Vehicle-mounted touch screen mistaken touch prevention method and device

Abstract

The application discloses a kind of vehicle-mounted touch screen's false touch prevention method and device, the method includes: through the multi-axis vibration sensor array of being arranged in touch screen cover plate, real-time acquisition solid conduction vibration signal, extract the multidimensional feature vector containing time domain, frequency domain and multi-sensor space correlation, and with the positive and negative class probability model matched in advance generation vibration confidence; While obtaining vehicle state parameter dynamic adjustment decision threshold, after receiving the touch coordinate reported by touch display integrated chip, vibration confidence is fused and adjusted by time window alignment and space consistency verification, finally based on fusion confidence determines real touch event.The application carries out space-time multidimensional fusion decision by vibration recognition and coordinate signal, effectively solve the false touch problem of touch screen in the strong electromagnetic interference environment of vehicle-mounted, significantly improve the accuracy and robustness of touch recognition.

Description

Technical Field

[0001] This invention relates to the field of touch screen technology, and in particular to a method and device for preventing accidental touches on an in-vehicle touch screen. Background Technology

[0002] Touch functionality in in-vehicle displays has become an important means for users to interact with their vehicles. Mainstream capacitive touchscreens achieve coordinate positioning by detecting changes in capacitance caused by finger touches; however, their signal processing links are extremely sensitive to electromagnetic interference (EMI). The in-vehicle environment contains multi-source electromagnetic interference from the powertrain, communication modules, and other sources, which can easily cause the Touch Display Integrated Chip (TDDIC) to output false coordinates, resulting in "false touches" or "no response" issues.

[0003] Existing solutions mainly fall into two categories: one is hardware filtering and shielding, but it is difficult to locate the interference source, the effectiveness of the countermeasures is limited, and the portability is poor; the other is software filtering algorithms, but debugging is complicated, the effect is not good when facing multi-frequency interference, and it also lacks cross-vehicle adaptability.

[0004] In recent years, multimodal fusion anti-mistouch solutions have emerged, such as dual confirmation by detecting touch sound or pressure signals and coordinates. Such solutions work well in quiet indoor environments or foldable screen devices, but the in-vehicle environment has the dual characteristics of strong background noise (wind noise, road noise, engine roar) and strong electromagnetic interference, making it difficult for a single sensor to cope with both types of interference at the same time.

[0005] Therefore, how to accurately distinguish between real touch and false touch caused by electromagnetic interference in the complex environment of an in-vehicle vehicle has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] The purpose of this invention is to disclose a method and device for preventing accidental touches on an in-vehicle touchscreen, which solves the technical problem that existing touchscreens cannot accurately distinguish false touches in the complex environment of an in-vehicle environment.

[0007] To achieve the above objectives, the present invention discloses a method for preventing accidental touches on an in-vehicle touchscreen, comprising: Real-time acquisition of solid-borne vibration signals caused by touch or interference; The collected vibration signals are preprocessed by framing and potential touch event frames are detected. A multi-dimensional feature vector is extracted from the detected potential touch event frames. The feature vector includes time-domain features, frequency-domain features, and spatial features reflecting the signal correlation between different vibration sensors. The feature vector is matched with pre-constructed positive and negative class templates to calculate the vibration confidence level, which characterizes the possibility that the vibration signal originates from a real touch. Obtain vehicle state parameters and dynamically adjust the confidence threshold for determining valid vibration events based on the vehicle state parameters; Receive touch coordinates reported from the touch display integrated chip, and verify whether the touch coordinates are time-aligned with vibration events that meet the confidence threshold; For time-aligned vibration events, the location of the vibration source is estimated based on the spatial characteristics, and spatial consistency is verified with the touch coordinates. Based on the results of the time alignment and spatial consistency verification, the vibration confidence is adjusted to generate a fusion confidence, and the touch coordinates are determined to be real touches based on the fusion confidence.

[0008] Specifically, the step of extracting multi-dimensional feature vectors from detected potential touch event frames includes: Extract time-domain features, including the peak amplitude of the X, Y, and Z axis signals, the minimum rise time of the three axis signals, the minimum decay time constant of the three axis signals, and the maximum pulse width of the three axis signals; Extract frequency domain features, including the energy proportion of each axis signal in the preset low frequency band, mid frequency band and high frequency band, the spectral flatness of each frequency band, and the main frequency and bandwidth of the synthesized spectrum; Extract cepstral features, including cepstral coefficients of a preset order obtained from the inverse Fourier transform after the logarithmic transform of the synthesized spectrum; Spatial features are extracted, including the time difference of arrival based on the reference sensor, the energy ratio of each sensor signal, the vibration source location and location confidence estimated based on the Z-axis energy weighting of each sensor, and the normalized cross-correlation peak value of the reference sensor and other sensor signals.

[0009] Specifically, the positive class template is a Gaussian mixture model trained based on positive sample feature vectors, and the negative class template is a single Gaussian model trained based on negative sample feature vectors. The calculation of vibration confidence includes: Calculate the weighted likelihood score of the feature vector under the positive Gaussian mixture model; Calculate the likelihood score of the feature vector under the negative class single Gaussian model; The normalized vibration confidence score is calculated based on the ratio of the weighted likelihood score to the negative class model likelihood score.

[0010] Specifically, the vehicle state parameters include vehicle speed and engine speed, and the step of dynamically adjusting the confidence threshold includes: According to the formula Calculate the dynamic decision threshold C th , where C base The base threshold is v, where v is vehicle speed, r is engine speed, and k is the engine speed.v and k r This is the preset positive compensation coefficient.

[0011] Specifically, adjusting the vibration confidence score to generate a fusion confidence score based on the spatial consistency verification result includes: If the distance between the estimated location of the vibration source and the touch coordinates is less than or equal to a preset distance threshold, and the location confidence level is greater than a preset confidence threshold, then spatial consistency is determined, and the formula is used. Generate fusion confidence C fusion C vib For vibration confidence, c pos For location reliability, 'a' is the preset reward coefficient; Otherwise, determine if the spaces are inconsistent and use the formula. Generate a fusion confidence score, where b is a preset penalty coefficient.

[0012] Furthermore, the method for preventing accidental touches on the in-vehicle touchscreen also includes an online adaptive learning step: Once a real touch is determined, the feature vector of the current event is used as a new positive sample, and the parameters of the model component that best matches it in the positive class template are fine-tuned and updated. When the system is falsely triggered and receives a user's cancellation command, the feature vector that caused the false trigger will be used as a new negative sample to update the model parameters of the negative class template.

[0013] On the other hand, in order to achieve the above objectives, the present invention discloses an anti-accidental touch device for an in-vehicle touchscreen, comprising: The vibration sensor array includes multiple triaxial accelerometers mounted on the touchscreen cover for real-time acquisition of solid-borne vibration signals caused by touch or interference. The signal acquisition and preprocessing module is used to perform frame-by-frame preprocessing on the vibration signal and detect potential touch event frames; The feature extraction module is used to extract multi-dimensional feature vectors from detected potential touch event frames. The feature vectors include time-domain features, frequency-domain features, and spatial features reflecting the signal correlation between different vibration sensors. The template library storage module is used to store pre-built positive class templates and negative class templates; The real-time matching module is used to match the feature vector with the positive class template and the negative class template to calculate the vibration confidence. The vehicle status acquisition module is used to acquire vehicle status parameters and provide them to the real-time matching module to dynamically adjust the confidence threshold. The fusion decision module is used to receive touch coordinates reported from the touch display integrated chip and perform the following operations: verify whether the touch coordinates are time-aligned with a vibration event that meets the confidence threshold; for time-aligned vibration events, estimate the vibration source position based on the spatial characteristics and perform spatial consistency verification with the touch coordinates; adjust the vibration confidence based on the verification result to generate a fusion confidence, and determine whether the touch coordinates are a real touch based on the fusion confidence.

[0014] Specifically, the feature extraction module is configured as follows: Extract time-domain features, including the peak amplitude of the X, Y, and Z axis signals, the minimum rise time of the three axis signals, the minimum decay time constant of the three axis signals, and the maximum pulse width of the three axis signals; Extract frequency domain features, including the energy proportion of each axis signal in the preset low frequency band, mid frequency band and high frequency band, the spectral flatness of each frequency band, and the main frequency and bandwidth of the synthesized spectrum; Extract cepstral features, including cepstral coefficients of a preset order obtained from the inverse Fourier transform after the logarithmic transform of the synthesized spectrum; Spatial features are extracted, including the time difference of arrival based on the reference sensor, the energy ratio of each sensor signal, the vibration source location and location confidence estimated based on the Z-axis energy weighting of each sensor, and the normalized cross-correlation peak value of the reference sensor and other sensor signals.

[0015] Specifically, the fusion decision module is configured as follows: If the distance between the estimated location of the vibration source and the touch coordinates is less than or equal to a preset distance threshold, and the location confidence level is greater than a preset confidence threshold, then spatial consistency is determined, and the formula is used. Generate fusion confidence C fusion C vib For vibration confidence, c pos For location reliability, 'a' is the preset reward coefficient; Otherwise, determine if the spaces are inconsistent and use the formula. Generate a fusion confidence score, where b is a preset penalty coefficient.

[0016] Furthermore, the anti-accidental touch device for the vehicle touchscreen also includes an adaptive learning module, used for: When the fusion decision module determines that it is a real touch, it uses the feature vector of the current event as a new positive sample to fine-tune and update the positive class template parameters in the template library storage module. When the system is falsely triggered and receives a user's cancellation command, the feature vector that caused the false trigger will be used as a new negative sample to update the negative class template parameters in the template library storage module.

[0017] The beneficial effects of this invention are as follows: This invention collects solid-borne vibration signals by deploying a multi-axis vibration sensor array on the touch screen cover, and combines multi-dimensional time-frequency domain feature extraction and multi-sensor spatial feature analysis to construct a statistical probability model of real touch and background interference. The vibration recognition results are then fused with the coordinate signals reported by the touch display integrated chip in a spatiotemporal manner to make a decision. This effectively solves the problem of false touch caused by electromagnetic interference in the vehicle environment. Under complex working conditions with strong background noise and multi-source electromagnetic interference, it significantly improves the recognition accuracy and anti-interference robustness of real touch events. At the same time, by introducing vehicle state parameters for dynamic threshold adjustment, the system's adaptability to different driving scenarios is enhanced, ensuring the reliability and safety of human-computer interaction during driving. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the structure of the vehicle-mounted touchscreen anti-accidental touch device of the present invention; Figure 2 This is a flowchart illustrating the method for preventing accidental touches on an in-vehicle touchscreen according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] In this invention, the terms "upper," "lower," "left," "right," "front," "rear," "top," "bottom," "inner," "outer," "middle," "vertical," "horizontal," "lateral," and "longitudinal" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for the purpose of better describing the invention and its embodiments, and are not intended to limit the indicated devices, elements, or components to having a specific orientation, or to be constructed and operated in a specific orientation.

[0022] Furthermore, in addition to indicating direction or positional relationship, some of the aforementioned terms may also have other meanings. For example, the term "above" may also be used in certain situations to indicate a dependency or connection. Those skilled in the art can understand the specific meaning of these terms in this invention based on the specific circumstances.

[0023] Furthermore, the terms "installation," "setup," "equipped with," "connection," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral structure; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium, or an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of these terms in this invention based on the specific circumstances.

[0024] Furthermore, the terms "first," "second," etc., are primarily used to distinguish different devices, components, or parts (which may be the same or different in specific type and construction), and are not intended to indicate or imply the relative importance or quantity of the indicated devices, components, or parts. Unless otherwise stated, "a plurality of" means two or more.

[0025] The technical solution of the present invention will be further described below with reference to the embodiments and accompanying drawings.

[0026] Example 1 like Figure 1 As shown, this embodiment provides an anti-accidental touch device for an in-vehicle touchscreen, including: a vibration sensor array, a signal acquisition and preprocessing module, a feature extraction module, a template library storage module, a vehicle status acquisition module, a real-time matching module, a fusion decision module, and an adaptive learning module.

[0027] The vibration sensor array includes four triaxial accelerometers, each positioned at one of the four corners of the touchscreen glass cover. Each triaxial accelerometer is used to collect solid-conducted vibration signals generated when a finger touches the screen, outputting X, Y, and Z-axis acceleration components. During installation and configuration, the Z-axis of all sensors is perpendicular to the screen plane to sense the vertical vibration generated by the touch. Furthermore, the local coordinate systems of all sensors are calibrated so that the X-axis of each sensor points in the same first direction, the Y-axis points in the same second direction, and the Z-axis is perpendicular to the screen plane and points in the same direction.

[0028] The signal acquisition and preprocessing module is connected to the vibration sensor array and is used to perform anti-aliasing filtering, analog-to-digital conversion (ADC), frame processing, background noise estimation, and activity detection on the analog signals output by the sensors in order to filter out potential touch event frames.

[0029] The feature extraction module is used to process the potential touch event signals of each frame and extract feature vectors of a preset dimension.

[0030] The template library storage module is a non-volatile memory used to store the parameters of the positive Gaussian Mixture Model (GMM) and the negative Gaussian Single Model obtained through offline training.

[0031] The vehicle status acquisition module is used to receive vehicle status parameters from the vehicle bus (such as the CAN bus), which include at least vehicle speed and engine speed.

[0032] The real-time matching module is used to match the feature vector of the current frame with the positive and negative class models in the template library to generate vibration confidence and dynamically adjust the judgment threshold according to the vehicle state parameters.

[0033] The fusion decision module is used to integrate the vibration confidence, the touch coordinates reported by TDDIC, the time window alignment result, and the spatial consistency verification result to calculate the fusion confidence and output the final real touch judgment signal.

[0034] In another embodiment of the present invention, the vehicle touchscreen anti-accidental touch device further includes: The adaptive learning module is used to update the model parameters in the template library online based on the confirmed touch results during system use.

[0035] Example 2 like Figure 2 As shown, this embodiment provides a method for preventing accidental touches on an in-vehicle touchscreen, including: Step 1: Construct a vibration fingerprint template library.

[0036] Before the vehicle leaves the factory or during the initial system initialization, a vibration fingerprint template library is built through a calibration acquisition process, and the trained model parameters are stored in non-volatile memory for direct loading and calling during system runtime.

[0037] In this embodiment, step 1 includes: Step 101: Collect positive sample data.

[0038] Under controlled conditions where the vehicle is stationary and there is no significant environmental vibration, the user taps multiple preset locations (e.g., nine evenly distributed coordinate points) on the touchscreen with preset light, medium, and heavy pressure levels. Each location-pressure combination is repeated a preset number of times (e.g., 10 times), and the data is recorded at a sampling rate f using triaxial accelerometers located at the four corners of the touchscreen. s =20kHz synchronous acquisition of the triaxial vibration signal generated by each click, a total of 9×3×10=270 positive sample data segments were obtained.

[0039] Step 102: Collect negative sample data.

[0040] Collect negative sample vibration signals representing various non-touch interference events, including but not limited to the following types: a. When the vehicle is idling, continuously collect vibration signals for a preset duration (e.g., 10 seconds); b. When the vehicle is driving on a bumpy road, continuously collect vibration signals for a preset duration (e.g., 10 seconds); c. Perform the door closing operation a preset number of times (e.g., 10 times) and collect the impact vibration signal at the moment the door closes; d. Perform a preset number of taps on the screen border (e.g., 10 times) and collect the resulting vibration signals; e. When the car audio system plays music, it continuously collects vibration signals for a preset duration (e.g., 10 seconds).

[0041] Step 103: Feature extraction.

[0042] For each acquired positive and negative sample segment, the signal is divided into frames of fixed length L and frame shift M. Each frame is used as a basic processing unit to extract a multi-dimensional feature vector. Specifically, the continuous triaxial vibration signal stream is divided into frames of length L = 400 sampling points (corresponding to a time window T = 20 ms), with a frame shift of M = 200 sampling points between adjacent frames (i.e., 10 ms overlap). For each frame, a feature vector containing time-domain features, frequency-domain features, cepstral features, and spatial features is extracted, totaling 45 dimensions.

[0043] The specific feature extraction process is as follows: a. Extracting temporal features (6 dimensions) a1. Extracting Peak Amplitude Features (3D): Calculate the maximum absolute value of the X, Y, and Z axis signals within the frame to obtain A. x A y A z .

[0044] a2. Extract rise time features (1D): Determine the time t required for each axis signal to rise from its 10% peak value to its 90% peak value. r,x ,t r,y ,t r,z The minimum value among the three is taken as the rise time feature t of the frame. r .

[0045] a3. Extract the decay time constant feature (1D): Determine the time τ required for each axis signal to decay from its peak value to its 1 / e peak value. x ,τ y ,τ zThe minimum value among the three is taken as the decay time constant feature τ of the frame.

[0046] a4. Extract pulse width features (1D): Calculate the duration w of the signal amplitude exceeding 50% of its peak value for each axis. x ,w y ,w z The maximum value among the three is taken as the pulse width feature w of the frame.

[0047] b. Extract frequency domain features (14 dimensions) Hanning windows are applied to the X, Y, and Z axis signals of each frame, and N is performed. FFT The 512-point Fast Fourier Transform (FFT) yields the corresponding spectra X(f), Y(f), and Z(f). Three frequency bands of interest are defined: low-frequency band B1 from 50Hz to 200Hz, mid-frequency band B2 from 500Hz to 1500Hz, and high-frequency band B3 from 2000Hz to 5000Hz.

[0048] b1. Extracting frequency band energy proportion features (9 dimensions): For each signal in the X, Y, and Z axes, calculate the proportion of its energy in the above three frequency bands to the total energy of that axis. The calculation formula is as follows:

[0049] Where i represents the axis index of the triaxial accelerometer, j represents the frequency band index (j=1,2,3 correspond to low frequency band B1, mid frequency band B2 and high frequency band B3 respectively), and f s Indicates the sampling frequency. The spectrum represents the i-axis signal.

[0050] This yields a 3-axis × 3-band = 9-dimensional feature.

[0051] b2. Extracting spectral flatness features (3D): First, calculate the amplitude S(f) of the triaxial synthesized spectrum. The formula is as follows:

[0052] Then for each frequency band B j (j=1,2,3), calculate its spectral flatness. j Spectral flatness is defined as the ratio of the geometric mean to the arithmetic mean of the spectral amplitudes within a frequency band. The formula is:

[0053] Where, N j For frequency band B j The number of frequency points within.

[0054] This yields 3D features.

[0055] b3. Extracting the main frequency parameter features (2D): In the full bandwidth (0 to f) of the triaxial synthesized spectrum S(f) s Within the range of / 2), determine the frequency f corresponding to the point of maximum amplitude. peak And calculate the -3dB bandwidth B at that peak. peak , as a 2D feature.

[0056] c. Extract cepstral features (13 dimensions) Take the logarithm of the amplitude of the triaxial synthesized spectrum S(f), and then perform an inverse Fourier transform to obtain the cepstral coefficients c[n]. The calculation formula is as follows:

[0057] Where S(k) is the synthesized spectral amplitude at discrete frequency point k, corresponding to the discrete sampling of S(f).

[0058] Take the first 12 order cepstral coefficients c[1] to c

[12] and the 0th order energy term c[0], for a total of 13 features.

[0059] d. Extract spatial features (12 dimensions) Based on signals collected by four vibration sensors S1, S2, S3, and S4, features reflecting the spatial distribution of the vibration source are extracted. The vibration event timestamp for each sensor channel is defined as: when the signal amplitude of that channel exceeds a preset activity detection threshold (e.g., calibrated to 1×10⁻⁶ based on the sensor's static noise level). - ³) And the time point corresponding to when the energy reaches a local maximum. All sensors are sampled synchronously by the same analog-to-digital converter to ensure time consistency across channels.

[0060] d1. Extracting arrival time difference features (3D): Using sensor S1 as a reference, calculate the difference in vibration event timestamps between it and the other sensors to obtain Δt. 12 =t1-t2, Δt 13 =t1-t3, Δt 14 =t1-t4.

[0061] d2. Extract energy ratio features (3D): Calculate the ratio of the peak energies e2, e3, e4 of sensors S2, S3, and S4 to the peak energy e1 of the reference sensor S1, i.e., e2 / e1, e3 / e1, e4 / e1.

[0062] d3. Extract the principal direction vector and confidence features (3-dimensional): Let e z,iThis represents the total energy of the Z-axis signal from the i-th sensor within the current event frame (i.e., the sum of the squares of the amplitudes at each sampling point). This is calculated using the known coordinates (x, y) of each sensor. i ,y i The weighted average of the touch point and its Z-axis energy is used to estimate the projected coordinates of the touch point on the screen plane. , The calculation formula is as follows:

[0063]

[0064] Simultaneously, the location confidence level c, used to characterize the reliability of this location estimation result, is calculated. pos The calculation formula is:

[0065] Where, μ e σ e The Z-axis energy values ​​(e) of the four sensors are respectively z,1 , e z,2 , e z,3 , e z,4 The mean and standard deviation of (). If Then directly let c pos = 0.

[0066] d4. Extract waveform similarity features (3D): Calculate the Z-axis signal s1(t) of the reference sensor S1 and the signals of the other sensors S1 and S2. i Z-axis signal s (i=2,3,4) i The normalized cross-correlation peak value ρ between (t) and 1i The calculation formula is as follows:

[0067] Step 104: Positive sample modeling.

[0068] The set of all positive sample feature vectors extracted in step 103 is used as training data, and a Gaussian Mixture Model (GMM) is used for probability density estimation. The number of Gaussian components is set to K=5, and the covariance matrix of each component is in diagonal form. The mean of each component is initialized using the K-means++ algorithm, and iterative optimization is performed using the Expectation-Maximization (EM) algorithm until the change in the log-likelihood function value is less than a preset threshold (e.g., 1e-5). After training, a positive class model is obtained. Its parameter set is , where μ k , Σ k w k Let be the mean vector, covariance matrix, and weight of the k-th Gaussian component, respectively, and satisfy . .

[0069] Step 105: Negative sample modeling.

[0070] The set of all negative sample feature vectors extracted in step 103 is used as training data, and a single Gaussian model is used for modeling. The mean vector μ of this Gaussian distribution is calculated through maximum likelihood estimation. bg The covariance matrix Σ bg Covariance matrix Σ bg It also uses a diagonal matrix form. The negative class model is denoted as... .

[0071] After training, the above positive class model Parameters and negative class model The parameters are stored in the system's template library storage module.

[0072] Step 2: Real-time signal preprocessing and event detection.

[0073] After the system is powered on, it continuously preprocesses the vibration sensor signals in real time to detect potential touch events. This step specifically includes: Step 201: Signal framing.

[0074] A sliding window with a length of L=400 sampling points is maintained in real time. A new signal frame is generated every time the window slides forward by M=200 sampling points.

[0075] Step 202: Background noise energy estimation.

[0076] Establish a circular buffer of historical frame energy with a length of 5 seconds. During the initial 5-second phase after system startup, a fixed initial background energy threshold E is used. bg (e.g., 1e-3) Perform activity detection. Once the buffer is full, initiate dynamic background noise estimation.

[0077] For each newly generated frame, its short-time energy E is first calculated using the following formula:

[0078] Where x[n], y[n], and z[n] are the X, Y, and Z axis acceleration values ​​of the nth sampling point in the frame, respectively.

[0079] Then, the 10% of historical frames with the lowest short-term energy values ​​are selected from the buffer, and the average energy of these frames is calculated. Finally, a first-order recursive filter is used to update the background noise energy estimate at time t. The updated formula is as follows:

[0080] Where α is the forgetting factor, typically taking the value of 0.95.

[0081] Step 203: Activity detection.

[0082] Combine the short-time energy E of the current frame with the updated background noise energy estimate. Compare them. If satisfied... (where β is the energy multiplier threshold, typically 3), then the frame is marked as a potential touch event frame and sent to the feature extraction module for further processing; otherwise, the frame is discarded.

[0083] Step 3: Real-time matching and confidence calculation.

[0084] Upon detecting a potential touch event frame, the following real-time matching steps are performed: Step 301: Extract the feature vector of the current frame.

[0085] Extract the feature vector V of the current potential touch event frame using the same method described in step 103.

[0086] Step 302: Calculate the positive similarity.

[0087] The feature vector V is input into the positive class GMM model, and its likelihood L under each Gaussian component is calculated. k (V). For the k-th Gaussian component, the likelihood is calculated as follows:

[0088] Where d is the dimension of the feature vector. Represents the covariance matrix Σ k The determinant of .

[0089] The likelihood of each component is assigned according to its weight w k We perform a weighted summation to obtain the positive class composite score S. pos :

[0090] Step 303: Calculate the negative similarity probability.

[0091] The feature vector V is input into the negative class single Gaussian model, and its likelihood S is calculated. neg The calculation formula is:

[0092] Step 304: Calculate the vibration confidence level.

[0093] Based on the positive class composite score S pos Similar to negative S negCalculate a normalized vibration confidence level C. vib This is used to characterize the probability that the current vibration event was caused by a real touch. The calculation formula is:

[0094] C vib The value range is [0,1]. The larger the value, the closer the current vibration signal is to the vibration fingerprint of a real touch.

[0095] Step 305: Obtain vehicle status parameters.

[0096] The system obtains the current vehicle speed v (unit: km / h) and engine speed r (unit: rpm) in real time through the vehicle bus (such as CAN bus).

[0097] Step 306: Dynamically adjust the judgment threshold.

[0098] Based on the acquired vehicle state parameters, the vibration confidence baseline threshold used for subsequent comparisons is dynamically compensated to generate a dynamic judgment threshold C. th The calculation formula is:

[0099] Among them, C base The base threshold is typically 0.7. v and k r These are compensation coefficients greater than zero. These two coefficients are determined through a real-vehicle calibration process. The calibration objective is to minimize the sum of the false alarm rate and the false alarm rate under different combinations of vehicle speed and engine speed. An exemplary calibration result could be k. v =0.005, k r =0.0001.

[0100] Step 4: Decision on fusion with TDDIC coordinate signal.

[0101] The main control integrated circuit (IC) simultaneously receives touch coordinate data reported from TDDIC (reporting frequency, for example, 100Hz). For each newly received coordinate point (x... tddic ,y tddic The following fusion decision steps will be executed: Step 401: Time window alignment verification.

[0102] Set a time verification window for the currently received TDDIC coordinates [t] coord -Δt,t coord +Δt], where t coordThe coordinate reporting time is Δt, which is a preset time tolerance, typically 15ms. Check if at least one vibration confidence level C calculated in step 304 exists within this time window. vib Greater than the dynamic judgment threshold C calculated in step 306 th Vibration events.

[0103] If multiple vibration events satisfy the conditions, then select the one with confidence level C. vib The highest-ranking event is selected as a candidate associated event and proceeds to step 402 for spatial verification. If no vibration event meets the conditions within the time window, the current TDDIC coordinate is determined to be "not verified by vibration" and is discarded without responding to any touch operation.

[0104] Step 402: Spatial consistency verification.

[0105] For candidate vibration events verified by time window alignment, obtain the estimated location of the vibration source calculated in part d3 of step 103. , ) and its location reliability c pos .

[0106] Calculate the estimated location and the coordinates reported by TDDIC (x tddic ,y tddic The Euclidean distance d between them:

[0107] The conditions for passing a space consistency verification are defined as the simultaneous satisfaction of the following two conditions: a. The Euclidean distance d is less than or equal to the preset distance threshold D. th D th It can be set to 10% of the diagonal length of the touchscreen screen; b. Locational reliability pos Greater than 0.5.

[0108] Step 403: Calculate the fusion confidence score.

[0109] Based on the results of spatial consistency verification, the original vibration confidence level C was... vib Adjustments are made to generate the fusion confidence score C. fusion .

[0110] If the spatial consistency verification passes, the confidence level is increased, calculated using the following formula:

[0111] If the spatial consistency verification fails, a penalty confidence score is applied, calculated using the following formula:

[0112] Among them, coefficients 0.3 and 0.5 are the optimal values ​​obtained through experimental calibration, which are used to control the reward intensity when the space is consistent and the penalty intensity when the space is inconsistent, respectively.

[0113] Step 404, Touch Response Decision.

[0114] Based on all the above information, a final decision is made. If both of the following conditions are met simultaneously, the current TDDIC coordinate is determined to be a real touch event, and the system's main control IC responds to the operation corresponding to that coordinate: a. There are TDDIC coordinates that have been verified by time window alignment; b. Calculated fusion confidence level C fusion The response value is greater than the preset response threshold, for example, 0.6.

[0115] Otherwise, the TDDIC coordinates are determined to be false touches caused by electromagnetic interference, and the system ignores the coordinates.

[0116] In another embodiment of the invention, the method further includes the following after step 4: Step 5: Online adaptive learning.

[0117] To maintain and optimize system performance during long-term use, this embodiment also includes an online adaptive learning step, as detailed below: Step 501: Update the positive class model.

[0118] When the system responds to a real touch event and the user does not undo, the feature vector V extracted from this event can be considered a high-quality new positive sample. This sample is then used to fine-tune the Gaussian component in the positive class GMM model that best matches it. The method for determining the best-matching component is as follows: calculate the posterior probability of feature vector V under each Gaussian component, and select the component with the highest probability value as the component to be updated. The mean vector μ of this component is then calculated using exponential smoothing. k When updating, use a small learning rate (e.g., 0.01) to avoid model mutation caused by a single abnormal sample.

[0119] Step 502: Update the negative class model.

[0120] If the system experiences a false trigger, and this false trigger is manually undone by the user via a physical button or other means, the feature vector V of the vibration event that caused the false trigger can be considered as a new negative sample. This sample is then used to calculate the mean vector μ of the negative class single Gaussian model. bg The update can also be performed using the exponential smoothing method, with the covariance matrix Σ... bg It can remain unchanged.

[0121] Through the above online learning mechanism, the system can continuously adapt to changes in vehicles and usage environments, and continuously improve the accuracy and robustness of preventing accidental touches.

[0122] The technical means disclosed in this invention are not limited to those disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications are also considered within the scope of protection of this invention.

Claims

1. A method for preventing accidental touches on an in-vehicle touchscreen, characterized in that, include: Real-time acquisition of solid-borne vibration signals caused by touch or interference; The collected vibration signals are preprocessed by framing and potential touch event frames are detected. A multi-dimensional feature vector is extracted from the detected potential touch event frames. The feature vector includes time-domain features, frequency-domain features, and spatial features reflecting the signal correlation between different vibration sensors. The feature vector is matched with pre-constructed positive and negative class templates to calculate the vibration confidence level, which characterizes the possibility that the vibration signal originates from a real touch. Obtain vehicle state parameters and dynamically adjust the confidence threshold for determining valid vibration events based on the vehicle state parameters; Receive touch coordinates reported from the touch display integrated chip, and verify whether the touch coordinates are time-aligned with vibration events that meet the confidence threshold; For time-aligned vibration events, the location of the vibration source is estimated based on the spatial characteristics, and spatial consistency is verified with the touch coordinates. Based on the results of the time alignment and spatial consistency verification, the vibration confidence is adjusted to generate a fusion confidence, and the touch coordinates are determined to be real touches based on the fusion confidence.

2. The method according to claim 1, characterized in that, The extraction of multi-dimensional feature vectors from detected potential touch event frames includes: Extract time-domain features, including the peak amplitude of the X, Y, and Z axis signals, the minimum rise time of the three axis signals, the minimum decay time constant of the three axis signals, and the maximum pulse width of the three axis signals; Extract frequency domain features, including the energy proportion of each axis signal in the preset low frequency band, mid frequency band and high frequency band, the spectral flatness of each frequency band, and the main frequency and bandwidth of the synthesized spectrum; Extract cepstral features, including cepstral coefficients of a preset order obtained from the inverse Fourier transform after the logarithmic transform of the synthesized spectrum; Spatial features are extracted, including the time difference of arrival based on the reference sensor, the energy ratio of each sensor signal, the vibration source location and location confidence estimated based on the Z-axis energy weighting of each sensor, and the normalized cross-correlation peak value of the reference sensor and other sensor signals.

3. The method according to claim 1, characterized in that, The positive class template is a Gaussian mixture model trained based on positive sample feature vectors, and the negative class template is a single Gaussian model trained based on negative sample feature vectors. The calculation of vibration confidence includes: Calculate the weighted likelihood score of the feature vector under the positive Gaussian mixture model; Calculate the likelihood score of the feature vector under the negative class single Gaussian model; The normalized vibration confidence score is calculated based on the ratio of the weighted likelihood score to the negative class model likelihood score.

4. The method according to claim 1, characterized in that, The vehicle state parameters include vehicle speed and engine speed, and the step of dynamically adjusting the confidence threshold includes: According to the formula Calculate the dynamic decision threshold C th , where C base The base threshold is v, where v is vehicle speed, r is engine speed, and k is the engine speed. v and k r This is the preset positive compensation coefficient.

5. The method according to claim 1, characterized in that, The step of adjusting the vibration confidence score to generate a fusion confidence score based on the spatial consistency verification result includes: If the distance between the estimated location of the vibration source and the touch coordinates is less than or equal to a preset distance threshold, and the location confidence level is greater than a preset confidence threshold, then spatial consistency is determined, and the formula is used. Generate fusion confidence C fusion C vib For vibration confidence, c pos For location reliability, 'a' is the preset reward coefficient; Otherwise, determine if the space is inconsistent and use the formula. Generate a fusion confidence score, where b is a preset penalty coefficient.

6. The method according to any one of claims 1 to 5, characterized in that, It also includes online adaptive learning steps: Once a real touch is determined, the feature vector of the current event is used as a new positive sample, and the parameters of the model component that best matches it in the positive class template are fine-tuned and updated. When the system is falsely triggered and receives a user's cancellation command, the feature vector that caused the false trigger will be used as a new negative sample to update the model parameters of the negative class template.

7. A device for preventing accidental touches on an in-vehicle touchscreen, characterized in that, include: The vibration sensor array includes multiple triaxial accelerometers mounted on the touchscreen cover for real-time acquisition of solid-borne vibration signals caused by touch or interference. The signal acquisition and preprocessing module is used to perform frame-by-frame preprocessing on the vibration signal and detect potential touch event frames; The feature extraction module is used to extract multi-dimensional feature vectors from detected potential touch event frames. The feature vectors include time-domain features, frequency-domain features, and spatial features reflecting the signal correlation between different vibration sensors. The template library storage module is used to store pre-built positive class templates and negative class templates; The real-time matching module is used to match the feature vector with the positive class template and the negative class template to calculate the vibration confidence. The vehicle status acquisition module is used to acquire vehicle status parameters and provide them to the real-time matching module to dynamically adjust the confidence threshold. The fusion decision module is used to receive touch coordinates reported from the touch display integrated chip and perform the following operations: verify whether the touch coordinates are time-aligned with a vibration event that meets the confidence threshold. For time-aligned vibration events, the vibration source location is estimated based on the spatial characteristics and spatially consistent with the touch coordinates; the vibration confidence is adjusted based on the verification results to generate a fusion confidence, and the touch coordinates are determined to be a real touch based on the fusion confidence.

8. The apparatus according to claim 7, characterized in that, The feature extraction module is configured as follows: Extract time-domain features, including the peak amplitude of the X, Y, and Z axis signals, the minimum rise time of the three axis signals, the minimum decay time constant of the three axis signals, and the maximum pulse width of the three axis signals; Extract frequency domain features, including the energy proportion of each axis signal in the preset low frequency band, mid frequency band and high frequency band, the spectral flatness of each frequency band, and the main frequency and bandwidth of the synthesized spectrum; Extract cepstral features, including cepstral coefficients of a preset order obtained from the inverse Fourier transform after the logarithmic transform of the synthesized spectrum; Spatial features are extracted, including the time difference of arrival based on the reference sensor, the energy ratio of each sensor signal, the vibration source location and location confidence estimated based on the Z-axis energy weighting of each sensor, and the normalized cross-correlation peak value of the reference sensor and other sensor signals.

9. The apparatus according to claim 7, characterized in that, The fusion decision module is configured as follows: If the distance between the estimated location of the vibration source and the touch coordinates is less than or equal to a preset distance threshold, and the location confidence level is greater than a preset confidence threshold, then spatial consistency is determined, and the formula is used. Generate fusion confidence C fusion C vib For vibration confidence, c pos For location reliability, 'a' is the preset reward coefficient; Otherwise, determine if the space is inconsistent and use the formula. Generate a fusion confidence score, where b is a preset penalty coefficient.

10. The apparatus according to any one of claims 7 to 9, characterized in that, It also includes an adaptive learning module, used for: When the fusion decision module determines that it is a real touch, it uses the feature vector of the current event as a new positive sample to fine-tune and update the positive class template parameters in the template library storage module. When the system is falsely triggered and receives a user's cancellation command, the feature vector that caused the false trigger will be used as a new negative sample to update the negative class template parameters in the template library storage module.

Images