Fatigue detection and early warning system

Abstract

The invention discloses a fatigue detection and early warning system, and relates to the field of physiological monitoring and safety management. The system comprises a body temperature acquisition module for continuously acquiring body temperature time sequence data of a user during exercise or driving; the multi-scale decomposition module is used for analyzing body temperature changes under different time scales by adopting a wavelet decomposition technology; the stability feature extraction module is used for extracting stability features from the decomposed waveforms and forming feature vectors; the pedigree clustering module is used for performing clustering analysis based on the feature vectors and constructing a body temperature stability pedigree database; the real-time pedigree matching module calculates and matches the current feature vector of the user to the pedigree category in the database in real time; and the fatigue risk judgment module is used for monitoring whether the pedigree category migrates from the low risk to the high risk, and generating early warning information when the migration is detected. According to the system, body temperature changes are dynamically analyzed, so that the real-time performance and convenience of fatigue monitoring are improved, and the safety in sports and driving scenes is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of physiological monitoring and safety management, and more particularly to a fatigue detection and early warning system. Background Art

[0002] Body temperature, as an important indicator of human physiological status, is widely used to assess health status, metabolic levels, and fatigue levels. In scenarios requiring sustained concentration, such as exercise and driving, changes in body temperature often reflect the accumulation of fatigue. For example, prolonged activity may cause elevated body temperature or abnormal fluctuations, while fatigue may reduce the ability to regulate body temperature, presenting an atypical pattern of changes. Currently, fatigue monitoring relies heavily on methods such as heart rate, EEG signals, or subjective assessments, but these methods are often limited in real-time data acquisition and portable applications. Although body temperature is easy to measure, it is underutilized, especially in dynamic analysis. Existing body temperature monitoring technologies often use data as a single static indicator, lacking in-depth exploration of time series characteristics. This makes it difficult to capture the patterns of change at different time scales and cannot effectively correlate fatigue risks. Therefore, accurately identifying fatigue states through dynamic changes in body temperature has become one of the key issues in improving sports and driving safety. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a fatigue detection and early warning system to solve the problems mentioned in the background technology.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A fatigue detection and early warning system, comprising:

[0006] The body temperature collection module is used to continuously collect the target user's body temperature time series data when exercising or driving;

[0007] A multi-scale decomposition module is used to perform multi-scale wavelet decomposition on the body temperature time series data to obtain body temperature waveform subsequences at different time scales;

[0008] a stability feature extraction module, configured to extract the waveform stability feature corresponding to each time scale based on the body temperature waveform subsequence to form a stability feature vector;

[0009] a pedigree clustering module, configured to perform cluster analysis based on the stability feature vector to form multiple body temperature stability pedigrees and construct a stability pedigree database;

[0010] A real-time pedigree matching module is used to calculate the stability feature vector of the current user in real time and match it to the corresponding pedigree category in the stability pedigree database;

[0011] The fatigue risk judgment module is used to judge whether the stability spectrum category of the target user has migrated from a low-risk spectrum to a high-risk spectrum, and generate fatigue warning information when migration is detected.

[0012] In some embodiments, the body temperature time series data collected by the body temperature collection module includes continuous skin temperature data or core body temperature data, and each data point is timestamped.

[0013] In some embodiments, the multi-scale decomposition module uses a discrete wavelet transform algorithm or a continuous wavelet transform algorithm to decompose the body temperature time series data into a plurality of body temperature waveform subsequences of different time scales.

[0014] In some embodiments, the waveform stability features extracted by the stability feature extraction module include at least one or more of the standard deviation of the fluctuation amplitude, the density of local extreme values, the mean of the waveform curvature, the distribution variance of the peak-to-valley distance, and the local coefficient of variation.

[0015] In some embodiments, the pedigree clustering module uses a hierarchical clustering algorithm or a density clustering algorithm to cluster the stability feature vectors to form a plurality of stability pedigree categories with obvious distinctions, and stores each stability pedigree category in the stability pedigree database.

[0016] In some embodiments, the pedigree clustering module further includes a pedigree category marking unit, which performs fatigue risk marking on the stability pedigree category based on historical sample data, and the fatigue risk marking includes at least high risk, medium risk and low risk.

[0017] In some embodiments, the real-time pedigree matching module periodically calculates the stability feature vector of the current user at set time intervals during the real-time monitoring process, and matches the stability feature vector of the current user to a specific pedigree category in the stability pedigree database in real time based on the nearest neighbor algorithm or the K-nearest neighbor algorithm.

[0018] In some embodiments, the fatigue risk judgment module further includes a pedigree migration detection unit, which is used to continuously monitor the pedigree category matching results within multiple time intervals. When the pedigree category continuously migrates from a low-risk pedigree to a high-risk pedigree within a predetermined number of times, it is determined to be a fatigue risk state.

[0019] In some embodiments, the fatigue warning information generated by the fatigue risk judgment module includes fatigue risk level information and corresponding intervention suggestion information, and the intervention suggestion information includes suggestions to remind the user to rest, adjust exercise intensity or reduce driving speed.

[0020] In some embodiments, the system further includes a pedigree database update module, which is used to regularly collect body temperature time series data of the target user and update the stability pedigree category in the stability pedigree database to achieve adaptive adjustment of the system to the target user.

[0021] The advantage of the present invention over existing technologies lies in that it continuously collects body temperature time series data and performs multi-scale wavelet decomposition, deeply analyzing waveform characteristics at different time scales, thus solving the problem that traditional methods have difficulty in dynamically assessing body temperature changes. This method not only improves the real-time performance of fatigue monitoring, but also significantly improves the accuracy of identifying fatigue risks through the extraction of stability features and pedigree matching, providing higher safety protection for sports and driving scenarios. In addition, the system further utilizes diverse stability features, such as fluctuation amplitude and local extreme value density, combined with hierarchical or density clustering algorithms to construct and update a personalized pedigree database. This enables the monitoring process to adapt to individual differences in users and promptly detect potential fatigue trends through periodic matching and pedigree migration detection. At the same time, the generated warning information includes specific intervention recommendations, such as rest or adjusting activity intensity, further enhancing the practicality and reliability of the system and providing users with comprehensive fatigue management support. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is the overall flow chart of the present invention;

[0023] Figure 2 is a flow chart of the multi-scale decomposition module of the present invention;

[0024] Figure 3 is a flow chart of the stability feature extraction module of the present invention;

[0025] Figure 4 is a flow chart of the lineage clustering module of the present invention. DETAILED DESCRIPTION

[0026] The specific embodiments of the present invention will be described below with reference to the accompanying drawings.

[0027] like Figure 1 The present invention relates to a fatigue detection and warning system designed to improve safety during exercise and driving by monitoring and analyzing changes in a user's body temperature in real time, detecting potential fatigue and issuing warnings when necessary. The system utilizes a combination of body temperature acquisition, multi-scale signal analysis, feature extraction, and spectral clustering technologies to accurately identify patterns in body temperature stability and correlate them with fatigue risk.

[0028] The present invention includes a body temperature acquisition module that continuously collects time-series body temperature data while the user is exercising or driving. This data can be either skin temperature or core body temperature, depending on the actual application scenario and device conditions. Skin temperature is typically measured using a sensor close to the skin, such as the wrist, forehead, or chest, while core body temperature may require a more in-depth measurement method, such as an ear canal or oral thermometer. Generally, skin temperature is sufficient.

[0029] In practice, temperature acquisition modules typically rely on high-precision digital sensors, such as thermistors or infrared temperature sensors. These sensors record temperature data at a fixed frequency (e.g., once per second or once per minute) and attach a precise timestamp to each data point, allowing subsequent analysis to track temperature changes over time. The collected data is transmitted to the system via wired or wireless methods (e.g., Bluetooth or Wi-Fi).

[0030] For example, while driving, a driver could wear a smartwatch or wristband with a built-in temperature sensor that continuously monitors skin temperature at the wrist. The wristband records this data every few seconds and transmits it to the vehicle's onboard system in real time. Meanwhile, during exercise, a runner or cyclist could monitor their body temperature with a similar wearable device, with the data transmitted to a mobile app or dedicated receiving device.

[0031] To ensure a smooth collection process, several key points need to be considered when designing the body temperature collection module. The sensor must be lightweight and fit snugly to the human body to avoid discomfort. At the same time, the measurement accuracy must be high enough, typically within ±0.1°C, to capture subtle changes in body temperature. Furthermore, the device needs to be waterproof and shock-resistant, as sweat or bumps may affect it during exercise or driving. The collected body temperature time series data is not just a single numerical value; it also includes a time dimension, forming a continuous dynamic dataset that lays the foundation for subsequent multi-scale analysis.

[0032] like Figure 2 As shown in the figure, the multi-scale decomposition module receives the temperature time series data from the temperature acquisition module and uses multi-scale wavelet decomposition technology to split it into temperature waveform subsequences at different time scales. This decomposition method can reveal the characteristics of the temperature signal in different frequency ranges, such as short-term rapid fluctuations and long-term slow trends.

[0033] Wavelet decomposition is widely used in time series analysis. In the system of the present invention, either discrete wavelet transform (DWT) or continuous wavelet transform (CWT) can be selected to achieve decomposition. Taking discrete wavelet transform as an example, this method decomposes the original signal into approximation coefficients and detail coefficients through a series of high-pass and low-pass filters. The approximation coefficients capture low-frequency changes in body temperature, such as overall trends within a few hours, while the detail coefficients reflect high-frequency changes, such as sudden fluctuations within a few minutes. Through multiple iterations, subsequences of multiple time scales can be generated.

[0034] Specifically, assume that the system performs a three-level decomposition on the body temperature time series. The first decomposition produces a low-frequency approximation subsequence and a high-frequency detail subsequence. The second decomposition further processes the low-frequency subsequence, generating new approximation and detail subsequences. This process is repeated, ultimately resulting in multiple subsequences, each corresponding to a different time scale. The Haar wavelet can be used as the wavelet basis function, which is simple and efficient and suitable for capturing sudden changes, while the Daubechies wavelet is more suitable for analyzing smooth signals.

[0035] In practice, the decomposed body temperature waveform subsequences can help the system identify distinct patterns of variation. For example, if a driver experiences abnormal temperature fluctuations due to fatigue while driving, these fluctuations may be particularly pronounced at a specific scale (such as the minute level). The advantage of multi-scale decomposition is that it not only provides rich signal information but also effectively suppresses noise, providing a reliable data foundation for subsequent feature extraction.

[0036] like Figure 3 As shown in Figure 2, the stability feature extraction module analyzes the stability of each temperature waveform subsequence obtained from multi-scale decomposition and extracts corresponding features. These features are ultimately integrated into a stability feature vector, which is used to describe the regularity and stability of temperature changes.

[0037] The extracted features are diverse and are designed to quantify the characteristics of the body temperature waveform from different perspectives. For example, the standard deviation of the fluctuation amplitude can reflect the degree of dispersion of the body temperature values. If the standard deviation is large, it means that the body temperature fluctuates violently, which may indicate that the user's state is unstable. The density of local extreme values ​​measures the frequency of body temperature changes by counting the number of peaks and valleys in the subsequence. The mean of the waveform curvature describes the smoothness of the waveform by calculating the second-order derivative. The smaller the curvature, the smoother the waveform. In addition, the distribution variance of the peak-to-valley distance focuses on whether the time interval between peaks and valleys is regular, while the local coefficient of variation calculates the degree of variation within the sliding window of the subsequence to capture local instability.

[0038] During implementation, the system computes one or more of the above features for each time-scale subsequence. For example, for minute-scale subsequences, the standard deviation and local extreme value density might be extracted; for hour-scale subsequences, the mean waveform curvature might be added. All these features are combined into a multidimensional vector, such as a six- or eight-dimensional stability feature vector. To ensure comparability, the features are typically normalized to remove dimensionality differences.

[0039] The advantage of this feature extraction method is that it not only focuses on the absolute value of body temperature, but also deeply analyzes the dynamic patterns of body temperature changes. Using these features, the system can more accurately determine the user's physiological state and provide a scientific basis for fatigue risk assessment.

[0040] like Figure 4 As shown in Figure 2, the pedigree clustering module uses stability feature vectors to perform cluster analysis, classifying similar body temperature stability patterns into multiple pedigree categories and constructing a stability pedigree database. These pedigree categories represent different body temperature change patterns and can be used for subsequent real-time monitoring and fatigue assessment.

[0041] Cluster analysis can use a variety of algorithms, such as hierarchical clustering or density clustering. Taking hierarchical clustering as an example, the system first calculates the distance between all feature vectors (such as Euclidean distance), and then gradually merges similar vectors in a bottom-up manner, eventually forming a cluster tree. Depending on actual needs, the cluster tree can be intercepted by setting the number of clusters or distance threshold to obtain several pedigree categories. In contrast, density clustering (such as DBSCAN) is more suitable for processing noisy data and can automatically identify cluster shapes.

[0042] The generated pedigree categories also need to be linked to fatigue risk. To this end, the pedigree clustering module has a built-in pedigree category labeling unit. This unit uses historical sample data, such as pre-collected body temperature time series with fatigue status labels, to analyze the correlation between each pedigree category and fatigue status. Based on the analysis results, each pedigree category is labeled as high risk, medium risk, or low risk. For example, if the feature vectors in a pedigree category mostly correspond to the user's body temperature pattern when fatigued, it is labeled as high risk.

[0043] All pedigree categories and their risk markers are stored in a stability pedigree database. This database is a dynamic structure that can be updated as new data is added, ensuring that the system can adapt to the needs of different users and scenarios.

[0044] The real-time pedigree matching module is a key component of the system's real-time monitoring. It collects the user's body temperature data at fixed intervals (e.g., every 5 minutes), calculates the corresponding stability feature vector, and matches it with the pedigree categories in the stability pedigree database to determine the user's current body temperature stability.

[0045] The specific process is as follows: the system first obtains the most recent body temperature time series and processes it with the multi-scale decomposition module and the stability feature extraction module to generate the current feature vector. This vector is then compared with the lineage categories in the database using either the nearest neighbor or K-nearest neighbor algorithm. Using the K-nearest neighbor algorithm as an example, the system finds the K lineage categories closest to the current vector in the database and, based on the category distribution of these neighbors, determines the lineage to which the current vector belongs.

[0046] The matching results are recorded and passed to the next module for analysis. To ensure real-time performance, the matching algorithm must run efficiently, typically completing calculations in milliseconds. Furthermore, the system must be robust to minor perturbations in the data to avoid misjudgments caused by noise.

[0047] The fatigue risk assessment module analyzes the user's pedigree category based on real-time pedigree matching results to determine whether it has shifted from low-risk to high-risk, and generates fatigue warning information when necessary. This module is the core of the system's warning output.

[0048] To detect lineage migration, the module has a built-in lineage migration detection unit. This unit continuously tracks changes in lineage categories over multiple time intervals, identifying potential risk trends. For example, if the system detects a change from low risk to medium risk and then to high risk at three consecutive time points, it will determine that the user may be experiencing fatigue. The specific number of migrations and time windows can be adjusted based on the application scenario.

[0049] Once fatigue risk is detected, the system generates a warning message, including the fatigue risk level (e.g., high risk) and intervention recommendations. These recommendations might include reminding the user to take a break, reduce exercise intensity, or slow down driving. This information can be conveyed to the user through audio prompts, screen displays, or vibration feedback, ensuring timely attention.

[0050] To adapt the system to individual differences and long-term changes in users, the system also includes a pedigree database update module. This module regularly collects users' body temperature data, re-decomposes, extracts features, and clusters them, and updates the pedigree categories in the stable pedigree database.

[0051] For example, the system might run an update process once a month, using data collected that month to adjust the center of the spectrum category and risk markers. This adaptive mechanism ensures that the system always reflects the user's latest status and improves the accuracy of fatigue detection.

[0052] Consider the following examples to illustrate the present invention:

[0053] Imagine a long-distance truck driver wearing a smart bracelet with a built-in temperature sensor. This bracelet records the skin temperature of the wrist every 10 seconds, such as "36.5°C", and then transmits this data to the vehicle's onboard system via Bluetooth in real time. The system analyzes these temperature changes to determine whether the driver is fatigued and issues reminders when necessary, helping to improve driving safety.

[0054] The wristband's sensor uses a thermistor, measuring the temperature every 10 seconds and generating a time-stamped data point, such as "2024-10-01 08:00:00, 36.5°C." This data is then linked together to form a series of temperature records, such as "08:00:10, 36.6°C" and "08:00:20, 36.4°C," which are continuously fed into the vehicle's system, laying the foundation for subsequent analysis.

[0055] To understand the short-term fluctuations and long-term trends in body temperature, the system splits this data into different levels of change. Wavelet decomposition methods can be used, such as the simple and efficient Haar wavelet basis, which can effectively capture sudden changes in body temperature signals. For example, taking the body temperature data of the past hour, there are a total of 360 points. The system uses a set of filters to divide the data into low-frequency and high-frequency parts. The low-frequency part reflects the slow change trend within a few hours, while the high-frequency part can show the rapid fluctuation within a few minutes. After several such splits, several subsequences of different time scales can be obtained. For example, the minute-level subsequence may show abnormal fluctuations in body temperature when the driver is tired, while the hour-level subsequence is more suitable for observing the overall trend.

[0056] After decomposition, the system extracts key information from these subsequences to characterize temperature fluctuations. For example, the system calculates the amplitude of temperature fluctuations within a subsequence—excessive fluctuations indicate instability. It also counts the number of highs and lows within a short period of time; frequent fluctuations may be cause for concern. Furthermore, it can draw conclusions by examining the smoothness of the waveform or the regularity of the intervals between highs and lows. This information, combined into a feature set, facilitates further processing by the system.

[0057] Next, the system groups these feature sets to identify similarities, using a hierarchical clustering approach. Initially, each feature set is grouped separately, then the two most similar groups are gradually merged, ultimately forming a tree-like structure. Based on the needs, the system selects three categories: low risk, medium risk, and high risk. This is based on historical data with fatigue markers, such as small and regular fluctuations classified as low risk, and large and unusually frequent fluctuations labeled as high risk. This gives each category a specific meaning.

[0058] During actual monitoring, the system takes the last hour's temperature data every five minutes, generates the current feature set, and then compares it with the categories in the database. During the comparison, the system searches for the five most similar categories. If most of them are low-risk, then everything is fine; if most of them are high-risk, caution may be warranted. The system also monitors trends in category changes. For example, if a driver experiences a change from low risk to medium risk and then to high risk several times in a row, this may indicate increasing fatigue. If a high risk or negative trend is detected, the system will immediately alert the driver, for example, by saying, "You may be a little tired. I suggest you take a break."

[0059] To maintain accuracy, the system updates its database once a month, using newly collected data to re-decompose, extract features, and group them to ensure it can adapt to the driver's individual body temperature changes.

[0060] Through this entire process, from body temperature collection to decomposition analysis, and then to comparison and early warning, the system can monitor the driver's status in real time, detect fatigue risks in a timely manner, and ensure driving safety.

[0061] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A fatigue detection and early warning system, characterized in that: include: The body temperature collection module is used to continuously collect the target user's body temperature time series data when exercising or driving; A multi-scale decomposition module is used to perform multi-scale wavelet decomposition on the body temperature time series data to obtain body temperature waveform subsequences at different time scales; a stability feature extraction module, configured to extract the waveform stability feature corresponding to each time scale based on the body temperature waveform subsequence to form a stability feature vector; a pedigree clustering module, configured to perform cluster analysis based on the stability feature vector to form multiple body temperature stability pedigrees and construct a stability pedigree database; A real-time pedigree matching module is used to calculate the stability feature vector of the current user in real time and match it to the corresponding pedigree category in the stability pedigree database; The fatigue risk judgment module is used to judge whether the stability spectrum category of the target user has migrated from a low-risk spectrum to a high-risk spectrum, and generate fatigue warning information when migration is detected.

2. The fatigue detection and early warning system according to claim 1, characterized in that: The body temperature time series data collected by the body temperature collection module includes continuous skin temperature data or core body temperature data, and each data point is timestamped.

3. The fatigue detection and early warning system according to claim 1, characterized in that: The multi-scale decomposition module uses a discrete wavelet transform algorithm or a continuous wavelet transform algorithm to decompose the body temperature time series data into a plurality of body temperature waveform subsequences of different time scales.

4. The fatigue detection and early warning system according to claim 1, characterized in that: The waveform stability features extracted by the stability feature extraction module include at least one or more of the standard deviation of fluctuation amplitude, density of local extreme values, mean value of waveform curvature, distribution variance of peak-to-valley distance, and local coefficient of variation.

5. The fatigue detection and early warning system according to claim 1, characterized in that: The pedigree clustering module uses a hierarchical clustering algorithm or a density clustering algorithm to cluster the stability feature vectors to form a plurality of stability pedigree categories with obvious distinctions, and stores each stability pedigree category in the stability pedigree database.

6. The fatigue detection and early warning system according to claim 5, characterized in that: The pedigree clustering module further includes a pedigree category marking unit, which performs fatigue risk marking on the stability pedigree category according to historical sample data, and the fatigue risk marking includes at least high risk, medium risk and low risk.

7. The fatigue detection and warning system according to claim 1, characterized in that: During the real-time monitoring process, the real-time pedigree matching module periodically calculates the stability feature vector of the current user at set time intervals, and matches the stability feature vector of the current user to a specific pedigree category in the stability pedigree database in real time based on the nearest neighbor algorithm or the K-nearest neighbor algorithm.

8. The fatigue detection and early warning system according to claim 7, characterized in that: The fatigue risk judgment module further includes a pedigree migration detection unit, which is used to continuously monitor the pedigree category matching results within multiple time intervals. When the pedigree category continuously migrates from a low-risk pedigree to a high-risk pedigree within a predetermined number of times, it is determined to be a fatigue risk state.

9. The fatigue detection and early warning system according to claim 1, characterized in that: The fatigue warning information generated by the fatigue risk judgment module includes fatigue risk level information and corresponding intervention suggestion information, and the intervention suggestion information includes suggestions to remind the user to take a rest, adjust exercise intensity or reduce driving speed.

10. The fatigue detection and early warning system according to claim 1, characterized in that: The system further includes a pedigree database update module, which is used to regularly collect body temperature time series data of the target user and update the stability pedigree category in the stability pedigree database to achieve adaptive adjustment of the system to the target user.

Images