Click detection method and device based on multi-axis sensor fusion

Abstract

This application discloses a click detection method and apparatus based on multi-axis sensor fusion, relating to the field of intelligent devices and human-computer interaction technology. The method includes: acquiring raw data streams collected by multi-axis sensors built into a target device, the raw data streams including acceleration data, angular velocity data, and pressure data; extracting feature data from the raw data streams and generating a standardized dataset based on the feature data; temporally aligning the acceleration data, angular velocity data, and pressure data in the standardized dataset to generate a multi-dimensional feature matrix; inputting the multi-dimensional feature matrix into a behavior detection model, and identifying click events in user operation behavior based on the behavior detection model; determining feature parameters of the click events, generating a feedback signal based on the feature parameters, and generating a target response strategy based on the feedback signal. This addresses the technical problems of insufficient data fusion accuracy, poor real-time performance, and weak adaptability to complex scenarios.

Description

Technical Field

[0001] This application relates to the field of intelligent devices and human-computer interaction technology, and in particular to a click detection method and device based on multi-axis sensor fusion. Background Technology

[0002] The rapid development of multi-axis sensor technology has made its application in click detection a research hotspot. By fusing data from multi-axis sensors, high-precision detection of user actions can be achieved, thereby improving the interactive experience and device response efficiency. However, current click detection methods still have limitations in data fusion accuracy, real-time performance, and adaptability to complex scenarios, which affect the accuracy and reliability of detection. Currently, commonly used techniques mainly rely on acquiring hand movement videos and extracting angular changes and angular velocity features of finger movements. This approach has limited detection accuracy in complex environments, resulting in low accuracy and failing to meet the diverse needs of practical applications.

[0003] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide a click detection method and device based on multi-axis sensor fusion, which aims to solve the technical problems of insufficient data fusion accuracy, poor real-time performance and weak adaptability to complex scenarios in the prior art.

[0005] To achieve the above objectives, this application provides a click detection method based on multi-axis sensor fusion, the method comprising: Acquire raw data streams collected by the multi-axis sensors built into the target device, the raw data streams including acceleration data, angular velocity data and pressure data; Extract the feature data from the original data stream, and generate a standardized dataset based on the feature data; The acceleration data, angular velocity data, and pressure data in the standardized dataset are time-aligned to generate a multidimensional feature matrix. The multidimensional feature matrix is ​​input into the behavior detection model, and click events in user operation behavior are identified based on the behavior detection model. Determine the characteristic parameters of the click event, generate a feedback signal based on the characteristic parameters, and generate a target response strategy based on the feedback signal.

[0006] In one embodiment, the step of extracting feature data from the original data stream and generating a standardized dataset based on the feature data includes: Identify the peak and trough points in the acceleration data, and mark the areas corresponding to the peak and trough points as potential click areas; The time periods in the angular velocity data where the rate of change exceeds a preset threshold are marked as potential click time periods; Extract the pressure abrupt change intervals from the pressure data within the potential click area and the potential click time period; The peak points of the acceleration data, the rate of change of the angular velocity data, and the pressure abrupt change range of the pressure data are normalized to generate a standardized dataset.

[0007] In one embodiment, the step of extracting feature data from the original data stream further includes: The acceleration data is subjected to zero-phase filtering based on a second-order Butterworth low-pass filter to obtain low-frequency acceleration data. The angular velocity data is forward filtered using a fourth-order Butterworth low-pass filter to obtain low-frequency angular velocity data. The pressure data is filtered using an FIR filter based on a Hamming window to obtain low-frequency pressure data. Frequency domain energy analysis is performed on the low-frequency acceleration data, the low-frequency angular velocity data, and the low-frequency pressure data to obtain filtered acceleration data, filtered angular velocity data, and filtered pressure data.

[0008] In one embodiment, the step of temporally aligning the acceleration data, angular velocity data, and pressure data in the standardized dataset to generate a multidimensional feature matrix includes: The sampling frequencies for the acceleration data, the angular velocity data, and the pressure data are determined respectively, and the timestamp deviations of the acceleration data, the angular velocity data, and the pressure data are calculated respectively based on the sampling frequencies; Using the timestamp deviation as a calibration benchmark, the time series of each data stream is adjusted to align the acceleration data, the angular velocity data, and the pressure data on the time axis. The aligned acceleration data, angular velocity data, and pressure data are mapped to different dimensions of the three-dimensional coordinate system to form initial feature vectors. The initial feature vectors are arranged in chronological order to generate a multidimensional feature matrix.

[0009] In one embodiment, the step of mapping the aligned acceleration data, angular velocity data, and pressure data to different dimensions of a three-dimensional coordinate system to form an initial feature vector includes: The triaxial components of the aligned acceleration data are mapped to the X-axis of the three-dimensional coordinate system after feature synthesis. The aligned angular velocity data is dynamically integrated and then mapped to the Y-axis of the three-dimensional coordinate system; The aligned pressure data is mapped to the Z-axis of the three-dimensional coordinate system; The mapped 3D coordinate points are arranged based on a time series to form an initial feature vector.

[0010] In one embodiment, the step of constructing a behavior detection model based on a multi-dimensional feature matrix and identifying click events in user operation behavior based on the behavior detection model includes: The multidimensional feature matrix is ​​input into the initial behavior detection model, the initial behavior detection model is transformed into a behavior detection model, and candidate features are extracted from the multidimensional feature matrix based on the behavior detection model. The confidence level of the click event is determined based on the peak acceleration, rate of change of angular velocity, and pressure mutation amount in the candidate features. Determine a confidence threshold, and mark the click events corresponding to candidate features whose confidence scores are higher than the confidence threshold as valid click events; Based on the timestamp and feature parameters of the valid click event, generate the recognition result of the click event.

[0011] In one embodiment, the steps of determining the feature parameters of the click event, generating a feedback signal based on the feature parameters, and generating a target response strategy based on the feedback signal include: Determine the event type of the click event, and determine the feature parameters of the click event based on the multidimensional feature matrix; First data for determining the feedback signal based on the event type of the click event, the first data being used to distinguish signal categories; The second data of the feedback signal is determined based on the characteristic parameters, and the second data is used to represent the parameters of the click event; A feedback signal is generated based on the first data and the second data, and a target response strategy is generated based on the feedback signal.

[0012] In one embodiment, before the step of inputting the multidimensional feature matrix into the initial behavior detection model, the method further includes: Obtain the model training dataset, which includes acceleration sample data, angular velocity sample data, pressure data, and behavior category labels for click events; The model training dataset is divided into a training set and a validation set. The training set is input into two one-dimensional convolutional layers for convolution processing to extract local spatiotemporal features. The training set is input into a two-layer LSTM model to obtain long-term dependent features; The long-term dependency features and the local spatiotemporal features are concatenated, and binary classification is performed based on the concatenated feature data to determine the predicted behavior category. The prediction error is determined based on the predicted behavior category and the behavior category label, and a correction parameter is generated based on the prediction error. A mapping equation is generated based on long-term dependency features, local spatiotemporal features, behavior category labels, and correction parameters; The predicted behavior category is validated based on the validation set to determine the equation error; The mapping equation is updated based on the error of the equation to generate an initial behavior detection model.

[0013] In one embodiment, before the step of temporally aligning the acceleration data, angular velocity data, and pressure data in the standardized dataset to generate a multidimensional feature matrix, the method further includes: Determine the attitude change vector based on the angular velocity data; The displacement acceleration is determined based on the time of change of the attitude change vector and the angular velocity data. An acceleration increment is generated based on the displacement acceleration; The acceleration data is corrected based on the acceleration increment.

[0014] Furthermore, to achieve the above objectives, this application also proposes a click detection device based on multi-axis sensor fusion, which includes: The data acquisition module is used to acquire the raw data stream collected by the multi-axis sensor built into the target device. The raw data stream includes acceleration data, angular velocity data and pressure data. The data processing module is used to extract feature data from the original data stream and generate a standardized dataset based on the feature data. The feature alignment module is used to perform temporal alignment of the acceleration data, angular velocity data and pressure data in the standardized dataset to generate a multidimensional feature matrix. The behavior recognition module is used to input the multidimensional feature matrix into the behavior detection model and identify click events in user operation behavior based on the behavior detection model; The feedback generation module is used to determine the feature parameters of the click event, generate a feedback signal based on the feature parameters, and generate a target response strategy based on the feedback signal.

[0015] Furthermore, to achieve the above objectives, this application also proposes a click detection device based on multi-axis sensor fusion. The click detection device based on multi-axis sensor fusion includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the click detection method based on multi-axis sensor fusion as described above.

[0016] In addition, to achieve the above objectives, the present invention also proposes a storage medium, which is a computer-readable storage medium, and stores a computer program on the storage medium. When the computer program is executed by a processor, it implements the steps of the click detection method based on multi-axis sensor fusion as described above.

[0017] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the click detection method based on multi-axis sensor fusion as described above.

[0018] This application provides a click detection method based on multi-axis sensor fusion. It acquires raw data streams from multi-axis sensors built into a target device, including acceleration, angular velocity, and pressure data. Feature data is extracted from the raw data stream, and a standardized dataset is generated based on these feature data. The acceleration, angular velocity, and pressure data in the standardized dataset are time-aligned to generate a multi-dimensional feature matrix. This multi-dimensional feature matrix is ​​input into a behavior detection model, which identifies click events in user actions. Feature parameters of the click events are determined, and a feedback signal is generated based on these parameters. Finally, a target response strategy is generated based on the feedback signal. This method addresses the technical problems of insufficient data fusion accuracy, poor real-time performance, and weak adaptability to complex scenarios. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating an embodiment of the click detection method based on multi-axis sensor fusion according to this application. Figure 2This is a schematic diagram of the click event determination process according to an embodiment of the click detection method based on multi-axis sensor fusion in this application; Figure 3 This is a schematic diagram illustrating the partitioning of an embodiment of the click detection method based on multi-axis sensor fusion according to this application; Figure 4 This is a schematic diagram of the module structure of the click detection device based on multi-axis sensor fusion according to an embodiment of this application; Figure 5 This is a schematic diagram of the device structure of the hardware operating environment involved in the click detection method based on multi-axis sensor fusion in the embodiments of this application.

[0022] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0024] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0025] The main solution of this application embodiment is as follows: Acquire raw data streams collected by a multi-axis sensor built into the target device, the raw data streams including acceleration data, angular velocity data, and pressure data; extract feature data from the raw data streams, and generate a standardized dataset based on the feature data; perform time-series alignment of the acceleration data, angular velocity data, and pressure data in the standardized dataset to generate a multi-dimensional feature matrix; input the multi-dimensional feature matrix into a behavior detection model, and identify click events in user operation behavior based on the behavior detection model; determine the feature parameters of the click events, generate feedback signals based on the feature parameters, and generate a target response strategy based on the feedback signals.

[0026] Currently, with the rapid development of multi-axis sensor technology, its application in click detection has gradually become a research hotspot. By fusing data from multi-axis sensors, high-precision detection of user actions can be achieved, thereby improving the interactive experience and device response efficiency. However, existing click detection methods still have limitations in terms of data fusion accuracy, real-time performance, and adaptability to complex scenarios. These shortcomings have a certain impact on the accuracy and reliability of detection. Currently, commonly used technical solutions mainly rely on acquiring hand movement videos and extracting the angle changes and angular velocity features of finger movements. This approach has limited detection accuracy in complex environments, resulting in low accuracy of detection results and making it difficult to meet the diverse needs of practical applications.

[0027] This application provides a solution to acquire raw data streams collected by multi-axis sensors built into a target device, including acceleration data, angular velocity data, and pressure data; extract feature data from the raw data stream and generate a standardized dataset based on the feature data; perform time-series alignment of the acceleration data, angular velocity data, and pressure data in the standardized dataset to generate a multi-dimensional feature matrix; input the multi-dimensional feature matrix into a behavior detection model to identify click events in user operations based on the behavior detection model; determine the feature parameters of the click events; generate feedback signals based on the feature parameters; and generate a target response strategy based on the feedback signals. This addresses the technical problems of insufficient data fusion accuracy, poor real-time performance, and weak adaptability to complex scenarios.

[0028] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, a click detection device based on multi-axis sensor fusion, etc. This embodiment does not specifically limit it. The following uses a click detection device based on multi-axis sensor fusion as an example to describe this embodiment and the following embodiments.

[0029] This application provides a click detection method based on multi-axis sensor fusion, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the click detection method based on multi-axis sensor fusion according to this application.

[0030] In this embodiment, the click detection method based on multi-axis sensor fusion includes steps S10~S40: Step S10: Obtain the raw data stream collected by the multi-axis sensor built into the target device. The raw data stream includes acceleration data, angular velocity data, and pressure data. It should be noted that the target device can be a smart device that can interact with people, such as a mobile phone or wearable device. In these devices, the built-in integrated sensor system includes multi-axis sensors that can read unprocessed low-level measurement data sequences, including acceleration data, angular velocity data, and pressure data.

[0031] It should be understood that acceleration data describes the linear rate of change of a device in three-dimensional space (X / Y / Z axes), such as the dynamic response under rapid movement or gravity; angular velocity data records the angular rate of rotation of the device around the three axes, reflecting attitude changes, such as tilting, flipping, etc.; pressure data is obtained by pressure sensors during human-computer interaction by measuring the click pressure.

[0032] Step S20: Extract feature data from the original data stream and generate a standardized dataset based on the feature data; It should be noted that extracting feature data from the raw data stream refers to extracting core indicators (such as mean, variance, spectral characteristics, etc.) that can characterize the device's motion mode from continuous raw sensor measurements through mathematical transformations and statistical methods. Subsequently, a standardized dataset is generated based on these feature data, which involves normalizing multidimensional features (eliminating dimensional differences), aligning data (unifying time series), and storing them in a structured manner to form a standardized data set that can be directly used by machine learning models.

[0033] In one feasible implementation, the step of extracting feature data from the original data stream further includes: The acceleration data is subjected to zero-phase filtering based on a second-order Butterworth low-pass filter to obtain low-frequency acceleration data. The angular velocity data is forward filtered using a fourth-order Butterworth low-pass filter to obtain low-frequency angular velocity data. The pressure data is filtered using an FIR filter based on a Hamming window to obtain low-frequency pressure data. Frequency domain energy analysis is performed on the low-frequency acceleration data, the low-frequency angular velocity data, and the low-frequency pressure data to obtain filtered acceleration data, filtered angular velocity data, and filtered pressure data.

[0034] In a practical implementation, the transfer function of a second-order Butterworth low-pass filter is:

[0035] in, For Laplace variables.

[0036] Zero-phase filtering of acceleration data is performed based on a second-order Butterworth low-pass filter, where zero-phase filtering is achieved by two filtering steps: forward and reverse.

[0037] First, forward filtering: ; Then, reverse the output: ; Filter again: ; Finally, in reverse: ; The output signal obtained after two filtering cycles, forward and reverse. With the original signal They are aligned in phase, achieving zero phase distortion.

[0038] When performing forward filtering on the angular velocity data based on a fourth-order Butterworth low-pass filter, the fourth-order Butterworth low-pass filter can be regarded as a cascade of two second-order Butterworth filters, and its transfer function is:

[0039] Forward filtering performs only one forward filtering step:

[0040] in, It is the input angular velocity data. This is the filtered output.

[0041] The pressure data is filtered using an FIR filter based on a Hamming window to obtain low-frequency pressure data. The impulse response of the FIR low-pass filter designed with a Hamming window is as follows:

[0042] in, The impulse response of an ideal low-pass filter has a cutoff frequency of . .

[0043]

[0044] exist hour, The Hamming window function is:

[0045] The filtered output is:

[0046] Frequency domain energy analysis is performed on the low-frequency acceleration data, low-frequency angular velocity data, and low-frequency pressure data to obtain filtered acceleration data, filtered angular velocity data, and filtered pressure data. Frequency domain energy analysis is then performed on the three filtered low-frequency signals (acceleration, angular velocity, and pressure) respectively. Windowing (of length L) is applied to reduce spectral leakage, followed by FFT:

[0047] in, It is a window function.

[0048] Then calculate the energy spectral density. :

[0049] in The sampling frequency is used, and the normalization factor is used to obtain the correct amplitude.

[0050] Integrating the energy spectral density over a specific frequency band yields the energy of that band.

[0051] in, Lower limit frequency corresponding to the target frequency band and upper limit frequency Index, satisfying:

[0052] After filtering, the energy characteristics of the three signals in the frequency domain can be obtained, along with the filtered acceleration data, filtered angular velocity data, and filtered pressure data.

[0053] In one feasible implementation, the step of extracting feature data from the original data stream and generating a standardized dataset based on the feature data includes: Identify the peak and trough points in the acceleration data, and mark the areas corresponding to the peak and trough points as potential click areas; The time periods in the angular velocity data where the rate of change exceeds a preset threshold are marked as potential click time periods; Extract the pressure abrupt change intervals from the pressure data within the potential click area and the potential click time period; The peak points of the acceleration data, the rate of change of the angular velocity data, and the pressure abrupt change range of the pressure data are normalized to generate a standardized dataset.

[0054] In the specific implementation, when determining acceleration data To ensure more accurate results when identifying peak and trough points, a sliding window approach can be used, combining first-order and second-order differencing methods. The first-order and second-order differencing methods are defined as follows: First-order difference: ; Second-order difference: ; The corresponding peak and trough points are as follows: Peak point: and ; Valley point: and ; Then, the area between each peak point and its adjacent valley point is marked as a potential click area. Specifically, a click action may manifest as a peak-valley pair, therefore, the area between each peak point and its nearest left and right valley points can be considered as a potential click area.

[0055] Based on angular velocity data In determining the potential click time period, the first-order difference can be used to determine the rate of change of angular velocity, i.e.:

[0056] Then set a threshold. ,exist At that time, it is considered that moment Within a potential click time period. Since clicks may continue for a period of time, we can merge areas that continuously exceed the threshold into a single time period.

[0057] After identifying the potential click region and potential click time period, determining the intersection of the potential click region and potential click time period allows us to identify the stress mutation interval of the potential click region within the potential click time period. Stress mutations can be detected using the first difference of the stress data.

[0058] Then set a threshold. ,exist At that time, it was assumed that a stress mutation had occurred, and the intervals where these mutation points were located were extracted.

[0059] The peak points of the acceleration data, the rate of change of the angular velocity data, and the pressure abrupt change range of the pressure data are normalized to generate a standardized dataset.

[0060] Step S30: Time-series alignment of the acceleration data, angular velocity data, and pressure data in the standardized dataset to generate a multidimensional feature matrix; It should be noted that each sensor data point has an original timestamp. Since the sensor sampling frequency may be different, all data needs to be resampled onto a unified time grid for time alignment.

[0061] It is understood that the step of temporally aligning the acceleration data, angular velocity data, and pressure data in the standardized dataset to generate a multidimensional feature matrix includes: The sampling frequencies for the acceleration data, angular velocity data, and pressure data are determined respectively. The timestamp deviations of the acceleration data, angular velocity data, and pressure data are calculated based on the sampling frequencies. Using the timestamp deviations as a calibration benchmark, the time series of each data stream is adjusted to align the acceleration data, angular velocity data, and pressure data on the time axis. The three-axis components of the aligned acceleration data are mapped to the X-axis of the three-dimensional coordinate system after feature synthesis. The aligned angular velocity data is mapped to the Y-axis of the three-dimensional coordinate system after dynamic integration. The aligned pressure data is mapped to the Z-axis of the three-dimensional coordinate system. The mapped three-dimensional coordinate points are arranged based on the time series to form an initial feature vector. The initial feature vectors are arranged in chronological order to generate a multi-dimensional feature matrix.

[0062] In the specific implementation, the sampling frequencies for the acceleration data, the angular velocity data, and the pressure data are determined respectively. The corresponding sampling interval can be expressed as: Acceleration data sampling interval: ; Angular data sampling interval: ; Pressure data sampling interval: ; Due to differences in sensor startup time and transmission delay, there is an initial timestamp discrepancy between the data streams. The acceleration data start timestamp is... The angular velocity data start timestamp is The pressure data start timestamp is The deviation of each data stream relative to the reference time is:

[0063] Using a reference timeline as a baseline, calibrate the timestamps of each sample in each data stream: The timestamp of the kth sample of the acceleration data: ; The timestamp of the kth sample of angular velocity data: ; The timestamp of the kth sample in the stress data: ; At this point, the timestamps of all data streams are aligned to the reference timeline. However, since the sampling rates may differ, the data needs to be resampled to a unified time grid. Let the unified sampling interval be... The unified time grid is then:

[0064] in, The minimum timestamp for all data streams. The maximum timestamp of all data streams. .

[0065] Interpolation methods are used to resample each data stream to a unified grid.

[0066] The aligned acceleration data is mapped to the X-axis of the three-dimensional coordinate system after feature synthesis of its three-axis components; the aligned angular velocity data is mapped to the Y-axis of the three-dimensional coordinate system after dynamic integration; and the aligned pressure data is mapped to the Z-axis of the three-dimensional coordinate system. Feature synthesis is performed on the three-axis components, and the synthesized acceleration amplitude is used as the X-axis feature. Dynamic integration is performed on the triaxial angular velocity data. Dynamic integration typically refers to integrating the magnitude of the angular velocity. First, the magnitude of the angular velocity is calculated: Then perform numerical integration:

[0067] The resampled pressure values ​​are directly mapped to the Z-axis. The mapped 3D coordinate points are then arranged based on a time series to form an initial feature vector. ,

[0068] Arrange the initial feature vectors in chronological order to obtain a matrix .

[0069]

[0070] In one feasible implementation, before the step of temporally aligning the acceleration data, angular velocity data, and pressure data in the standardized dataset to generate a multidimensional feature matrix, the method further includes: Determine the attitude change vector based on the angular velocity data; The displacement acceleration is determined based on the time of change of the attitude change vector and the angular velocity data. An acceleration increment is generated based on the displacement acceleration; The acceleration data is corrected based on the acceleration increment.

[0071] In the specific implementation, at the time interval Inside, the angular velocity is Then the attitude change vector can be approximated by integrating the angular velocity vector over that time interval, i.e.:

[0072] Displacement acceleration is determined based on the time of change of the attitude change vector and angular velocity data. , During the time interval Inside, the object's displacement acceleration from Change to Then the acceleration increment It can be represented as Due to factors such as sensor error, noise, and cumulative error during integration, directly measured acceleration data may contain deviations. By calculating and analyzing the acceleration increment, the original acceleration data can be corrected, thereby improving the accuracy and reliability of the data.

[0073] Step S40: Input the multidimensional feature matrix into the behavior detection model, and identify click events in user operation behavior based on the behavior detection model.

[0074] It should be noted that, referring to Figure 2 , Figure 2 This is a flowchart illustrating the click event determination process. A multi-dimensional feature matrix (a three-dimensional spatiotemporal sequence composed of composite acceleration data, cumulative rotation angle of angular velocity integral, and pressure data) is input into the behavior detection model. This model extracts local motion pattern features through a one-dimensional convolutional layer, models temporal dependencies through a bidirectional LSTM layer, and uses an attention mechanism to focus on keyframes. Finally, it outputs a click probability sequence through a Sigmoid activation function. When the fusion judgment conditions are met, the user's click event is determined.

[0075] It is understood that, prior to the step of inputting the multidimensional feature matrix into the initial behavior detection model, the following steps are also included: Obtain the model training dataset, which includes acceleration sample data, angular velocity sample data, pressure data, and behavior category labels for click events; The model training dataset is divided into a training set and a validation set. The training set is input into two one-dimensional convolutional layers for convolution processing to extract local spatiotemporal features. The training set is input into a two-layer LSTM model to obtain long-term dependent features; The long-term dependency features and the local spatiotemporal features are concatenated, and binary classification is performed based on the concatenated feature data to determine the predicted behavior category. The prediction error is determined based on the predicted behavior category and the behavior category label, and a correction parameter is generated based on the prediction error. A mapping equation is generated based on long-term dependency features, local spatiotemporal features, behavior category labels, and correction parameters; The predicted behavior category is validated based on the validation set to determine the equation error; The mapping equation is updated based on the error of the equation to generate an initial behavior detection model.

[0076] In the specific implementation, let the dataset be... ,in, Let be the multidimensional feature matrix of the i-th sample (containing acceleration, angular velocity, and pressure data). The data is labeled with behavior categories (0 for non-click, 1 for click). The training and validation sets are randomly divided in a 7:3 ratio. The training data is then processed through two one-dimensional convolutional layers to extract local spatiotemporal features.

[0077] First convolutional layer: ; in, , The kernel size is [size]. This represents the number of output channels.

[0078] Second convolutional layer: ; in, , The kernel size is [size]. This represents the number of output channels.

[0079] At the same time, the original input The input is fed into a two-layer LSTM to extract long-term dependency features. The first LSTM layer can be represented as:

[0080]

[0081]

[0082] The second LSTM layer: The input is the hidden state of the first layer, and it is also bidirectional. Take the hidden state at the last time step as the representation of the entire sequence: .

[0083] Long-term dependency features and local spatiotemporal features are concatenated to obtain concatenated feature data. Next, binary classification is performed to determine the predicted behavior category. At this point, the output of the fully connected layer is... Predicted probability The prediction error is calculated using cross-entropy loss, and corrected parameters are generated. The gradient is calculated through backpropagation, and the parameters are updated using the Adam optimizer to generate a mapping equation. The predicted behavior category is validated based on the validation set to determine the equation error. The mapping equation is then updated based on the equation error to generate an initial behavior detection model.

[0084] In one feasible implementation, the step of constructing a behavior detection model based on a multidimensional feature matrix and identifying click events in user operation behavior based on the behavior detection model includes: The multidimensional feature matrix is ​​input into the initial behavior detection model, the initial behavior detection model is transformed into a behavior detection model, and candidate features are extracted from the multidimensional feature matrix based on the behavior detection model. The confidence level of the click event is determined based on the peak acceleration, rate of change of angular velocity, and pressure mutation amount in the candidate features. Determine a confidence threshold, and mark the click events corresponding to candidate features whose confidence scores are higher than the confidence threshold as valid click events; Based on the timestamp and feature parameters of valid click events, generate click event recognition results. In the specific implementation, the initial behavior detection model needs to be transformed before deployment. Based on a multi-dimensional feature matrix, the initial behavior detection model is transformed into the final behavior detection model. The confidence level of the click event is determined based on the acceleration peak, angular velocity change rate, and pressure mutation amount in the candidate features. For each time step... Not only do you get the click probability Furthermore, the confidence level needs to be calculated by combining the peak acceleration, rate of change of angular velocity, and pressure mutation amount from the candidate features. A confidence threshold is then determined, and click events corresponding to candidate features with confidence levels higher than the threshold are marked as valid click events.

[0085] Step S50: Determine the feature parameters of the click event, generate a feedback signal based on the feature parameters, and generate a target response strategy based on the feedback signal.

[0086] In the specific implementation, the event type of the click event is determined, and the feature parameters of the click event are determined based on the multi-dimensional feature matrix. First data of the feedback signal is determined according to the event type of the click event; the first data is used to distinguish signal categories. Second data of the feedback signal is determined according to the feature parameters; the second data is used to represent the parameters of the click event. A feedback signal is generated based on the first data and the second data, and a target response strategy is generated based on the feedback signal. Specifically, the time parameter, position parameter, and pressure parameter in the click event feature parameters are determined. At this time, the first data of the feedback signal can be determined according to the time type of the click event; the first data is used to distinguish signal categories. Simultaneously, the second data is determined according to the feature parameters; the second data is used to represent the parameters of the click event. The click event can be classified to determine whether it is a valid click event and whether it is a human operation. See the specific classification diagram for reference. Figure 3A feedback signal is generated based on the first data and the second data, and a target response strategy is generated based on the feedback signal.

[0087] This embodiment provides a click detection method based on multi-axis sensor fusion. It acquires raw data streams from multi-axis sensors built into the target device, including acceleration, angular velocity, and pressure data. Feature data is extracted from the raw data stream, and a standardized dataset is generated based on these feature data. The acceleration, angular velocity, and pressure data in the standardized dataset are time-aligned to generate a multi-dimensional feature matrix. This multi-dimensional feature matrix is ​​input into a behavior detection model, which identifies click events in user actions. Feature parameters of the click events are determined, and a feedback signal is generated based on these parameters. Finally, a target response strategy is generated based on the feedback signal. This method addresses the technical problems of insufficient data fusion accuracy, poor real-time performance, and weak adaptability to complex scenarios.

[0088] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the click detection method based on multi-axis sensor fusion in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0089] This application also provides a click detection device based on multi-axis sensor fusion, please refer to... Figure 4 The click detection device based on multi-axis sensor fusion includes: Data acquisition module 10 is used to acquire raw data streams collected by the multi-axis sensors built into the target device, the raw data streams including acceleration data, angular velocity data and pressure data; Data processing module 20 is used to extract feature data from the original data stream and generate a standardized dataset based on the feature data; The feature alignment module 30 is used to perform temporal alignment of the acceleration data, the angular velocity data and the pressure data in the standardized dataset to generate a multidimensional feature matrix. The behavior recognition module 40 is used to input the multidimensional feature matrix into the behavior detection model and identify click events in user operation behavior based on the behavior detection model; The feedback generation module 50 is used to determine the feature parameters of the click event, generate a feedback signal based on the feature parameters, and generate a target response strategy based on the feedback signal.

[0090] In one feasible implementation, the data processing module 20 is further configured to determine the peak points and valley points in the acceleration data, mark the regions corresponding to the peak points and valley points as potential click regions; mark the time periods in the angular velocity data where the rate of change exceeds a preset threshold as potential click time periods; extract the pressure mutation intervals of the pressure data in the potential click regions and potential click time periods; and normalize the peak points of the acceleration data, the rate of change of the angular velocity data, and the pressure mutation intervals of the pressure data to generate a standardized dataset.

[0091] In one feasible implementation, the data processing module 20 is further configured to perform zero-phase filtering on the acceleration data based on a second-order Butterworth low-pass filter to obtain low-frequency acceleration data; perform forward filtering on the angular velocity data based on a fourth-order Butterworth low-pass filter to obtain low-frequency angular velocity data; filter the pressure data based on a Hamming window FIR filter to obtain low-frequency pressure data; and perform frequency domain energy analysis on the low-frequency acceleration data, the low-frequency angular velocity data, and the low-frequency pressure data to obtain filtered acceleration data, filtered angular velocity data, and filtered pressure data.

[0092] In one feasible implementation, the feature alignment module 30 is further configured to: determine the sampling frequencies of the acceleration data, the angular velocity data, and the pressure data respectively; calculate the timestamp deviations of the acceleration data, the angular velocity data, and the pressure data respectively based on the sampling frequencies; adjust the time series of each data stream using the timestamp deviations as a calibration benchmark to align the acceleration data, the angular velocity data, and the pressure data on the time axis; map the aligned acceleration data, angular velocity data, and pressure data to different dimensions of the three-dimensional coordinate system respectively to form initial feature vectors; and arrange the initial feature vectors in chronological order to generate a multidimensional feature matrix.

[0093] In one feasible implementation, the feature alignment module 30 is further configured to map the three-axis components of the aligned acceleration data to the X-axis of the three-dimensional coordinate system after feature synthesis. The aligned angular velocity data is dynamically integrated and mapped to the Y-axis of the three-dimensional coordinate system; the aligned pressure data is mapped to the Z-axis of the three-dimensional coordinate system; the mapped three-dimensional coordinate points are arranged based on a time series to form an initial feature vector.

[0094] In one feasible implementation, the behavior recognition module 40 is further configured to input the multidimensional feature matrix into an initial behavior detection model, transform the initial behavior detection model into a behavior detection model, extract candidate features from the multidimensional feature matrix based on the behavior detection model, determine the confidence level of a click event based on the peak acceleration, rate of change of angular velocity, and pressure mutation amount in the candidate features, determine a confidence level threshold, mark the click events corresponding to the candidate features with confidence levels higher than the confidence level threshold as valid click events, and generate the recognition result of the click event based on the timestamp and feature parameters of the valid click event.

[0095] In one feasible implementation, the feedback generation module 50 is further configured to determine the event type of the click event, determine the feature parameters of the click event based on the multidimensional feature matrix, determine first data of the feedback signal according to the event type of the click event, the first data being used to distinguish signal categories, determine second data of the feedback signal according to the feature parameters, the second data being used to represent the parameters of the click event, generate a feedback signal based on the first data and the second data, and generate a target response strategy based on the feedback signal.

[0096] In one feasible implementation, the behavior recognition module 40 is further configured to acquire a model training dataset, which includes acceleration sample data, angular velocity sample data, pressure data, and behavior category labels for click events; divide the model training dataset into a training set and a validation set; input the training set into two one-dimensional convolutional layers for convolution processing to extract local spatiotemporal features; input the training set into a two-layer LSTM model to obtain long-term dependency features; concatenate the long-term dependency features and the local spatiotemporal features, and perform binary classification based on the concatenated feature data to determine the predicted behavior category; determine the prediction error based on the predicted behavior category and the behavior category label, and generate correction parameters based on the prediction error; generate a mapping equation based on the long-term dependency features, the local spatiotemporal features, the behavior category label, and the correction parameters; validate the predicted behavior category based on the validation set to determine the equation error; update the mapping equation based on the equation error to generate an initial behavior detection model.

[0097] In one feasible implementation, the feature alignment module 30 is further configured to determine an attitude change vector based on the angular velocity data; determine a displacement acceleration based on the time of change between the attitude change vector and the angular velocity data; generate an acceleration increment based on the displacement acceleration; and correct the acceleration data based on the acceleration increment.

[0098] The click detection device based on multi-axis sensor fusion provided in this application, employing the click detection method based on multi-axis sensor fusion in the above embodiments, can solve the technical problems of insufficient data fusion accuracy, poor real-time performance, and weak adaptability to complex scenarios. Compared with the prior art, the beneficial effects of the click detection device based on multi-axis sensor fusion provided in this application are the same as those of the click detection method based on multi-axis sensor fusion provided in the above embodiments, and other technical features in the click detection device based on multi-axis sensor fusion are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0099] This application provides a click detection device based on multi-axis sensor fusion. The click detection device based on multi-axis sensor fusion includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the click detection method based on multi-axis sensor fusion in the above embodiment 1.

[0100] The following is for reference. Figure 5 This document illustrates a schematic diagram of a click detection device based on multi-axis sensor fusion suitable for implementing embodiments of this application. The click detection device based on multi-axis sensor fusion in these embodiments may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 5 The click detection device based on multi-axis sensor fusion shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0101] like Figure 5As shown, the click detection device based on multi-axis sensor fusion may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in ROM (Read Only Memory) 1002 or a program loaded from storage device 1003 into RAM (Random Access Memory) 1004. RAM 1004 also stores various programs and data required for the operation of the click detection device based on multi-axis sensor fusion. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. Input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, LCDs (Liquid Crystal Displays), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the multi-axis sensor fusion-based click detection device to wirelessly or wiredly communicate with other devices to exchange data. Although a multi-axis sensor fusion-based click detection device with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0102] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0103] The click detection device based on multi-axis sensor fusion provided in this application, employing the click detection method based on multi-axis sensor fusion in the above embodiments, can solve the technical problems of click detection based on multi-axis sensor fusion. Compared with the prior art, the beneficial effects of the click detection device based on multi-axis sensor fusion provided in this application are the same as those of the click detection method based on multi-axis sensor fusion provided in the above embodiments, and other technical features in this click detection device based on multi-axis sensor fusion are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0104] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0105] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0106] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the click detection method based on multi-axis sensor fusion in the above embodiments.

[0107] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable Read Only Memory or Flash Memory), optical fibers, CD-ROM (CD-Read Only Memory), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0108] The aforementioned computer-readable storage medium may be included in a click detection device based on multi-axis sensor fusion; or it may exist independently and not assembled into a click detection device based on multi-axis sensor fusion.

[0109] The aforementioned computer-readable storage medium carries one or more programs that, when executed by a multi-axis sensor fusion-based click detection device, cause the multi-axis sensor fusion-based click detection device to: acquire raw data streams collected by the multi-axis sensors built into the target device, the raw data streams including acceleration data, angular velocity data, and pressure data; extract feature data from the raw data streams and generate a standardized dataset based on the feature data; perform time-series alignment of the acceleration data, angular velocity data, and pressure data in the standardized dataset to generate a multi-dimensional feature matrix; input the multi-dimensional feature matrix into a behavior detection model and identify click events in user operation behavior based on the behavior detection model; determine feature parameters of the click events, generate feedback signals based on the feature parameters, and generate a target response strategy based on the feedback signals.

[0110] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including LAN (Local Area Network) or WAN (Wide Area Network)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0111] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0112] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0113] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described click detection method based on multi-axis sensor fusion, thereby solving the technical problem of click detection based on multi-axis sensor fusion. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the click detection method based on multi-axis sensor fusion provided in the above embodiments, and will not be repeated here.

[0114] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the click detection method based on multi-axis sensor fusion as described above.

[0115] The computer program product provided in this application can solve the technical problem of click detection based on multi-axis sensor fusion. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the click detection method based on multi-axis sensor fusion provided in the above embodiments, and will not be repeated here.

[0116] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A click detection method based on multi-axis sensor fusion, characterized in that, The click detection method based on multi-axis sensor fusion includes: Acquire raw data streams collected by the multi-axis sensors built into the target device, the raw data streams including acceleration data, angular velocity data and pressure data; Extract the feature data from the original data stream, and generate a standardized dataset based on the feature data; The acceleration data, angular velocity data, and pressure data in the standardized dataset are time-aligned to generate a multidimensional feature matrix. The multidimensional feature matrix is ​​input into the behavior detection model, and click events in user operation behavior are identified based on the behavior detection model. Determine the feature parameters of the click event, generate a feedback signal based on the feature parameters, and generate a target response strategy based on the feedback signal; The step of extracting feature data from the original data stream and generating a standardized dataset based on the feature data includes: Identify the peak and trough points in the acceleration data, and mark the areas corresponding to the peak and trough points as potential click areas; The time periods in the angular velocity data where the rate of change exceeds a preset threshold are marked as potential click time periods; Extract the pressure abrupt change intervals from the pressure data within the potential click area and the potential click time period; The peak points of the acceleration data, the rate of change of the angular velocity data, and the pressure abrupt range of the pressure data are normalized to generate a standardized dataset. The step of inputting the multidimensional feature matrix into the behavior detection model and identifying click events in user operation behavior based on the behavior detection model includes: The multidimensional feature matrix is ​​input into the initial behavior detection model, the initial behavior detection model is transformed into a behavior detection model, and candidate features are extracted from the multidimensional feature matrix based on the behavior detection model. The confidence level of the click event is determined based on the peak acceleration, rate of change of angular velocity, and pressure mutation amount in the candidate features. Determine a confidence threshold, and mark the click events corresponding to candidate features whose confidence scores are higher than the confidence threshold as valid click events; Based on the timestamp and feature parameters of the valid click event, generate the recognition result of the click event.

2. The method as described in claim 1, characterized in that, Prior to the step of extracting feature data from the original data stream, the method further includes: The acceleration data is subjected to zero-phase filtering based on a second-order Butterworth low-pass filter to obtain low-frequency acceleration data. The angular velocity data is forward filtered using a fourth-order Butterworth low-pass filter to obtain low-frequency angular velocity data. The pressure data is filtered using an FIR filter based on a Hamming window to obtain low-frequency pressure data. Frequency domain energy analysis is performed on the low-frequency acceleration data, the low-frequency angular velocity data, and the low-frequency pressure data to obtain filtered acceleration data, filtered angular velocity data, and filtered pressure data.

3. The method as described in claim 1, characterized in that, The step of temporally aligning the acceleration data, angular velocity data, and pressure data in the standardized dataset to generate a multidimensional feature matrix includes: The sampling frequencies for the acceleration data, the angular velocity data, and the pressure data are determined respectively, and the timestamp deviations of the acceleration data, the angular velocity data, and the pressure data are calculated respectively based on the sampling frequencies; Using the timestamp deviation as a calibration benchmark, the time series of each data stream is adjusted to align the acceleration data, the angular velocity data, and the pressure data on the time axis. The aligned acceleration data, angular velocity data, and pressure data are mapped to different dimensions of the three-dimensional coordinate system to form initial feature vectors. The initial feature vectors are arranged in chronological order to generate a multidimensional feature matrix.

4. The method as described in claim 3, characterized in that, The step of mapping the aligned acceleration data, angular velocity data, and pressure data to different dimensions of the three-dimensional coordinate system to form an initial feature vector includes: The triaxial components of the aligned acceleration data are mapped to the X-axis of the three-dimensional coordinate system after feature synthesis. The aligned angular velocity data is dynamically integrated and then mapped to the Y-axis of the three-dimensional coordinate system; The aligned pressure data is mapped to the Z-axis of the three-dimensional coordinate system; The mapped 3D coordinate points are arranged based on a time series to form an initial feature vector.

5. The method as described in claim 1, characterized in that, The steps of determining the feature parameters of the click event, generating a feedback signal based on the feature parameters, and generating a target response strategy based on the feedback signal include: Determine the event type of the click event, and determine the feature parameters of the click event based on the multidimensional feature matrix; First data for determining the feedback signal based on the event type of the click event, the first data being used to distinguish signal categories; The second data of the feedback signal is determined based on the characteristic parameters, and the second data is used to represent the parameters of the click event; A feedback signal is generated based on the first data and the second data, and a target response strategy is generated based on the feedback signal.

6. The method as described in claim 1, characterized in that, Before the step of inputting the multidimensional feature matrix into the initial behavior detection model, the method further includes: Obtain the model training dataset, which includes acceleration sample data, angular velocity sample data, pressure data, and behavior category labels for click events; The model training dataset is divided into a training set and a validation set. The training set is input into two one-dimensional convolutional layers for convolution processing to extract local spatiotemporal features. The training set is input into a two-layer LSTM model to obtain long-term dependent features; The long-term dependency features and the local spatiotemporal features are concatenated, and binary classification is performed based on the concatenated feature data to determine the predicted behavior category. The prediction error is determined based on the predicted behavior category and the behavior category label, and a correction parameter is generated based on the prediction error. A mapping equation is generated based on long-term dependency features, local spatiotemporal features, behavior category labels, and correction parameters; The predicted behavior category is validated based on the validation set to determine the equation error; The mapping equation is updated based on the error of the equation to generate an initial behavior detection model.

7. The method according to any one of claims 1 to 6, characterized in that, Before the step of temporally aligning the acceleration data, angular velocity data, and pressure data in the standardized dataset to generate a multidimensional feature matrix, the method further includes: Determine the attitude change vector based on the angular velocity data; The displacement acceleration is determined based on the time of change of the attitude change vector and the angular velocity data. An acceleration increment is generated based on the displacement acceleration; The acceleration data is corrected based on the acceleration increment.

8. A click detection device based on multi-axis sensor fusion, characterized in that, The click detection device based on multi-axis sensor fusion includes: The data acquisition module is used to acquire the raw data stream collected by the multi-axis sensor built into the target device. The raw data stream includes acceleration data, angular velocity data and pressure data. The data processing module is used to extract feature data from the original data stream and generate a standardized dataset based on the feature data. The feature alignment module is used to perform temporal alignment of the acceleration data, angular velocity data and pressure data in the standardized dataset to generate a multidimensional feature matrix. The behavior recognition module is used to input the multidimensional feature matrix into the behavior detection model and identify click events in user operation behavior based on the behavior detection model; The feedback generation module is used to determine the feature parameters of the click event, generate a feedback signal based on the feature parameters, and generate a target response strategy based on the feedback signal. The step of extracting feature data from the original data stream and generating a standardized dataset based on the feature data includes: Identify the peak and trough points in the acceleration data, and mark the areas corresponding to the peak and trough points as potential click areas; The time periods in the angular velocity data where the rate of change exceeds a preset threshold are marked as potential click time periods; Extract the pressure abrupt change intervals from the pressure data within the potential click area and the potential click time period; The peak points of the acceleration data, the rate of change of the angular velocity data, and the pressure abrupt range of the pressure data are normalized to generate a standardized dataset. The step of inputting the multidimensional feature matrix into the behavior detection model and identifying click events in user operation behavior based on the behavior detection model includes: The multidimensional feature matrix is ​​input into the initial behavior detection model, the initial behavior detection model is transformed into a behavior detection model, and candidate features are extracted from the multidimensional feature matrix based on the behavior detection model. The confidence level of the click event is determined based on the peak acceleration, rate of change of angular velocity, and pressure mutation amount in the candidate features. Determine a confidence threshold, and mark the click events corresponding to candidate features whose confidence scores are higher than the confidence threshold as valid click events; Based on the timestamp and feature parameters of the valid click event, generate the recognition result of the click event.

Images