it also includes a Wi-Fi communication unit.

[0042] The vibration module 106 includes a vibration motor and its drive circuit, which is used to send a low-interference tactile vibration alert to the wearer under the control of the processor 101.

[0043] The power management unit 107 is used to provide stable voltage conversion for the above-mentioned units and, together with the processor 101, implements a microampere-level multi-level sleep and wake-up mechanism.

[0044] Optionally, the smart wearable device may also include the following auxiliary components: a display module 108, such as an LCD screen, LED screen, or touch screen, for displaying device status or prompt information; an input module 109, such as a touch screen or keyboard, for receiving user input; a speaker 110 for outputting audio information; a microphone 111 for collecting ambient audio information; and an audio-visual warning module 112 for providing audio-visual warnings under the control of the processor 101, including flashing lights and buzzer alerts. In some embodiments, these auxiliary modules may be omitted to reduce cost and power consumption.

[0045] Those skilled in the art should understand that Figure 1The structure shown does not constitute a limitation on smart wearable devices and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Figure 1 The smart wearable devices shown are for illustrative purposes only and do not constitute any limitation on the scope of protection, application scenarios, or uses of this disclosure.

[0046] Figure 2 This is a schematic diagram of the hardware architecture of a commuting pattern recognition and intervention system based on multi-sensor fusion according to an embodiment of this disclosure.

[0047] Figure 2 The overall hardware shown can be integrated into lightweight smart wearable devices, such as smart bracelets or smart glasses. The system revolves around low-power edge computing and multimodal perception, without relying on a high-power standalone GPS module. Through the coordinated operation of environmental perception and inertial measurement, it achieves accurate identification and intervention during commutes. The system mainly includes an edge computing and main control module, a multimodal perception module, an interactive feedback and communication module, and a power management module. These modules interact collaboratively through data and control signals.

[0048] In this embodiment, the edge computing and main control module is implemented using a low-power microcontroller (MCU). The microcontroller is preferably based on the Cortex-M4 architecture or has equivalent computing power, natively supports tflite-micro, and supports DSP extended instruction sets and a contiguous memory pool (Arena) management mechanism. This microcontroller receives raw data from the multimodal perception module, including inertial measurement data and environmental beacon scan data, and performs feature extraction, model inference, and temporal smoothing algorithms on the edge side to generate commuting trajectory slices with semantic information.

[0049] The multimodal sensing module includes an inertial measurement unit (IMU) and a low-frequency scanning and auxiliary positioning unit. The IMU is preferably a 6-axis or 9-axis sensor used to acquire the wearer's three-axis acceleration, three-axis angular velocity, and optional three-axis geomagnetic data to reflect their motion posture and force characteristics. The IMU is a high-power device in the system, and it is in a deep sleep state by default, only being awakened to perform high-frequency data acquisition when specific trigger conditions are met.

[0050] The low-frequency scanning and assisted positioning unit can integrate Bluetooth Low Energy (BLE) or Wi-Fi probes to acquire environmental beacon information through passive scanning. In specific implementations, this unit operates in BLE's Observer mode or Wi-Fi's sniffing mode, only receiving beacon data (including MAC addresses and RSSI signal strength) periodically broadcast by surrounding devices, without initiating active scanning requests or connection handshakes. This reduces RF power consumption while avoiding frequent exposure of the device's own identification information.

[0051] Furthermore, the low-frequency scanning and auxiliary positioning unit operates periodically with a preset low duty cycle, for example, waking up every few seconds to perform a scan every few hundred milliseconds, and immediately entering a sleep state after completing the scan. By recording the scan results with timestamps and performing sequence analysis, environmental fingerprint feature sets for "home" and "school" can be constructed, thereby forming a low-frequency electronic fence.

[0052] In this embodiment, the low-frequency scanning and auxiliary positioning unit and the inertial measurement unit (IMU) form a master-slave wake-up relationship. That is, in non-commuting mode, the IMU is always in a sleep state, and only the low-frequency scanning unit performs periodic monitoring. When a preset commuting trigger condition is detected, the low-frequency scanning unit sends a wake-up trigger signal to the microcontroller, which then sends a wake-up control command to the IMU.

[0053] Specifically, the commuting trigger condition is preferably a dual constraint condition based on time and environmental characteristics. For example, within a preset commuting time window, when multiple consecutive scans show that the "home" environmental beacon features have significantly weakened or disappeared, it is determined as "leaving home trigger," thereby waking up the inertial measurement unit to start high-frequency data acquisition; when the "school" environmental beacon features are detected, and the identification results combined with the edge-side algorithm show that the state changes from walking to stationary and remains stable for a period of time, it is determined as "arriving at school trigger." At this time, the microcontroller sends a sleep command to the inertial measurement unit, thereby stopping data acquisition.

[0054] Through the above mechanism, the inertial measurement unit is only in working condition on the "home-school commute" route, and remains dormant at other times (such as at home or at school), thereby avoiding the power consumption and redundant data problems caused by all-day data collection.

[0055] Furthermore, the microcontroller communicates with the inertial measurement unit (IMU) and dynamically configures the IMU's operating mode, including wake-up and sleep control, by writing control instructions to its internal registers. For example... Figure 2 As shown, the microcontroller outputs a "wake-up / sleep control command" to the inertial measurement unit to achieve on-demand sampling, thereby reducing the overall system power consumption.

[0056] The interactive feedback and communication module includes a vibration feedback unit and a low-power data synchronization unit. The vibration feedback unit is preferably a miniature vibration motor. The microcontroller sends a "vibration trigger command" to the motor drive circuit by outputting GPIO control signals or PWM drive signals to provide tactile feedback to the wearer, such as: (1) When the wearer is detected to have reached the spatial anchor point corresponding to the intervention strategy, the first vibration prompt is triggered to guide the wearer to switch from taking a vehicle to walking. (2) When it is detected that the wearer has met the target walking conditions corresponding to the intervention strategy, a second vibration prompt is triggered to provide feedback on the completion of the target behavior.

[0057] The low-power data synchronization unit is preferably implemented based on the BLE communication protocol. It is used to upload structured data such as commuting trajectory slices and effective walking equivalents generated by the edge computing to the user terminal or the cloud, while receiving intervention strategies. Compared with transmitting raw sensor data streams, this method significantly reduces the amount of communication data and power consumption.

[0058] The power management module provides a stable voltage supply to all hardware units and works with the microcontroller to implement a multi-level sleep management mechanism. During non-commuting periods, in addition to the low-frequency scanning unit operating at a low duty cycle, the microcontroller itself is periodically and briefly woken up via an internal low-power timer (RTC) to perform tasks only when necessary, thus keeping the overall system at a microampere-level power consumption. This mechanism is manifested as a "tiered sleep" strategy where each functional module operates at different sleep depths according to task requirements.

[0059] In summary, this embodiment, through the collaborative design of low-frequency environmental perception and high-frequency inertial measurement, combined with a master-slave wake-up control mechanism, significantly reduces system power consumption while ensuring commuting recognition accuracy, providing a reliable data foundation for subsequent edge-side algorithm processing and health intervention.

[0060] Figure 3 This is a flowchart of a commuting pattern recognition and intervention method according to an embodiment of this disclosure.

[0061] Firstly, referring to Figure 3 This disclosure provides a commuting pattern recognition and intervention method, including: S11. When the preset commuting triggering conditions determined based on the preset time window and the changes in environmental beacon signals obtained through passive scanning are met, multimodal sensing data of the wearer is collected, wherein the multimodal sensing data includes at least inertial sensing data. S12. The multimodal sensing data is segmented and the time-frequency domain features are extracted while retaining the data amplitude information to obtain a feature data sequence. S13. Input the feature data sequence into the machine learning model to obtain traffic state recognition results; S14. Perform time-series constraint processing on the traffic state recognition results to generate commuter trajectory slices; S15. Determine the walking equivalent based on the commuting trajectory slice, and generate an intervention strategy when the walking equivalent is lower than a preset threshold. S16. During the subsequent commute, detect whether the triggering conditions corresponding to the intervention strategy are met based on the commuting trajectory slice, and output prompt feedback information when they are met.

[0062] This disclosure provides a method for commuting pattern recognition and intervention. First, when preset commuting trigger conditions are met, multimodal sensor data of the wearer is collected. The preset commuting trigger conditions are determined based on a preset commuting time window and changes in environmental beacon signals obtained through low-duty-cycle passive scanning. When a change in the beacon signal corresponding to the home or school environment is detected, and the current time falls within the commuting time window, it is determined that the user has entered a commuting state, thereby waking up the inertial measurement unit and initiating data collection.

[0063] The acquired multimodal sensing data includes at least inertial sensing data, wherein the inertial measurement unit can be a 6-axis or 9-axis sensor, outputting acceleration, angular velocity, and optional geomagnetic data, respectively. In other embodiments, the multimodal sensing data may also include geomagnetic sensor data, optical sensor data, or data from a low-frequency global navigation satellite system for auxiliary correction.

[0064] Subsequently, the multimodal sensing data is segmented, and time-domain and frequency-domain features are extracted while preserving the data amplitude information to form a feature data sequence. Then, the feature data sequence is input into a pre-deployed machine learning model to obtain traffic state recognition results for each time window.

[0065] Furthermore, the traffic state recognition results are subjected to temporal constraint processing to eliminate transient noise interference and generate continuous commuter trajectory slices with semantic consistency.

[0066] Based on the commuting trajectory slices, the walking equivalent during the commute is determined, and an intervention strategy is generated when the walking equivalent falls below a preset threshold. During subsequent commutes, the system checks whether the triggering conditions corresponding to the intervention strategy are met based on the real-time generated commuting trajectory slices. When met, a prompt feedback message is output through the wearable device to guide the user to complete a preset action. The feedback message may include one or more of the following: vibration, sound, graphical interface prompts, flashing lights, or voice announcements.

[0067] In some embodiments, satisfying preset commuting trigger conditions includes at least one of the following situations: First scenario: When the strength of the beacon signal corresponding to the first environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within the first commuting time window, it is determined that the first commuting start condition is met, and the collection of the multimodal sensing data begins. When a beacon signal corresponding to the second environmental feature is detected to change from an undetectable state to a detectable state in multiple consecutive scans, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state for a preset duration, it is determined that the first commuting end condition is met, and the collection of the multimodal sensing data is stopped. Second scenario: When the strength of the beacon signal corresponding to the second environmental feature is detected to be lower than the preset threshold in multiple consecutive scans, and the current time is within the second commuting time window, it is determined that the second commuting start condition is met, and the collection of the multimodal sensing data begins. When a beacon signal corresponding to the first environmental feature is detected to change from an undetectable state to a detectable state in multiple consecutive scans, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state for a preset duration, it is determined that the second commuting end condition is met, and the collection of the multimodal sensing data is stopped.

[0068] In this embodiment, the preset commuting trigger condition is determined by changes in environmental beacon signals. Specifically, within the preset commuting time window, the device periodically performs passive scanning with a preset duty cycle, only receiving Bluetooth, Wi-Fi, or other low-power wireless beacon broadcast signals from the surrounding environment, without actively initiating a connection. In other embodiments, the environmental beacon signals may also include ultra-wideband signals, RFID tag signals, or broadcast signals from fixedly deployed dedicated beacon devices.

[0069] By analyzing the results of multiple consecutive scans, environmental feature information is extracted, including the presence of beacon signals and changes in their signal strength.

[0070] In the first scenario, when the beacon signal strength corresponding to the first environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within the first commuting time window, it can be determined that the wearer has left the starting position (first environmental position), thereby meeting the commuting start condition and triggering multimodal sensor data acquisition. When a beacon signal corresponding to the second environmental feature is detected to change from an undetectable state to a detectable state, and the traffic state identified based on inertial data changes from a walking state to a stationary state and remains stable within a preset time, it can be determined that the destination (second environmental location) has been reached, thus satisfying the commuting end condition and stopping data collection.

[0071] In the second scenario, when the beacon signal strength corresponding to the second environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within the second commuting time window, it can be determined that the wearer has left the second environmental location, thereby meeting the second commuting start condition and starting multimodal sensing data acquisition. When a beacon signal corresponding to the first environmental feature is detected to change from an undetectable state to a detectable state during multiple consecutive scans, and the traffic state changes from a walking state to a stationary state and remains stable within a preset time, it can be determined to return to the first environmental location, thereby satisfying the second commuting end condition and stopping data collection.

[0072] In practical applications, the commuting triggering condition can be either the first or the second scenario described above, or both scenarios can be used simultaneously to detect one-way or two-way commuting processes. This disclosure does not limit this. For example, using both the first and second scenarios simultaneously can achieve complete coverage of the "home-school round trip" two-way commuting process.

[0073] In terms of triggering mechanisms, in addition to determining based on changes in signal strength, other methods can be used, such as determining based on changes in the matching degree of signal sets, event detection based on the appearance or disappearance of signals, changes in signal statistical characteristics, or classification and determination of environmental states based on machine learning models.

[0074] In some embodiments, the multimodal sensing data is segmented, and time-frequency domain features are extracted while retaining the data amplitude information to obtain a feature data sequence, including: The multimodal sensing data is segmented based on a sliding window of fixed length, with a preset overlap between adjacent windows; For the sensing data within each window, time-domain features and frequency-domain features are extracted while retaining the original sensing data amplitude information, and corresponding feature vectors are generated. The feature vectors are arranged in chronological order to obtain the feature data sequence.

[0075] In this embodiment, the segmentation of multimodal sensing data is achieved using a sliding window mechanism. Specifically, continuously acquired time-series data is segmented into fixed-length windows with a preset overlap ratio between adjacent windows, such as a window length of 2.56 seconds and an overlap rate of 50%. In other embodiments, time windows of different lengths or adaptive lengths can be used, or window segmentation strategies with no overlap or multiple overlaps can be employed.

[0076] Within each window, feature extraction is performed on the raw sensor data, preserving the amplitude information of the data during the feature extraction process without global normalization, to avoid losing crucial physical force difference information. In other implementations, the sensor data can also be denoised to adapt to different scenario requirements.

[0077] The extracted features include time-domain features (such as mean, variance, peak-to-average ratio, and zero-crossing rate) and frequency-domain features (spectral energy distribution obtained through Fast Fourier Transform). In other implementations, wavelet transform can be used for joint time-frequency analysis, or other feature forms such as statistical features and morphological features can be introduced. Each window generates a feature vector, which is arranged in chronological order to form a feature data sequence.

[0078] In some embodiments, the feature data sequence is input into a machine learning model to obtain traffic state recognition results, including: The feature data sequence is input into a lightweight machine learning model deployed on the edge. The lightweight machine learning model outputs the transient traffic state probability distribution for each feature vector in the feature data sequence within the corresponding time window. The probability distribution represents the confidence level of each traffic state and constitutes the traffic state identification result; The traffic conditions include at least one or more of the following: stationary, smooth travel, bumpy travel, cycling, and walking.

[0079] In this embodiment of the disclosure, the machine learning model is a lightweight model deployed on an edge device. This model can run on an inference framework suitable for microcontrollers and undergoes operator pruning and integer quantization optimization to meet low power consumption and real-time requirements.

[0080] The model infers from each input feature vector and outputs a probability distribution of traffic states for the corresponding time window. This probability distribution characterizes the confidence level of each traffic state. In other implementations, classification labels can be directly output.

[0081] Traffic state categories include at least stationary, smooth travel, bumpy travel, cycling, and walking, enabling fine-grained identification of different commuting behaviors. In other embodiments, the model may employ one or more of the following: one-dimensional convolutional neural network, recurrent neural network, long short-term memory network, lightweight variant of Transformer, random forest, support vector machine, or gradient boosting tree.

[0082] In terms of model deployment, in addition to edge microcontrollers, it can also be deployed on mobile terminals.

[0083] In some embodiments, the traffic state recognition results are subjected to temporal constraint processing to generate commuter trajectory slices, including: The traffic state identification results are processed in a time sequence based on state transition constraints to filter out isolated state jumps caused by transient interference. Based on the smoothed sequence, continuous time periods with a single traffic semantic are aggregated to generate the commuting trajectory slice.

[0084] In this embodiment of the disclosure, the traffic state recognition results are subjected to time-series constraint processing to improve the stability of the recognition results.

[0085] Specifically, the probability distribution sequence arranged in chronological order is input into the time-series smoothing module and filtered based on preset state transition constraints to eliminate isolated state jumps caused by short-term noise.

[0086] The temporal smoothing can be implemented based on a hidden Markov model or a finite state machine, determining the optimal state sequence through global path optimization. Other implementations may employ a sliding window majority voting mechanism, a state filtering method based on a confidence threshold, or a sequence modeling method based on deep learning.

[0087] Based on the smoothed results, continuous time periods with consistent traffic semantics are aggregated to generate commuter trajectory slices.

[0088] In some embodiments, a walking equivalent is determined based on the commuting trajectory slice, and an intervention strategy is generated when the walking equivalent is lower than a preset threshold, including: A user commuting profile is constructed based on historical commuting trajectory slices or current and historical commuting trajectory slices. The user commuting profile includes commuting mode combination, commuting distance and walking equivalent. When the walking equivalent is lower than a preset threshold, an intervention strategy containing spatial anchor points is generated based on the user commuting profile.

[0089] In this embodiment of the disclosure, the user's walking equivalent is determined based on commuting trajectory slices, and an intervention strategy is generated accordingly.

[0090] Specifically, the system constructs a user commuting profile based on historical commuting trajectory slices or current and historical trajectory data. The commuting profile includes commuting mode combinations, commuting time distribution, and walking equivalent information.

[0091] When a user's walking equivalent is detected to be below a preset threshold, an intervention strategy including spatial anchor points is generated based on the commuting profile to guide the user to increase walking activity. In other embodiments, the intervention strategy can also be dynamically optimized based on a reinforcement learning model, recommended based on group data for similar users, or dynamically adjusted in conjunction with external data such as weather, air quality, and traffic conditions.

[0092] In some embodiments, generating an intervention strategy that includes spatial anchor points includes: If the commuting distance is less than or equal to a preset threshold, a full-distance walking intervention task is generated. If the commuting distance is greater than a preset threshold, then based on the time distribution of the commuting status in historical commuting trajectory slices or current and historical commuting trajectory slices, the recommended spatial anchor point for ending the commuting early is calculated in reverse, and a phased walking intervention task is generated to switch from commuting to walking.

[0093] In this embodiment of the disclosure, the intervention strategy includes a hierarchical generation mechanism.

[0094] When the commuting distance is less than or equal to a preset threshold, a full-walking intervention task is generated to encourage users to commute entirely on foot.

[0095] When the commuting distance exceeds a preset threshold, the system reverse-engineers the recommended spatial anchor points for ending the journey early based on the time distribution of the riding status in the historical commuting trajectory slices, and generates a phased walking intervention task accordingly, so that the user can switch from riding status to walking status during the commuting process.

[0096] In one specific embodiment of this disclosure, a core data processing and edge-side algorithm flow deployed on the edge of a wearable device is provided. This flow may include four consecutive steps, achieving accurate identification and trajectory segmentation of commuting status while ensuring ultra-low power consumption operation.

[0097] Step 1: Spatiotemporal Anchor Point Triggering and Intelligent Wake-up The system defaults to low-power mode, with the high-frequency inertial measurement unit (IMU) in deep sleep. The low-frequency scanning unit periodically performs passive scans at an extremely low duty cycle (e.g., waking up every few seconds for hundreds of milliseconds) to continuously detect Bluetooth or Wi-Fi beacon signals in the surrounding environment.

[0098] In this embodiment, the preset commuting time window is the school start time window (06:30-08:00) and the school end time window (16:00-18:00). When the system detects that the current time is within the school start time window, and multiple consecutive scans show that the Wi-Fi beacon signal strength corresponding to the home environment is lower than the preset threshold (i.e. the signal disappears), it determines that the wearer has left the home geofence and entered commuting mode.

[0099] At this point, the main control microcontroller (MCU) writes control commands to the IMU's registers via the I2C or SPI bus and sends a wake-up signal. The IMU is awakened from deep sleep and begins to continuously acquire raw inertial data at a sampling rate of 50Hz. For a 6-axis IMU, it outputs three-axis acceleration data (acc_x, acc_y, acc_z) and three-axis angular velocity data (gyro_x, gyro_y, gyro_z); for a 9-axis IMU, it further outputs three-axis magnetometer data (mag_x, mag_y, mag_z), forming a continuous time-series data stream.

[0100] Step 2: Multidimensional Inertial Signal Preprocessing and Feature Extraction After being awakened, the system preprocesses and constructs features from the acquired raw inertial data stream. In this embodiment, a fixed-length sliding window is used to segment the continuous time series, with the window length set to 2.56 seconds and an overlap rate of 50% between adjacent windows. This results in multiple time window data segments: window 1 covers 0-2.56 seconds, window 2 covers 1.28-3.84 seconds, window 3 covers 2.56-5.12 seconds, and so on.

[0101] In the feature extraction process, this embodiment explicitly adopts an absolute physical quantity preservation mechanism. The system does not perform global normalization on the original triaxial acceleration data because distinguishing between "the severe jolt of a bus going over a speed bump" and "the impact on the soles of the feet when walking to school" requires a rigorous assessment of the absolute stability and true amplitude of the physical forces. Blindly normalizing would erase this crucial physical quantity information, causing the classifier to fail in complex riding environments.

[0102] For each window, the system extracts time-domain and frequency-domain features in real time at the MCU. Time-domain features include statistics such as mean, variance, peak-to-average ratio, and zero-crossing rate; frequency-domain features are extracted using a lightweight fast Fourier transform (FFT) to extract the energy distribution of a specific frequency band, used to characterize the periodic patterns of motion. After feature extraction, each window generates a high-dimensional feature vector, denoted as F1, F2, F3, ..., which are arranged in chronological order to form a feature vector sequence [F1, F2, F3, ..., Fn].

[0103] Step 3: End-side model inference based on TFLite Micro After feature extraction, the feature vector sequence is fed into a lightweight machine learning model pre-deployed on the main microcontroller. In this embodiment, the microcontroller uses a Cortex-M4 architecture, and the model is deployed based on the TensorFlow Lite Micro (TFLM) framework. During training, the model undergoes rigorous operator pruning and INT8 full integer quantization to adapt to edge resource constraints. At runtime, the model performs tensor operations without dynamic memory allocation in the microcontroller's contiguous memory pool (Arena), ensuring execution efficiency and stability.

[0104] The model independently outputs the transient traffic state probability distribution for each feature vector within the corresponding time window. In this embodiment, the model employs a one-dimensional convolutional neural network (1D-CNN) structure, and the recognized categories include: stationary, smooth travel, bumpy travel (such as on a bus), cycling, and walking. For example, for feature vector F1, the model outputs the probability distribution P1 = [stationary: 0.1, smooth travel: 0.7, bumpy travel: 0.15, cycling: 0.03, walking: 0.02]; for feature vector F2, it outputs the probability distribution P2, and so on. This forms the probability distribution sequence [P1, P2, P3, …, Pn].

[0105] Step 4: Temporal smoothing and trajectory segmentation Since window-by-window independent inference is susceptible to sudden noise interference that can cause state jitter, this embodiment introduces a timing smoothing post-processing module. For example, when a bus stops for 30 seconds due to a red light, the transient model may misjudge this stationary signal as a "stationary" state, resulting in an abnormal jump in the output sequence of "riding → stationary → riding".

[0106] To address the aforementioned issues, this embodiment employs the Viterbi algorithm of an Hidden Markov Model (HMM) to apply temporal constraints and smoothing filters to the probability distribution sequence. The system pre-defines a state transition probability matrix, constraining the likelihood of transitions between different traffic states. For example, the probability of consecutive occurrences of the "riding" state is relatively high, while the probability of a direct transition from "riding" to "walking" and back to "riding" is set to an extremely low level. Based on this state transition matrix and the observed probabilities (i.e., the probability distribution output by the model), the Viterbi algorithm solves for the globally optimal path of the sequence, eliminating frame-level state jitter caused by sudden environmental changes.

[0107] After smoothing, the system aggregates continuous time periods with single traffic semantics based on the optimized state sequence to generate commuter trajectory slices. For example, it aggregates potentially incorrectly segmented sequences like "10 minutes by car → 30 seconds stationary → 5 minutes by car" into a continuous "15 minutes by car". The final output commuter trajectory slices have clear traffic semantics, providing a reliable basis for subsequent walking equivalent statistics and intervention strategy generation.

[0108] Figure 4 This is a schematic diagram of the workflow of the scenario-based health intervention and recommendation engine according to an embodiment of the present disclosure, showing the closed-loop mechanism of constructing a commuting profile based on historical commuting trajectory slices, generating dynamic intervention strategies, and performing edge-side triggering and feedback during actual commuting.

[0109] In this embodiment of the disclosure, based on the completion of commuting trajectory slice recognition, the system further constructs a scenario-based health intervention and recommendation engine, which is used to transform the objective behavioral data obtained from the edge recognition into executable health behavior intervention strategies, thereby forming a closed loop of data perception, strategy generation and behavior feedback.

[0110] 1. Commuter Profile Construction and Baseline Modeling In this embodiment of the disclosure, the construction of the commuting profile is preferably performed in the cloud. Because user profile construction and long-term statistical analysis involve high computational complexity, the limited computing power on the edge device makes it impossible to complete such modeling tasks independently.

[0111] Specifically, the system continuously receives users' commuting trajectory slice data during the initial observation period (e.g., 1 to 2 weeks) and performs statistical analysis on the data to construct a user's regular commuting profile. The commuting profile includes at least the following information: Commuting mode combination characteristics, such as "all by car" accounting for 90% and "car + 3-minute walk" accounting for 10%; The time percentage of each traffic condition; One-way commute time span; Distance estimation information corresponding to commuting routes; The effective walking equivalent for walking behavior includes average daily walking time, average daily steps, and indicators of moderate to vigorous physical activity.

[0112] The commuting distance can be estimated based on low-frequency electronic fence information (home and school locations) or the home-school distance input by the user terminal.

[0113] Based on the above information, the system further establishes a user's walking equivalent baseline to characterize the user's daily exercise level during regular commutes and to serve as a reference for generating subsequent intervention strategies.

[0114] 2. Dynamic intervention decision-making mechanism In this embodiment of the disclosure, the generation of the intervention strategy is also preferably performed in the cloud.

[0115] The system pre-sets a target walking equivalent or a threshold for moderate to high intensity physical activity. When it detects that a user's actual walking equivalent is lower than the threshold during the current assessment period (e.g., a day), the intervention strategy generation process is triggered.

[0116] To avoid giving vague and difficult-to-implement suggestions like "walk more every day," the system adopts a tiered recommendation mechanism, generating intervention strategies with clear spatial or temporal anchors based on the user's commuting profile, including but not limited to the following two categories: (1) Walking strategy throughout the entire journey When a user's commuting distance is determined to be less than a preset threshold based on their commuting profile (e.g., commuting time less than 15 minutes and the route does not contain highway or expressway features), a full-process walking intervention strategy is generated to guide the user to complete their commute on foot in subsequent commutes.

[0117] (2) Phased walking strategy When a user is determined to be in a medium-to-long-distance commuting scenario (e.g., consistently displaying long-term (over 20 minutes) continuous travel trajectory slices), the system reverse-engineers the recommended time or spatial anchor point for ending the journey early based on the time distribution of the travel phase in historical commuting trajectory slices. For example, the system generates an intervention suggestion: "It is recommended to get off the bus at the 15-minute mark of the regular journey tomorrow (or at the intersection approximately 800 meters from the school) and complete the last 10 minutes of walking."

[0118] Based on the aforementioned anchor points, a phased walking intervention strategy is generated to guide users to switch from a commuting state to a walking state during their commute, thereby increasing effective walking time. This strategy breaks down long-distance commutes into a combination of "commuting + walking," ensuring commuting efficiency while effectively increasing walking equivalent.

[0119] 3. Intervention task issuance and execution at the endpoint In this embodiment, the generated intervention strategy is distributed to the wearable device and user terminal via the user terminal device or cloud system at a preset time (e.g., the night before or before the start of commuting), so that the user is aware of it in advance and forms behavioral expectations. The intervention task is simultaneously synchronized to the wearable device as the execution target for the next day.

[0120] During subsequent commutes, the edge computing and main control modules of the wearable device restart the commuting recognition process, run the traffic status recognition algorithm in real time, and continuously monitor the current commuting trajectory slices.

[0121] When the current commuting status is detected to meet the triggering conditions corresponding to the intervention strategy (e.g., switching from a car state to a walking state, and the walking duration reaches the threshold set by the intervention task, such as walking for 10 minutes), the intervention goal is determined to be achieved, and the wearable device outputs a prompt message through the feedback module.

[0122] The prompting information is preferably low-interference tactile feedback, such as generating a vibration signal at a preset frequency (e.g., two consecutive short vibrations) through a micro-vibration motor, to prompt the user to complete the target behavior without affecting their normal activities. In other embodiments, the feedback information may also include one or more of the following: sound prompts, graphical interface prompts, flashing lights, or voice broadcasts.

[0123] At the same time, user terminal devices can provide positive incentive feedback, including but not limited to virtual badges, achievement prompts, points or progress prompts, thereby enhancing users' willingness to perform actions.

[0124] 4. Intervention cycle management and system state switching In this embodiment of the disclosure, when the system determines that the user has reached the preset walking equivalent or health goal within the current period, no new intervention strategy will be triggered.

[0125] At this point, the system ends the current intervention decision cycle and enters the default monitoring state, handing control back to the multimodal perception module. The system's main control program returns to the default standby state, waiting for the next commuting time window to arrive and for the commuting trigger conditions to be met again, thereby initiating a new round of data collection and intervention processes.

[0126] In this state, the device maintains low power consumption and only performs low duty cycle scanning and time window detection of environmental beacons until the next commuting time window is triggered, achieving the effect of "not triggering intervention branches, directly resetting and entering the daily perception loop of the next day".

[0127] Through the above mechanism, this embodiment constructs a complete intervention closed loop: edge recognition: generating commuting trajectory slices; cloud modeling: constructing user commuting profiles; strategy generation: outputting intervention strategies with spatial anchors; task distribution: synchronizing to wearable devices via user terminals; edge execution: real-time monitoring and determination of behavior achievement; instant feedback: combining tactile feedback with virtual incentives; data feedback: updating the commuting profile model.

[0128] This closed-loop mechanism realizes the transformation from "passive monitoring" to "active intervention". The intervention strategy has clear spatial and temporal anchors, which effectively improves feasibility and user participation.

[0129] Through the aforementioned scenario-based health intervention and recommendation engine, this embodiment achieves real-time identification and immediate feedback on the device side, while the cloud side is responsible for historical profiles and strategy decisions, balancing real-time performance and computational accuracy. Based on commuting distance, it generates either full-distance walking or phased walking tasks, avoiding a one-size-fits-all approach. Intervention strategies are bound to specific time points or geographical locations, significantly improving executability. Vibration feedback combined with virtual badges forms a low-interference, high-stickiness behavior guidance mechanism. Commuting profiles are continuously updated through data feedback, allowing intervention strategies to gradually adapt to user behavior patterns. After the intervention cycle ends, the system automatically returns to a low-power monitoring state, only waking up under triggered conditions.

[0130] In other embodiments of this disclosure, the above-described intervention mechanism can be modified as follows: The initial observation period can be adjusted to 1 week, 3 weeks or other durations based on the stability of user behavior, or it can be continuously updated using a sliding window method. In addition to rule-based decision trees, reinforcement learning models can also be used to dynamically optimize strategies based on historical performance, or to combine group data for similar user recommendations. Intervention strategies can be dynamically adjusted based on external data such as weather, air quality, and traffic conditions; for example, walking targets can be automatically reduced during severe weather. In addition to vibration feedback, it may also include one or more of the following: sound prompts, graphical interface prompts, flashing lights, or voice broadcasts; In addition to step count targets, calorie consumption, activity duration, or other health indicators can be set as intervention triggers. In addition to being deployed in the cloud, the intervention engine can also be deployed on user terminal devices (such as mobile phones) or edge gateway devices to adapt to different computing power and network conditions. All of the above alternative methods fall within the protection scope of this disclosure.

[0131] Figure 5 This is a block diagram of a commuting pattern recognition and intervention system according to an embodiment of the present disclosure.

[0132] Secondly, referring to Figure 5 This disclosure provides a commuting pattern recognition and intervention system, including: Wearable devices, including: The multimodal sensing module is used to perform passive scanning periodically with a preset duty cycle to detect environmental beacon signals. When a preset commuting trigger condition determined based on a preset time window and the change of the environmental beacon signals is met, the module is awakened from a dormant state to collect the wearer's multimodal sensing data. The multimodal sensing data includes at least inertial sensing data. The microcontroller, connected to the multimodal sensing module, is used for: The multimodal sensing data is segmented, and time-frequency domain features are extracted while retaining the amplitude information of the data to obtain a feature data sequence. The feature data sequence is input into a machine learning model to obtain traffic state recognition results; The traffic state recognition results are subjected to temporal constraint processing to generate commuter trajectory slices; The interactive feedback and communication module is used to upload the commuting trajectory slice and receive the intervention strategy. During the subsequent commuting process, when the microcontroller detects that the triggering condition corresponding to the intervention strategy is met based on the commuting trajectory slice, it outputs prompt feedback information. The server is used to acquire the commuting trajectory slices, determine the walking equivalent based on the commuting trajectory slices, generate an intervention strategy based on the walking equivalent, and send the intervention strategy to the wearable device.

[0133] In this embodiment of the disclosure, a commuting pattern recognition and intervention system based on multi-sensor fusion is provided, including a wearable device and a server, wherein: The wearable device includes: The multimodal sensing module is used to perform passive scanning periodically with a preset duty cycle to detect environmental beacon signals, and generate a wake-up trigger signal when a preset commuting trigger condition determined based on a preset time window and the change of the environmental beacon signals is met; after receiving a wake-up control command, it switches from a sleep state to a working state to collect the wearer's multimodal sensing data, which includes at least inertial sensing data; The multimodal sensing module includes a low-frequency scanning and auxiliary positioning unit (such as a Bluetooth / Wi-Fi probe) and an inertial measurement unit (6-axis or 9-axis IMU). A microcontroller, connected to the multimodal sensing module, is configured to: segment the multimodal sensing data, extract time-frequency domain features while retaining the data amplitude information, and obtain a feature data sequence; input the feature data sequence into a machine learning model to obtain traffic state recognition results; perform temporal constraint processing on the traffic state recognition results to generate commuting trajectory slices; and during subsequent commutes, based on the real-time generated commuting trajectory slices, detect whether the triggering conditions corresponding to the intervention strategy are met, wherein the triggering conditions include at least traffic state transition features and / or the duration of the target behavior; An interactive feedback and communication module, connected to the microcontroller, is used to upload the commuting trajectory slices and receive intervention strategies, and output prompt feedback information when the microcontroller detects that the triggering conditions corresponding to the intervention strategy are met; The interactive feedback and communication module includes a vibration feedback motor and a wireless communication unit. The server is communicatively connected to the wearable device to acquire the commuting trajectory slices, determine the walking equivalent based on the commuting trajectory slices, generate an intervention strategy when the walking equivalent is lower than a preset threshold, and send the intervention strategy to the wearable device.

[0134] Preferably, the feedback information is a tactile cue signal output by a vibration feedback motor.

[0135] In some embodiments, the system further includes: The user terminal device is communicatively connected to the wearable device and the server, respectively, and is used to receive the commuting trajectory slices and / or the intervention strategy, and to display commuting information and intervention strategy to the user.

[0136] In other embodiments of this disclosure, the system further includes a user terminal device, which is communicatively connected to the wearable device and the server, respectively, for receiving the commuting trajectory slices and / or the intervention strategy, and displaying the corresponding commuting information and intervention content to the user.

[0137] Specifically, the wearable device establishes a communication connection with the user terminal device via Bluetooth Low Energy, and the user terminal device communicates with the server via cellular network or Wi-Fi, thereby enabling the uploading of commuting trajectory slices and the distribution of intervention strategies. The user terminal device is also used to visualize the commuting trajectory slices and to provide prompts or interactive confirmations for the intervention strategies.

[0138] In this implementation, wearable devices, user terminal devices, and servers together constitute an edge-cloud collaborative architecture.

[0139] In some embodiments, the multimodal sensing module includes: The low-frequency scanning and auxiliary positioning unit is used to perform passive scanning periodically with a preset duty cycle to detect environmental beacon signals; The inertial measurement unit is initially in a dormant state; The microcontroller is also used for: When the strength of a beacon signal corresponding to the first environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within the first commuting time window, a wake-up command is generated and sent to the inertial measurement unit. When a beacon signal corresponding to the second environmental feature is detected to change from an undetectable state to a detectable state in multiple consecutive scans, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state for a preset duration, a sleep command is generated and sent to the inertial measurement unit. When the strength of the beacon signal corresponding to the second environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within the second commuting time window, a wake-up command is generated and sent to the inertial measurement unit. When a beacon signal corresponding to the first environmental feature is detected to change from an undetectable state to a detectable state in multiple consecutive scans, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state for a preset duration, a sleep command is generated and sent to the inertial measurement unit. The inertial measurement unit switches from a sleep state to a working state in response to the wake-up command to start the acquisition of multimodal sensing data, and stops the acquisition of data and returns to the sleep state in response to the sleep command.

[0140] In a further embodiment of this disclosure, the multimodal sensing module includes a low-frequency scanning and assisted positioning unit and an inertial measurement unit.

[0141] The low-frequency scanning and auxiliary positioning unit is used to perform passive scanning periodically with a preset duty cycle to detect environmental beacon signals and obtain corresponding signal strength information; the inertial measurement unit is used to collect the wearer's inertial sensing data. The inertial measurement unit is in a sleep state in the initial state of the system to reduce overall power consumption.

[0142] The microcontroller is connected to the low-frequency scanning and assisted positioning unit and the inertial measurement unit, and is used to control the operating state of the inertial measurement unit based on the environmental beacon signal and time information, specifically including: When the beacon signal strength corresponding to the first environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within the first commuting time window, it is determined that the wearer has left the first environmental area, a wake-up command is generated and sent to the inertial measurement unit, so that the inertial measurement unit switches from the dormant state to the working state and starts the acquisition of multimodal sensing data. When a beacon signal corresponding to the second environmental feature is detected to change from an undetectable state to a detectable state in multiple consecutive scans, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state within a preset time and continues for a preset duration, it is determined that the wearer has reached the second environmental area, a sleep command is generated and sent to the inertial measurement unit to stop the acquisition of the multimodal sensing data. When the beacon signal strength corresponding to the second environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within the second commuting time window, it is determined that the wearer has left the second environmental area, a wake-up command is generated and sent to the inertial measurement unit to restart the acquisition of multimodal sensing data; When a beacon signal corresponding to the first environmental feature is detected to change from an undetectable state to a detectable state during multiple consecutive scans, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state within a preset time and continues for a preset duration, it is determined that the wearer has returned to the first environmental area, a sleep command is generated and sent to the inertial measurement unit to stop the acquisition of the multimodal sensing data.

[0143] In the above process, the inertial measurement unit switches from sleep state to working state in response to the wake-up command, and collects inertial sensing data at a preset sampling frequency; in response to the sleep command, it stops data acquisition and returns to sleep state, thereby realizing a low-power dynamic acquisition control mechanism based on environmental perception and time constraints.

[0144] Thirdly, embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements any of the methods described above.

[0145] Fourthly, embodiments of this disclosure provide a computer program product, including a computer program or instructions, which, when executed by a processor, implement any of the methods described above.

[0146] In this embodiment of the disclosure, the commuting pattern recognition and intervention method and system described above have at least the following technical effects: 1) Achieve low-power operation and enhance privacy protection capabilities This disclosure combines passive scanning of environmental beacons with low duty cycle with high-frequency inertial sensing data acquisition, waking up the inertial measurement unit only when preset commuting trigger conditions are met, thereby avoiding the power consumption overhead caused by continuous high-frequency sampling. At the same time, the processing of traffic status recognition and trajectory slicing generation is carried out on the wearable device side, reducing the dependence on continuous positioning data in the cloud and reducing the frequency of uploading sensitive location information, thereby improving the system's privacy protection capabilities while ensuring the realization of functions.

[0147] 2) Improve recognition accuracy and stability in complex commuting scenarios This disclosure preserves the amplitude information of multimodal sensing data during feature extraction and combines time-domain and frequency-domain features to model traffic conditions, effectively distinguishing the differences between walking and vehicle bumpy conditions. Furthermore, by performing temporal constraint processing on the traffic condition recognition results, it suppresses state jumps caused by instantaneous noise and generates continuous and semantically consistent commuter trajectory slices, thereby improving the recognition accuracy and result stability in mixed traffic scenarios.

[0148] 3) Achieve executable intervention strategy generation and closed-loop feedback based on commuting scenarios. This disclosure determines the walking equivalent based on commuting trajectory slices and generates an intervention strategy when it falls below a preset threshold, providing data-driven intervention decisions. By introducing a strategy generation method using spatial anchors, intervention suggestions that match the actual commuting path can be output. Simultaneously, during subsequent commutes, the triggering conditions of the strategy are detected based on real-time trajectory slices, and feedback information is output through wearable devices, thus forming a closed loop of "identification—evaluation—decision—execution—feedback," improving the feasibility and timeliness of intervention measures.

[0149] 4) Supports system scalability for end-edge-cloud collaboration This disclosure achieves functional division between the edge and cloud by deploying data collection and basic identification on wearable devices and user profile building and strategy generation on the server side. This reduces the computing burden on the edge while improving the overall processing capacity of the system. Information can be displayed and interacted with through user terminal devices, thus providing good system scalability and application flexibility.

[0150] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0151] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in connection with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in connection with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of this disclosure as set forth by the appended claims.

Claims

1. A commuting pattern recognition and intervention method, comprising: When the preset commuting triggering conditions determined based on a preset time window and changes in environmental beacon signals obtained through passive scanning are met, multimodal sensing data of the wearer is collected, and the multimodal sensing data includes at least inertial sensing data. The multimodal sensing data is segmented, and time-frequency domain features are extracted while retaining the amplitude information of the data to obtain a feature data sequence. The feature data sequence is input into a machine learning model to obtain traffic state recognition results; The traffic state recognition results are subjected to temporal constraint processing to generate commuter trajectory slices; Based on the commuting trajectory slices, walking equivalents are determined, and an intervention strategy is generated when the walking equivalents are lower than a preset threshold. During the subsequent commute, the system detects whether the triggering conditions corresponding to the intervention strategy are met based on the commuting trajectory slices, and outputs prompt feedback information when the conditions are met.

2. The method according to claim 1, wherein, Meet the preset commuting trigger conditions, including at least one of the following situations: First scenario: When the strength of the beacon signal corresponding to the first environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within the first commuting time window, it is determined that the first commuting start condition is met, and the collection of the multimodal sensing data begins. When a beacon signal corresponding to the second environmental feature is detected to change from an undetectable state to a detectable state in multiple consecutive scans, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state for a preset duration, it is determined that the first commuting end condition is met, and the collection of the multimodal sensing data is stopped. Second scenario: When the strength of the beacon signal corresponding to the second environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within the second commuting time window, it is determined that the second commuting start condition is met, and the collection of the multimodal sensing data begins. When a beacon signal corresponding to the first environmental feature is detected to change from an undetectable state to a detectable state in multiple consecutive scans, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state for a preset duration, it is determined that the second commuting end condition is met, and the collection of the multimodal sensing data is stopped.

3. The method according to claim 1, wherein, The feature data sequence is input into a machine learning model to obtain traffic state recognition results, including: The feature data sequence is input into a lightweight machine learning model deployed on the edge. The lightweight machine learning model outputs the transient traffic state probability distribution for each feature vector in the feature data sequence within the corresponding time window. The probability distribution represents the confidence level of each traffic state and constitutes the traffic state identification result; The traffic conditions include at least one or more of the following: stationary, smooth travel, bumpy travel, cycling, and walking.

4. The method according to claim 1, wherein, The traffic state recognition results are subjected to temporal constraint processing to generate commuter trajectory slices, including: The traffic state identification results are processed in a time sequence based on state transition constraints to filter out isolated state jumps caused by transient interference. Based on the smoothed sequence, continuous time periods with a single traffic semantic are aggregated to generate the commuting trajectory slice.

5. The method according to claim 1, wherein, Based on the commuting trajectory slices, walking equivalents are determined, and when the walking equivalents are below a preset threshold, an intervention strategy is generated, including: A user commuting profile is constructed based on historical commuting trajectory slices or current and historical commuting trajectory slices. The user commuting profile includes commuting mode combination, commuting distance and walking equivalent. When the walking equivalent is lower than a preset threshold, an intervention strategy containing spatial anchor points is generated based on the user commuting profile.

6. The method according to claim 5, wherein, Generate intervention strategies that include spatial anchor points, including: If the commuting distance is less than or equal to a preset threshold, a full-distance walking intervention task is generated. If the commuting distance is greater than a preset threshold, then based on the time distribution of the commuting status in historical commuting trajectory slices or current and historical commuting trajectory slices, the recommended spatial anchor point for ending the commuting early is calculated in reverse, and a phased walking intervention task is generated to switch from commuting to walking.

7. A commuting pattern recognition and intervention system, comprising: Wearable devices, including: The multimodal sensing module is used to perform passive scanning periodically with a preset duty cycle to detect environmental beacon signals. When a preset commuting trigger condition determined based on a preset time window and the change of the environmental beacon signals is met, the module is awakened from a dormant state to collect the wearer's multimodal sensing data. The multimodal sensing data includes at least inertial sensing data. The microcontroller, connected to the multimodal sensing module, is used for: The multimodal sensing data is segmented, and time-frequency domain features are extracted while retaining the amplitude information of the data to obtain a feature data sequence. The feature data sequence is input into a machine learning model to obtain traffic state recognition results; The traffic state recognition results are subjected to temporal constraint processing to generate commuter trajectory slices; The interactive feedback and communication module is used to upload the commuting trajectory slice and receive the intervention strategy. During the subsequent commuting process, when the microcontroller detects that the triggering condition corresponding to the intervention strategy is met based on the commuting trajectory slice, it outputs prompt feedback information. The server is used to acquire the commuting trajectory slices, determine the walking equivalent based on the commuting trajectory slices, generate an intervention strategy based on the walking equivalent, and send the intervention strategy to the wearable device.

8. The system according to claim 7, wherein, Also includes: The user terminal device is communicatively connected to the wearable device and the server, respectively, and is used to receive the commuting trajectory slices and / or the intervention strategy, and to display commuting information and intervention strategy to the user.

9. The system according to claim 7, wherein, The multimodal sensing module includes: The low-frequency scanning and auxiliary positioning unit is used to perform passive scanning periodically with a preset duty cycle to detect environmental beacon signals; The inertial measurement unit is initially in a dormant state; The microcontroller is also used for: When the strength of a beacon signal corresponding to the first environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within the first commuting time window, a wake-up command is generated and sent to the inertial measurement unit. When a beacon signal corresponding to the second environmental feature is detected to change from an undetectable state to a detectable state in multiple consecutive scans, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state for a preset duration, a sleep command is generated and sent to the inertial measurement unit. When the strength of the beacon signal corresponding to the second environmental feature is detected to be lower than a preset threshold in multiple consecutive scans, and the current time is within the second commuting time window, a wake-up command is generated and sent to the inertial measurement unit. When a beacon signal corresponding to the first environmental feature is detected to change from an undetectable state to a detectable state in multiple consecutive scans, and the traffic state identified based on the multimodal sensing data continuously and stably changes from a walking state to a stationary state for a preset duration, a sleep command is generated and sent to the inertial measurement unit. The inertial measurement unit switches from a sleep state to a working state in response to the wake-up command to start the acquisition of multimodal sensing data, and stops the acquisition of data and returns to the sleep state in response to the sleep command.

10. A computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images