Multifunctional system based on single accelerometer

Abstract

The invention discloses a multifunctional system based on a single accelerometer, which is applied to wearable equipment and comprises an accelerometer module, a processor module and a control module, the data acquisition and preprocessing module is used for carrying out vector norm calculation and data standardization processing on the acquired three-axis acceleration data and extracting related features according to different functions; the threshold value detection module is used for carrying out threshold value judgment according to the acceleration change and angle characteristics of the arm movement and filtering out invalid actions; the LDA classification module is used for classifying the data subjected to preprocessing and threshold value testing through a linear discriminant analysis model and identifying different arm actions; and the function module has the functions of turning on and turning off the screen when the user lifts the wrist, counting steps, identifying a motion mode and evaluating sleep, and performs corresponding response control according to a classification result. Through a single accelerometer, multiple functions such as screen turn-on and turn-off by lifting the wrist, step counting, motion mode recognition and sleep evaluation are accurately realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of wearable devices and relates to a multifunctional system based on a single accelerometer, and more specifically, to a system for step counting, real-time screen lighting and screen turning off by wrist raising, motion pattern recognition, and sleep assessment. Background Art

[0002] With the rapid development of science and technology, wearable device technology has gradually entered a more efficient and intelligent stage. Smart watches and bracelets are typical representatives of wearable devices, which have been widely used in daily life due to their portability and versatility. These devices usually rely on multi-sensor systems, such as accelerometers, gyroscopes, and magnetometers, to achieve gesture recognition, motion pattern analysis, health monitoring and other functions.

[0003] However, although multi-sensor collaboration can significantly improve the accuracy and comprehensiveness of data, this technical solution also brings many challenges. First, the complexity of the multi-sensor system increases the difficulty of hardware design and integration, which increases the R&D and production costs of the equipment. At the same time, the simultaneous operation of multiple sensors also significantly increases the power consumption of the device, which poses a severe test to the device's battery life and user experience.

[0004] In terms of function implementation, the traditional threshold method is widely used in the implementation of interactive functions such as raising the wrist to light up the screen. However, this method is prone to false triggering in complex movements or environments, resulting in poor user experience. In addition, existing pedometer algorithms and motion pattern recognition technologies also face the problem of insufficient accuracy. Pedometer algorithms often rely on simple acceleration changes or gait patterns, and calculate the number of steps through fixed thresholds or maximum value detection based on vibration amplitude. It is difficult to adapt to variable motion states (such as slow walking and fast running), especially in low-frequency movements. It is easy to miscount or miss the number of steps. Motion pattern recognition technology usually relies on simple feature analysis and fixed rule classification, such as distinguishing between walking, running, and stationary states by analyzing the mean and fluctuation amplitude of acceleration data. However, this type of rule classification method shows great limitations when facing diverse user behaviors, especially in scenarios where multiple motion modes appear alternately, which is easy to cause confusion.

[0005] Therefore, existing wearable device technologies have obvious deficiencies in terms of the complexity of multi-sensor systems, the limitations of threshold methods, the precision of step counting algorithms, and the accuracy of motion pattern recognition. These challenges limit the performance of the device and the user experience. Summary of the invention

[0006] The object of the present invention is to provide a multifunctional system based on a single accelerometer to solve the problems raised in the above background technology.

[0007] The objective of the present invention is achieved through the following technical solutions:

[0008] A multifunctional system based on a single accelerometer for use on a wearable device, comprising:

[0009] Accelerometer module: used to collect three-axis acceleration data of the user's arm;

[0010] Data acquisition and preprocessing module: performs vector norm calculation and data standardization on the collected three-axis acceleration data, and extracts relevant features according to different functions;

[0011] Threshold detection module: It makes threshold judgment based on the acceleration change and angle characteristics of the arm movement to filter out invalid movements;

[0012] LDA classification module: classifies the pre-processed and threshold-tested data through the linear discriminant analysis model to identify different arm movements;

[0013] Functional modules: including raising the wrist to light up and turn off the screen, step counting, exercise pattern recognition and sleep assessment functions, and corresponding response control based on the classification results.

[0014] As a further improvement of an embodiment of the present invention, the data acquisition and preprocessing module includes a data processing module, which collects acceleration data in three-axis directions and calculates the vector norm of the three-axis acceleration in each time window, and then normalizes the collected data to ensure that data of different dimensions can be compared within the same range, and then performs feature extraction on the normalized data in each time window.

[0015] As a further improvement of an embodiment of the present invention, the threshold verification module includes a data window for time threshold verification, and the data after feature extraction is filtered through the data window to obtain valid data. The acceleration change of the Z axis and the inclination change of the Y axis are detected based on the valid data to achieve a preliminary judgment on the lifting and lowering actions of the arm.

[0016] As a further improvement of an embodiment of the present invention, the LDA classification module is based on the multidimensional features in the acceleration data, performs classification training through a linear discriminant analysis model according to the feature vectors of different actions, and generates a classification model for real-time prediction of the user's arm movements.

[0017] As a further improvement of an embodiment of the present invention, the function module determines whether the current arm action is to raise the wrist or to lower the wrist through the LDA model to determine whether to turn on the screen or turn off the screen, which specifically includes the following steps:

[0018] Step 1.1, the smart watch uses a three-axis accelerometer with a frequency of 100Hz;

[0019] Step 1.2, the accelerometer collects the three-axis acceleration data of the arm 100 times per second, and transmits it to the data processing module in real time for vector norm calculation to obtain the vector norm of the acceleration, and obtains the standardized eigenvalue through the standardization processing formula. The standardization processing formula is as follows:

[0020]

[0021] Among them, feature i is the i-th original feature, MEAN i is the mean of feature i, SCALE i is the standard deviation of feature i, scaled_feature i is the standardized eigenvalue;

[0022] Step 1.3: In each time window, extract the angle and acceleration features of the arm; when the Z-axis acceleration is negative and the Y-axis inclination is less than 0.6, it is judged as a wrist-raising action;

[0023] Step 1.4: Input the features extracted in step 1.3 into the LDA classification module, and determine whether it is a wrist-raising action to turn on the screen or a wrist-lowering action to turn off the screen based on the LDA classification results.

[0024] As a further improvement of an embodiment of the present invention, the function module identifies whether the current action is a gait action through an LDA model, specifically comprising the following steps:

[0025] Step 2.1, the smart watch uses a three-axis accelerometer with a frequency of 100 Hz;

[0026] Step 2.2, collect three-axis acceleration data and transmit it to the data processing module in real time, analyze the acceleration signal through fast Fourier transform, and extract the frequency characteristics;

[0027] Step 2.3, input the frequency features extracted in step 2.2 into the LDA classification module to identify whether the current action is a gait;

[0028] Step 2.4: After identifying the gait movement, the peak value of each step is identified through the local maximum detection algorithm. After removing the outliers, the frequency domain analysis is performed for anti-interference processing, and the moving average filter is used to distinguish between running and walking.

[0029] As a further improvement of an embodiment of the present invention, the functional module determines the motion mode classification steps as follows:

[0030] Step 3.1, the smart watch uses a three-axis accelerometer with a frequency of 100 Hz;

[0031] Step 3.2, collect triaxial acceleration data every three seconds, and perform feature extraction on the data in each time window;

[0032] Step 3.3: Normalize the features extracted in step 3.2. The formula is as follows:

[0033]

[0034] Where y represents the extracted features; normalization is performed to ensure that the feature values ​​are uniformly processed in different dimensions; the extracted features are then input into the LDA model to optimize the classification accuracy, and the formula is as follows:

[0035]

[0036] Where W is the coefficient matrix, X is the motion eigenvalue, and L is the intercept vector;

[0037] Step 3.4: By judging each element of each output value, values ​​greater than 0 are considered positively correlated, values ​​less than 0 are considered negatively correlated, and screening is performed in combination with the motion mode determination rules; based on 20 consecutive statistical outputs, the current motion state is judged to achieve mode switching and response.

[0038] As a further improvement of an embodiment of the present invention, the rules for determining the motion mode are as follows: when there are more than 2 positive correlations in the output or all negative correlation values ​​are less than a set threshold, the output will be excluded from recognition, but its statistics will still be accumulated.

[0039] As a further improvement of an embodiment of the present invention, the functional module performs the following steps for performing the sleep assessment function:

[0040] Step 4.1, the smart watch uses an accelerometer to monitor the user's nighttime movement, and the sampling frequency of the sensor is set to 50 Hz;

[0041] Step 4.2, collect the three-axis acceleration data of the user's arm every 30 seconds to monitor the movement and movement frequency of the arm;

[0042] Step 4.3: Calculate the average square wavelength and standard deviation of the low-frequency motion features extracted in step 3.2, and then determine whether the user is in a deep sleep, light sleep or wakefulness state by analyzing the motion amplitude, frequency and duration; the formula for the average square wavelength is as follows:

[0043]

[0044] And the standard deviation formula is as follows:

[0045]

[0046] Among them, yk represents acceleration data, n is the total number of data in the sliding window;

[0047] Step 4.4: Record and analyze the data obtained in step 4.3 to generate a sleep report.

[0048] As a further improvement of one embodiment of the present invention, the threshold judgment standard in step 4.3 is as follows: based on the extracted low-frequency motion features and the set threshold, square wavelength and s are comprehensively considered, different thresholds are set to obtain a score for the current motion intensity, and the light sleep, deep sleep or wakefulness state is judged based on the average value of the score.

[0049] The above technical solution has the following beneficial effects: The core innovation of this patent is to use a single accelerometer, combined with advanced signal processing and existing machine learning technologies such as linear discriminant analysis (LDA) models, to accurately realize multiple functions such as raising the wrist to light up and turn off the screen, step counting, motion pattern recognition and sleep assessment. This method not only simplifies the hardware structure and significantly reduces power consumption and system complexity, but also optimizes the algorithm so that the device can still maintain efficient and accurate motion recognition and state judgment under limited sensor input.

[0050] Compared with existing technologies that rely on multiple sensors, the outstanding advantage of this patent lies in the ability to efficiently process and classify limited input data. Through feature extraction and machine learning models, it ensures more reliable motion recognition and status monitoring in complex environments and under different user behavior patterns. This technology not only improves the user experience of the device, but also makes the device more suitable for portable and low-power application scenarios, greatly promoting the application of single-sensor devices in the field of smart wearables. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0052] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0053] Figure 1 This is a schematic diagram of the system architecture provided by the present invention. DETAILED DESCRIPTION

[0054] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0055] It should be noted that, unless otherwise specified, all technical and scientific terms used in this application have the same meanings as commonly understood by ordinary technicians in the technical field to which this application belongs.

[0056] In the present invention, unless otherwise specified, the directional words used, such as "up, down, top, bottom", usually refer to the directions shown in the drawings, or to the components themselves in the vertical, perpendicular or gravity directions; similarly, for ease of understanding and description, "inside and outside" refer to the inside and outside relative to the outline of each component itself, but the above-mentioned directional words are not used to limit the present invention.

[0057] A multifunctional system based on a single accelerometer realizes multiple functions such as screen lighting and screen off by wrist raising, step counting, motion pattern recognition and sleep assessment through a single accelerometer. The solution is designed for wearable devices and has the advantages of simple structure, low power consumption and high precision, and is suitable for long-term daily use. The technical solution of the present invention and its implementation steps are described in detail below.

[0058] The system architecture of the present invention is as follows Figure 1 As shown, it mainly includes the following modules:

[0059] Accelerometer module 10: collects the three-axis acceleration data of the user's arm as the main input source.

[0060] Data acquisition and preprocessing module 20: performs vector norm calculation and data standardization processing on the collected three-axis acceleration data, and extracts relevant features according to different functions.

[0061] Threshold detection module 30: performs threshold determination based on the acceleration change and angle characteristics of the arm movement to filter out invalid movements.

[0062] LDA classification module 40: The pre-processed and threshold-checked data is classified through a linear discriminant analysis model to identify different arm movements, such as wrist raising, wrist dropping, gait and movement patterns.

[0063] Functional module 50: includes functions of turning on and off the screen by raising the wrist, step counting, exercise pattern recognition and sleep assessment, and performs corresponding response control according to the classification results.

[0064] The data acquisition and preprocessing module 20 in this embodiment is introduced as follows:

[0065] The accelerometer is installed on the watch device and is responsible for collecting acceleration data in three axes. It samples 100 times per second and transmits the data to the upper processing module (data processing module).

[0066] Vector norm calculation: In each time window, the system calculates the vector norm of the three-axis acceleration to reflect the overall movement intensity of the arm.

[0067] Data standardization: The collected data is normalized to ensure that data of different dimensions can be compared within the same range.

[0068] Feature extraction: In each time window, nine features are extracted, including the mean, variance, root mean square value, peak value, direction change, spectrum energy, wavelength, waveform coefficient, and average absolute value of acceleration for subsequent LDA classification.

[0069] The threshold value checking module 30 in this embodiment is introduced as follows:

[0070] Angle and acceleration threshold judgment: By detecting the acceleration change of the Z axis and the inclination change of the Y axis, the system makes a preliminary judgment on the arm raising and lowering action. For example, when the Z axis acceleration is less than 0 and the absolute value of the Y axis inclination is less than 0.6, it is judged as a wrist raising action.

[0071] Time threshold check: To prevent the screen from turning on and off frequently, the system sets a time threshold. There are at least 15 data windows (about 1.5 seconds) between each action. The corresponding operation will only be triggered if there is continuous valid data.

[0072] The LDA classification module 40 in this embodiment is introduced as follows:

[0073] LDA model training: The system uses the linear discriminant analysis (LDA) model to perform classification training based on the multi-dimensional features in the acceleration data to generate a classification model for real-time prediction of the user's arm movements. The model's training data covers different arm movements, such as raising the wrist, lowering the wrist, and gait.

[0074] Real-time classification: During system operation, the LDA model classifies the data in each time window and classifies it into corresponding action categories, such as wrist raising, gait, or stillness, based on the feature vectors of different actions.

[0075] The linear discriminant analysis (English: Linear discriminant analysis, abbreviation: LDA) in this embodiment is a widely used data analysis method, which is often used for feature dimension reduction and classification problems.

[0076] The functional module 50 in this embodiment is described as follows:

[0077] 1. Raise your wrist to turn on the screen and turn off the screen

[0078] Screen light control: When the LDA model determines that the current arm movement is a wrist-raising movement, the system confirms the validity of the movement through a threshold test and triggers the screen to light up.

[0079] Screen off control: When the wrist drop action is detected (LDA classification result is wrist drop category) and passes the threshold test, the system will trigger the screen to turn off. By setting the time threshold, frequent screen on and screen off operations can be prevented to save power.

[0080] 2. Pedometer function

[0081] Gait signal detection: The system identifies gait signals by analyzing the frequency domain of acceleration signals. FFT (Fast Fourier Transform) is used to extract gait features such as step frequency and amplitude.

[0082] LDA gait classification: The system inputs gait data into the LDA model and classifies the actions into gait or other actions, such as running, walking, etc., through feature analysis.

[0083] Step Counting: Combined with the local maximum detection algorithm, the system calculates the number of steps in real time by analyzing the change in stride length when recognizing gait. For non-gait signals (such as static or interfering movements), the system filters them out to avoid miscounting.

[0084] 3. Sports mode recognition function

[0085] Motion feature extraction: The system extracts features of different motion modes, such as walking, running, and stillness, through acceleration data. The extracted features include acceleration mean, standard deviation, direction change, etc.

[0086] LDA classifier: By using the LDA model to classify the extracted features, the system can accurately identify different motion modes and perform corresponding processing based on the classification results, such as recording motion data or prompting the user's current motion status.

[0087] 4. Sleep assessment function

[0088] Low-frequency motion feature analysis: At night, the system determines the user's sleep state by analyzing the low-frequency motion features of the arm. Sleep states are divided into deep sleep, light sleep and wakefulness. Based on the smoothness and frequency changes of acceleration data, the system can distinguish these states.

[0089] Time threshold: In order to avoid misjudgment in a short period of time, the system analyzes the status of multiple consecutive data windows to ensure the smoothness of state changes and generate a detailed sleep report.

[0090] In summary, the system architecture of the present invention has the following functions:

[0091] Data acquisition: The accelerometer module collects three-axis acceleration data in real time.

[0092] Data preprocessing: The system calculates the vector norm and standardizes the data to extract features.

[0093] Threshold judgment and LDA classification: Through the LDA model and threshold test, the system recognizes the user's arm movements.

[0094] Function realization: According to the classification results, the system triggers the functions of raising the wrist to light up the screen, counting steps, exercise pattern recognition or sleep assessment.

[0095] The technical solution proposed in the present invention uses a single accelerometer in combination with signal processing and a linear discriminant analysis model (LDA model) to provide a smart watch system with integrated pedometer, screen lighting and screen off upon wrist raise, motion pattern recognition and sleep assessment functions.

[0096] Compared with the prior art, the present invention has significant technical progress in many aspects and brings the following beneficial effects:

[0097] 1. Reduce system complexity and power consumption, and improve device endurance

[0098] Existing technologies generally rely on multi-sensor systems, which increases hardware complexity and energy consumption, affecting the portability and user experience of the device. The present invention achieves multi-functional applications by using only a single accelerometer combined with optimized signal processing technology. Since multiple sensors are no longer required to work simultaneously, the power consumption of the system is significantly reduced, the device can maintain efficient operation for a long time, and the user's endurance experience is improved. In addition, the hardware structure of the system is more concise, which reduces the design and manufacturing costs, making the device have significant advantages in portability and performance.

[0099] 2. Improve the accuracy of the wrist-raising screen-on and screen-off functions to reduce false triggering

[0100] The existing wrist-raising screen-on and screen-off functions mostly rely on simple threshold methods, which are easily triggered by mistake in complex action scenes, causing the screen to frequently turn on and off. The present invention combines the LDA model with multi-dimensional feature extraction to accurately identify effective wrist-raising actions, and at the same time significantly reduces false triggering through time and angle threshold tests. Since the system can adapt to the changes in the actions of different users, users will not frequently consume batteries due to false triggering in daily use, further improving the device's efficacy and user experience.

[0101] 3. Significantly improve the accuracy of the pedometer function

[0102] The pedometer algorithm in the prior art is prone to missing or miscounting steps in different gaits (such as slow walking, fast running, etc.), especially in complex environments where the pedometer error is large. The present invention accurately identifies gait signals through LDA model classification, and combines the local maximum detection algorithm to maintain high-precision step calculation under variable motion states. Whether it is slow walking, fast walking or running, the present invention can ensure the accuracy of step recording through feature extraction and classification models, avoid missing and miscounting, and meet users' high requirements for motion data.

[0103] 4. Improve the reliability and adaptability of motion pattern recognition

[0104] Most existing motion pattern recognition systems rely on rule classification algorithms, which cannot effectively cope with complex motion scenes and are prone to confuse different motion patterns. The present invention can accurately distinguish different motion patterns, such as walking, running, and stillness, through multi-dimensional feature analysis based on the LDA model. Since the LDA model can automatically learn the characteristics of motion patterns from data, the system can adapt to diverse user behaviors, reduce errors and confusion in pattern recognition, and improve the overall recognition accuracy and stability of the system.

[0105] 5. Optimize the accuracy and power consumption of the sleep assessment function

[0106] Existing sleep assessment systems usually rely on multiple sensors or complex machine learning models. Although they are highly accurate, they also consume relatively large amounts of power. Especially when used for a long time, the battery consumption of the device becomes a major problem. The present invention analyzes the low-frequency motion characteristics of the accelerometer to determine the user's sleep state, such as deep sleep, light sleep, and wakefulness, without relying on complex machine learning models. This method not only reduces the power consumption of the system, but also ensures the accuracy of sleep assessment, enabling the device to provide reliable sleep monitoring reports under low power consumption conditions.

[0107] 6. Improve the cost-effectiveness of equipment

[0108] Since the present invention realizes multiple functions through a single accelerometer, the dependence on other sensors is reduced, the system hardware structure is simplified, and the design and manufacturing costs are significantly reduced. The overall cost-effectiveness of the equipment is greatly improved, and it has strong market competitiveness. In addition, the simplified design also shortens the production cycle of the equipment, further improving the economic benefits.

[0109] In summary, the present invention has greatly improved the accuracy of functions and user experience by optimizing hardware structure and algorithm design, while reducing system complexity and power consumption. This technical solution not only has significant advantages in wrist-raising screen lighting, step counting, motion pattern recognition and sleep assessment, but also provides a new direction for the development of future smart wearable devices.

[0110] The present invention provides a variety of smart watch implementations based on a single accelerometer. Through the organic cooperation of multiple algorithms, multiple functions such as step counting, motion pattern recognition, screen lighting when wrist raised, and screen-off sleep monitoring are realized. We not only designed highly adaptable algorithms in each functional module, but also combined anti-interference processing, feature extraction, LDA classification, and statistical analysis technologies to make the system have high reliability and accuracy in complex environments. The following examples describe in detail the specific implementation steps and conditions of these functional modules.

[0111] Example 1: Raise wrist to light up screen and turn off screen

[0112] Step 1.1: Hardware Preparation

[0113] Device: Use a smartwatch with a three-axis accelerometer.

[0114] Installation: The accelerometer is embedded in the wristwatch. The sampling frequency of the sensor is set to 100 Hz to ensure high-precision detection of arm movements.

[0115] Step 1.2: Data collection and processing

[0116] Data acquisition: The accelerometer collects the arm's three-axis acceleration data 100 times per second and transmits it to the data processing module in real time.

[0117] Vector norm calculation: The vector norm of acceleration is calculated by the following formula:

[0118]

[0119] Where v is a vector containing multiple components a x ,a y ,a z , square each component and then take the square root of the result to get the Euclidean norm. This vector norm is used to measure the amplitude of the overall arm movement.

[0120] Standardization: The standardization formula is used to scale the data to the same scale, usually zero mean and unit variance. After training with the training set, the mean MEAN and standard deviation SCALE arrays with the number of features as the scale are obtained. The formula is as follows:

[0121]

[0122] Among them, feature i is the i-th original feature, MEAN i is the mean of feature i, SCALE i is the standard deviation of feature i (it can also be the range of the feature or other scaling value), scaled_featurei is the standardized eigenvalue.

[0123] Step 1.3: Feature extraction and threshold determination

[0124] Feature extraction: In each time window, extract the angle and acceleration features of the arm, focusing on the change in Z-axis acceleration (consistent with the vertical direction).

[0125] Threshold judgment: When the Z-axis acceleration is negative and the Y-axis inclination is less than 0.6, it is initially judged as a wrist-raising action. By setting a time threshold (such as at least 1.5 seconds between each action), frequent triggering of the screen lighting can be avoided.

[0126] Step 1.4: LDA classification and control

[0127] LDA classification: The extracted features are input into the LDA classification module, and the classifier identifies whether it is a wrist raising or wrist dropping action.

[0128] Screen-on and screen-off control: Based on the LDA classification results, the system will trigger the screen-on function after confirming that wrist-raising is a valid action; if the classification result is wrist-dropping, the screen-off function will be triggered.

[0129] Example 2: Pedometer function

[0130] Step 2.1: Hardware Preparation

[0131] Device: A smartwatch with an accelerometer was used. The sampling frequency of the sensor was kept at 100 Hz.

[0132] Step 2.2: Data acquisition and signal processing

[0133] Data collection: The watch collects three-axis acceleration data and transmits it to the data processing module. The acceleration signal is analyzed through FFT (Fast Fourier Transform) to extract frequency features for gait recognition.

[0134] Step 2.3: Feature extraction and gait recognition

[0135] Feature extraction: Extract features such as frequency, amplitude, and stride length from gait signals as the basis for gait recognition.

[0136] LDA classification: The extracted features are input into the LDA classification module to identify whether the current gait action is walking or running.

[0137] Step 2.4: Step Count

[0138] Local maximum detection: After identifying the gait action, the system uses the local maximum detection algorithm to identify the peak value of each step. The system updates the step count in real time based on the detected step increments.

[0139] Anti-interference processing: After removing outliers, frequency domain analysis is performed, and a small number of irregular interference movements are screened through thresholds. Specific interference movements are filtered through three-axis acceleration feature extraction, and running and walking are distinguished through moving average filtering to ensure the accuracy of gait recognition.

[0140] Example 3: Motion pattern recognition function

[0141] Step 3.1: Hardware Preparation

[0142] Device: A smartwatch with an accelerometer was used, and the sampling frequency of the sensor was set to 100 Hz.

[0143] Step 3.2: Data collection and processing

[0144] Data collection: The watch continuously collects triaxial acceleration data and performs feature extraction on the data in each time window;

[0145] Motion mode recognition: Extract three-axis acceleration signals every 3s.

[0146] Step 3.3: Movement pattern classification

[0147] Feature extraction: Extract multi-dimensional features such as the arm's acceleration mean, direction change, standard deviation, etc., without filtering and normalization:

[0148]

[0149] Where y represents the various features extracted, and this process ensures that the feature values ​​are uniformly processed within different dimensions.

[0150] LDA classification: Input the extracted features into the LDA model through the formula:

[0151]

[0152] Where W is the coefficient matrix, X is the motion eigenvalue, and L is the intercept vector, which further optimizes the classification accuracy. Different from the traditional probability conversion, each output value is directly judged to perform rule-based classification.

[0153] Step 3.4: Positive and negative correlation judgment and mode switching

[0154] Adaptive judgment of positive and negative correlation: By judging each element of each output value, values ​​greater than 0 are considered positive correlations, values ​​less than 0 are considered negative correlations, and screening is performed in combination with specific judgment rules for motion modes. For example, when there are more than 2 positive correlations in the output or all negative correlation values ​​are less than the set threshold, the output will be excluded from recognition, but its statistics will still be accumulated.

[0155] Mode switching and response: The system determines the current motion state based on 20 consecutive statistical outputs. This judgment method based on cumulative statistics enhances the stability of recognition, reduces short-term misjudgments, and improves the accuracy of the system in long-term motion recognition. The system will record the change and perform corresponding data processing, such as motion data recording and prompting the user's current motion state.

[0156] Example 4: Sleep assessment function

[0157] Step 4.1: Hardware Preparation

[0158] Device: The user’s nighttime movement is monitored using an accelerometer, with the sampling frequency of the sensor set to 50 Hz.

[0159] Step 4.2: Data collection and processing

[0160] Data collection: At night, the watch continuously collects the three-axis acceleration data of the user's arm, mainly monitoring the arm's micro-movements and movement frequency. Sleep assessment: Extract the three-axis acceleration signal every 30 seconds.

[0161] Step 4.3: Feature extraction and sleep state judgment

[0162] Feature extraction: Extract low-frequency motion features of the arm, including the average square wavelength:

[0163]

[0164] And the standard deviation:

[0165]

[0166] Among them, y k Represents acceleration data, and n is the total number of data in the sliding window. By analyzing the movement amplitude, frequency, and duration, it is determined whether the user is in deep sleep, light sleep, or awake state.

[0167] ·Threshold judgment: Based on the extracted low-frequency motion features and the set threshold, the system judges the user's sleep stage. Combining squarewavelength and s, different thresholds are set to obtain the score of the current motion intensity. If the long-term score average is lower than threshold 1, it is judged that the user has fallen asleep, and falling asleep is judged as light sleep by default; after a specified period of time after light sleep, if the score average for a period of time is lower than threshold 2, it is judged to have entered deep sleep; after a period of time after deep sleep, if the score average for a period of time is greater than threshold 3, it is judged to have entered light sleep; in light or deep sleep state, if the score average for a period of time is greater than threshold 4, it is judged to have entered awakening.

[0168] Step 4.4: Sleep report generation

[0169] Data recording and analysis: The system records the user's sleep status throughout the night and generates a detailed sleep report, including deep sleep, light sleep and wake-up time. The report can be used to analyze the user's sleep quality and provide personalized health advice.

[0170] Summary of algorithm coordination and technical effects

[0171] The above four embodiments not only realize functions through their own independent algorithms, but also embody the technical advantages of the present invention through the collaborative work between algorithms:

[0172] 1. The consistent application of anti-interference processing: Whether it is raising the wrist to light up the screen, counting steps or motion pattern recognition, anti-interference processing (such as eliminating outliers, frequency domain analysis, and threshold judgment) is reflected in all functional modules to ensure stability in complex environments.

[0173] 2. Unified application of LDA classification module: LDA model is widely used in wrist-raising screen lighting, gait recognition and motion pattern recognition. Through unified feature extraction and classification strategy, the system architecture is simplified, and the recognition accuracy and response speed are improved.

[0174] 3. Effective coordination of multi-level algorithms: Multiple algorithms such as feature extraction, frequency domain analysis, moving average filtering, local maximum detection, etc. work together to ensure the system's high efficiency in data collection, processing, identification and control.

[0175] 4. Statistical-based misjudgment correction mechanism: The judgment method of multiple statistics and cumulative output (such as the 20-time statistical rule) effectively reduces short-term misjudgments and ensures the stability of the system during long-term use.

[0176] Through the organic coordination of these algorithms, the present invention not only improves the multi-functional realization capability of the smart watch, but also reduces power consumption and hardware complexity through a unified algorithm architecture, providing an ideal technical solution for multi-functional, low-power wearable devices.

[0177] The technical solution of the present invention realizes multiple functions through a single accelerometer, including step counting, screen lighting and screen off by wrist raising, motion pattern recognition and sleep assessment, and has the characteristics of simplified hardware design, low power consumption and high precision. Through the organic coordination of multiple algorithms, such as anti-interference processing, LDA classification, feature extraction and time threshold management, the system can operate stably in complex environments. The following are the key technical points of the present invention and the technical solutions that need to be protected:

[0178] 1. Technical solution to achieve multi-function based on single accelerometer

[0179] Key technical points:

[0180] 1. Use a single accelerometer to collect three-axis acceleration data, and provide multiple functional support (wrist-raising screen lighting, step counting, motion pattern recognition, and sleep assessment) through feature extraction and signal processing.

[0181] 2. Extract features from acceleration data through signal processing techniques (such as vector norm calculation, FFT frequency domain analysis, etc.).

[0182] 3. Combine the linear discriminant analysis (LDA) model to classify and recognize the multi-function action, and realize high-precision functions based on limited sensor input.

[0183] 4. Anti-interference algorithms (eliminating outliers, frequency domain analysis) and feature extraction algorithms are applicable to all functional modules, reducing the impact of noise and interference on system performance.

[0184] 2. Accurate recognition and control method of turning on and off the screen by raising the wrist

[0185] Key technical points:

[0186] 1. By extracting the Z-axis acceleration and Y-axis inclination features and combining them with threshold judgment, the wrist raising and wrist lowering movements can be preliminarily identified.

[0187] 2. Use the LDA model to accurately classify the wrist-raising action to ensure the accuracy of action recognition and reduce false triggering.

[0188] 3. Set the time threshold to avoid frequent switching between the screen on and off.

[0189] 4. Reduce misjudgment and enhance system stability by continuously counting classification results (such as 20 cumulative statistics).

[0190] 3. High-precision step counting algorithm based on LDA model

[0191] Key technical points:

[0192] 1. Collect gait signals through accelerometers and use FFT frequency domain analysis to extract gait features (such as step frequency, amplitude, etc.).

[0193] 2. Use the LDA model to classify gait signals and combine it with the local maximum detection algorithm to calculate the number of steps to reduce the impact of interference signals on step calculation.

[0194] 3. Real-time gait classification and error correction to ensure the accuracy of step counting in various gaits (slow walking, fast walking, running).

[0195] 4. The moving average filter distinguishes between walking and running, and is combined with the local maximum detection algorithm to improve the accuracy of step count calculation.

[0196] 4. Motion pattern recognition method

[0197] Key technical points:

[0198] 1. Extract multidimensional features (such as mean, standard deviation, direction change, etc.) from the accelerometer, and combine the LDA model to classify motion patterns and distinguish between walking, running, and stationary motion states.

[0199] 2. Dynamically adapt to the motion behavior patterns of different users to ensure accurate identification of different motion states in complex motion scenes.

[0200] 3. Instead of using the traditional probability conversion step, each element of each output is directly judged, and positive correlation and negative correlation are classified according to specific rules. Compared with probability conversion, this method reduces the computational complexity and improves the real-time response speed and recognition accuracy.

[0201] 4. Based on the LDA model, motion classification and positive and negative correlation judgment are carried out, and multiple statistical outputs are continuously performed to reduce misjudgments in the short term and improve the stability of recognition.

[0202] 5. Low-power sleep assessment function

[0203] Key technical points:

[0204] 1. Extract the micro-movement characteristics of the arm through low-frequency acceleration signals to determine the user's sleep state (deep sleep, light sleep, wakefulness).

[0205] 2. It does not rely on complex machine learning algorithms. By setting thresholds and time windows, it can accurately determine different sleep stages and generate sleep reports.

[0206] 3. Low-power algorithms ensure long-term continuous monitoring, enabling efficient sleep assessment and health advice generation.

[0207] 6. Time threshold and anti-false trigger mechanism

[0208] Key technical points:

[0209] 1. By setting time thresholds, frequent triggering of various functions (such as raising the wrist to light up the screen, turning off the screen, and step counting) can be avoided to ensure system stability.

[0210] 2. Combine action classification with time window management to prevent the system from over-responding to the user's short-term or invalid actions and reduce misoperations.

[0211] 3. The combination of time threshold and action classification ensures that false triggering is reduced while improving system response speed and stability.

[0212] 7. Integration and implementation of the overall system

[0213] Key technical points:

[0214] 1. The overall system integrates a single accelerometer, a multifunctional signal processing module, an LDA classification model, and control logic to form a multifunctional wearable device with a simplified hardware structure.

[0215] 2. The system architecture design combines the collaborative work of multiple functions, reduces multi-sensor dependence, and achieves a balance between function and performance.

[0216] 3. By sharing accelerometer data, efficient use and flexible switching of data between different functional modules can be achieved, thereby improving the overall performance of the system.

[0217] in conclusion

[0218] The key technical points and protection points of the present invention are mainly concentrated on using a single accelerometer to realize multifunctional applications such as step counting, screen on and off when wrist is raised, motion pattern recognition and sleep assessment through signal processing and LDA model. By simplifying the hardware structure, reducing power consumption, introducing anti-interference processing and feature extraction strategies, and combining multi-dimensional feature extraction with machine learning classification models, a high-precision, low-power, and portable smart watch system is provided.

[0219] Obviously, the above-described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0220] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0221] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0222] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multifunctional system based on a single accelerometer for use on a wearable device, characterized in that: include: Accelerometer module: used to collect three-axis acceleration data of the user's arm; Data acquisition and preprocessing module: performs vector norm calculation and data standardization on the collected three-axis acceleration data, and extracts relevant features according to different functions; Threshold detection module: It makes threshold judgment based on the acceleration change and angle characteristics of the arm movement to filter out invalid movements; LDA classification module: classifies the pre-processed and threshold-tested data through the linear discriminant analysis model to identify different arm movements; Functional modules: including raising the wrist to light up and turn off the screen, step counting, exercise pattern recognition and sleep assessment functions, and corresponding response control based on the classification results.

2. The multifunctional system according to claim 1, characterized in that: The data acquisition and preprocessing module includes a data processing module, which acquires acceleration data in three-axis directions and calculates the vector norm of the three-axis acceleration in each time window, and then normalizes the acquired data to ensure that data of different dimensions can be compared within the same range, and then performs feature extraction on the normalized data in each time window.

3. The multifunctional system according to claim 1, characterized in that: The threshold test module includes a data window for time threshold test. The data after feature extraction is filtered through the data window to obtain valid data. The acceleration change of the Z axis and the inclination change of the Y axis are detected based on the valid data to realize the preliminary judgment of the arm raising and lowering action.

4. The multifunctional system according to claim 1, characterized in that: The LDA classification module is based on the multi-dimensional features in the acceleration data and the feature vectors of different actions, and performs classification training through a linear discriminant analysis model to generate a classification model for real-time prediction of the user's arm movements.

5. The multifunctional system according to claim 1, characterized in that: The function module determines whether the current arm action is to raise the wrist or to lower the wrist through the LDA model to determine whether the screen is on or off. Specifically, the steps include: Step 1.1, the smart watch uses a three-axis accelerometer with a frequency of 100Hz; Step 1.2, the accelerometer collects the three-axis acceleration data of the arm 100 times per second, and transmits it to the data processing module in real time for vector norm calculation to obtain the vector norm of the acceleration, and obtains the standardized eigenvalue through the standardization processing formula. The standardization processing formula is as follows: Among them, feature i is the i-th original feature, MEAN i is the mean of feature i, SCALE i is the standard deviation of feature i, scaled_feature i is the standardized eigenvalue; Step 1.3: In each time window, extract the angle and acceleration features of the arm; when the Z-axis acceleration is negative and the Y-axis inclination is less than 0.6, it is judged as a wrist-raising action; Step 1.4: Input the features extracted in step 1.3 into the LDA classification module, and determine whether it is a wrist-raising action to turn on the screen or a wrist-lowering action to turn off the screen based on the LDA classification results.

6. The multifunctional system according to claim 1, characterized in that: The functional module identifies whether the current action is a gait action through the LDA model, specifically including the following steps: Step 2.1, the smart watch uses a three-axis accelerometer with a frequency of 100 Hz; Step 2.2, collect three-axis acceleration data and transmit it to the data processing module in real time, analyze the acceleration signal through fast Fourier transform, and extract the frequency characteristics; Step 2.3, input the frequency features extracted in step 2.2 into the LDA classification module to identify whether the current action is a gait; Step 2.4: After identifying the gait movement, the peak value of each step is identified through the local maximum detection algorithm. After removing the outliers, the frequency domain analysis is performed for anti-interference processing, and the moving average filter is used to distinguish between running and walking.

7. The multifunctional system according to claim 1, characterized in that: The functional module determines the motion mode classification steps as follows: Step 3.1, the smart watch uses a three-axis accelerometer with a frequency of 100 Hz; Step 3.2, collect triaxial acceleration data every three seconds, and perform feature extraction on the data in each time window; Step 3.3: Normalize the features extracted in step 3.

2. The formula is as follows: Where y represents the extracted features; normalization is performed to ensure that the feature values ​​are uniformly processed in different dimensions; the extracted features are then input into the LDA model to optimize the classification accuracy, and the formula is as follows: Where W is the coefficient matrix, X is the motion eigenvalue, and L is the intercept vector; Step 3.4, by judging each element of each output value, a value greater than 0 is considered as a positive correlation, and a value less than 0 is considered as a negative correlation, and screening is performed in combination with the judgment rules of the motion mode; The current motion state is judged based on 20 consecutive statistical outputs to achieve mode switching and response.

8. The multifunctional system according to claim 7, characterized in that: The rules for determining the motion mode are as follows: when there are more than 2 positive correlations in the output or all negative correlation values ​​are less than the set threshold, the output will be excluded and not identified, but its statistics will still be accumulated.

9. The multifunctional system according to claim 1, characterized in that: The functional module performs the following steps to perform sleep assessment: Step 4.1, the smart watch uses an accelerometer to monitor the user's nighttime movement, and the sampling frequency of the sensor is set to 50 Hz; Step 4.2, collect the three-axis acceleration data of the user's arm every 30 seconds to monitor the movement and movement frequency of the arm; Step 4.3: Calculate the average square wavelength and standard deviation of the low-frequency motion features extracted in step 3.2, and then determine whether the user is in a deep sleep, light sleep or wakefulness state by analyzing the motion amplitude, frequency and duration; the formula for the average square wavelength is as follows: And the standard deviation formula is as follows: Among them, y k represents acceleration data, n is the total number of data in the sliding window; Step 4.4: Record and analyze the data obtained in step 4.3 to generate a sleep report.

10. The multifunctional system according to claim 9, characterized in that: The threshold judgment standard in step 4.3 is as follows: based on the extracted low-frequency motion features and the set threshold, square wavelength and s are combined to set different thresholds to obtain a score for the current motion intensity, and the light sleep, deep sleep or wakefulness state is judged based on the average value of the score.

Images