Steering wheel alarm method and system based on HOD hand-off detection

Abstract

The present application belongs to the technical field of vehicle safety driving assistance, and particularly relates to a steering wheel alarm method and system based on HOD hand-off detection, which comprises the following steps: collecting steering column vibration signals, wheel speed signals and steering torque signals; determining a band-pass filter set by using a spectral kurtosis maximization criterion, and screening effective sub-bands by second-order cyclostationarity; performing envelope spectrum analysis on the effective sub-bands, setting a search interval centered on wheel rotation orders and their harmonics, extracting spectral peak amplitudes and performing weighted fusion to obtain hand-related characteristic quantities; then adaptively fusing the characteristic quantities and steering torque sliding variances according to vehicle speed and steering wheel rotation rate to generate a hand-off discriminant value; when the hand-off discriminant value is lower than a hand-off threshold compensated by the number of effective sub-bands within a continuous time window, it is determined that the steering wheel is off the hands, and a hierarchical alarm is triggered according to the duration. The present application can improve the accuracy of hand-off detection and the stability of the alarm.

Description

Technical Field

[0001] This invention relates to the field of vehicle safety driving assistance technology. More specifically, this invention relates to a steering wheel alarm method and system based on HOD (Hands-of-Driving Detection). Background Technology

[0002] Steering wheel hands-off detection is typically based on capacitive sensing or steering torque detection. Capacitive solutions require an internal sensing structure within the steering wheel, increasing manufacturing costs and structural complexity, and are susceptible to factors such as in-vehicle temperature and humidity, driver hand moisture, and the wearing of gloves. Single steering torque detection solutions rely on changes in the torque applied to the steering wheel by the driver; however, at high speeds with straight-line driving and hands lightly on the wheel, the input torque is small, leading to potential misjudgments. Furthermore, at low speeds and large steering angles, the road feedback torque, steering assist torque, and driver's hand operating torque are coupled, further affecting the accuracy of hands-off detection. Therefore, detection methods relying solely on torque signals or a single sensor are insufficient to meet the demands of hands-off detection under complex driving conditions.

[0003] By fusing the steering column vibration signal with the steering torque signal, the mechanical response features related to hand contact can be extracted by utilizing the added mass and damping effect created when a person holds the steering wheel. Envelope spectrum analysis of the vibration signal can obtain modulation and attenuation information of the wheel rotation excitation during transmission. However, changes in vehicle speed cause frequency drift and order ambiguity in the time-domain vibration signal. Without an angle-domain resampling mechanism based on wheel speed, the stability of subsequent feature extraction will be affected. Furthermore, existing filtering and demodulation frequency bands mostly rely on fixed settings or empirical selection, lacking an adaptive frequency band determination method based on the spectral kurtosis maximization criterion, and also failing to introduce a processing mechanism that uses a second-order cyclostationarity index to eliminate invalid noise subbands.

[0004] Furthermore, in the envelope spectrum harmonic feature extraction process, if a fixed search interval is used and different order harmonics are not weighted, the influence of spectral kurtosis on the peak concentration and the differences in contribution of different order harmonics can easily be overlooked. In the multi-source feature fusion and state discrimination stage, if fixed fusion weights and static alarm thresholds are used, it is difficult to adaptively adjust the discrimination strategy according to changes in vehicle speed, steering wheel angle rate, and the number of effective sub-bands, thereby increasing the risk of false alarms or missed alarms. Summary of the Invention

[0005] To address the technical problems of poor anti-interference of single sensors and frequency ambiguity caused by changes in vehicle speed, this invention proposes a steering wheel alarm method and system based on HOD (Hands-off Detection), which can improve the accuracy of hands-off detection and the stability of alarms.

[0006] In a first aspect, the present invention provides a steering wheel alarm method based on HOD (Hands-Off Detection) detection, comprising: S1, acquiring steering column vibration signal, wheel speed signal, and steering torque signal; resampling the vibration signal in the angle domain according to the wheel speed signal to obtain order domain signals with the same angle interval; S2, decomposing the order domain signals through a bandpass filter bank whose center frequency and bandwidth are determined by the spectral kurtosis maximization criterion, extracting the envelope of each sub-band, calculating the second-order cyclic stability index, and removing sub-bands whose index is lower than a set threshold; S3, taking the square of the modulus of each remaining sub-band envelope to obtain a square envelope, subtracting the local mean, and performing a Fourier transform to obtain the envelope spectrum of each sub-band; S4, in each sub-band envelope spectrum, taking the first-order wheel rotation and... S5. A narrow-band search interval is set with the harmonic as the center, and the width of the interval is adjusted inversely by the sub-band spectrum kurtosis value; S6. The peak amplitude values ​​of each order harmonic spectrum are extracted in each sub-band search interval, and the sum is weighted according to the order attenuation weight that decreases with the increase of the harmonic order. The results of each remaining sub-band are then weighted and fused according to the second-order cyclic stability index to obtain the hand-related feature quantity; S7. The hand-related feature quantity and the steering torque sliding variance are weighted and fused to obtain the hand-off discrimination value. The fusion weight is determined by the vehicle speed and the steering wheel turning rate; S8. When the hand-off discrimination value is lower than the hand-off threshold within a continuously set time window, the steering wheel is judged to be off and a graded alarm is triggered. The alarm level is determined by the duration of the value being below the threshold. The hand-off threshold is compensated for with safety sensitivity based on the number of remaining sub-bands.

[0007] By adopting the above technical solution, the vibration signals of the steering column, wheel speed, and steering torque are collected and resampled in the angle domain to reduce the impact of vehicle speed changes on vibration frequency analysis. The bandpass filter bank is determined by the spectral kurtosis maximization criterion, and effective sub-bands are selected by combining the second-order cyclic stability index to reduce background noise interference. The envelope spectrum harmonic features are weighted and fused to obtain the hand-related feature quantity, and adaptively fused with the steering torque sliding variance to generate the hand-off discrimination value. Combined with continuous time window judgment and threshold safety sensitivity compensation, a graded alarm is realized, thereby improving the overall accuracy and alarm stability of steering wheel hand-off detection.

[0008] Preferably, the step of resampling the vibration signal in the angle domain based on the wheel rotation speed signal to obtain an order domain signal with the same angle interval includes: extracting the pulse sequence from the wheel rotation speed signal, calculating the time interval between adjacent pulses to obtain the instantaneous rotation speed; integrating time based on the instantaneous rotation speed to obtain the cumulative angle value corresponding to each vibration sampling point; setting a fixed angle sampling interval, and using a spline interpolation algorithm to interpolate the vibration signal in the time domain into a vibration signal sequence with the same angle interval distribution, and using the sequence as the order domain signal.

[0009] By adopting the above technical solution, the instantaneous rotational speed is calculated by extracting the pulse sequence of the wheel rotational speed signal, and the cumulative angle value is obtained based on the integral of the instantaneous rotational speed. Then, the time-domain vibration signal is converted into an order-domain signal with equal angular intervals by spline interpolation with a fixed angle sampling interval, which effectively eliminates the frequency drift and order ambiguity caused by changes in vehicle speed and improves the stability of subsequent feature extraction.

[0010] Preferably, the step of decomposing the order domain signal through a bandpass filter bank with center frequency and bandwidth determined by the spectral kurtosis maximization criterion includes: calculating a fast spectral kurtosis map of the order domain signal; calculating the spectral kurtosis value of the bandpass filtered signal under different combinations of center frequency and bandwidth; identifying multiple feature nodes with local maxima of spectral kurtosis value in the fast spectral kurtosis map; obtaining the center frequency and bandwidth parameters corresponding to each feature node; constructing a set of finite impulse response bandpass filters based on the extracted center frequency and bandwidth parameters; and inputting the order domain signal into the bandpass filter bank to obtain multiple independent sub-band signals.

[0011] By adopting the above technical solution, a fast spectral kurtosis map of the order domain signal is calculated, feature nodes with local maxima of spectral kurtosis are identified, and center frequency and bandwidth parameters are extracted. Based on this, a finite impulse response bandpass filter bank is constructed to decompose the signal, realizing adaptive selection of feature frequency bands and avoiding the limitations caused by fixed frequency band settings or empirical selection.

[0012] Preferably, the step of extracting the envelope of each sub-band and calculating the second-order cyclostationarity index includes: performing a Hilbert transform on each sub-band signal to obtain an analytic signal, and taking the amplitude of the analytic signal to obtain the sub-band envelope; calculating the time-varying autocorrelation function of the sub-band envelope signal, and performing a two-dimensional Fourier transform on the time-varying autocorrelation function along the time axis and the time delay axis to obtain the cyclic spectral density function; extracting the maximum amplitude of the cyclic spectral density function in the cyclic frequency domain, and performing a ratio calculation between the amplitude and the average power spectral density amplitude at the non-cyclic frequency, and using the obtained ratio as the second-order cyclostationarity index.

[0013] By adopting the above technical solution, the envelope of the subband signal is extracted by Hilbert transform, the time-varying autocorrelation function is calculated and the two-dimensional Fourier transform is performed to obtain the cyclic spectral density function. The ratio of the maximum amplitude in the cyclic frequency domain to the amplitude of the average power spectral density at the non-cyclic frequency is used as the second-order cyclic stationarity index, which can effectively quantify the significance of the cyclic modulation component in the subband and facilitate the removal of invalid noise subbands.

[0014] Preferably, the step of extracting the peak amplitudes of each harmonic spectrum within the search interval of each sub-band, weighting and summing them according to the order attenuation weight that decreases with increasing harmonic order, and then weighting and fusing the results of each remaining sub-band according to the second-order cyclostationarity index to obtain the hand-related feature quantity includes: within the narrow band search interval of each harmonic spectrum of each remaining sub-band, searching for the local maximum point of the envelope spectrum amplitude as the peak amplitude of the corresponding order harmonic spectrum; multiplying the peak amplitudes of each harmonic spectrum within the same sub-band by the arithmetic square root of the reciprocal of the corresponding order, summing the product results to obtain the feature value of each sub-band; multiplying the feature value of each sub-band by the second-order cyclostationarity index of the sub-band, accumulating the product results of all remaining sub-bands, and dividing by the sum of the second-order cyclostationarity indices of all remaining sub-bands, and using the resulting quotient as the hand-related feature quantity.

[0015] By adopting the above technical solution, the peak amplitude of each harmonic spectrum is extracted within the narrow search interval where the kurtosis value of each sub-band spectrum is adjusted in the opposite direction. The peak amplitude is then weighted and summed according to the order attenuation weight. The results of each sub-band are then weighted and fused according to the second-order cyclic stationarity index to obtain the hand-related feature quantity. This can comprehensively reflect the differences in the contribution of different sub-bands and different order harmonics, and improve the stability and representativeness of the hand-related feature quantity.

[0016] Preferably, the step of weightedly fusing the hand-related feature quantity with the steering torque sliding variance to obtain the hand-off discrimination value, wherein the fusion weight is determined by the vehicle speed and the steering wheel turning rate, includes: using the amplitude of the vehicle speed signal and the absolute value of the steering wheel turning rate as input variables of the fuzzy inference system; using the weight coefficients output by the fuzzy inference system to linearly weight and sum the normalized hand-related feature quantity and the normalized steering torque sliding variance to obtain the hand-off discrimination value.

[0017] Preferably, the step of outputting weight coefficients using the fuzzy inference system includes: outputting weight coefficients based on a fuzzy rule base, wherein the fuzzy rule base is set such that when the vehicle speed is below a preset low-speed threshold and the turning rate is above a preset large turning threshold, the weight coefficient of the output hand-related feature quantity is greater than the weight coefficient of the steering torque sliding variance; when the vehicle speed is above a preset high-speed threshold and the turning rate is below a preset small turning threshold, the weight coefficient is determined based on whether the steering torque sliding variance is within a preset effective fluctuation range; when the steering torque sliding variance is within the preset effective fluctuation range, the weight of the steering torque sliding variance is increased; when the steering torque sliding variance is below a preset effective fluctuation lower limit, the weight of the hand-related feature quantity is maintained.

[0018] Preferably, when the hand-off detection value is lower than the hand-off threshold within a continuously set time window, the steering wheel is determined to be out of control and a graded alarm is triggered. The alarm level is determined by the duration of the value being below the threshold, including: continuously monitoring the hand-off detection value; when the duration of the hand-off detection value being lower than the hand-off threshold reaches the first-level alarm time threshold, triggering a visual prompt alarm on the instrument panel; when the duration reaches the second-level alarm time threshold, triggering a combined visual and auditory alarm; when the duration reaches the third-level alarm time threshold, triggering a pre-tensioned seat belt vibration intervention and activating the hazard warning flasher.

[0019] Preferably, the dropout threshold is used for safety sensitivity compensation based on the number of remaining sub-bands, including: setting a baseline dropout threshold, counting the number of remaining sub-bands currently participating in feature extraction, and calculating the ratio of the number of remaining sub-bands to the total number of sub-bands initially decomposed in the bandpass filter bank; determining a threshold compensation coefficient based on the ratio, and correcting the baseline dropout threshold based on the threshold compensation coefficient to obtain the current dropout threshold.

[0020] Secondly, the present invention provides a steering wheel alarm system based on HOD (Hands-Off Detection) detection, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned steering wheel alarm method based on HOD detection is implemented.

[0021] By adopting the above technical solution, the above-mentioned steering wheel alarm method based on HOD hands-off detection is generated into a computer program and stored in a memory so that it can be loaded and executed by a processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0022] The present invention has the following beneficial effects:

[0023] This invention resamples the steering column vibration signal in the angular domain based on the wheel rotation speed signal, reducing the impact of vehicle speed changes on vibration frequency analysis; it determines the characteristic frequency band by maximizing the spectral kurtosis criterion, and selects effective sub-bands by combining the second-order cyclic stability index, reducing the interference of background noise on feature extraction; and it adjusts the harmonic search interval width in reverse according to the spectral kurtosis, and performs weighted fusion of the results of each order harmonic spectral peak and different sub-bands to obtain stable hand-related feature quantities, thereby improving the accuracy and alarm stability of steering wheel hand-off detection.

[0024] Furthermore, by combining vehicle speed and steering wheel angle rate, differentiated weight allocation is applied to hand-related feature quantities and steering torque slip variance, enabling the hand-off judgment to adapt to different driving conditions such as high-speed straight driving and low-speed steering; based on the continuous time window judgment mechanism and the hand-off threshold adjustment strategy that compensates for safety sensitivity according to the number of effective sub-bands, the risk of misjudgment caused by transient signal fluctuations is reduced, and graded alarms are triggered according to the duration of hand-off. Attached Figure Description

[0025] Figure 1 This is a flowchart of a steering wheel alarm method based on HOD (Hands-Off Detection) detection. Figure 2 This is a schematic diagram of the original time-domain vibration signal of the steering column of the steering wheel; Figure 3 This is a diagram comparing the performance of different hand-release detection schemes. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0027] One or more embodiments of the present invention provide a steering wheel alarm method based on HOD (Hands-Off Detection) for steering wheel warning, such as... Figure 1 As shown, it includes the following steps: S1 collects the vibration signal of the steering column, the wheel speed signal, and the steering torque signal. Based on the wheel speed signal, the vibration signal is resampled in the angle domain to obtain the order domain signal with the same angle interval.

[0028] High-frequency vibration signals are continuously acquired by a piezoelectric accelerometer mounted on the steering column. Wheel speed signals from the wheel speed sensor and steering torque signals from the torque sensor are read in real-time via the vehicle control local area network bus. A synchronous data acquisition card ensures the time synchronization of multi-channel signals. For the angle domain resampling process, the cumulative trapezoidal integral algorithm is used to process the wheel speed signal sequence to calculate the instantaneous cumulative rotation angle of the wheel. An interpolation algorithm is used to construct a mapping relationship between the original time series and the cumulative rotation angle sequence. A fixed angle increment constant is used as the resampling interval. A one-dimensional linear interpolation function is used to calculate the vibration signal amplitude at corresponding equal-angle interval angle nodes, thereby converting the discrete-time domain vibration signal into an equal-angle interval order domain signal.

[0029] In an optional embodiment, the vibration signal is resampled in the angle domain based on the wheel rotation speed signal to obtain an order domain signal with the same angular interval, including: Extract the pulse sequence from the wheel speed signal and calculate the time interval between adjacent pulses to obtain the instantaneous speed; By integrating time with respect to the instantaneous rotational speed, the cumulative angle value corresponding to each vibration sampling point is obtained; By setting a fixed angle sampling interval, the vibration signal in the time domain is interpolated into a vibration signal sequence with the same angle interval distribution using a spline interpolation algorithm, and the sequence is used as the order domain signal.

[0030] The system acquires rotational speed signals in real time within the frequency range of 0-2000Hz, extracting rising or falling edges to obtain continuous pulse sequences. Based on the 48 pulses output per revolution of the sensor, a timer is used to calculate the time interval between the arrival of two adjacent pulses. Calculate the instantaneous angular velocity of the wheel .

[0031] Perform time integration on the instantaneous angular velocity curve to obtain time-domain sampling points synchronized with a high-frequency vibration accelerometer with a sampling frequency set to 12.8 kHz or 25.6 kHz. Corresponding cumulative rotation angle value Based on the Nyquist-Shannon sampling theorem and the target highest analysis order covering up to the 100th harmonic, the number of resampling points N per cycle is set to 1024, thereby determining a fixed angular sampling interval. .

[0032] Based on cumulative angle values The independent variable is the amplitude of the original vibration signal. As the dependent variable, a piecewise cubic polynomial curve function is constructed using the cubic spline interpolation algorithm, and the curve is plotted at equally spaced angled grid nodes. The signal is resampled and output as a sequence of vibration signals with equal angular intervals. The original time-domain vibration signal of the steering column is shown below. Figure 2 As shown.

[0033] S2 decomposes the order domain signal through a bandpass filter bank whose center frequency and bandwidth are determined by the spectral kurtosis maximization criterion, extracts the envelope of each sub-band, calculates the second-order cyclic stationarity index, and removes sub-bands whose index is lower than the set threshold.

[0034] A signal processing program based on the FastKurtogram algorithm is invoked to perform time-frequency analysis on the order domain signal. Spectral kurtosis values ​​at multiple scale levels and in different frequency bands are calculated. Multiple non-overlapping characteristic frequency band nodes that meet threshold requirements are extracted through local maximum search or sorting by spectral kurtosis values. The center frequency and bandwidth parameters corresponding to each characteristic frequency band node are obtained. A finite impulse response bandpass filter bank is constructed based on these parameters, filter coefficients are generated, and the order domain signal is decomposed into multiple sub-band signals through multi-channel filtering. A Hilbert transform is applied to each sub-band signal to calculate the analytic signal. The absolute value of this analytic signal is used to extract the sub-band envelope of each frequency band. For the calculation of the second-order cyclostationarity index, a time-varying autocorrelation function is calculated for each sub-band envelope sequence, simultaneously retaining the time variable and the time delay variable. A Fourier transform is performed on the time-varying autocorrelation function to obtain the cyclic spectral density function. The amplitude of the spectral peak near the cyclic frequency corresponding to the fundamental frequency order of wheel rotation or its harmonics is extracted as the second-order cyclostationarity index of that sub-band. A constant is set as the minimum quality judgment threshold. The second-order cyclic stability index of each sub-band is compared with the value of the threshold. Sub-band signals with values ​​lower than the set threshold are deleted from the storage matrix to achieve the removal operation.

[0035] In an optional embodiment, the order-domain signal is decomposed using a bandpass filter bank whose center frequency and bandwidth are determined by the spectral kurtosis maximization criterion, including: Calculate the fast spectral kurtosis plot of the order domain signal, and calculate the spectral kurtosis value of the band-filtered signal under different combinations of center frequency and bandwidth; In the fast spectral kurtosis map, multiple feature nodes with local maxima of spectral kurtosis values ​​are identified, and the center frequency and bandwidth parameters corresponding to each feature node are obtained. Based on the extracted center frequency and bandwidth parameters, a set of finite impulse response bandpass filters is constructed. The order domain signals are then input into the bandpass filter group to obtain multiple independent sub-band signals.

[0036] A fast spectral kurtosis algorithm is employed, and the original signal is divided into binary or tri-band segments using short-time Fourier transform or a tree-structured multi-rate finite impulse response filter bank, with the decomposition level L set to 5 to 7 levels. At each decomposition node, the frequency derived from the center frequency is calculated. and bandwidth The spectral kurtosis value K of the frequency band signal is represented. K is set as the ratio of the fourth central moment to the square of the second central moment of the signal envelope amplitude within the target frequency band minus 3, and a three-dimensional spectral kurtosis map is constructed.

[0037] The two-dimensional local nonmaximum suppression algorithm is applied to locate local maxima of spectral kurtosis in the spectrum, with a preset judgment threshold. From higher Extract the first 3 to 5 non-overlapping extreme nodes from the peaks to obtain the center frequency, such as... Rank, Order, and bandwidth, such as Rank, Based on the aforementioned frequency and bandwidth parameters, shape parameters are used. A Caesar window is used, and a linear phase finite impulse response bandpass filter with an order M of 512 or 1024 and a stopband attenuation greater than 60dB is set. The order domain sequence is input into the above 3 to 5 groups of FIR bandpass filter groups to extract the subband signal.

[0038] In an optional embodiment, the envelope of each subband is extracted and a second-order cyclic stability index is calculated, including: Perform a Hilbert transform on each sub-band signal to obtain an analytic signal, and take the amplitude of the analytic signal to obtain the sub-band envelope; Calculate the time-varying autocorrelation function of the subband envelope signal, and perform a two-dimensional Fourier transform along the time axis and the time delay axis on the time-varying autocorrelation function to obtain the cyclic spectral density function; The maximum amplitude of the cyclic spectral density function is extracted in the cyclic frequency domain, and the ratio of the amplitude to the average power spectral density amplitude at non-cyclic frequencies is calculated. The resulting ratio is used as a second-order cyclic stability index.

[0039] Perform a Discrete Hilbert Transform on any of the separated sub-band signals x(t) to generate orthogonal time-domain components. Based on this, a complex analytic signal sequence is synthesized. Determine the absolute magnitude of the analytic signal. The subband envelope signal e(t) is obtained by using a low-pass anti-aliasing filter with a cutoff frequency set below the 20th order.

[0040] Calculate the envelope signal e(t) and utilize the time delay. Instantaneous autocorrelation function between subsequent sequences The superscript asterisk indicates complex conjugate; the autocorrelation function is analyzed using a two-dimensional fast Fourier transform on the time axis t and the time delay axis. The direction is changed to obtain the cycle frequency. The two-dimensional cyclic spectral density function matrix formed by the carrier frequency f In the process of eliminating After the DC plane, the maximum local energy peak of the cyclic spectral density function is extracted. At the same time Planar calculation of the average power spectral density amplitude at all frequency points Calculate the ratio. As a second-order cyclostationarity index for the current frequency band, it is used to characterize the significance of the cyclomodulation component relative to the background power in the corresponding subband.

[0041] S3. Take the square of the envelope of each remaining sub-band to obtain the square envelope. Subtract the local mean and then perform a Fourier transform to obtain the envelope spectrum of each sub-band.

[0042] The algorithm iterates through all the remaining sub-band envelope data matrices after filtering. For each floating-point element in the envelope data sequence, it performs a self-multiplication and exponentiation operation to obtain the modulus squared value, generating the corresponding sub-band squared envelope time series. A moving average filtering algorithm is used to calculate the local mean of this squared envelope. The moving average of the sequence is calculated by calling the `convolve` function in conjunction with a full-one rectangular window function. The original squared envelope sequence is then subtracted point-by-point from this local mean sequence to filter out interference from DC components and ultra-low frequency trend terms. A discrete fast Fourier transform is performed on the de-meaned time series signal, and the absolute value of the complex domain transform result is taken to calculate the true amplitude spectrum. The corresponding envelope spectrum data arrays for each remaining sub-band are output.

[0043] S4. In each sub-band envelope spectrum, a narrow band search interval is set with the first order of wheel rotation and harmonics as the center. The interval width is adjusted inversely by the sub-band spectrum kurtosis value.

[0044] Based on the prior-defined first-order frequency of wheel rotation as the fundamental frequency parameter, a sequence of positive integer multiples of this fundamental frequency parameter is generated as the center frequency positions of the second to higher harmonics. The spectral kurtosis feature values ​​of the corresponding sub-band recorded in the fast spectral kurtosis algorithm are extracted. An inverse proportional adjustment function is set to calculate the frequency interval width. The frequency interval width is the result of dividing the fundamental interval width with order units by a dimensionless adjustment term composed of the spectral kurtosis value and a smoothing constant. This ensures that when the code reads a large spectral kurtosis value, the frequency interval span calculated by the function becomes smaller, and when a small spectral kurtosis value is read, the frequency interval span calculated by the function becomes larger. Using the calculated center frequency positions of each harmonic as the midpoint and the interval width value output by the aforementioned inverse proportional adjustment function as the total span, a frequency index range matrix is ​​divided on the corresponding sub-band envelope spectrum frequency axis sequence using logical judgment statements of greater than or equal to and less than or equal to. This index range is used as the narrowband search interval for that harmonic signal.

[0045] S5. Extract the peak amplitude of each harmonic spectrum within each sub-band search interval, sum them by weighting according to the order attenuation weight that decreases with the increase of the harmonic order, and then fuse the results of each remaining sub-band by weighting according to the second-order cyclic stability index to obtain the hand-related feature quantity.

[0046] Within the narrow-band search intervals defined on the frequency axis, the local maximum peak points of the envelope spectrum amplitude sequence are identified as the spectral peak amplitudes of the corresponding harmonics. A one-dimensional array of order attenuation weight coefficients is constructed, where each element is the square root of the reciprocal of the natural number of the corresponding harmonic order index. A linear inner product algebra operation is performed between the extracted column vector of harmonic spectral peak amplitudes and the order attenuation weight coefficient vector to obtain the weighted summation scalar result within each sub-band. The second-order cyclic stability index numerical sequence of each remaining sub-band, calculated and retained in the previous steps, is retrieved and normalized by dividing this sequence by the sum of its elements to obtain the system fusion weight vector for each sub-band. The weighted summation scalar result of each sub-band is multiplied by the normalized fusion weight element at the corresponding mapping position, and all product results are summed and accumulated to calculate and output a scalar value that comprehensively represents the steering wheel vibration characteristics as a hand-related feature quantity.

[0047] In an optional embodiment, the peak amplitudes of each harmonic spectrum are extracted within each sub-band search interval, and weighted and summed according to an order attenuation weight that decreases with increasing harmonic order. The results of each remaining sub-band are then weighted and fused according to a second-order cyclostationarity index to obtain hand-related feature quantities, including: Within the narrow band search interval of each order of harmonics in each remaining subband, the local maxima of the search envelope spectrum amplitude are taken as the peak amplitude of the corresponding order of harmonic spectrum. The characteristic values ​​of each sub-band are obtained by multiplying the peak amplitude of each harmonic spectrum within the same sub-band by the arithmetic square root of the reciprocal of the corresponding order, and summing the product results. Multiply the eigenvalues ​​of each subband by the second-order cyclic stationarity index of the subband, sum the products of all remaining subbands, and divide by the sum of the second-order cyclic stationarity indices of all remaining subbands. Use the resulting quotient as the hand-related feature quantity.

[0048] First principal order and expand to =5th order harmonic order For the i-th remaining subband, define a narrow-band search interval in the envelope spectrum for the k-th harmonic. ; Interval half width Based on the spectral kurtosis value of the sub-band Inversely proportional adjustment, for example when =3.0 setting Step, when =1.5 setting The peak amplitude of the k-th harmonic spectrum is extracted within the interval using a cubic spline parabola fitting extreme point detection method. Extraction of a single subband The amplitude of each harmonic spectral peak is multiplied by the arithmetic square root weighting coefficient of the reciprocal of the corresponding order. Summing yields the characteristic scalar of the current frequency band. The second-order cyclic stability index of each subband is used. As a fusion coefficient, the feature scalar and After multiplication Accumulate the remaining subband intervals. Divide by the sum of the stability indices of all remaining sub-bands. Perform the normalization operation. Calculate the resulting quotient. As a hand-related feature quantity output with a period of 100ms.

[0049] S6, the hand-related feature quantity and the steering torque sliding variance are weighted and fused to obtain the hand-off discrimination value. The fusion weight is determined by the vehicle speed and the steering wheel turning rate.

[0050] The local sliding variance sequence of torque data within a specified time window is calculated from the raw steering torque signal acquired by the synchronous acquisition card. The current vehicle speed is parsed from the vehicle control local area network bus message, and the absolute steering wheel angle signal is read. Differential differentiation is performed on the absolute steering wheel angle signal to obtain the real-time steering wheel angle rate. A two-dimensional fuzzy logic controller mechanism is used to allocate feature weights, constructing a fuzzy inference system. The current vehicle speed and steering wheel angle rate are used as the state input variables of the fuzzy system. A fuzzy inference rule table is formulated such that when the input state is evaluated as low vehicle speed and high steering angle rate, the fuzzy inference output has a hand association weight coefficient greater than 0.5 and a torque variance weight coefficient less than 0.5. When the input state is assessed as high vehicle speed and low steering angle rate, if the steering torque slip variance is within the preset effective fluctuation range, the torque variance weight is appropriately increased; if the steering torque slip variance is below the preset effective fluctuation lower limit, the hand-related feature weight is maintained or a balanced weight is adopted to reduce the risk of false alarms under light steering wheel grip conditions. The algebraic sum of the two weight coefficients is constrained to always be equal to 1. The normalized hand-related feature calculated at the current moment is multiplied by the corresponding assigned hand-related weight coefficient, and the normalized steering torque slip variance is multiplied by the corresponding assigned torque variance weight coefficient. The two products are then summed to generate the system's hands-off discrimination value at the current sampling moment. The hand-related feature participating in the fusion is the feature after calibration mapping and direction consistency processing, and its value increases with the degree of hand grip correlation.

[0051] In an optional embodiment, a hand-related feature quantity and the steering torque sliding variance are weighted and fused to obtain a hand-off discrimination value. The fusion weight is determined by the vehicle speed and the steering wheel turning rate, including: A two-dimensional fuzzy inference system is established, with the amplitude of the vehicle speed signal and the absolute value of the steering wheel angular rate as the input variables of the fuzzy inference system; Based on the weight coefficients output by the fuzzy rule base, the fuzzy rule base is set such that when the vehicle speed is lower than the preset low speed threshold and the turning rate is higher than the preset large turning angle threshold, the weight coefficient of the output hand-related feature quantity is greater than the weight coefficient of the steering torque sliding variance. When the vehicle speed is higher than the preset high speed threshold and the turning rate is lower than the preset small turning angle threshold, the weight coefficient is determined based on whether the steering torque sliding variance is within the preset effective fluctuation range; when the steering torque sliding variance is within the preset effective fluctuation range, the weight of the steering torque sliding variance is increased; when the steering torque sliding variance is lower than the preset effective fluctuation lower limit, the weight of the hand-related feature quantity is maintained. By using weighted coefficients, a linear weighted sum is performed on the normalized hand-related features and the normalized steering torque slip variance to obtain the hand-off discrimination value.

[0052] Real-time acquisition of vehicle speed signal V with a range of 0-150km / h and absolute value of steering wheel angular rate with a range of 0-360° / s As input variables, fuzzification is performed using Gaussian or triangular membership functions. Based on a pre-set fuzzy rule base, it is determined that when V is below a low-speed threshold and... When the value exceeds the large turning angle threshold, output the weights of the hand-related features. The weighting of the sliding variance of the steering torque is in the range of 0.7 to 0.9. The range is 0.1 to 0.3; when V is higher than the high-speed threshold and When the steering angle is less than the small turning angle threshold, if the steering torque slip variance is within the preset effective fluctuation range, then It can be set to the range of 0.4 to 0.6. It can be set to the range of 0.4 to 0.6 or increased appropriately. If the steering torque slip variance is lower than the preset effective fluctuation lower limit, then It remains in the range of 0.5 to 0.7. Maintain a value between 0.3 and 0.5, and meet the conditions. The minimax limiting normalization method is employed to compare the hand-related features with a set length. The sliding variance of the steering torque within the ms sliding time window is scaled; according to the formula... Normalized hand features Normalize the torque variance, perform linear weighted summation, and output the fusion release discriminant value for a single-step cycle.

[0053] S7: When the hand-off judgment value is lower than the hand-off threshold within a continuously set time window, it is determined that the steering wheel is out of control and a graded alarm is triggered. The alarm level is determined by the duration of the time it is below the threshold. The hand-off threshold is compensated for with safety sensitivity based on the number of remaining sub-bands.

[0054] A compensation mapping function is established between the baseline dropout threshold and the reliability of the remaining sub-band quantity. The state variable of the current remaining sub-band quantity in memory is obtained. When the remaining sub-band quantity decreases, a threshold compensation coefficient is determined based on the ratio of the remaining sub-band quantity to the initial total number of sub-bands, and this is adjusted based on the baseline dropout threshold to ensure that the current dropout threshold is not lower than the baseline dropout threshold, thus avoiding a reverse decrease in alarm sensitivity when the effective sub-band quantity decreases. Simultaneously, a time window delay discrimination logic based on an accumulator counter is established. The dropout discrimination value updated at the current moment is read frame by frame at a fixed sampling frequency and compared with the calculated updated dropout threshold using a relational operator. If the dropout discrimination value is lower than the threshold, the global counter is incremented by one step; otherwise, the global counter variable is immediately reset to zero. The accumulated counter value is multiplied by the single sampling time period to convert it into a duration below the threshold in seconds. The preset time interval of this duration is determined. If the duration exceeds the first-level time limit threshold but has not yet reached the second-level time limit threshold, a command is sent to the vehicle's in-vehicle infotainment system via the bus to trigger a visual flashing of the instrument panel icon as a first-level warning alarm. If the duration exceeds the second-level time limit threshold but has not yet reached the third-level time limit threshold, a high-frequency audio prompt triggered by a message will be sent on top of the visual flashing as a second-level warning alarm. If the duration exceeds the third-level time limit threshold, a periodic vibration of the pre-tensioned seat belt or seat vibration will be triggered as a reminder, and the hazard warning lights will be activated, thus completing the continuous judgment and graded warning actions for the steering wheel being out of control.

[0055] In an optional embodiment, when the hand-off detection value is lower than the hand-off threshold within a continuously set time window, it is determined that the steering wheel has been released and a tiered alarm is triggered. The alarm level is determined by the duration of the value being below the threshold. The hand-off threshold is compensated for with safety sensitivity based on the number of remaining sub-bands, including: Set a baseline dropout threshold, count the number of remaining subbands currently participating in feature extraction, and calculate the ratio of the number of remaining subbands to the total number of subbands initially decomposed in the bandpass filter bank; determine the threshold compensation coefficient based on the ratio, and correct the baseline dropout threshold based on the threshold compensation coefficient to obtain the current dropout threshold; Continuously monitor the release judgment value. When the release judgment value is continuously lower than the release threshold for a continuous period of time that reaches the first-level alarm time threshold, trigger a visual prompt alarm on the dashboard. When the duration reaches the Level 2 alarm time threshold, a combined visual and auditory alarm is triggered; when the duration reaches the Level 3 alarm time threshold, a pre-tensioned seat belt vibration intervention is triggered and the hazard warning flasher is activated.

[0056] Set the baseline drop threshold If set to 0.3 or 0.35, the actual number of remaining sub-bands used in the calculation will be counted. Compare the theoretical total number of subbands for bandpass separation If the value is set to 5, calculate the quantity ratio. According to the formula Calculate the threshold compensation coefficient, and based on... The current sell threshold is obtained, where The preset compensation coefficient is preferably between 0 and 0.5, and Not less than For example, when =2、 =5 and When =0.3, =1.18, the current sell threshold is adjusted to 1.18 times the baseline sell threshold. A 20ms resolution timer is used to continuously track the sell decision value D. When D continuously falls below... At that time, the cumulative duration of being released from the hands variable .when Reaching the Level 1 alarm time threshold When the steering wheel is not gripped firmly, the visual warning indicator on the dashboard is triggered; if the duration reaches the level 2 alarm time threshold... Simultaneously trigger a buzzer warning with a sound pressure level between 75 and 85 dB; if the duration reaches the level three alarm time threshold... When activated, the motor-driven pretensioning seatbelt in the driver's seat is triggered to perform periodic vibration or tightening reminders according to a preset duty cycle, and the vehicle's external two-way continuous synchronous flashing hazard warning lights are activated.

[0057] The test collected 800 real slip-out events and 1200 normal grip events. The baseline system using traditional fixed-weight fusion detected 640 slip-outs with 410 false positives, achieving an overall accuracy of 71.5%. The ablation group, with only the 2D fuzzy inference weight module removed, detected 720 slip-outs with 190 false positives, achieving an accuracy of 86.5%. The ablation group, with only the order domain resampling and cyclic stationarity feature extraction modules removed, detected 690 slip-outs with 230 false positives, achieving an accuracy of 83.0%. When using the complete scheme presented in this paper, 785 slip-outs were detected with 12 false positives, achieving an overall accuracy of approximately 98.7%. Figure 3 As shown.

[0058] Under the aforementioned test conditions, the order-domain resampling and cyclic stability feature extraction modules help reduce modulation interference caused by time-varying vehicle speeds, improving the stability of subsequent hand-related feature extraction. The two-dimensional fuzzy inference weight module can adjust the sensor feature weights according to the driving state, thereby reducing the risk of false triggering caused by torque oscillations during low-speed sharp turns. After the multi-dimensional algorithm modules operate in conjunction, the number of false alarms in the test samples is significantly reduced, indicating that the core technical components of the proposed scheme have a synergistic effect in improving the accuracy of hands-off detection and the stability of alarms.

[0059] One or more embodiments of the present invention also provide a steering wheel alarm system based on HOD (Hands-Off Detection) detection, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the steering wheel alarm method based on HOD detection according to the present invention.

[0060] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0061] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise expressly and specifically defined.

Claims

1. A steering wheel alarm method based on HOD (Hands-off Detection) detection, characterized in that, include: S1. Acquire steering column vibration signal, wheel speed signal, and steering torque signal. Resample the vibration signal in the angle domain based on the wheel speed signal to obtain order domain signals with the same angle interval. S2. Decompose the order domain signals using a bandpass filter bank whose center frequency and bandwidth are determined by the spectral kurtosis maximization criterion. Extract the envelope of each sub-band and calculate the second-order cyclic stability index. Remove sub-bands whose index is below a set threshold. S3. Squaring the envelope of each remaining sub-band to obtain the squared envelope, subtracting the local mean, and then performing a Fourier transform to obtain the envelope of each sub-band. S4, In the envelope spectrum of each sub-band, a narrow band search interval is set with the first order of wheel rotation and harmonics as the center. The width of the interval is adjusted inversely by the kurtosis value of the sub-band spectrum. S5, The peak amplitude values ​​of each harmonic spectrum are extracted in each sub-band search interval. The results are weighted and summed according to the order attenuation weight that decreases with the increase of the harmonic order. The results of each remaining sub-band are then weighted and fused according to the second-order cyclic stability index to obtain the hand-related feature quantity. S6, The hand-related feature quantity and the steering torque sliding variance are weighted and fused to obtain the hand-off discrimination value. The fusion weight is determined by the vehicle speed and the steering wheel angle rate. S7. When the hand-off judgment value is lower than the hand-off threshold within a continuously set time window, it is determined that the steering wheel is out of control and a graded alarm is triggered. The alarm level is determined by the duration of the time it is below the threshold. The hand-off threshold is compensated for with safety sensitivity based on the number of remaining sub-bands.

2. The method according to claim 1, characterized in that, The step of resampling the vibration signal in the angle domain based on the wheel speed signal to obtain order domain signals with the same angle interval includes: Extract the pulse sequence from the wheel speed signal and calculate the time interval between adjacent pulses to obtain the instantaneous speed; By integrating time with respect to the instantaneous rotational speed, the cumulative angle value corresponding to each vibration sampling point is obtained; A fixed angular sampling interval is set, and the vibration signal in the time domain is interpolated into a vibration signal sequence with the same angular interval distribution using a spline interpolation algorithm. The sequence is then used as the order domain signal.

3. The method according to claim 1, characterized in that, The process of decomposing the order-domain signal using a bandpass filter bank with a center frequency and bandwidth determined by the spectral kurtosis maximization criterion includes: Calculate the fast spectral kurtosis plot of the order domain signal, and calculate the spectral kurtosis value of the band-filtered signal under different combinations of center frequency and bandwidth; In the fast spectral kurtosis map, multiple feature nodes with local maxima of spectral kurtosis values ​​are identified, and the center frequency and bandwidth parameters corresponding to each feature node are obtained. Based on the extracted center frequency and bandwidth parameters, a set of finite impulse response bandpass filters is constructed. The order domain signals are then input into the bandpass filter group to obtain multiple independent sub-band signals.

4. The method according to claim 1, characterized in that, The extraction of the envelope of each sub-band and the calculation of the second-order cyclic stability index include: Perform a Hilbert transform on each sub-band signal to obtain an analytic signal, and take the amplitude of the analytic signal to obtain the sub-band envelope; Calculate the time-varying autocorrelation function of the subband envelope signal, and perform a two-dimensional Fourier transform along the time axis and the time delay axis on the time-varying autocorrelation function to obtain the cyclic spectral density function; The maximum amplitude of the cyclic spectral density function is extracted in the cyclic frequency domain, and the ratio of the amplitude to the average power spectral density amplitude at the non-cyclic frequency is calculated. The resulting ratio is used as the second-order cyclic stability index.

5. The method according to claim 3 or 4, characterized in that, The process involves extracting the peak amplitudes of each harmonic spectrum within each sub-band search interval, weighting and summing them according to a decreasing order attenuation weight, and then fusing the results of each remaining sub-band according to the second-order cyclostationarity index to obtain the hand-related feature quantity, including: Within the narrow band search interval of each order of harmonics in each remaining subband, the local maxima of the search envelope spectrum amplitude are taken as the peak amplitude of the corresponding order of harmonic spectrum. The characteristic values ​​of each sub-band are obtained by multiplying the peak amplitude of each harmonic spectrum within the same sub-band by the arithmetic square root of the reciprocal of the corresponding order, and summing the product results. Multiply the eigenvalues ​​of each subband by the second-order cyclic stationarity index of that subband, sum the products of all remaining subbands, and divide by the sum of the second-order cyclic stationarity indices of all remaining subbands. Use the resulting quotient as the hand-related feature quantity.

6. The method according to claim 1, characterized in that, The process of weightedly fusing hand-related features with the steering torque sliding variance to obtain a hand-off discrimination value, where the fusion weights are determined by vehicle speed and steering wheel angle rate, includes: The amplitude of the vehicle speed signal and the absolute value of the steering wheel angle rate are used as input variables for the fuzzy inference system. By using the weight coefficients output by the fuzzy inference system, the normalized hand-related features and the normalized steering torque sliding variance are linearly weighted and summed to obtain the release-of-hand discrimination value.

7. The method according to claim 6, characterized in that, The output weight coefficients using the fuzzy inference system include: Based on the weight coefficients output by the fuzzy rule base, the fuzzy rule base is set such that when the vehicle speed is lower than a preset low speed threshold and the turning rate is higher than a preset large turning angle threshold, the weight coefficient of the output hand-related feature quantity is greater than the weight coefficient of the steering torque sliding variance. When the vehicle speed is higher than the preset high speed threshold and the turning rate is lower than the preset small turning angle threshold, the weighting coefficient is determined based on whether the steering torque slip variance is within the preset effective fluctuation range. When the steering torque sliding variance is within the preset effective fluctuation range, the weight of the steering torque sliding variance is increased; when the steering torque sliding variance is below the preset effective fluctuation lower limit, the weight of the hand-related feature quantity is maintained.

8. The method according to claim 1, characterized in that, The system determines that the steering wheel has been released from the hands and triggers a tiered alarm when the release threshold is below the threshold value for a continuous set time window. The alarm level is determined by the duration the value remains below the threshold, including: The system continuously monitors the release judgment value. When the release judgment value is continuously lower than the release threshold for a duration that reaches the first-level alarm time threshold, a visual prompt alarm is triggered on the dashboard. When the duration reaches the level 2 alarm time threshold, a combined visual and auditory alarm is triggered; when the duration reaches the level 3 alarm time threshold, a pre-tensioned seat belt vibration intervention is triggered and a hazard warning flasher is activated.

9. The method according to claim 8, characterized in that, The release threshold is compensated for with security sensitivity based on the number of remaining sub-bands, including: Set a baseline dropout threshold, count the number of remaining subbands currently participating in feature extraction, and calculate the ratio of the number of remaining subbands to the total number of subbands initially decomposed in the bandpass filter bank; The threshold compensation coefficient is determined based on the ratio of the quantities, and the benchmark sell threshold is corrected based on the threshold compensation coefficient to obtain the current sell threshold.

10. A steering wheel alarm system based on HOD (Hands-off Detection) detection, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the steering wheel alarm method based on HOD hands-off detection according to any one of claims 1-9.

Images