Remote data acquisition and analysis method and system for stroller

Abstract

The invention provides a remote data acquisition and analysis method and system for a baby stroller, and the method comprises the steps: synchronously acquiring environmental stimulation and passenger physiological reaction data through multiple modules on the baby stroller, locally processing the environmental data into cumulative stress indexes, and processing the physiological data such as sound production into state classification labels. And when the passenger has negative stress states such as crying, screaming and the like, the system automatically packs the accumulated stress index sequence in the previous period of time to form a stress event data packet. And the data packet and baseline data in a normal state are transmitted to a remote server together, and a specific environment mode causing the negative stress of the passenger is identified through comparative analysis. According to the method, through event-triggered data acquisition and comparative analysis, specific external factors causing discomfort of infants can be accurately identified and understood.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to a remote data acquisition and analysis method and system for children's strollers. Background Technology

[0002] With advancements in technology and evolving parenting philosophies, strollers are no longer simply transportation tools, but comprehensive childcare products integrating safety, comfort, and multifunctionality. In recent years, smart strollers integrating audio-visual monitoring, intelligent rocking, and even environmental monitoring have emerged on the market. These innovations aim to provide caregivers with greater convenience, allowing them to understand their child's condition anytime, anywhere, and thus provide more timely care.

[0003] However, most existing intelligent childcare methods remain at the level of one-way, descriptive status monitoring. For example, caregivers can use a mobile app to see if a child is crying or hear sounds around the stroller. Infants, especially babies who cannot yet express themselves clearly, may cry due to internal physiological needs such as hunger or drowsiness, or it may be caused by external environmental factors. When a child cries in a stroller, caregivers can only guess the reason based on experience; this judgment relies entirely on subjective feelings and lacks objective data support. Therefore, how to accurately identify and understand the specific external factors causing discomfort in infants is a pressing technical problem that needs to be solved in the field of intelligent childcare devices. Summary of the Invention

[0004] This invention provides a remote data acquisition and analysis method and system for children's strollers to solve the above-mentioned technical problems.

[0005] In a first aspect, the present invention provides a remote data acquisition and analysis method for a child stroller, the method comprising the following steps: By deploying an environmental acquisition module and a directional microphone on the stroller, environmental stimulus data including the stroller's movement posture and physiological response data including the rider's vocalizations are collected and time-stamped. In the local processing unit of the stroller, environmental stimulus data is processed in real time into standardized cumulative stress indicators, and physiological response data is processed into state classification labels representing the rider's state through sound source separation and pre-deployed audio classification models. When the state classification label meets the preset negative stress conditions, the cumulative stress index sequence within the preset time window before the occurrence of the negative stress conditions is extracted and associated to generate a stress event data packet. The caregiver transmits stress event data packets and baseline data packets representing the passenger's normal riding status under non-negative stress conditions to a remote server via a mobile terminal. By aggregating stress event data packets and baseline data packets on a remote server, and analyzing the statistical differences in cumulative stress indicators between the stress event data packets and baseline data packets, the target stress patterns that cause negative stress in passengers are identified.

[0006] Optionally, the step of synchronously collecting and timestamping environmental stimulus data including the stroller's movement posture and physiological response data including the occupant's vocalizations using an environmental acquisition module and a directional microphone deployed on the stroller includes the following steps: The inertial measurement unit in the environmental acquisition module collects the three-axis acceleration data of the stroller at a preset frequency; The ambient background audio and the passenger's vocal audio are collected using the omnidirectional and directional microphones in the environmental acquisition module at preset sampling rates. The stroller's local processing unit assigns synchronized timestamps from a unified clock to the collected triaxial acceleration data, ambient background audio, and sound audio. Integrate time-stamped triaxial acceleration data into an environmental stimulus data stream; Integrate time-stamped ambient background audio and vocal audio into a physiological response data stream.

[0007] Optionally, in the local processing unit of the stroller, environmental stimulus data is processed in real time into standardized cumulative stress indicators, and physiological response data is processed into state classification labels representing the rider's state through sound source separation and a pre-deployed audio classification model, including the following steps: Triaxial acceleration data is extracted in real time from the environmental stimulus data stream and combined with the environmental background audio in the physiological response data stream to calculate and generate cumulative stress indicators. A sound source separation algorithm is used to process the environmental background audio and vocal audio in the physiological response data stream to obtain the denoised vocal audio signal of the rider. The audio signal is input into a pre-deployed audio classification model, which outputs a state classification label representing the rider's state. The generated cumulative stress index and state classification label are stored in a circular buffer with a preset time length.

[0008] Optionally, the calculation of the cumulative stress index includes the following steps: A preset frequency weighting algorithm is applied to the triaxial acceleration data to highlight the vibration frequency components in the triaxial acceleration data; The weighted vibration data is integrated over time to calculate the vibration dose value that quantifies the cumulative vibration effect. Real-time analysis of ambient background audio is performed to calculate the A-weighted equivalent sound level value that conforms to human auditory perception. The vibration dose value and the A-weighted equivalent sound level value are normalized, and then the normalized vibration dose value and the A-weighted equivalent sound level value are combined through a weighted fusion function to generate a single-dimensional cumulative stress index.

[0009] Optionally, the step of using a sound source separation algorithm to process the environmental background audio and vocal audio in the physiological response data stream to obtain the denoised vocal audio signal of the passenger includes the following steps: The ambient background audio is used as a reference noise signal, and the emitted audio is used as a mixed signal containing the target signal and noise, which is then input to a preset audio filter. The filtering parameters inside the audio filter are dynamically adjusted based on the correlation between the reference noise signal and the mixed signal. The estimated ambient noise component is subtracted from the mixed signal using the adjusted filtering parameters, and the denoised audio signal of the passenger is output.

[0010] Optionally, when the state classification label meets the preset negative stress condition, extracting and associating the cumulative stress index sequence within the preset time window before the occurrence of the negative stress condition to generate a stress event data packet includes the following steps: The real-time output values ​​of the status classification labels are continuously monitored in the local processing unit; When the output value of the state classification label continuously matches a negative emotional state within a preset duration threshold, the state classification label is determined to meet the preset negative stress condition. Lock the trigger time and extract all target cumulative stress indicators within a preset time window prior to the trigger time from the circular buffer; Construct a data packet structure, encapsulate the state classification label of the negative stress condition, the timestamp of the trigger time, and the extracted target cumulative stress index into the data packet structure to generate a structured stress event data packet.

[0011] Optionally, the step of aggregating stress event data packets and baseline data packets in a remote server, and identifying the target stress pattern that causes negative stress in passengers by analyzing the statistical differences between the stress event data packets and baseline data packets in terms of cumulative stress indicators, includes the following steps: All received stress event data packets were classified into a negative sample set, and all received baseline data packets were classified into a normal sample set. Cumulative stress index sequences were extracted from the negative sample set and the normal sample set, and multi-dimensional statistical characteristics, including mean, variance, peak value, and duration, were calculated between the two cumulative stress index sequences. The gradient boosting tree classification model is trained using multi-dimensional statistical features. The training objective of the gradient boosting tree classification model is to maximize the ability to distinguish between negative sample sets and normal sample sets. After the gradient boosting tree classification model is trained, the statistical features that contribute the most to the classification results and their corresponding decision thresholds are extracted. The statistical features with the highest contribution and the decision threshold are combined and quantified into a target stress mode.

[0012] Optionally, the method further includes the following steps: Provide a feedback interface on the mobile device carried by the caregiver to allow the caregiver to provide feedback and mark the accuracy of the identified target stress patterns; Receive and store caregiver feedback tags via a remote server; The feedback tags are associated with the corresponding stress event data packets to form a validation dataset; Periodically use the validation dataset to incrementally train or retrain the gradient boosting tree classification model.

[0013] In a second aspect, the present invention also provides a remote data acquisition and analysis system for a stroller, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the remote data acquisition and analysis method for a stroller as described in the first aspect.

[0014] Thirdly, the present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, configure the processor to perform the remote data acquisition and analysis method for a stroller as described in the first aspect.

[0015] The beneficial effects of this invention are: This invention achieves real-time capture of causal relationships by simultaneously collecting environmental stimuli and the physiological responses of passengers, and performing intelligent preprocessing locally. Its core innovation lies in a trigger-based data packaging mechanism for negative stress events. This mechanism does not indiscriminately upload all data, but rather precisely encapsulates the cumulative environmental stress indicator sequence over a previous period when a negative state is detected in the passenger, thus efficiently locking in potential triggering data. By statistically comparing this data with baseline data under normal conditions on a remote server, the specific environmental patterns causing negative stress in passengers can be objectively and scientifically separated from massive amounts of data, such as whether it is due to bumps, excessively fast turns, or continuous uphill driving. This invention accurately identifies and understands the specific external factors causing discomfort in infants and young children, transforming vague subjective feelings into quantifiable and avoidable clear guidance, fundamentally improving passenger comfort and the scientific nature of care. Attached Figure Description

[0016] Figure 1This is a flowchart illustrating a remote data acquisition and analysis method for a child stroller in one embodiment of this application.

[0017] Figure 2 This is a schematic diagram of the system structure of a remote data acquisition and analysis system for a stroller in one embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0019] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0020] Figure 1 This is a flowchart illustrating a remote data acquisition and analysis method for a stroller in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise explicitly stated herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps. For example Figure 1 As shown, the remote data acquisition and analysis method for a children's stroller disclosed in this invention specifically includes the following steps: S101. By deploying an environmental acquisition module and a directional microphone on the stroller, environmental stimulus data including the stroller's movement posture and physiological response data including the occupant's vocalizations are simultaneously acquired and time-stamped.

[0021] The stroller integrates a sophisticated environmental data acquisition module and a directional microphone. These two modules work together: the former focuses on recording external physical stimuli, particularly the stroller's motion and posture information, such as bumps, vibrations, and tilts; the latter focuses on capturing the rider's direct physiological responses, primarily their vocalizations. Crucially, all collected data, whether describing the environment or the rider's reactions, must be strictly timestamped. Each data point is assigned a time stamp accurate to the millisecond, derived from a unified central clock. This synchronization mechanism ensures a direct temporal correlation between an environmental stimulus (such as a violent bump) and the subsequent physiological response (such as a cry). In this way, what is collected is no longer isolated fragments of information, but two interconnected, precisely time-corresponding data streams: one containing environmental stimulus data detailing the stroller's movement, and the other containing physiological response data including the rider's vocalizations.

[0022] S102. In the local processing unit of the stroller, environmental stimulus data is processed in real time into standardized cumulative stress indicators, and physiological response data is processed into state classification labels representing the rider's state through sound source separation and a pre-deployed audio classification model.

[0023] The raw data stream collected needs to be intelligently refined in real time by the local processing unit before it can be transformed into effective information for analysis. This processing is completed directly on the stroller to ensure immediacy and efficiency. Environmental stimulus data, mainly the stroller's motion posture information, is converted into a standardized cumulative stress index. This index aims to quantify the overall intensity of environmental stimulation exerted on the rider over a period of time. Simultaneously, physiological response data, including the rider's vocalizations and environmental noise, undergoes complex audio processing. First, sound source separation technology is applied to accurately separate the rider's voice from the noisy background. Then, this purified audio of the rider's vocalization is fed into an audio classification model pre-deployed in the local processing unit. This model, trained on a large amount of data, can identify the emotional state contained in the voice and output it as a clear state classification label, such as crying, irritability, calmness, or pleasure.

[0024] S103. When the state classification label meets the preset negative stress conditions, extract and associate the cumulative stress index sequence within the preset time window before the occurrence of the negative stress conditions, and generate a stress event data packet.

[0025] The local processing unit continuously monitors the output of the state classification labels after generating cumulative stress indicators and state classification labels in real time. Not every brief cry is recorded; instead, a preset negative stress condition is used to ensure that only sustained and significant negative emotional responses are captured. For example, this condition might be defined as the state classification label continuously displaying crying or agitation for more than a specific time threshold. Once this condition is met, the system determines that a negative stress event has occurred and immediately locks onto the precise moment that triggered the determination. At this point, the system uses a circular buffer to trace back the environmental conditions before the event. This circular buffer is an efficient memory management structure that always stores the cumulative stress indicator sequence for a recent period. When a negative stress event is triggered, the system extracts and associates all cumulative stress indicator data within a preset time window prior to the trigger time from this buffer. Finally, this key information, including the state classification label of the triggering event, the precise trigger timestamp, and the associated stress indicator sequence, is encapsulated into a structured stress event data packet.

[0026] S104. The stress event data packet and the baseline data packet representing the normal riding state of the rider, generated under non-negative stress conditions, are transmitted to a remote server via a mobile terminal carried by the caregiver.

[0027] To enable deeper aggregation analysis and pattern mining, locally generated stress event data packets need to be transmitted to a remote server. This process cleverly utilizes the caregiver's mobile device (such as a smartphone) as a data transmission bridge. When the local processing unit on the stroller generates a stress event data packet, it sends it to the paired mobile device via low-power wireless technology (such as Bluetooth). However, analyzing only negative event data is incomplete, as it cannot determine whether these environmental stimuli necessarily lead to a negative response. To establish a scientific baseline, the system also generates and packages data under non-negative stress conditions—that is, when the rider is calm or in a pleasant state—according to a certain strategy (such as fixed time intervals or random sampling). These are called baseline data packets. Structurally, baseline data packets are similar to stress event data packets, also containing a cumulative stress index sequence over a period of time, but their associated state labels are normal. Upon receiving these two types of data packets, the mobile device uses its own network connection (such as a cellular network or Wi-Fi) to upload them to the remote server. The advantage of this design is that it shifts the burden of data transmission from the stroller itself to a more powerful mobile terminal, while ensuring the integrity of the dataset used for final analysis.

[0028] S105. Aggregate stress event data packets and baseline data packets in a remote server, and identify the target stress patterns that cause negative stress in passengers by analyzing the statistical differences between the stress event data packets and the baseline data packets in terms of cumulative stress indicators.

[0029] On the remote server side, after a large number of stress event data packets and baseline data packets are collected, the final core analysis stage begins. The goal is to automatically identify and quantify the target stress patterns that cause negative stress in passengers. First, the server categorizes all received stress event data packets into a negative sample set and all baseline data packets into a normal sample set. Next, feature engineering is performed on the cumulative stress index sequence in each data packet to calculate multi-dimensional statistical features that describe the dynamic characteristics of the sequence, such as the mean, variance, peak size, and duration of high-order values. These statistical features constitute the key basis for distinguishing between negative events and normal situations. Subsequently, these features are used to train a powerful classification model, such as a Gradient Boosting Trees (GBT) model. The training objective of this model is to learn a decision rule that maximizes the distinction between the negative and normal sample sets. The GBT model iteratively constructs a series of simple decision trees, with each new tree working to correct the prediction error of the previous tree. After training, by analyzing the internal structure of the model, the statistical features that contribute most to the classification results and their corresponding decision thresholds can be extracted. For example, the model might find that when the variance of a stress indicator sequence exceeds a certain value and the peak value is also above a specific threshold, it is highly likely to correspond to a negative stress event. This combination of features and thresholds is precisely quantified as the target stress pattern that causes discomfort to the rider.

[0030] In one embodiment, the synchronous acquisition and timestamping of environmental stimulus data including the stroller's movement posture and physiological response data including the occupant's vocalizations, using an environmental acquisition module and a directional microphone deployed on the stroller, includes the following steps: The inertial measurement unit in the environmental acquisition module collects the three-axis acceleration data of the stroller at a preset frequency; The ambient background audio and the passenger's vocal audio are collected using the omnidirectional microphone in the environmental acquisition module and the directional microphone in the directional microphone at a preset sampling rate, respectively. The stroller's local processing unit assigns synchronized timestamps from a unified clock to the collected triaxial acceleration data, ambient background audio, and sound audio. Integrate time-stamped triaxial acceleration data into an environmental stimulus data stream; Integrate time-stamped ambient background audio and vocal audio into a physiological response data stream.

[0031] In this embodiment, to accurately capture the dynamic characteristics of the stroller during movement, the inertial measurement unit (IMU), a core component of the environmental acquisition module, is used to continuously collect the stroller's three-axis acceleration data at a preset, relatively high frequency. This unit can sense and quantify the stroller's acceleration changes in three mutually perpendicular directions—forward, left-right, and up-down—in real time. Setting a high acquisition frequency, such as collecting data 100 times per second, ensures that no brief but severe impacts that might cause discomfort to the rider are missed, such as the bumps caused by quickly running over a pebble or moving along a curb. This raw data is output in digital form, accurately depicting the stroller's motion posture, including the amplitude and frequency of its vibration, tilt, and swaying. To comprehensively record the acoustic environment in which the rider is located and their own vocal responses, the system employs a dual-microphone acquisition scheme. An omnidirectional microphone is deployed in the environmental acquisition module, designed to indiscriminately pick up all sounds from all directions, thus completely recording the background audio of the surrounding environment, such as traffic noise on the street and the noise of people in a shopping mall. Meanwhile, a directional microphone is positioned close to the occupant, its pickup direction precisely aimed at them. This microphone design makes it most sensitive to sounds directly in front, while effectively suppressing sounds from the sides and rear. The core purpose of this design is to maximize the capture of the occupant's vocal audio, such as cries, laughter, or babbling, while minimizing interference from ambient noise. Both microphones operate at a preset audio sampling rate to ensure high fidelity of sound information.

[0032] To ensure meaningful correlation analysis of data collected from different sensors, strict time synchronization of all data is required. The stroller's local processing unit maintains a unified, high-precision system clock. Whether it's each set of triaxial acceleration data collected by the inertial measurement unit or each audio data block collected by the two microphones, the local processing unit immediately appends a synchronization timestamp from this unified clock the moment the audio data is recorded. This precisely aligns data points from different sources and at different sampling frequencies onto the same timeline, ensuring that the stroller's motion, ambient sounds, and the rider's voice at any given time can be accurately correlated. After completing the acquisition and timestamp assignment of the stroller's triaxial acceleration data, these discrete data points need to be integrated into a structured environmental stimulus data stream. Logically, this process involves organizing a series of timestamped acceleration measurements in chronological order to form a continuous data sequence that can be efficiently read and processed by a computer program. In practice, each data unit contains two core components: a precise timestamp and a vector consisting of three values, representing the acceleration values ​​on the x, y, and z axes at that moment. This data stream can be viewed as a digitized trajectory, recording all the motion details of the cart over time.

[0033] Corresponding to the construction of the environmental stimulus data stream, the two acquired audio data streams also need to be integrated into a unified physiological response data stream. This process logically merges the environmental background audio from the omnidirectional microphone and the passenger's vocal audio from the directional microphone—two data sources tagged with synchronized timestamps—into a parallel data sequence. In specific implementation, each basic unit in the data stream contains a unified timestamp and a short audio signal acquired from each of the two microphones at that time point. This means that at any given time, samples of both environmental sound and passenger voice can be obtained simultaneously. This parallel structure is crucial for subsequent sound source separation and noise cancellation algorithms, as it provides the necessary reference signal (environmental audio) and mixed signal (audio containing both passenger voice and environmental sound), making it an indispensable step in accurately extracting the passenger's pure voice and thus accurately determining their emotional state.

[0034] In one embodiment, within the local processing unit of the stroller, environmental stimulus data is processed in real time into standardized cumulative stress indicators, and physiological response data is processed into state classification labels characterizing the rider's state through sound source separation and a pre-deployed audio classification model, including the following steps: Triaxial acceleration data is extracted in real time from the environmental stimulus data stream and combined with the environmental background audio in the physiological response data stream to calculate and generate cumulative stress indicators. A sound source separation algorithm is used to process the environmental background audio and vocal audio in the physiological response data stream to obtain the denoised vocal audio signal of the rider. The audio signal is input into a pre-deployed audio classification model, which outputs a state classification label representing the rider's state. The generated cumulative stress index and state classification label are stored in a circular buffer with a preset time length.

[0035] In this embodiment, the local processing unit extracts triaxial acceleration data from the environmental stimulus data stream in real time and simultaneously acquires environmental background audio from the physiological response data stream. These two distinct physical quantities, vibration and noise, are considered core stressors affecting rider comfort, thus requiring fusion into a comprehensive cumulative stress index. For vibration data, a specific frequency-weighted algorithm is first used to simulate the differences in sensitivity of infants to vibrations of different frequencies. Then, the weighted vibration value is integrated over time to obtain a vibration dose value that quantifies the cumulative impact of vibration. For environmental background audio, its spectral characteristics are analyzed in real time to calculate an A-weighted equivalent sound level that conforms to human auditory perception. Finally, these two values, representing the impact of vibration and noise respectively—vibration dose value and A-weighted equivalent sound level value—are normalized to scale them into a unified numerical range and combined using a weighted fusion function to generate a single-dimensional cumulative stress index. This fusion process can be represented as: ,in It is the ultimate cumulative stress indicator. and These are the normalized vibration dose value and the A-weighted equivalent sound level value, respectively. and These are preset weighting coefficients, representing the relative importance of vibration and noise in the final index.

[0036] The physiological response data stream provides two key signals: one is the audio signal primarily containing the passenger's vocalization but also mixed with environmental noise, and the other is a reference signal containing only environmental background audio. The core task of the sound source separation algorithm is to use the latter to purify the former. An efficient implementation method is to use adaptive noise cancellation technology. This technology uses the environmental background audio as the reference input and the passenger's vocalization audio as the main input. The algorithm dynamically adjusts the internal parameters of a digital filter, aiming to make the filtered reference noise signal infinitely close in morphology to the noise components contained in the main input signal. Once the optimal match is achieved, this estimated noise component is subtracted from the original main input signal. Ideally, this subtraction process can completely preserve the passenger's original vocalization signal while eliminating most of the background noise. This process is achieved through iterative optimization, and the filter adjustment logic aims to minimize the energy of the output signal. After obtaining the passenger's clean vocalization audio signal, the signal is immediately fed into an audio classification model pre-deployed in a local processing unit. Its task is to analyze the deep information contained in the sound and output a classification label that can clearly represent the passenger's current state. The model is typically a deep learning network trained on a large amount of labeled audio data (e.g., containing thousands of hours of audio clips of babies crying, laughing, breathing calmly).

[0037] Its working principle consists of two steps: First, the model extracts features from the input audio signal, transforming the original sound wave pattern into a set of digital features that can effectively distinguish different sound types. For example, it converts the sound into a spectrogram and extracts Mel-frequency cepstral coefficients. These features can capture key acoustic attributes such as pitch, timbre, energy distribution, and rhythmic variations. Then, these extracted features are fed into the model's classifier. The classifier, through its complex network structure, matches the input feature patterns with various state patterns learned during training, ultimately outputting the classification result with the highest probability. This result is a discrete state classification label, such as crying, agitated, calm, or happy.

[0038] A circular buffer is a first-in, first-out (FIFO) data structure with a fixed storage size. During system operation, the local processing unit stores the latest calculated cumulative stress index and the state classification label output by the audio classification model as a data pair into this circular buffer at fixed time intervals. When a new data pair is written, it is placed at the end of the buffer; if the buffer is full, the oldest data pair is automatically overwritten by the new data. This design ensures that the buffer always maintains a complete historical record of the intensity of environmental stimuli and the rider's state response, spanning a fixed duration (e.g., five minutes) from the current moment back.

[0039] In one implementation, calculating the cumulative stress index includes the following steps: A preset frequency weighting algorithm is applied to the triaxial acceleration data to highlight the vibration frequency components in the triaxial acceleration data; The weighted vibration data is integrated over time to calculate the vibration dose value that quantifies the cumulative vibration effect. Real-time analysis of ambient background audio is performed to calculate the A-weighted equivalent sound level value that conforms to human auditory perception. The vibration dose value and the A-weighted equivalent sound level value are normalized, and then the normalized vibration dose value and the A-weighted equivalent sound level value are combined through a weighted fusion function to generate a single-dimensional cumulative stress index.

[0040] In this embodiment, the triaxial acceleration data needs to undergo preprocessing to simulate the biomechanical response of the human body. The core of this process is the application of a pre-defined frequency-weighted algorithm, based on the conclusion drawn from extensive experimental research that the human body exhibits significant differences in sensitivity to vibrations at different frequencies. Typically, for infants and young children exposed to whole-body vibrations, low-frequency, generalized shaking and jolts (e.g., in the 1-10 Hz range) are more likely to cause discomfort than high-frequency, subtle humming. The frequency-weighted algorithm is essentially a digital filter designed to amplify the frequency components that have the greatest impact on human comfort while attenuating those with less significant impact. In practice, the original triaxial acceleration time-series signal is processed through this filter. The filter's response characteristics are set according to international standards (such as the Wk or Wd weighted curves defined in ISO 2631-1), ultimately outputting a new weighted acceleration time-series. This new sequence is no longer a purely physical quantity but is transformed into a more physiologically meaningful indicator, whose numerical fluctuations more accurately reflect the vibration intensity perceived by the rider.

[0041] After obtaining frequency-weighted vibration data, directly averaging the data ignores brief but intense impact events, which are crucial causes of discomfort. Therefore, a method for calculating vibration dose is used, a cumulative measure more sensitive to peak impacts. The calculation process first obtains the weighted vibration data time series generated in the previous step, then calculates the fourth power of the value of each data point in the series. The fourth power, rather than the square (energy), is used to significantly amplify the weight of high-intensity impacts, making a single violent jolt account for a much larger proportion of the final result than a prolonged, gentle vibration, which better aligns with the human body's stress response pattern. Subsequently, this fourth-powered data series is integrated within a preset time window, and finally, the fourth root of the integral is taken. The calculation formula can be expressed as: in, It is the calculated vibration dose value. It is in time Frequency-weighted acceleration value, It is the time period for performing integral calculations.

[0042] Similar to processing vibration data, analyzing environmental background audio also requires simulating human perception characteristics, rather than simply measuring its physical intensity. The human ear has varying sensitivities to different frequencies, being most sensitive to mid-frequency sounds (such as human speech) and relatively less sensitive to extremely low and high frequencies. To ensure noise measurement results match human auditory perception, the raw audio signal needs to be A-weighted. A-weighting is a standardized frequency weighting curve applied to the audio signal using a filter to adjust the weights of its spectral components, matching them to the frequency response characteristics of the human ear. After obtaining the A-weighted audio signal, its equivalent sound level over a period of time is calculated, i.e., the A-weighted equivalent sound level value. This value represents the sound level that a steady-state sound would need to produce the same sound energy as the changing noise within that time period; essentially, it is an average energy value. This value is measured in decibels (dBA) and is an internationally recognized authoritative indicator for measuring the impact of environmental noise. Its calculation formula is: in It is the A-weighted equivalent sound level value. It is the time period for averaging. It is the instantaneous sound pressure level after A-weighting, and It is the reference sound pressure level for human hearing threshold.

[0043] After calculating the vibration dose value, which quantifies the cumulative impact of vibration, and the A-weighted equivalent sound level value, which characterizes the auditory impact, respectively, the final step is to fuse these two indicators from different sources and with different units into a single-dimensional, comprehensive cumulative stress indicator. Since the numerical ranges and physical units of the vibration dose value and the A-weighted equivalent sound level value are completely different, direct arithmetic operations are not possible; therefore, normalization must first be performed. Normalization aims to map the two values ​​to the same dimensionless interval, such as 0 to 1. A common normalization method is based on preset comfort and discomfort thresholds, comparing the current measurement value to these thresholds to calculate a percentage of relative position. Once both indicators are converted to the same scale, they can be combined using a weighted fusion function. This function is typically a simple weighted sum, where the weighting coefficients are set based on prior knowledge or experimental data, reflecting the relative importance of vibration and noise stimuli in inducing negative stress responses in infants and young children.

[0044] In one embodiment, the remote data acquisition and analysis method for a stroller further includes the following steps: In addition to calculating cumulative environmental stress indicators, non-acoustic feature sequences characterizing the rider's physiological rhythms are also extracted from the physiological response data stream through a local processing unit. Using a pre-defined temporal correlation analysis algorithm, the synchronous correlation score between the changing trends of cumulative environmental stress indicators and the changing trends of non-acoustic feature sequences is calculated in real time in the local processing unit. When the synchronous correlation score exceeds the preset correlation threshold and the classification label does not meet the negative stress condition, it is determined that a latent stress event has occurred. In response to the determination that a latent stress event has occurred, a causal event data packet labeled as a latent stress type is generated; Causal event data packets labeled as latent stress types are used together with causal event data packets labeled as negative stress types for training machine learning models on remote servers to enhance the comprehensiveness of identification of key cumulative environmental stress patterns.

[0045] In this embodiment, in addition to assessing the environment and the rider's state through sound and motion, the local processing unit extracts non-acoustic biorhythmic features from the physiological response data stream to capture physiological responses that are not easily detected through overt behavior at a deeper level. This typically requires the use of non-invasive sensors integrated into the stroller seat or seatbelt, such as piezoelectric thin-film sensors or photoplethysmography (PPG) sensors. These sensors can detect subtle physical changes caused by heartbeat and respiration, such as minute vibrations of the body or pulsations of subcutaneous capillaries. The local processing unit performs sophisticated signal processing on the raw signals acquired by these sensors, including filtering out noise interference caused by the stroller's movement itself, and then applies peak detection and periodic analysis algorithms to calculate real-time physiological rhythm indicators, such as heart rate, respiratory rate, and heart rate variability, which measures the balance of the autonomic nervous system. Finally, these quantified, time-varying physiological indicators are organized into a continuous sequence of non-acoustic features.

[0046] The local processing unit performs real-time correlation analysis on two data streams: one is a cumulative environmental stress index sequence representing overall environmental pressure, and the other is a newly extracted non-acoustic feature sequence, such as a heart rate sequence. This process is implemented using a pre-defined time-series correlation analysis algorithm, which calculates within a continuously sliding time window to ensure the immediacy of the analysis. One implementation method is to calculate the Pearson correlation coefficient, which measures the degree of linear correlation between two continuous variables. Within a very short, continuously sliding time window (e.g., thirty seconds), the algorithm simultaneously retrieves the environmental stress index data and heart rate data within that window, and then calculates the correlation between other variables. The calculation formula is as follows: in, This is the calculated synchronization correlation score, where W is the length of the time window. and These are the k-th data points of the environmental stress index and the heart rate sequence within the window, respectively. and These are their average values ​​within the window. The benefit of this step is that it generates a score that varies continuously between -1 and +1, quantifying in real time the synchronicity between environmental stimuli and the rider's circadian rhythm. A high score close to +1 means that whenever the environmental stress index increases, the rider's heart rate also increases significantly, revealing a strong positive correlation between the two.

[0047] After obtaining real-time synchronization correlation scores, the system gains the ability to identify a more subtle stress state. When a passenger's circadian rhythm is highly synchronized with external environmental stress, but their outward emotions have not yet manifested through crying or other means, a potential, suppressed negative state may exist. Therefore, the system establishes a judgment condition to capture this latent stress event. This judgment logic consists of two sub-conditions that must be met simultaneously: first, the calculated synchronization correlation score exceeds a preset, high correlation threshold, such as 0.7, for a sustained period of time; second, within the same time period, the state classification label derived from audio analysis does not meet the predefined negative stress condition, i.e., the passenger does not exhibit obvious crying or agitation. When both conditions are met simultaneously, the system determines that a latent stress event has occurred. This judgment logic no longer relies solely on explicit negative signals such as crying, but rather, through the inherent correlation between data, it discerns that the passenger may be in a critical state where they are physiologically stressed but behaviorally enduring it.

[0048] Once the local processing unit determines that a latent stress event has occurred based on preset logic, the system immediately responds and generates a causal event data packet specifically designed for this type of event. The structure of this data packet is similar to the data packets previously used to record overt negative stress events, but it contains richer and more unique evidentiary information. Its core content is first clearly marked as a latent stress type to distinguish it from negative stress data packets triggered by crying on the server side. Secondly, the data packet encapsulates key evidence that triggered this determination. This includes not only a cumulative environmental stress indicator sequence over a period prior to the event, but also a synchronized non-acoustic feature sequence (such as a heart rate sequence) that caused the correlation score to exceed the threshold. Furthermore, the specific synchronized correlation score value exceeding the threshold is also recorded in the data packet as a direct quantitative basis for the strength of this determination. This step creates a completely new data output with extremely high information density. It not only records the potential environmental trigger but also directly links and packages this trigger with a silent, purely physiological negative reaction, providing deeper causal chain samples for subsequent machine learning analysis.

[0049] Ultimately, on the remote server, these newly generated causal event data packets, labeled as latent stress types, will be given equal importance to negative stress data packets triggered by crying, and both will be used in the training process of the machine learning model. The fundamental purpose of this approach is to significantly enhance the model's comprehensiveness in recognizing key environmental stress patterns. If the model is trained using only explicit crying events as negative samples, it may miss stress patterns that cause significant physiological discomfort but do not immediately lead to crying, such as persistent, subtle vibrations at a specific frequency. By including latent stress data packets in the training set, the model is effectively provided with a broader and more diverse range of negative cases. During the learning process, the model will treat the environmental stress indicator sequences in these latent events as targets that need to be distinguished from the normal baseline data. This forces the model to learn and recognize more subtle and covert combinations of stimuli. The final trained model will have stronger generalization ability and higher sensitivity. The target stress patterns it identifies will no longer be limited to those strong stimuli that can trigger intense emotional outbursts, but will be extended to all combinations of environmental factors that can be proven to cause significant negative physiological reactions in riders, thereby bringing the warning and protection capabilities of the entire system to a whole new level.

[0050] In one embodiment, the process of using a sound source separation algorithm to process the environmental background audio and vocal audio in the physiological response data stream to obtain the denoised vocal audio signal of the passenger includes the following steps: The ambient background audio is used as a reference noise signal, and the emitted audio is used as a mixed signal containing the target signal and noise, which is then input to a preset audio filter. The filtering parameters inside the audio filter are dynamically adjusted based on the correlation between the reference noise signal and the mixed signal. The estimated ambient noise component is subtracted from the mixed signal using the adjusted filtering parameters, and the denoised audio signal of the passenger is output.

[0051] In this embodiment, time-synchronized audio streams captured from two different microphones are fed into a preset adaptive audio filter. The ambient background audio captured by the omnidirectional microphone is designated as the reference noise signal. The value of this signal lies in its provision of a sample of the current ambient noise characteristics (such as traffic noise or wind noise). Simultaneously, the audio captured by the directional microphone, emanating from a nearby occupant, is defined as a mixed signal containing both the target signal and noise. This signal needs to be cleaned because it contains both the occupant's voice (i.e., the target signal) and inevitably mixed with ambient noise highly similar to the reference noise signal. Once the input signal is ready, the filter's internal filtering parameters are dynamically and intelligently adjusted based on the real-time correlation between the reference noise signal and the mixed signal. The core principle is that the filter continuously attempts to generate a "simulated" noise signal using the reference noise signal and constantly compares this simulated noise with the mixed signal, aiming to minimize the energy of the difference between the two (i.e., the error signal). Ideally, this error signal is the pure target sound. A widely adopted algorithm for implementing this dynamic adjustment is the Least Mean Square (LMS) algorithm. This algorithm updates the filter parameters through a concise iterative formula, and its core update logic can be expressed as: In this formula, It is the vector of internal parameters (coefficients) of the filter at time k. It is a step size factor that controls the adjustment speed and stability. It is not the input reference noise signal vector, but the error signal obtained by subtracting the simulated noise signal currently generated by the filter from the mixed signal. The significance of this formula is that with each calculation, the filter will fine-tune its parameters according to the current error magnitude and direction, so that it can generate a more realistic simulated noise in the next calculation.

[0052] Once the filter's internal parameters reach their optimal state through dynamic adjustment, noise cancellation is performed, and the final result is output. Specifically, using the adjusted optimal filter parameters, the reference noise signal is processed into an optimal estimate of the noise components in the mixed signal. Then, this accurately estimated environmental noise component is subtracted from the original mixed signal. The result of this subtraction operation is the final output denoised audio signal of the passenger. Mathematically, this output signal is actually the error signal used in the previous step to drive the filter parameter adjustment. When the filter is well-fitted, the resulting simulated noise signal... Will be with mixed signals Noise components included Infinitely close. Therefore, The result will be approximately equal to Ultimately, only the pure target signal remains. The final result was an audio recording with an extremely high signal-to-noise ratio, in which the details of the rider's vocalizations were clearly preserved, while background noise was significantly suppressed or completely eliminated.

[0053] In one implementation, when the state classification label meets a preset negative stress condition, the cumulative stress index sequence within a preset time window before the occurrence of the negative stress condition is extracted and associated to generate a stress event data packet, including the following steps: The real-time output values ​​of the status classification labels are continuously monitored in the local processing unit; When the output value of the state classification label continuously matches a negative emotional state within a preset duration threshold, the state classification label is determined to meet the preset negative stress condition. Lock the trigger time and extract all target cumulative stress indicators within a preset time window prior to the trigger time from the circular buffer; Construct a data packet structure, encapsulate the state classification label of the negative stress condition, the timestamp of the trigger time, and the extracted target cumulative stress index into the data packet structure to generate a structured stress event data packet.

[0054] In this implementation, the local processing unit continuously monitors the state classification labels output by the audio classification model in real time. Each state classification label is a discrete output value, such as crying, agitation, calmness, or pleasure, and is refreshed at a fixed frequency. The monitoring mechanism is essentially a continuously running loop detection program that acquires the latest label value at each time point and compares it with a predefined set of negative emotional states. This set serves as the starting point for analysis, identifying which emotional states are considered negative stress responses requiring attention. Merely detecting brief negative emotional labels is insufficient, as these could be accidental or meaningless reactions. To ensure that the captured events represent genuine and persistent stress states, the system introduces a time-dimensional verification mechanism. When the monitoring program first detects that a state classification label matches a preset negative emotional state (e.g., crying), an internal timer or counter is activated. At each subsequent time point, the system continuously checks whether the label remains in that negative state. Only when the cumulative duration of such continuous matching negative states exceeds a pre-set duration threshold (e.g., ten consecutive seconds) will the system finally determine that a valid negative stress condition has been met.

[0055] Once the negative stress condition is confirmed to be met, the system immediately locks the current moment as the trigger moment of the event and immediately initiates the data backtracking and extraction process. At this time, the circular buffer, which is maintained in memory beforehand, plays a crucial role as a short-term memory. This buffer stores the complete sequence of target cumulative stress indicators within a recent period (e.g., the past five minutes). Using the locked trigger moment as a reference point, the system traces back to the past according to a preset time window (e.g., one minute before the event), accurately extracting all cumulative stress indicator data contained within this time window from the circular buffer. This extraction process is not a simple data copying, but rather a logically extracted segment of the environmental stimulus history that is closely related to the negative event in time. After extracting the key environmental data, the final step is to locally encapsulate all relevant information in a structured manner, generating a complete and uniformly formatted stress event data packet. This process is like writing a standard case report for an investigation of an event. First, a clear data packet structure is constructed, which defines several standardized fields. These fields must contain the core elements of this event: first, the state classification label of the negative stress condition that triggered the event (such as crying); second, the trigger timestamp recording the precise time the event occurred; and finally, and most importantly, the sequence of target cumulative stress indicators, extracted from the circular buffer as potential triggers. All this information is systematically populated into a predefined data packet structure.

[0056] In one implementation, aggregating stress event data packets and baseline data packets in a remote server, and identifying the target stress pattern that causes negative stress in passengers by analyzing the statistical differences in cumulative stress indicators between the stress event data packets and the baseline data packets, includes the following steps: All received stress event data packets were classified into a negative sample set, and all received baseline data packets were classified into a normal sample set. Cumulative stress index sequences were extracted from the negative sample set and the normal sample set, and multi-dimensional statistical characteristics, including mean, variance, peak value, and duration, were calculated between the two cumulative stress index sequences. The gradient boosting tree classification model is trained using multi-dimensional statistical features. The training objective of the gradient boosting tree classification model is to maximize the ability to distinguish between negative sample sets and normal sample sets. After the gradient boosting tree classification model is trained, the statistical features that contribute the most to the classification results and their corresponding decision thresholds are extracted. The statistical features with the highest contribution and the decision threshold are combined and quantified into a target stress mode.

[0057] In this implementation, after a massive amount of data packets are collected on the remote server, they are classified based on the inherent labels assigned to each data packet when it is generated locally. All data packets marked as recording explicit negative reactions such as crying or agitation in passengers are systematically grouped into a special set called the negative sample set. Conversely, all baseline data packets collected when passengers are in a calm or pleasant state, representing a normal riding experience, are grouped into another set, the normal sample set. After dividing the data into negative and normal sample sets, it is necessary to extract key information that can quantify their core characteristics from these raw data. Directly comparing two cumulative stress index sequences of different lengths is difficult and inefficient; therefore, feature engineering must be performed on each sequence to transform it into a set of fixed, multi-dimensional statistical features. For each data packet in both the negative and normal sample sets, the cumulative stress index sequence contained within it is extracted and calculated. These calculations encompass multiple dimensions, aiming to depict the sequence's shape from different perspectives. For example, they calculate the sequence mean to reflect average stimulus intensity; calculate variance to measure the stationarity or volatility of the stimulus; identify peaks in the sequence to capture the strongest single impact; and calculate the duration for which indicators remain above a certain threshold to quantify the persistence of high-intensity stimuli. For instance, the variance calculation can be expressed as: in, It is the calculated variance of the sequence, which is the total number of data points in the sequence. It is the i-th cumulative stress index value in the sequence, and It is the average value of the sequence.

[0058] After each data packet is transformed into a multi-dimensional statistical feature vector, these features are used to train a gradient boosting tree classification model. The fundamental goal is to learn a decision rule that can distinguish between negative and normal samples with the highest accuracy. Gradient boosting trees are a powerful and efficient ensemble learning algorithm. Instead of building a large and complex model all at once, it iteratively generates a series of simple decision trees. In each iteration, the algorithm builds a new decision tree, whose core task is to correct errors in the cumulative predictions of all previous trees. The final model is a weighted combination of all these trees, and its predictive power far exceeds that of any single decision tree. The essence of this training process is to find an optimal decision boundary in a high-dimensional feature space, thereby maximizing the separation of points representing negative events from those representing normal events.

[0059] After the gradient boosting tree classification model is trained, it needs to be analyzed in depth to extract the key information that contributes most to the classification results. Since the gradient boosting tree model evaluates the role of each feature at each split node during its construction, the contribution of each statistical feature to improving the overall classification accuracy of the model can be quantified. By analyzing these contribution scores, it is possible to clearly identify which environmental features, such as drastic fluctuations in stimulus (high variance) or a particularly high impact (peak value), are the most critical factors in distinguishing whether a rider experiences negative stress. After determining the feature with the highest contribution, the optimal decision threshold related to that feature is further extracted from the model structure; that is, the specific cutoff value used by the model when making judgments.

[0060] Finally, the key insights extracted from the model are transformed into a quantifiable target stress pattern. This process involves meaningfully combining the statistically most contributing features obtained in the previous step with their corresponding decision thresholds. For example, if the analysis reveals that the most contributing feature is the variance of the cumulative stress indicator sequence, with an optimal decision threshold of 2.5, then this finding can be concretized into a definition of a target stress pattern: when the quantified variance of the environmental stimuli experienced by a stroller over a certain period exceeds 2.5, it is highly likely to cause a negative stress response in the rider. This quantified combination constitutes a data-driven, statistically significant target stress pattern.

[0061] In one embodiment, the remote data acquisition and analysis method for a stroller further includes the following steps: Provide a feedback interface on the mobile device carried by the caregiver to allow the caregiver to provide feedback and mark the accuracy of the identified target stress patterns; Receive and store caregiver feedback tags via a remote server; The feedback tags are associated with the corresponding stress event data packets to form a validation dataset; Periodically use the validation dataset to incrementally train or retrain the gradient boosting tree classification model.

[0062] In this embodiment, an interactive feedback interface is provided on a mobile terminal application carried by the caregiver. When a remote server identifies and pushes a target stress pattern that may cause discomfort to the passenger, the interface presents it to the caregiver in a concise and intuitive manner. For example, it might display a brief environmental summary before and after the event and ask a specific question, such as: "Do you think the continuous bumps in the road were the main cause before the crying?" The caregiver can mark the accuracy of this inference with a simple button click; options include confirming, denying, or marking as uncertain. Every feedback mark provided by the caregiver on the mobile terminal must be reliably collected and integrated into a remote central data system. This is specifically accomplished through a communication link between the mobile terminal and the remote server. After the caregiver completes a feedback mark, the mobile application immediately encapsulates this feedback information into a standardized data message. This message contains at least three key elements: an ID uniquely identifying the stress event, the specific feedback mark selected by the caregiver (e.g., confirming, denying), and a timestamp when the feedback was submitted. The data message is then sent via a secure network protocol, such as HTTPS, to a dedicated application programming interface (API) on a remote server for receiving feedback. Upon receiving the message, the server verifies it to ensure data integrity and validity, and then persistently stores the record containing the event ID and corresponding feedback in a dedicated database table.

[0063] Simply storing the feedback tags themselves is insufficient; these tags must be precisely associated with their corresponding original event data. This process is completed at the database level on a remote server, and its core is a data connection operation. The system uses newly received feedback tag records as indexes, employing the stored event IDs, to retrieve and locate the complete stress event data packet that initially triggered the feedback. This data packet contains a sequence of cumulative environmental stress indicators and multi-dimensional statistical features calculated from this sequence. By pairing and binding these two elements—the event data with its original features and the authoritative feedback tags from caregivers—a new, high-quality dataset is formed, called the validation dataset. A sample from this dataset... It can be represented as a tuple: , It is a multi-dimensional statistical feature vector extracted from the original stress event data packet, and It is a feedback mark that represents the true label of the event and has been personally confirmed or corrected by the caregiver.

[0064] Ultimately, the validation dataset will be used to periodically improve the performance of the core gradient boosting tree classification model. Specifically, at predetermined intervals, such as weekly or monthly, the system will initiate a training task. This task can use one of two strategies: incremental training, which involves fine-tuning the existing model using only the new validation dataset to correct its erroneous perceptions—suitable for scenarios requiring rapid iteration; or a more thorough retraining, which involves merging the new validation dataset with some of the original, high-quality training data, using this richer and more accurate new dataset to train a new, more powerful model from scratch to replace the old one. Regardless of the strategy used, the training goal is always to make the model's predictions on new data as close as possible to the actual feedback labels provided by caregivers.

[0065] The present invention also discloses a remote data acquisition and analysis system for a stroller, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the remote data acquisition and analysis method for a stroller as described in any of the above embodiments.

[0066] The processor can be a central processing unit (CPU). Of course, depending on the actual use, it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it.

[0067] The memory can be an internal storage unit of a computer device, such as a hard disk or RAM, or an external storage device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) provided on the computer device. Furthermore, the memory can be a combination of internal storage units and external storage devices of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0068] The present invention also discloses a computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the remote data acquisition and analysis method for a stroller described in any of the above embodiments.

[0069] The computer program can be stored in a machine-readable medium. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or certain middleware. The machine-readable medium includes any entity or device capable of carrying computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the machine-readable medium includes, but is not limited to, the above-mentioned components.

[0070] The remote data acquisition and analysis method for a stroller described in the above embodiments is stored in the computer-readable storage medium and loaded and executed on the processor to facilitate the storage and application of the above method.

[0071] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0072] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A remote data acquisition and analysis method for a child stroller, characterized in that, The method includes the following steps: By deploying an environmental acquisition module and a directional microphone on the stroller, environmental stimulus data including the stroller's movement posture and physiological response data including the rider's vocalizations are collected and time-stamped. In the local processing unit of the stroller, environmental stimulus data is processed in real time into standardized cumulative stress indicators, and physiological response data is processed into state classification labels representing the rider's state through sound source separation and pre-deployed audio classification models. When the state classification label meets the preset negative stress conditions, the cumulative stress index sequence within the preset time window before the occurrence of the negative stress conditions is extracted and associated to generate a stress event data packet. The caregiver transmits stress event data packets and baseline data packets representing the passenger's normal riding status under non-negative stress conditions to a remote server via a mobile terminal. By aggregating stress event data packets and baseline data packets on a remote server, and analyzing the statistical differences in cumulative stress indicators between the stress event data packets and baseline data packets, the target stress patterns that cause negative stress in passengers are identified.

2. The remote data acquisition and analysis method for a child stroller according to claim 1, characterized in that, The process of synchronously collecting and timestamping environmental stimulus data, including the stroller's movement posture, and physiological response data, including the occupant's vocalizations, using an environmental acquisition module and directional microphone deployed on the stroller, includes the following steps: The inertial measurement unit in the environmental acquisition module collects the three-axis acceleration data of the stroller at a preset frequency; The ambient background audio and the passenger's vocal audio are collected using the omnidirectional and directional microphones in the environmental acquisition module at a preset sampling rate. The stroller's local processing unit assigns synchronized timestamps from a unified clock to the collected triaxial acceleration data, ambient background audio, and sound audio. Integrate time-stamped triaxial acceleration data into an environmental stimulus data stream; Integrate time-stamped ambient background audio and vocal audio into a physiological response data stream.

3. The remote data acquisition and analysis method for a stroller according to claim 1, characterized in that, The local processing unit of the stroller processes environmental stimulus data in real time into standardized cumulative stress indicators, and processes physiological response data into state classification labels characterizing the rider's state through sound source separation and a pre-deployed audio classification model, including the following steps: Triaxial acceleration data is extracted in real time from the environmental stimulus data stream and combined with the environmental background audio in the physiological response data stream to calculate and generate cumulative stress indicators. A sound source separation algorithm is used to process the environmental background audio and vocal audio in the physiological response data stream to obtain the denoised vocal audio signal of the rider. The audio signal is input into a pre-deployed audio classification model, which outputs a state classification label representing the rider's state. The generated cumulative stress index and state classification label are stored in a circular buffer with a preset time length.

4. The remote data acquisition and analysis method for a child stroller according to claim 3, characterized in that, The calculation and generation of the cumulative stress index includes the following steps: A preset frequency weighting algorithm is applied to the triaxial acceleration data to highlight the vibration frequency components in the triaxial acceleration data; The weighted vibration data is integrated over time to calculate the vibration dose value that quantifies the cumulative vibration effect. Real-time analysis of ambient background audio is performed to calculate the A-weighted equivalent sound level value that conforms to human auditory perception. The vibration dose value and the A-weighted equivalent sound level value are normalized, and then the normalized vibration dose value and the A-weighted equivalent sound level value are combined through a weighted fusion function to generate a single-dimensional cumulative stress index.

5. The remote data acquisition and analysis method for a stroller according to claim 3, characterized in that, The process of using a sound source separation algorithm to process the environmental background audio and vocal audio in the physiological response data stream to obtain the denoised vocal audio signal of the passenger includes the following steps: The ambient background audio is used as a reference noise signal, and the emitted audio is used as a mixed signal containing the target signal and noise, which is then input to a preset audio filter. The filtering parameters inside the audio filter are dynamically adjusted based on the correlation between the reference noise signal and the mixed signal. The estimated ambient noise component is subtracted from the mixed signal using the adjusted filtering parameters, and the denoised audio signal of the passenger is output.

6. The remote data acquisition and analysis method for a stroller according to claim 3, characterized in that, When the state classification label meets the preset negative stress condition, the process of extracting and associating the cumulative stress index sequence within the preset time window before the occurrence of the negative stress condition to generate a stress event data packet includes the following steps: The real-time output values ​​of the status classification labels are continuously monitored in the local processing unit; When the output value of the state classification label continuously matches a negative emotional state within a preset duration threshold, the state classification label is determined to meet the preset negative stress condition. Lock the trigger time and extract all target cumulative stress indicators within a preset time window prior to the trigger time from the circular buffer; Construct a data packet structure, encapsulate the state classification label of the negative stress condition, the timestamp of the trigger time, and the extracted target cumulative stress index into the data packet structure to generate a structured stress event data packet.

7. The remote data acquisition and analysis method for a stroller according to claim 1, characterized in that, The process of aggregating stress event data packets and baseline data packets on a remote server, and identifying the target stress pattern that causes negative stress in passengers by analyzing the statistical differences between the stress event data packets and baseline data packets in terms of cumulative stress indicators, includes the following steps: All received stress event data packets were classified into a negative sample set, and all received baseline data packets were classified into a normal sample set. Cumulative stress index sequences were extracted from the negative sample set and the normal sample set, and multi-dimensional statistical characteristics, including mean, variance, peak value, and duration, were calculated between the two cumulative stress index sequences. The gradient boosting tree classification model is trained using multi-dimensional statistical features. The training objective of the gradient boosting tree classification model is to maximize the ability to distinguish between negative sample sets and normal sample sets. After the gradient boosting tree classification model is trained, the statistical features that contribute the most to the classification results and their corresponding decision thresholds are extracted. The statistical features with the highest contribution and the decision threshold are combined and quantified into a target stress mode.

8. The remote data acquisition and analysis method for a stroller according to claim 7, characterized in that, The method further includes the following steps: Provide a feedback interface on the mobile device carried by the caregiver to allow the caregiver to provide feedback and mark the accuracy of the identified target stress patterns; Receive and store caregiver feedback tags via a remote server; The feedback tags are associated with the corresponding stress event data packets to form a validation dataset; Periodically use the validation dataset to incrementally train or retrain the gradient boosting tree classification model.

9. A remote data acquisition and analysis system for a stroller, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the remote data acquisition and analysis method for a children's stroller as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When executed by a processor, the instruction causes the processor to be configured to perform the remote data acquisition and analysis method for a stroller according to any one of claims 1 to 8.

Images