A passive commuting weather warning pendant and method based on bag action triggering

Abstract

The application discloses a kind of passive commuting weather warning pendants and methods based on bag action trigger, belong to intelligent wearable device technical field, solve the technical problem that current weather warning scheme operation friction is big, rely on mobile phone, privacy protection is insufficient.The application determines to pick up backpack action and wakes up master control chip by the combination output of accelerometer and gyroscope;Comprehensive stay length, current time and the history of travel law that has been learned, travel event is identified from multiple bag actions;Only for travel event, infer matching commuting mode, query weather along the way node with time in anonymized rough coordinate, and determine whether to warn, warning is output by output unit.Output unit is output.The personalized commuting weather warning of user less operation, device independent operation, privacy localization protection is realized.

Description

Technical Field

[0001] This invention relates to the field of smart wearable device technology, and in particular to a passive commuter weather warning pendant and method triggered by carrying a bag. Background Technology

[0002] Weather alert solutions for daily commuting scenarios mainly fall into two technical categories. The first category consists of general weather applications based on administrative regions or single coordinates, which cannot accurately match the dynamic changes in time and space along long-distance, cross-regional commuting routes. The second category comprises route-based weather applications, which, while solving the spatial distribution of weather along the route at the algorithm level, require users to actively perform a sequence of operations—"unlock phone - open application - enter destination"—before each trip, resulting in high operational friction and making it difficult to establish stable daily usage habits.

[0003] Existing patented technologies also have significant technical flaws. For example, US9851471B2 discloses a weather forecasting scheme based on expected location, which automatically predicts future locations and pushes corresponding weather based on the user's location history in the cloud. However, this scheme relies on smartphone platforms and centralized collection and processing of user location data in the cloud, posing risks of user trajectory uploading and privacy leaks. Furthermore, it does not involve independent hardware at the widget level and lacks a seamless early warning mechanism triggered by specific actions. US20160116640A1 aims to improve the accuracy of micro-regional forecasts rather than achieving seamless early warning triggering. The routes disclosed in US7486201B2 need to be predefined by the user and stored in the user's profile, still relying on the user to actively provide route information, and cannot automatically learn and update according to changes in the user's travel patterns. While Apple's motion gesture wake-up patents (such as US10101793B2 and US9933833B2) disclose general methods for triggering device wake-up by detecting gestures such as wrist raises using accelerometers and gyroscopes, they lack triple constraints for the specific complex motion pattern of "lifting a backpack," including vertical acceleration pulses, a posture stabilization window, and horizontal reverse pulse exclusion. Furthermore, they lack the logic to accurately identify "travel events" from repeated everyday bag-lifting actions. In addition, existing pendant-level products, such as the wearable motion-sensing pendant disclosed in US20100245078A1, are designed for motion data collection rather than predictive inference triggering; the anti-loss alarm pendant disclosed in US9165447B2 requires radio frequency pairing with a monitoring device placed inside the monitored item and lacks the complete technical capability to independently perform path prediction and weather inference.

[0004] In summary, existing technologies suffer from five major shortcomings: requiring active user operation, reliance on the mobile phone platform and its battery level, dependence on users actively providing route information, uploading user trajectory data to the cloud, and the lack of independent predictive capabilities in the accessory-level hardware. The core technical problem this invention aims to solve is how to achieve a complete technical loop—one that operates independently, consumes little power, and has significantly extended battery life—that requires no active user operation, does not rely on a mobile phone, does not upload user trajectory data, and can accurately predict and issue warnings about severe weather along a user's personal commuting route in the future. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a passive commuting weather warning pendant and method based on bag-carrying actions, aiming to solve the technical problems of high user operation friction, reliance on external devices, insufficient privacy protection, and lack of independent prediction capabilities of pendant-level hardware.

[0006] Technical solution: A passive commuter weather warning method based on bag-carrying actions, the method comprising the following steps: Step S1: Determine a backpack lifting action (hereinafter referred to as backpack lifting action) by combining the output of the accelerometer and gyroscope. This determination includes the amplitude and duration constraints of the vertical acceleration pulse, the subsequent attitude stability constraints, and the exclusion constraints of the horizontal reverse acceleration pulse.

[0007] Step S2: When the backpack lifting action is determined to be successful, the main control chip is woken up from deep sleep.

[0008] Step S3: Based on the timing of this bag-lifting action, the duration of the device's stay at the current location, the time interval between this bag-lifting action and the last determined bag-lifting action, and the learned historical travel patterns, determine whether this bag-lifting action constitutes a travel event, and execute subsequent steps only if it is determined to be a travel event.

[0009] Step S4: Based on the passive observation sequence of bag-picking events over the previous few days, infer the user's commuting pattern that matches the current time and current approximate location. The commuting pattern is represented by a time-location sequence, including the sequence of nodes along the route and the typical arrival time of each node. This inference can be performed by the device's local neural network processing unit or by other computing carriers with corresponding computing power.

[0010] Step S5: For the path node sequence in the commuting mode, obtain the weather data of each path node at the corresponding time from the weather data interface via wireless communication with a minimum query parameter containing only coarse coordinates and time; the acquisition can be requested one by one for each path node or in batches at once.

[0011] Step S6: Summarize the weather data of the route nodes to determine whether the current commute will encounter severe weather.

[0012] Step S7: When it is determined that a warning is needed, a warning is output through an output unit, wherein the output unit includes at least one of a display screen and a speaker.

[0013] Step S8: From the time the warning decision is made for a travel event until the device enters a dwell state that lasts for more than the preset time again, the warning will not be repeated for subsequent bag-collecting actions.

[0014] It should be noted that the inference of the commuting mode and the acquisition of weather data can also be performed in advance and the results can be cached before the travel event is determined. When the travel event is determined, the early warning decision and output can be completed directly based on the cached results, so as to shorten the response delay from the self-collection action to the early warning output.

[0015] In a preferred embodiment, the passive observation sequence, user commuting mode, and motion state recognition results are all stored and processed locally on the device and are not uploaded to a server outside the weather data interface.

[0016] A backpack-mountable smart accessory device is characterized by comprising an accelerometer, a gyroscope, a neural network processing unit, a fuzzy positioning module, a wireless communication module, an output unit, and a power management module configured as described above. The output unit includes at least one of a display screen and a speaker. The weather warning function of the device can be completed without the user operating any buttons, and the device does not rely on any external handheld device to complete the above method steps.

[0017] Beneficial effects: The benefits of technology that requires less user interaction: Users only need to pick up their backpacks as usual, without any additional human-computer interaction actions (such as unlocking their phones, opening apps, or pressing buttons), to receive accurate weather warnings along their commuting route, eliminating the operational friction of existing technologies.

[0018] Personalized and accurate forecasts: The device learns users’ unique commuting routes and time patterns through passive learning, so that the forecasts are accurately matched with the user’s actual journey, avoiding interference from invalid information.

[0019] Independent operation of the device: As an independent hardware device, this invention does not rely on a mobile phone, does not consume mobile phone power, and does not require Bluetooth pairing, completing the entire technical process from perception and decision-making to output.

[0020] Privacy protection technology implementation: All raw data such as the user's commuting route, movement status, and location sequence are processed and stored locally on the device. Only anonymized coordinates and time query parameters that cannot be associated with the user are sent to the weather service interface, which significantly reduces the risk of cloud leakage from the cyber-physical level.

[0021] Significantly extended battery life: Through the hardware partitioning of the three power domains of "always-awake domain - wake-up domain - communication and display domain" and the wake-up mechanism driven by hardware interrupts, combined with the zero-power static display characteristics of the display screen, ultra-long standby and usage time are achieved.

[0022] Intelligent and non-intrusive user experience: The device displays passive decorative patterns by default, and only actively issues warnings when the device predicts a risk of severe weather on the commuting route through model inference, thus avoiding frequent push notifications that may disturb the user. Attached Figure Description

[0023] Figure 1 This is a system overall structure block diagram of the device of the present invention; Figure 2 This is a flowchart illustrating the overall workflow of the method of the present invention; Figure 3 This is a timing diagram for determining the bag-lifting action in this invention; Figure 4 This is a flowchart of the commuting mode learning process in this invention; Figure 5 This is a flowchart of the path weather forecasting and early warning determination process in this invention; Figure 6 This is a diagram showing the division of the three power domains in this invention; Figure 7 This is a schematic diagram of the overall appearance of the device in this invention; Figure 8 This is a flowchart of the travel event determination process in this invention. Detailed Implementation

[0024] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] Example 1

[0026] This embodiment provides a passive commuter weather warning pendant and method triggered by carrying a bag. The device is shaped like a pendant that can be attached to a backpack, and integrates a main control chip (containing a neural network processing unit, NPU), an accelerometer, a gyroscope, a fuzzy positioning module based on cellular base stations and Wi-Fi fingerprint, an eSIM cellular communication module, a display screen (memory LCD screen, e-ink screen, or LCD screen), a speaker, and a graded power management module.

[0027] I. Hardware Architecture and Power Management (with Figure 1 , Figure 6 (For example) The device's components are divided into three power domains to achieve ultra-low power consumption. The always-on domain contains the accelerometer and power management IC, maintaining continuous operation. The wake-up domain contains the main control chip, NPU, and gyroscope. The communication and display domain contains the cellular communication module, fuzzy positioning module, screen driver, and speaker. In deep sleep mode, both the wake-up domain and the communication and display domain are completely powered off; only the accelerometer in the always-on domain operates with a microamplitude current, awaiting the trigger action of lifting the bag. The display screen does not consume sustaining current when displaying statically.

[0028] II. Perception and Triggering of Specific Bag-Carrying Actions (with Figure 3 (For example) The accelerometer integrates a finite state machine to continuously monitor triaxial acceleration data. Considering the arbitrary suspension posture of the pendant and the non-coincidence of the device's coordinate system and the geographic coordinate system, the device estimates the direction of gravity using a low-pass filtered acceleration vector and projects the triaxial acceleration in real time onto this direction of gravity and a horizontal plane orthogonal to it. The terms "vertical direction" and "horizontal direction" below refer to the projected directions. A valid "backpack lifting action" requires the simultaneous fulfillment of the following three technical conditions: Condition A (Vertical Acceleration Pulse): An acceleration pulse is detected in the vertical direction (opposite to the direction of gravity). In this embodiment, the peak amplitude of the pulse is set to 1.5g to 2.5g, and the pulse duration window is set to 200 to 500 milliseconds.

[0029] Condition B (Attitude Stability): During a 1-second attitude stabilization window following the end of the acceleration pulse, the angular velocities output by the three-axis gyroscopes are all below 30° / s.

[0030] Condition C (False Trigger Exclusion): No horizontal reverse acceleration pulse with an amplitude exceeding 0.5g was detected between the acceleration pulse and the attitude stabilization window, in order to exclude non-commuting scenarios such as taking a backpack out of the vehicle trunk.

[0031] When all three conditions are met, the accelerometer sends a wake-up signal to the main control chip via the hardware interrupt pin.

[0032] III. Localized Passive Learning in Commuting Modes (with) Figure 4 (For example) After the device is powered on for the first time, it enters a 14-day "learning period". Each time step S1 is triggered, the system wakes up and collects data: records the trigger timestamp (e.g., 2025-05-15 08:02:34) and the current approximate location (coordinates obtained through cellular base station positioning and a 1km error radius), and continuously records the motion state until it comes to rest (classified by the motion state recognition module as walking, cycling, riding in a vehicle, etc.).

[0033] After several working days of data collection, the device's local NPU runs a lightweight sequence model to learn the commuting pattern. In this embodiment, the sequence model is a pre-trained Long Short-Term Memory (LSTM) autoencoder, which takes a sequence of vectors [time, coarse location, motion state] as input and outputs a fixed-dimensional trip embedding vector. The device does not train the network itself, but only uses its reasoning ability to extract trip features. Then, a density-based clustering algorithm (such as DBSCAN) clusters the embedding vectors of each trip. Each cluster corresponds to a candidate commuting pattern, and its path node sequence and typical arrival time of each node are determined by the statistics of the samples within the cluster (such as the median). The confidence of the candidate commuting pattern is defined as the proportion of the number of samples in the cluster to the total number of trip samples at the same departure and stay point and at the same departure time during the observation period. The algorithm found that between 07:30 and 08:30 on weekdays, the location sequence after a bag-picking event always starts from coordinate A (clustered as "home"), follows specific path nodes (coordinates B1, B2, B3...), and arrives at coordinate C (clustered as "company") after approximately 45 minutes. When the confidence level of this pattern exceeds the 80% threshold, the device solidifies it as a valid commuting pattern, represented as: {ID: W01; Departure and rest point: Home; Departure time: 07:30-08:30; Path node sequence: [{coordinate A, T0}, {coordinate B1, T0+12min}, {coordinate B2, T0+25min}, {coordinate C, T0+45min}]; Time taken: 45min}. The device then switches to normal operating mode; thereafter, for each completed trip, the device incorporates the embedding vector of that trip into the corresponding cluster and updates the path node sequence and typical arrival times of each node using a sliding window statistical method, achieving incremental updates to the commuting pattern library. All calculations are performed locally on the NPU, with no data uploaded.

[0034] IV. Determination of Travel Events and Constraints on Single Trips (based on...) Figure 8 (For example) One morning during normal working hours (07:45), the device was awakened again by a bag-carrying action. The travel event determination module makes a judgment based on the following factors: After arriving at its "home" coordinates at 21:00 yesterday, the device has remained stationary for more than 2 hours (the dwell time threshold). This dwell time is accumulated by the continuous stationary detection of the accelerometer in the always-awake domain. During deep sleep, there is no need to periodically start the positioning module, which is compatible with the low-power design of the three power domains.

[0035] The current time 07:45 falls within the typical departure time of 07:30-08:30 for the learned commuting mode W01.

[0036] The time interval between this bag pickup and the previous bag pickup event (the morning of the previous day) exceeded 4 hours (the bag pickup interval threshold).

[0037] The commuting pattern library contains W01, whose departure point (home) matches the current approximate location, and whose departure time matches the current time.

[0038] Based on the above factors, the module determines that this bag-carrying action constitutes a "travel event," triggering the subsequent weather warning process. Simultaneously, the state machine switches to "traveling" and activates a single-travel constraint: until the device re-enters a state lasting more than 2 hours, no further warnings will be issued for any subsequent bag-carrying actions, thus preventing repeated interruptions to the user after they leave home (e.g., while moving their backpack at the office).

[0039] V. Weather Forecasting and Anonymized Early Warning Based on Path (using Figure 5 (For example) After successfully matching commuting mode W01, the system extracts three key nodes from the path node sequence in sequence: {Node B1, 08:00}, {Node B2, 08:13}, and {Node C, 08:33}.

[0040] The cellular communication module is activated, and the eSIM connects to the mobile network. The system constructs a minimal API request for each node. This request does not contain any device ID, user ID, or historical trajectory information; it only contains anonymized coordinates and time.

[0041] The system received data from three nodes: nodes B1 and B2 both had a 20% probability of precipitation, but node C had a 70% probability of precipitation at 08:33. Since the warning threshold was set at 60%, node C's data exceeded the threshold, and the system determined that a warning was needed. The determination immediately caused the display to switch from the default pixelated pattern to a prominent warning icon (such as an umbrella), while the speaker emitted a short alert. The user received a valid warning about rain along their commute and the need to carry rain gear without taking any action.

[0042] VI. Completeness of the Technical Solution Through the aforementioned hardware architecture, hierarchical power management, multi-constraint bag-carrying action judgment, localized commuting mode learning, accurate identification of travel events, anonymized weather query, and collaborative work of single-trip constraints, this embodiment fully realizes the technical effects claimed by the present invention: "fewer user operations, independent device operation, localized privacy protection, and ultra-long battery life," constituting a complete, natural, and repeatable technical solution.

[0043] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A passive commuter weather warning method based on bag-carrying actions, characterized in that, include: The backpack lifting action (hereinafter referred to as the bag lifting action) is determined by the combined output of the accelerometer and gyroscope. When the determination is successful, the main control chip is woken up. Based on the time of this bag-carrying action, the duration of stay at the current location, the time interval since the last bag-carrying action, and the historical travel patterns in the learned commuting pattern database, it is determined whether a travel event has been constituted. Based on previous passive observation sequences, a commuting pattern is inferred that matches the current time and location. The commuting pattern is represented by a time-location sequence, including a sequence of nodes along the route and typical arrival times of each node. For the path node sequence, the weather data of each path node at the corresponding time is obtained from the weather data interface via wireless communication with a minimum query parameter containing only coarse coordinates and time. The system aggregates weather data from the route nodes to determine whether the current commute will encounter severe weather. If so, it outputs a warning through the output unit.

2. The method according to claim 1, characterized in that, The determination of the bag-lifting action specifically includes: An acceleration pulse opposite to the direction of gravity is detected in the vertical direction, and the acceleration pulse has a preset amplitude range and duration range. Within a preset attitude stabilization window after the acceleration pulse ends, the angular velocities output by the three-axis gyroscope are all below a preset threshold. No horizontal reverse acceleration pulse with an amplitude exceeding a preset threshold was detected between the acceleration pulse and the attitude stabilization window.

3. The method according to claim 1, characterized in that, Determining whether this act of carrying a bag constitutes a travel incident involves considering the following factors: Prior to this bag-lifting action, the device remained at its current approximate location for a duration exceeding a certain threshold. The current moment of this bag-collecting action falls within the typical departure time period that has been learned for this location; The time interval between this bag-lifting action and the last successfully judged bag-lifting action exceeds a preset threshold; There is a commuting pattern in the learned commuting pattern library whose typical departure and dwell point is consistent with the current approximate location of the equipment, and the current time falls within its typical departure time period.

4. The method according to claim 1, characterized in that, The process of inferring the commuting pattern includes: During the observation period, for each bag-carrying action that is determined to be valid, the timestamp of the bag-carrying action and the current approximate location are recorded, and the location sequence and motion state sequence immediately following the bag-carrying action until the dwell state that lasts for more than the preset time is detected again (i.e. the commuting is completed) are recorded. The device's local neural network processing unit runs a sequence model to model and cluster the observed samples, and outputs a set of user commuting patterns. Each commuting pattern is represented as a time-location sequence, along with the typical departure and stay points, typical departure time periods, and typical travel time for that commuting pattern. When the confidence level of the sequence model for a certain commuting pattern reaches a preset threshold, the commuting pattern is included in the learned commuting pattern library. Once the device enters its normal operating period, the commuting pattern library is updated incrementally.

5. The method according to claim 1, characterized in that, The method further includes a motion state recognition step: during the movement observation period after each bag-carrying action is determined, features are continuously extracted from the outputs of the accelerometer and gyroscope, and a lightweight classifier running on the device's local neural network processing unit is used to classify the user's current motion state as walking, cycling, riding in a vehicle, or stationary; the motion state recognition result is used to reverse the determination of the path node sequence in the learning and inference process of the commuting mode, and to adjust the speed model used in the time-space mapping when predicting path weather.

6. The method according to claim 1, characterized in that, The determination of whether the current commute will encounter severe weather is as follows: if the probability of precipitation at any node along the route exceeds a preset threshold, it is determined that the current commute requires an early warning; or if any weather indicator at any node along the route exceeds the corresponding preset warning threshold, it is determined that the current commute requires an early warning.

7. The method according to claim 1, characterized in that, The device has two states: a learning period and a normal working period. After the device is powered on for the first time, it enters the learning period. During this period, each time the bag is picked up, only data collection and incremental updates of the commuting mode are completed, and weather warnings are not executed. When the confidence level of at least one commuting mode reaches a preset threshold for the first time, the commuting mode is included in the available mode library, the device status is switched to normal working period, and the travel event determination and weather warning steps are executed during the normal working period.

8. The method according to claim 1, characterized in that, The method further includes a single trip constraint step: from the time an early warning decision is made for a trip event until a dwelling state that lasts for more than a preset time is detected again, the early warning will not be repeated for subsequent bag-collecting actions.

9. A passive commuter weather warning pendant triggered by the action of carrying a bag, characterized in that, include: Accelerometers and gyroscopes are used to collect motion data to determine the action of lifting the backpack; A neural network processing unit is used to perform commuting pattern inference and path weather prediction locally on the device. A fuzzy positioning module is used to obtain the approximate location of the device. The fuzzy positioning module obtains location data through cellular base station positioning or wireless local area network fingerprint positioning. The wireless communication module is used to request weather data from the weather data interface with minimal query parameters containing only coarse coordinates and time. The wireless communication module is a cellular communication module that uses an embedded user identity module, namely eSIM. An output unit, including at least one of a display screen and a speaker, is used to output warning information; The power management module is used to divide the device into multiple power domains and perform hierarchical wake-up and sleep modes; The main control chip is connected to each of the above modules and is configured to perform the steps of the method described in any one of claims 1 to 8; The weather warning function of the device can be completed without the user pressing any buttons, and the device does not rely on any external handheld device to complete the above method steps.

10. The device according to claim 9, characterized in that, The power management module divides the device components into the following three power domains: The always-awake domain, which includes an accelerometer and a power management integrated circuit, remains operational indefinitely, waiting for the bag-lifting action to trigger it. The wake-up domain, which includes the main control chip, neural network processing unit and gyroscope, is lit up after the always-awakened domain detects the bag-lifting action and is used to perform model inference. The communication display area, which includes a wireless communication module, a fuzzy positioning module, a screen driver, and a speaker, is only lit up when positioning, weather inquiries, or information output is required. In deep sleep mode, both the wake-up domain and the communication display domain are turned off, and only the always-awake domain is active.

Images