CN-122020395-A - Precise identification method and system for touch operation of intelligent bracelet

CN122020395ACN 122020395 ACN122020395 ACN 122020395ACN-122020395-A

Abstract

The application relates to the technical field of data processing and discloses a precise identification method and a precise identification system for touch operation of an intelligent bracelet, wherein the method comprises the steps of collecting multi-dimensional data of a bracelet touch screen capacitor, motion sensing and pressure, establishing an analysis time window, extracting all dimensional characteristics, completing time stamp alignment and normalization, constructing multi-mode feature vectors, constructing a joint judgment space to obtain a space classification judgment result through causal examination and output a time sequence causal judgment result, combining a dynamic threshold adjustment value of motion intensity to output a dynamic threshold judgment result, finally integrating three types of results based on a Bayesian decision theory, outputting a final identification result, collecting feedback data to optimize model parameters, updating a sample library and forming a closed-loop identification mechanism. According to the intelligent bracelet touch control system, true and false touch control is effectively defined through multi-mode data processing and multi-dimensional judgment, recognition accuracy is continuously improved through closed loop optimization, misoperation is reduced, and reliability of the intelligent bracelet touch control system is greatly improved.

Inventors

ZHOU SIYI
JIANG YANFENG
HE SHUJING
XIE ZHEN

Assignees

中巢(深圳)商业服务有限公司

Dates

Publication Date: 20260512
Application Date: 20260212

Claims (10)

1. The accurate touch operation identification method of the intelligent bracelet is characterized by comprising the following steps of: Acquiring capacitance signals, motion sensing data and pressure data of an intelligent bracelet touch screen, establishing an analysis time window based on touch event triggering time, extracting frequency domain and time domain characteristics from the capacitance signals, filtering the motion sensing data to extract motion characteristics, aligning and normalizing all characteristic time stamps, and forming a multi-mode characteristic vector; Aligning the multi-mode feature vector to a unified reference coordinate system according to time dimension, analyzing time sequence relation of touch control and motion signals, judging causal directions of the touch control and the motion signals by combining a causal detection algorithm, and outputting a time sequence causal judgment result; Based on the multi-mode feature vector, extracting touch track, motion gesture and signal energy features, constructing a joint judgment space, constructing a judgment boundary by using a history sample, mapping the current touch event features to judgment space classification, and outputting a space classification judgment result; Determining the motion intensity according to the multi-mode feature vector, dynamically adjusting a judging threshold value, analyzing and judging by combining with the time sequence mode of the touch event, and outputting a dynamic threshold value judging result; integrating the three types of judgment results, calculating posterior probability based on Bayesian decision theory, outputting a final recognition result, collecting feedback data optimization model parameters and judgment thresholds, and updating a sample library to form a closed-loop recognition mechanism.
2. The precise identification method for touch operation of an intelligent bracelet according to claim 1, wherein the steps of aligning the multi-modal feature vectors to a unified reference coordinate system according to time dimension, analyzing time sequence relation between touch and motion signals, judging causal directions of the touch and motion signals by combining a causal detection algorithm, and outputting time sequence causal judgment results comprise the following steps: Extracting a motion mode, a motion intensity and a motion mutation duration, and establishing a mapping relation among the three, an initial time delay threshold, a causal inspection threshold and a time sequence analysis window; Setting various parameters according to the mapping relation, positioning the peak time of the touch signal and the abrupt change time of the motion signal in a dynamic analysis window, calculating the time difference of the two to finish initial time sequence judgment, and synchronously constructing a multi-mode weighted time sequence characteristic sequence; the multi-mode weighted time sequence feature sequence is fused into a causal inspection vector autoregressive model, the model order is adjusted according to the initial time sequence judgment result, and the causal judgment result is obtained through an inspection algorithm; and integrating the initial time sequence judgment result and the causal judgment result, and if the confidence coefficient of the judgment result does not reach a preset threshold value, dynamically adjusting the threshold value and iteratively optimizing the length of the analysis window according to the motion intensity until the result is consistent, and outputting the time sequence causal judgment result.
3. The precise identification method for touch operation of an intelligent bracelet according to claim 2, wherein the steps of setting parameters according to the mapping relation, positioning a touch signal peak time and a motion signal abrupt change time in a dynamic analysis window, calculating a time difference between the two to complete initial time sequence judgment, and synchronously constructing a multi-mode weighted time sequence feature sequence comprise the following steps: A multidimensional mapping relation is constructed by fusing the motion mode, the motion intensity and the historical data, and a gradient optimization algorithm is adopted to carry out iterative optimization on the dynamic analysis window and the characteristic weight parameters; According to the optimized parameters, combining the pressure signal stability calibration characteristic weights, extracting multi-mode time sequence data, and constructing a correlation enhancement type weighted time sequence characteristic sequence; In the dynamic analysis window, positioning the peak time of the touch signal and the abrupt change time of the motion signal by taking the effective pressure contact time as a reference, and calculating the comprehensive time difference of the peak time and the abrupt change time of the motion signal to finish the initial time sequence judgment; and inputting the calibrated weighted time sequence characteristic sequence and the initial time sequence judgment result into a causal inspection flow, and adjusting the characteristic attention according to the motion intensity.
4. The precise recognition method for touch operation of an intelligent bracelet according to claim 1, wherein the steps of extracting touch track, motion gesture and signal energy feature based on the multi-modal feature vector, constructing a joint judgment space, constructing a judgment boundary by using a history sample, mapping the current touch event feature to a judgment space classification, and outputting a space classification judgment result comprise: dynamically distributing the touch track, the motion gesture and the signal energy characteristic weight according to the motion mode and the motion intensity, fusing the pressure characteristics to construct a multi-dimensional characteristic set, and establishing a mapping relation between the motion intensity and a kernel function; Adopting a feature dimension reduction algorithm to reduce dimension of the multi-dimensional feature set to three dimensions, reserving core distinguishing information, mapping the multi-dimensional feature set to an enhanced three-dimensional joint judgment space, selecting an adaptive kernel function according to motion intensity, and initializing a support vector machine; Establishing exclusive judgment boundaries according to touch intention layering, pre-judging the touch intention to match the corresponding boundaries to finish preliminary classification, and optimizing hyperplane parameters by cross-verifying other intention boundaries and combining history samples; and calculating the classification accuracy, the pressure characteristic dimension reduction retention and the cross verification conflict rate, if any index does not reach a preset standard, readjusting the characteristic weight, the kernel function and the boundary parameter iterative optimization until the index reaches the standard, mapping the current touch event characteristic to the joint judgment space to complete classification, and outputting a space classification judgment result.
5. The precise identification method for touch operation of an intelligent bracelet according to claim 4, wherein the step of constructing dedicated decision boundaries hierarchically according to touch intention, pre-judging that the touch intention matches the corresponding boundary to finish preliminary classification, cross-verifying through other intention boundaries, and optimizing hyperplane parameters in combination with historical samples comprises the following steps: Refining the characteristic weight and kernel function parameters of the touch intention according to the motion strength, and adding the touch duration to assist in pre-judging the touch intention; adjusting exclusive boundary parameters of corresponding intentions by combining real-time motion intensity and prejudged touch intentions, completing preliminary classification by adopting a support vector machine, and synchronously calculating classification confidence; Cross-verifying the preliminary classification result by using the judgment boundaries of other touch intents, and checking the consistency of classification confidence coefficient, the touch duration and the matching degree of the touch intents; If the confidence degree consistency or the duration intention matching degree does not reach the preset standard, the historical misjudgment sample is combined to optimize the hyperplane punishment coefficient and the kernel function parameter, the exclusive judgment boundary is retrained and is subjected to iterative optimization until the index reaches the standard.
6. The precise identification method for touch operation of an intelligent bracelet according to claim 5, wherein the step of combining real-time motion intensity with a pre-determined touch intention, adjusting an exclusive boundary parameter of the corresponding intention, completing preliminary classification by using a support vector machine, and synchronously calculating classification confidence comprises the following steps: According to the motion strength and the motion mode, a mapping model of a confidence coefficient threshold value and a misjudgment risk is established, and a correlation library of historical confidence coefficient and classification accuracy is synchronously established; adjusting exclusive boundary parameters of corresponding intentions and adaptive kernel function types by referring to the mapping model by combining real-time motion intensity and prejudged touch intentions; Performing preliminary classification on the current touch event characteristics by adopting a support vector machine, calculating classified original confidence coefficient, calibrating the original confidence coefficient by using the history association library, and evaluating misjudgment risk by combining the calibrated confidence coefficient; If the classification confidence coefficient meets the preset standard and the misjudgment risk does not meet the preset standard, entering a boundary cross-validation link of other touch intentions, if the classification confidence coefficient does not meet the standard, marking the sample to be optimized, recording a classification confidence coefficient fluctuation value, collecting the sample to be optimized and fluctuation data, readjusting boundary parameters, confidence coefficient thresholds and kernel function types, and iterating the training model until the index meets the standard.
7. The precise identification method for touch operation of an intelligent bracelet according to claim 1, wherein the step of determining the motion intensity according to the multi-mode feature vector, dynamically adjusting a judgment threshold value, analyzing and judging in combination with a touch event time sequence mode, and outputting a judgment result of the dynamic threshold value comprises the following steps: Based on the multi-mode feature vector recognition motion mode, setting a special motion intensity grading threshold and a threshold adjustment rule according to the motion mode, and synchronously constructing an optimized sample library comprising historical threshold adjustment, judgment accuracy and boundary sample features; Setting the duration of a sliding window, continuously collecting touch events, calculating the statistical characteristics of the motion intensity and the distribution of the touch characteristics in the window, calibrating an initial judgment threshold value by combining an optimized sample library, dynamically adjusting the threshold value according to the motion mode and the real-time motion intensity, and synchronously labeling boundary samples in a preset threshold value interval; performing preliminary judgment on each touch event according to the adjusted judgment threshold value, strengthening judgment conclusion if the judgment results are consistent for a plurality of times, analyzing the periodicity of the motion mode if the judgment results are inconsistent or contain boundary samples, and matching the judgment logic of the similar historical scenes in the sample library; if the regular movement period exists, the movement interference is judged to be systematic false touch, if the regular movement period exists, the mixed touch scene is defined as irregular touch, all data, boundary samples and judgment results are supplemented to an optimized sample library, the special threshold rule and the calibration logic of the movement pattern are iterated and optimized, and the dynamic threshold judgment result is output.
8. The precise identification method for touch operation of an intelligent bracelet according to claim 7, wherein the steps of setting the duration of a sliding window, continuously collecting touch events, calculating the statistical characteristics of the motion intensity and the distribution of the touch characteristics in the window, calibrating an initial judgment threshold value by combining an optimized sample library, dynamically adjusting the threshold value according to the motion mode and the real-time motion intensity, and synchronously labeling boundary samples in a preset threshold value interval comprise the following steps: subdividing the motion modes, constructing an optimized sample library subset corresponding to each subdivision mode, screening historical misjudgment samples, establishing a mapping relation between misjudgment scenes and adaptation parameters, and setting a motion intensity fluctuation threshold, a window and a threshold collaborative adaptation rule; Matching a real-time subdivision motion mode, setting the duration of an initial sliding window by combining the mapping relation between the misjudgment scene and the adaptive parameters, continuously acquiring touch events, and calculating the motion intensity statistical characteristics and the touch characteristic distribution in the window; If the fluctuation of the motion intensity exceeds the standard, adjusting the window duration and the threshold value adjusting amplitude, combining the sample library subset corresponding to the subdivision mode to calibrate the initial judgment threshold value, dynamically adjusting the judgment threshold value according to the real-time motion intensity, and synchronously labeling boundary samples in a preset threshold value interval; and feeding back the fluctuation adaptation effect, the threshold deviation, the boundary sample and the misjudgment condition to the mapping relation between the optimized sample library and the misjudgment scene adaptation parameter, and iterating and optimizing a window, a threshold rule and a collaborative adaptation logic exclusive to the subdivision mode.
9. The precise identification method for touch operation of an intelligent bracelet according to claim 8, wherein the step of matching real-time subdivision motion modes, setting initial sliding window duration in combination with the mapping relation between the misjudgment scene and the adaptive parameters, continuously collecting touch events, and calculating the distribution of motion intensity statistical features and touch features in a window comprises the following steps: clustering historical misjudgment samples according to subdivision motion modes, giving high weight to high-frequency misjudgment scenes, constructing a weighted mapping relation between the misjudgment scenes and window duration, extracting historical statistical data of the same-mode same-intensity intervals, and setting a touch characteristic pre-screening rule; Matching a real-time subdivision motion mode, setting initial sliding window time length by combining the mapping relation between the misjudgment scene and the adaptive parameter and the weighted mapping between the misjudgment scene and the window time length, acquiring effective touch events according to a touch characteristic pre-screening rule, synchronously calculating touch frequency, and checking and fine-adjusting the sliding window time length; Calculating the motion intensity statistical characteristics in the sliding window after fine adjustment and the screened touch characteristic distribution, comparing and calibrating the motion intensity statistical characteristics with the historical statistical data of the same intensity interval of the corresponding subdivision mode, and marking the characteristic data similar to the misjudgment scene; And feeding back the feature similarity, the motion intensity fluctuation value, the calibration deviation and the touch frequency data to a weighted mapping and sample library of the misjudgment scene and the window duration, and iteratively optimizing window adaptation rules, touch feature pre-screening rules and historical data migration logic.
10. Accurate identification system of touch operation of intelligent bracelet, its characterized in that includes: The feature extraction module is used for collecting capacitance signals, motion sensing data and pressure data of the intelligent bracelet touch screen, establishing an analysis time window based on touch event triggering time, extracting frequency domain and time domain features from the capacitance signals, filtering the motion sensing data to extract motion features, aligning and normalizing all feature time stamps, and forming a multi-mode feature vector; the time sequence causal judgment module is used for aligning the multi-mode feature vector to a unified reference coordinate system according to time dimension, analyzing the time sequence relation between touch control and motion signals, judging causal directions of the touch control and the motion signals by combining a causal detection algorithm, and outputting a time sequence causal judgment result; The space classification judging module is used for extracting touch track, motion gesture and signal energy characteristics based on the multi-mode feature vector, constructing a joint judging space, constructing a judging boundary by utilizing a history sample, mapping the current touch event characteristics to judging space classification and outputting a space classification judging result; the dynamic threshold judgment module is used for determining the motion intensity according to the multi-mode feature vector, dynamically adjusting the judgment threshold, analyzing and judging in combination with the time sequence mode of the touch event, and outputting a dynamic threshold judgment result; And the fusion decision module is used for integrating the three types of decision results, calculating posterior probability based on a Bayesian decision theory, outputting a final recognition result, collecting feedback data optimization model parameters and decision thresholds, and updating a sample library to form a closed-loop recognition mechanism.

Description

Precise identification method and system for touch operation of intelligent bracelet Technical Field The application relates to the technical field of data processing, in particular to a touch operation accurate identification method and system of an intelligent bracelet. Background The intelligent bracelet is used as portable wearable equipment and is widely applied to daily sports and health monitoring scenes, and a touch screen is a core component for realizing functional interaction of users. However, in the daily body movement process of running, walking, arm swinging and the like, vibration and acceleration changes generated by movement can be transmitted to the touch screen, so that the touch sensor receives interference signals of non-user active touch. The existing intelligent bracelet touch control system lacks an effective signal identification and distinguishing mechanism, is difficult to accurately distinguish false touch control signals caused by active touch control operation and motion of a user, so that the problem of misoperation of the bracelet in a motion scene is very easy to occur, and the interaction experience of the user and the use reliability of equipment are seriously affected. Disclosure of Invention The application mainly aims to provide a precise identification method and a precise identification system for touch operation of an intelligent bracelet, and aims to solve the technical problems that an interference signal is generated by vibration and acceleration change of the intelligent bracelet in movement, a touch system cannot distinguish true touch from false touch, misoperation is easy to occur, and use experience and reliability are affected. The first aspect of the present application provides a method for precisely identifying touch operation of an intelligent bracelet, comprising: Acquiring capacitance signals, motion sensing data and pressure data of an intelligent bracelet touch screen, establishing an analysis time window based on touch event triggering time, extracting frequency domain and time domain characteristics from the capacitance signals, filtering the motion sensing data to extract motion characteristics, aligning and normalizing all characteristic time stamps, and forming a multi-mode characteristic vector; Aligning the multi-mode feature vector to a unified reference coordinate system according to time dimension, analyzing time sequence relation of touch control and motion signals, judging causal directions of the touch control and the motion signals by combining a causal detection algorithm, and outputting a time sequence causal judgment result; Based on the multi-mode feature vector, extracting touch track, motion gesture and signal energy features, constructing a joint judgment space, constructing a judgment boundary by using a history sample, mapping the current touch event features to judgment space classification, and outputting a space classification judgment result; Determining the motion intensity according to the multi-mode feature vector, dynamically adjusting a judging threshold value, analyzing and judging by combining with the time sequence mode of the touch event, and outputting a dynamic threshold value judging result; integrating the three types of judgment results, calculating posterior probability based on Bayesian decision theory, outputting a final recognition result, collecting feedback data optimization model parameters and judgment thresholds, and updating a sample library to form a closed-loop recognition mechanism. Further, the step of aligning the multi-modal feature vector to a unified reference coordinate system according to time dimension, analyzing time sequence relation of touch control and motion signals, judging causal directions of the touch control and the motion signals by combining a causal inspection algorithm, and outputting a time sequence causal judgment result comprises the following steps: Extracting a motion mode, a motion intensity and a motion mutation duration, and establishing a mapping relation among the three, an initial time delay threshold, a causal inspection threshold and a time sequence analysis window; Setting various parameters according to the mapping relation, positioning the peak time of the touch signal and the abrupt change time of the motion signal in a dynamic analysis window, calculating the time difference of the two to finish initial time sequence judgment, and synchronously constructing a multi-mode weighted time sequence characteristic sequence; the multi-mode weighted time sequence feature sequence is fused into a causal inspection vector autoregressive model, the model order is adjusted according to the initial time sequence judgment result, and the causal judgment result is obtained through an inspection algorithm; and integrating the initial time sequence judgment result and the causal judgment result, and if the confidence coefficient of the judgment result does not reach a preset