Driver state detection method, apparatus, device, and computer program product

Abstract

The application provides a driver state detection method, device, equipment and computer program product to improve the accuracy of the detected driver state information. The method comprises: collecting vehicle driving signals; the vehicle driving signals correspond to the driving state of the driver; based on the collection frequency of the vehicle driving signals, classifying the vehicle driving signals to obtain a plurality of type signals; slicing the plurality of type signals to obtain a plurality of type target slice signals; the time window of the target slice signal is negatively correlated with the collection frequency; determining a first feature vector of the target slice signal; determining the state information of the driver based on the first feature vector; the state information includes the fatigue level of the driver.

Description

Technical Field

[0001] This invention relates to the field of vehicle safety technology, and in particular to a driver status detection method, device, equipment, and computer program product. Background Technology

[0002] Fatigue driving is one of the main causes of traffic accidents. To improve driving safety, related technologies typically collect single-modal signals (e.g., physiological signals such as electrocardiograms, vehicle control signals such as steering wheel angles, and visual signals such as the driver's eyelid closure frequency) to assess the driver's condition and detect driver fatigue. However, the accuracy of these methods in assessing driver condition still needs improvement. Summary of the Invention

[0003] This invention provides a driver status detection method, apparatus, device, and computer program product to improve the accuracy of detected driver status information.

[0004] In a first aspect, embodiments of this application provide a driver state detection method, including: Collect vehicle driving signals; wherein the vehicle driving signals correspond to the driver's driving status; Based on the acquisition frequency of the vehicle driving signals, the vehicle driving signals are classified to obtain multiple types of signals; The multiple types of signals are sliced ​​to obtain multiple types of target slice signals; wherein, the time window of the target slice signal is negatively correlated with the acquisition frequency; Determine the first feature vector of the target slice signal; Based on the first feature vector, the driver's state information is determined; wherein, the state information includes the driver's fatigue level.

[0005] The driving state detection method provided in this application first classifies vehicle driving signals. Vehicle driving signals from different sampling frequency bands (e.g., high-frequency and low-frequency bands) are sliced ​​to obtain target slice signals whose time windows match the acquisition frequencies of each type of signal. By using time windows negatively correlated with the acquisition frequency, shorter window processing yields high-frequency target slice signals to capture instantaneous signal fluctuations; longer window processing yields low-frequency target slice signals to capture gradual trend changes in driving state. This ensures that each type of signal can accurately extract its first feature vector at its appropriate time scale, thereby effectively improving the expressive power and accuracy of the first feature vector and enhancing the driver's state information.

[0006] In some embodiments, the slicing process of the plurality of signal types to obtain multiple types of target slice signals includes: Obtain preset slice parameters; the slice parameters correspond to the types of the multiple signal types, and the slice parameters include preset time sliding windows and preset sliding step sizes; Based on the slicing parameters, the multiple types of signals are sliced ​​to obtain multiple types of initial slice signals; the time window of the initial slice signal corresponds to the time sliding window; Based on the time window, the initial slice signals of multiple types are time-aligned to obtain the target slice signal.

[0007] In this embodiment, multiple types of signals are sliced ​​using slicing parameters corresponding to the type, and the target slice signals are unified in the time dimension by time alignment of the initial slice signals. This improves the problems of time sequence offset and insufficient accuracy of the first feature vector caused by large differences in acquisition time and acquisition frequency, thereby improving the accuracy of the driver's state information.

[0008] In some embodiments, determining the first feature vector of the target slice signal includes: According to preset rules, a second feature vector corresponding to each type of target slice signal is determined; wherein, the preset rules include extracting the time domain features and / or frequency domain features of the target slice signal using the time window as a processing unit; The first feature vector is obtained by fusing the second feature vector.

[0009] In this embodiment, corresponding feature vectors are extracted for the target slice signal by preset rules, and the changes in the driver's driving performance are captured from at least two dimensions such as time domain and frequency domain, which enhances the comprehensiveness of the expression of the second feature vector. The second feature vector is then fused to achieve feature-level fusion of multi-source vehicle driving signals, further improving the expressive power of the first feature vector.

[0010] In some embodiments, the second feature vector corresponds to the target subclass slice signal in the target slice signal; The process of fusing the second feature vector to obtain the first feature vector includes: Determine the uncertainty of the target sub-class slice signal; Based on the uncertainty, determine the weights corresponding to the second feature vector; The second feature vector is weighted and then concatenated to obtain the first feature vector.

[0011] In this embodiment, by determining the uncertainty representing reliability for the target subclass slice signals in various target slice signals, and determining the weights accordingly, the noise in the first feature vector is further reduced. This achieves adaptive fusion with reliable signals as the main component and partial suppression of inferior signals, thereby improving the expressive power and accuracy of the first feature vector and improving the accuracy of the state information obtained through the first feature vector.

[0012] In some embodiments, determining the driver's state information based on the first feature vector includes: The first feature vector is preprocessed to obtain a first vector to be processed; the preprocessing includes scaling and normalization. Determine the target combined quantum state corresponding to the first vector to be processed; The target combined quantum state is input into a pre-constructed quantum neural network to obtain a fatigue identifier; wherein, the quantum neural network includes H sequentially connected quantum layers, and a measurement layer connected to the Hth quantum layer, where H is a positive integer; Based on a preset mapping relationship, the fatigue level corresponding to the fatigue identifier is determined.

[0013] In this embodiment, after preprocessing the first feature vector to obtain the first vector to be processed, a pre-constructed quantum neural network is used for quantum computing to efficiently determine the fatigue level. At the same time, H sequentially connected quantum layers are used to refine the deep abstract representation related to fatigue layer by layer. Through the measurement layer, the quantum collapses into a fatigue identifier, realizing the mapping from the quantum feature space to the fatigue level, effectively improving the accuracy of driving status containing fatigue level.

[0014] In some embodiments, determining the target combined quantum state corresponding to the first vector to be processed includes: The data points in the first vector to be processed are quantum-encoded to obtain the first combined quantum state; Based on the first combined quantum state, the density matrix is ​​determined, and principal components are extracted from the density matrix; The data points in the principal components are quantum encoded to obtain the target combined quantum state.

[0015] In this embodiment, after determining the first combined quantum state, the principal components are extracted and redundant and noise dimensions are eliminated, thereby reducing the quantum bit overhead and improving the feature purity.

[0016] In some embodiments, the quantum layer includes a quantum convolutional layer and a quantum pooling layer; The step of inputting the target combined quantum state into a pre-constructed quantum neural network to obtain a fatigue identifier includes: The quantum convolution layer performs convolution operations on adjacent bits in the target combined quantum state, and the quantum pooling layer discards bits to obtain the output quantum state. The output quantum state is converted into the fatigue identifier using the measurement layer.

[0017] In this embodiment, adjacent bits are convolved in the quantum convolution layer, entanglement is established between local bits using quantum gates, local quantum features between adjacent bits are extracted, and downsampling is performed through the quantum pooling layer to compress the quantum state dimension and improve generalization ability. The high-dimensional quantum features are then collapsed into fatigue classification identifiers through the measurement layer, achieving accurate mapping and effectively improving the accuracy of fatigue level.

[0018] In some embodiments, the fatigue identifier is the Pauli Z operator; The determination of the fatigue level corresponding to the fatigue identifier based on the preset mapping relationship includes: Within the preset mapping relationship, query the target interval corresponding to the Pauli Z operator. Read the fatigue level corresponding to the target range.

[0019] Secondly, embodiments of this application provide a driver state detection device, comprising: The acquisition unit is used to acquire vehicle driving signals; wherein the vehicle driving signals correspond to the driver's driving status. The classification unit is used to classify the vehicle driving signal based on the acquisition frequency of the vehicle driving signal to obtain multiple types of signals; A slicing unit is used to slice the multiple types of signals to obtain multiple types of target slice signals; wherein the time window of the target slice signal is negatively correlated with the acquisition frequency; A vector unit is used to determine the first feature vector of the target slice signal; A state unit is used to determine the driver's state information based on the first feature vector; wherein the state information includes the driver's fatigue level.

[0020] In some embodiments, the slicing unit is specifically used to obtain preset slicing parameters; the slicing parameters correspond to the types of the plurality of signal types, and the slicing parameters include a preset time sliding window and a preset sliding step size; based on the slicing parameters, the plurality of signal types are sliced ​​to obtain initial slice signals of the plurality of types; the time window of the initial slice signal corresponds to the time sliding window; based on the time window, the initial slice signals of the plurality of types are time-aligned to obtain the target slice signal.

[0021] In some embodiments, the vector unit is specifically used to determine a second feature vector corresponding to each type of target slice signal according to a preset rule; wherein the preset rule includes extracting the time domain features and / or frequency domain features of the target slice signal using the time window as a processing unit; and fusing the second feature vector to obtain the first feature vector.

[0022] In some embodiments, the second feature vector corresponds to a target sub-slice signal in the target slice signal; the vector unit is further used to determine the uncertainty of the target sub-slice signal; based on the uncertainty, a weight corresponding to the second feature vector is determined; the second feature vector is weighted and then concatenated to obtain the first feature vector.

[0023] In some embodiments, the state unit is specifically used to preprocess the first feature vector to obtain a first vector to be processed; the preprocessing includes scaling and normalization; determine the target combined quantum state corresponding to the first vector to be processed; input the target combined quantum state into a pre-constructed quantum neural network to obtain a fatigue identifier; wherein the quantum neural network includes H sequentially connected quantum layers and a measurement layer connected to the Hth quantum layer, where H is a positive integer; and determine the fatigue level corresponding to the fatigue identifier based on a preset mapping relationship.

[0024] In some embodiments, the state unit is further configured to quantum encode the data points in the first vector to be processed to obtain a first combined quantum state; determine a density matrix based on the first combined quantum state, and extract principal components for the density matrix; and quantum encode the data points in the principal components to obtain the target combined quantum state.

[0025] In some embodiments, the quantum layer includes a quantum convolution layer and a quantum pooling layer; the state unit is further configured to perform convolution operations on adjacent bits in the target combined quantum state through the quantum convolution layer, and to discard bits using the quantum pooling layer to obtain an output quantum state; and to convert the output quantum state into the fatigue identifier using the measurement layer.

[0026] In some embodiments, the fatigue identifier is a Pauli Z operator; the state unit is specifically used to query the target interval corresponding to the Pauli Z operator in a preset mapping relationship, and read the fatigue level corresponding to the target interval.

[0027] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and used to run on the processor, wherein the processor executes the computer program to implement the methods described in the first aspect and any embodiment.

[0028] Fourthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the methods described in the first aspect and any embodiment. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0030] Figure 1 This is a flowchart illustrating a driver state detection method in one embodiment; Figure 2A This is a schematic diagram of a sliding window in the slice parameters of one embodiment sliding according to the sliding step size to obtain a first type of target slice signal and a second target slice signal; Figure 2B for Figure 2A A schematic diagram showing the correspondence between the first type of target slice signal and the second type of target slice signal after time alignment; Figure 2C To Figure 2B A schematic diagram of extracting the second feature vector from the time-aligned target slice signal; Figure 3 This is a flowchart illustrating a method for determining a first feature vector in one embodiment; Figure 4 This is a flowchart illustrating a method for quantum encoding a first feature vector to determine a fatigue level in one embodiment. Figure 5 This is a flowchart illustrating a method for determining a target combination quantum state in one embodiment; Figure 6 This is a structural block diagram of a driver state detection device in one embodiment; Figure 7 This is a diagram of the internal structure of an electronic device in one embodiment. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0032] It should be noted that, in this document, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the present invention. The appearance of this phrase in various places throughout the document does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0033] As illustrated herein, unless the context clearly indicates otherwise, the words “a,” “an,” “an,” and / or “the” do not specifically refer to the singular and may also include the plural. Generally, the terms “comprising” and “including” only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements. It should be understood that the terms “comprising” or “having” as used herein indicate the presence of the features, numbers, steps, operations, elements, components, or combinations thereof described in the specification, without excluding the presence or addition of one or more other features, numbers, steps, operations, elements, components, or combinations thereof.

[0034] In this embodiment of the invention, prefixes such as "first" and "second" are used merely to distinguish different descriptive objects and do not limit the position, order, priority, quantity, or content of the described objects. The use of ordinal numbers and other prefixes to distinguish descriptive objects in this application does not constitute a limitation on the described objects. For statements regarding the described objects, please refer to the claims or the context of the embodiments. The use of such prefixes should not constitute unnecessary limitations. Furthermore, in the description of this embodiment, unless otherwise stated, "multiple" means two or more.

[0035] It should be noted that the processing of vehicle driving signals and the vehicle driving data they carry in this embodiment is authorized by the user. In particular, for physiological signals such as heart rate, if the user revokes their consent via the corresponding button on the vehicle, the sensors used to collect vehicle driving signals will be physically powered off. Furthermore, the collection of these signals is not used for a physiological database.

[0036] To address the issue of insufficient accuracy in driver fatigue detection, this application provides a driver state monitoring method: by classifying and slicing the collected vehicle driving signals corresponding to the driving state, the target slice signals obtained from the slicing are used as feature extraction units to generate a first feature vector, thereby obtaining the driver's state information.

[0037] Since there are various types of vehicle driving signals, each with a different acquisition frequency, target slice signals are obtained by slicing based on the acquisition frequency, with the time window matching the acquisition frequency. In this way, the target slice signals within each time window can reflect the driver's state changes within that time window, thereby improving the accuracy of the first feature vector and thus improving the accuracy of fatigue information within the state information.

[0038] The driver state detection method provided in the embodiments of this application will be described in detail below. Please refer to the following: Figure 1 The method includes at least the following implementation steps: Step 101: Collect vehicle driving signals.

[0039] Among them, the vehicle driving signals correspond to the driver's driving status.

[0040] The vehicle driving signal may include at least one of an analog voltage signal corresponding to sitting pressure and / or a physiological signal. The physiological signal may include a capacitively coupled physiological signal and / or a skin conductance signal.

[0041] In one embodiment, an analog voltage signal corresponding to the sitting posture pressure carries voltage data output from a seat pressure sensor. This voltage data can be used to reflect the pressure distribution on the driver's back, waist, and hips. Using this voltage data, sitting posture pressure data, including the sitting tilt angle and center of gravity shift, can be determined. The sampling frequency of the analog voltage signal can be, for example, 10 Hz.

[0042] In one embodiment, physiological signals can be obtained using a capacitive heart rate sensor and a skin conductivity sensor embedded in a preset position on the steering wheel. Heart rate and respiratory rate data can be determined through this capacitive coupling of physiological signals. Skin conductivity data can be determined through the aforementioned skin conductivity signal. The acquisition frequency of the capacitively coupled physiological and skin conductivity signals can be, for example, 250 Hz.

[0043] Furthermore, the vehicle driving signal may also include a CAN (Controller Area Network) signal containing at least one of a vehicle speed signal, a steering wheel angle signal, a steering wheel angular velocity signal, a vehicle yaw rate signal, and a lane departure signal; and / or a hand operation signal containing at least one of a steering wheel grip force signal, a steering wheel vibration signal, and a lever operation frequency signal.

[0044] In one embodiment, the vehicle speed signal, steering wheel angle signal, steering wheel angular velocity signal, and vehicle yaw angle signal in the aforementioned CAN signals can be obtained via the CAN bus. The CAN signal acquisition frequency can be, for example, 50Hz. Correspondingly, the various data carried are obtained by real-time parsing of each CAN signal: vehicle speed, steering wheel angle, steering wheel angular velocity, and vehicle yaw rate. The lane departure warning signal in the aforementioned CAN signals can be obtained by acquiring the LDW (Lane Departure Warning) CAN signal from the forward-facing camera. The lane departure amount can then be determined through the LDW CAN message carried by this lane departure warning signal.

[0045] In one embodiment, the steering wheel grip force signal in the aforementioned hand operation signal can be acquired by a strain gauge pressure sensor located beneath the foam layer at a preset position on the steering wheel. The steering wheel vibration signal can be a triaxial acceleration signal and angular velocity signal acquired by an IMU (Inertial Measurement Unit) within the steering wheel. The steering wheel vibration frequency can be determined accordingly. When the hand muscles are under normal control, the steering wheel vibration frequency is less than a preset low-frequency threshold (e.g., 5Hz). When the driver is fatigued, their hand muscle control ability decreases, and the high-frequency (less than the preset low-frequency threshold) micro-vibration of the steering wheel increases. The lever operation frequency signal can be obtained by polling the electrical state of the steering lever at a preset frequency using a microcontroller. For example, when the electrical state changes from an off state to an on state, and the on state lasts for 50 milliseconds after de-jittering, it is recorded as a valid operation. The steering lever operation frequency is determined based on the total number of valid operations within a preset time period (e.g., 30 milliseconds). The acquisition frequency of the aforementioned steering wheel grip force signal can be, for example, 100Hz.

[0046] Furthermore, the aforementioned vehicle driving signals have undergone noise reduction processing. The noise reduction processing for each vehicle driving signal is explained below: When denoising analog voltage signals, 2D median filtering can be used to remove single-point spike noise from the pressure matrix corresponding to the analog voltage signal. Then, bilinear interpolation is used to upsample to a preset dimension (e.g., 64×64) to obtain the sitting pressure data. The corresponding seat sensor can be the pressure matrix output by a 32×32 pressure window sensor array.

[0047] When denoising capacitively coupled physiological signals, baseline drift and high-frequency electromyographic interference can be removed using a 4th-order Butterworth bandpass filter with a frequency of 0.5~40Hz.

[0048] When denoising skin conductance signals, the skin conductance signals can be decomposed into two components: SCL (Skin Conductance Level) and SCR (Skin Conductance Response). The former is extracted by a 0.05Hz low-pass filter, and the latter is extracted by a 0.05~1.5Hz band-pass filter.

[0049] When denoising CAN signals, the 3σ criterion can be used to remove jump values ​​caused by CAN bus transmission anomalies, and linear interpolation can be used to fill in missing values.

[0050] When denoising hand operation signals, the steering wheel grip force signal can be processed by a 50Hz power frequency notch filter and then smoothed by a 10Hz low-pass filter.

[0051] Step 102: Based on the acquisition frequency of the vehicle driving signal, classify the vehicle driving signal to obtain multiple types of signals.

[0052] In one embodiment, vehicle driving signals can be divided into two categories: a first type of signal and a second type of signal. Multiple signal types include both the first and second type signals. Based on the acquisition frequency of the vehicle driving signals, classifying the vehicle driving signals to obtain multiple signal types includes: acquiring a preset first frequency threshold; determining the corresponding vehicle driving signal as a first type signal in response to an acquisition frequency greater than the first frequency threshold; or determining the corresponding vehicle driving signal as a second type signal in response to an acquisition frequency less than or equal to the first frequency threshold.

[0053] The first type of signal means that a signal extracted within a short period of time can be used to determine the driver's fatigue state. The second type of signal, however, requires a longer time period (e.g., tens of seconds) to be statistically significant. The second type of signal means that a signal extracted over a longer time period is needed to determine the driver's fatigue state.

[0054] For example, the first type of signal includes physiological signals (acquisition frequency may be, for example, 250 Hz) and hand operation signals (acquisition frequency may be, for example, 100 Hz). The second type of signal includes analog voltage signals corresponding to sitting pressure (acquisition frequency may be, for example, 10 Hz) and CAN signals (acquisition frequency may be, for example, 50 Hz).

[0055] Alternatively, in one embodiment, vehicle signals can be divided into three categories. Multiple signal types include a first type signal, a second type signal, and a third type signal. The sampling frequency range corresponding to the third type signal lies between the sampling frequencies of the first and second type signals. Based on the sampling frequency of the vehicle driving signal, the vehicle driving signal is classified to obtain multiple signal types, including: acquiring a preset second frequency threshold and a third frequency threshold. The second frequency is greater than the third frequency threshold. In response to a sampling frequency greater than the second frequency threshold, the corresponding vehicle driving signal is determined to be a first type signal. Alternatively, in response to a sampling frequency less than or equal to the second frequency threshold and greater than the third frequency threshold, the corresponding vehicle driving signal is determined to be a third type signal. Alternatively, in response to a sampling frequency less than or equal to the third frequency threshold, the corresponding vehicle driving signal is determined to be a second type signal. For example, the first type signal includes physiological signals (sampling frequency, for example, 250Hz). The second type signal includes an analog voltage signal corresponding to sitting pressure (sampling frequency, for example, 10Hz). The third type signal includes hand operation signals (sampling frequency, for example, 100Hz) and CAN signals (sampling frequency, for example, 50Hz).

[0056] Among the multiple signal types obtained from the above classification, each signal type may include one or more sub-type signals. Taking the above multiple signal types as composed of first-type signals and second-type signals as an example, physiological signals and hand operation signals in the first-type signals are respectively considered as sub-type signals of the first-type signals. In the second-type signals, the analog voltage signal and CAN signal corresponding to sitting pressure are respectively considered as sub-type signals of the second-type signals.

[0057] Step 103: Slice the signals of multiple types to obtain target slice signals of multiple types.

[0058] The time window of the target slice signal is negatively correlated with the acquisition frequency.

[0059] In one embodiment, preset slicing parameters can be obtained first. The slicing parameters include a preset time window and a preset sliding step size. These slicing parameters correspond one-to-one with the types of signals mentioned above. That is, each type of signal has its own time window (length) and sliding step size.

[0060] Then, based on the above slicing parameters, multiple signal types can be sliced ​​to obtain initial slice signals. The time window of this initial slice signal is consistent with the aforementioned time sliding window corresponding to the type. The only difference is that the time sliding window slides forward at a fixed time, while the time window determined by the sliding of the time sliding window has a fixed start time and an end time corresponding to the start time. The following provides further examples of slicing: When multiple signal types include a first type of signal containing physiological signals and hand operation signals, and a second type of signal containing analog voltage signals and CAN signals corresponding to sitting pressure, the slicing parameters corresponding to the first type of signal include a sliding window w1 of 5 seconds and a sliding step s1 of 1 second. The slicing parameters corresponding to the second type of signal include a sliding window w2 of 30 seconds and a sliding step s2 of 5 seconds. For further details, please refer to... Figure 2A The time windows for the first type of target slice signal obtained after slicing are [t, t+5], [t+1, t+6], [t+2, t+7], and so on. The time windows for the second type of target slice signal obtained after slicing are [t, t+30], [t+5, t+35], [t+10, t+40], and so on. Where t is the starting acquisition time of the multi-type signal.

[0061] Finally, based on the time window, the initial slice signals of multiple types are time-aligned to obtain the target slice signals. The target slice signal for each reference time window (i.e., the largest time window) corresponds to one or more target slice signals for smaller time windows. That is, among the multiple types of initial slice signals, those whose time windows fall outside the reference time window are excluded, and the initial slice signals with an inclusion relationship between their time windows are selected as the target slice signals. This improves the problem of inaccurate driver status identification caused by processing vehicle driving signals with significantly different acquisition times together. Specifically, the type signal with the largest time window can be selected as the reference signal; the time window of this reference signal becomes the reference window. Then, for each type of initial slice signal, target slice signals falling within this reference window are selected. The start point of each target slice signal is no earlier than the start point of the corresponding reference window, and the end point of each target slice signal is no later than the end point of the corresponding reference window. Thus, in the time-aligned target slice signals, each reference signal corresponds to the remaining target slice signals of multiple time windows, and these multiple time windows are a subset of the reference windows of the reference signal.

[0062] Please continue to refer to this. Figure 2AThe second type of initial slice signal serves as the reference signal, with a time window w2 = 30s. Taking t~t+30 and t+5~t+35 within the reference window as examples: within the first reference window (t~t+30), the corresponding time windows of the first type of target slice signals are t~t+5, t+1~t+6, t+2~t+7, ..., t+24~t+29, t+25~t+30. Therefore, in this example, the second type of target slice signal serves as the reference signal, and its time window w2 corresponds to the time windows w1 of 26 first type target slice signals. Please refer to... Figure 2B Similarly, the second reference window t+5~t+35 corresponds to the time windows of the first type of target slice signals, which are t+5~t+10, t+6~t+11, ..., t+29~t+34, and t+30~t+35 seconds respectively. It can be seen that among the selected target slice signals, the time windows of the second type of target slice signals correspond to the time windows of multiple first type target slice signals.

[0063] Step 104: Determine the first feature vector of the target slice signal.

[0064] In one embodiment, weighting coefficients corresponding to the type of the target slice signal can be obtained. Then, based on these weighting coefficients, the second feature vectors of each target slice signal are fused to obtain a first feature vector.

[0065] The following embodiment further illustrates the determination of the first feature vector. Please refer to [reference needed]. Figure 3 : Step 1041: Determine the second feature vector corresponding to the target slice signal according to a preset rule. The preset rule includes extracting the time-domain features and / or frequency-domain features of the target slice signal using a time window as the processing unit.

[0066] The aforementioned second feature vector corresponds one-to-one with the time window of the target slice signal.

[0067] To reduce instantaneous fluctuations, in one embodiment, the aforementioned time-aligned target slice signals can first be pooled according to a time window. Since the target slice signals contain multiple types of target slice signals, and the time windows have an inclusion relationship, a larger time window can be used as a reference window, and the feature vectors of the target slice signals corresponding to the smaller time windows within that reference window can be pooled and averaged to obtain the second feature vector of the corresponding type of target slice signal. Still using... Figure 2B For example, if the time window of the second type of target slice signal corresponds to the time windows of 26 first type target slice signals, please refer to... Figure 2CAfter extracting feature vectors from the target slice signals of the first type according to the aforementioned preset rules, the vectors are averaged along the column direction to obtain a second feature vector of 1×q. Figure 2C In this context, p and q are both positive integers. p represents the number of data points in the second feature vector extracted from the reference signal (i.e., the second type of target slice signal) in the reference window, and q represents the number of data points in the feature vector extracted from the first type of target slice signal in each time window.

[0068] In one embodiment, the preset rules may include the order in which the features to be extracted (e.g., time-domain features and frequency-domain features) are arranged in the second feature vector.

[0069] In one embodiment, the aforementioned time-domain features may include any one or more of the following: mean, standard deviation, skewness, kurtosis, and root mean square (RMS). The mean reflects the central trend of the corresponding target slice signal. The standard deviation reflects the fluctuation amplitude of the corresponding target slice signal. Skewness reflects the asymmetry of the distribution of the corresponding target slice signal. Skewness increases in physiological signals under fatigue conditions. Kurtosis reflects the frequency of extreme values ​​in the corresponding target slice signal. The RMS reflects the energy of the corresponding target slice signal. This time-domain feature may also include the zero-crossing rate of the grip force signal. The zero-crossing rate is the number of times the grip force signal alternates between releasing and gripping.

[0070] The aforementioned frequency domain features may include low-frequency features, high-frequency features, the low-frequency to high-frequency power ratio, and the low-frequency power of the steering wheel angle, extracted after performing a Fourier transform on the target slice signal within the corresponding time window. The low-frequency power reflects the sympathetic-parasympathetic balance. The high-frequency power reflects respiratory sinus arrhythmia. The low-frequency to high-frequency power ratio (LF / HF) is a classic fatigue indicator; the ratio increases as fatigue deepens. The low-frequency power of the steering wheel angle reflects the frequency of steering wheel fine-tuning operations. For example, the low-frequency power may be 0.04–0.15 Hz, the high-frequency power may be 0.15–0.4 Hz, and the low-frequency power of the steering wheel angle may be 0.1–0.5 Hz.

[0071] Furthermore, the aforementioned preset rules may also include extracting the target physiological signal sample entropy, and the trajectory length and / or central envelope area of ​​the target simulated voltage signal corresponding to the sitting pressure. Since the complexity of the physiological signal decreases as fatigue deepens, the sample entropy can be used to measure the regularity of the target physiological signal. The trajectory length of the central coordinates is the cumulative distance the center of the seat pressure moves on the seat plane within a preset time period (e.g., 30 seconds). The central envelope area is the area of ​​the minimum circumscribed ellipse of the pressure center trajectory. This central envelope area increases as the driver's fatigue level intensifies.

[0072] Step 1042: Fuse the second feature vector to obtain the first feature vector.

[0073] The aforementioned multiple signal types can each have one or more sub-type signals within each type of slice signal. Furthermore, corresponding to the multiple signal types, the aforementioned target slice signal is a target slice signal of multiple types. In one embodiment, fusing the second feature vector to obtain the first feature vector includes: firstly, determining weight coefficients for the multiple types of target slice signals. The weight coefficients correspond to the quality of the corresponding initial slice signals. Then, according to the arrangement order in the aforementioned preset rules, concatenating the product of the weight coefficients and the second feature vector to obtain the first feature vector. This first feature vector has the same dimension as the second feature vector.

[0074] In one embodiment, each type of target slice signal includes multiple target sub-slice signals. For example, the first sub-slice signal in the first type of signal is a physiological signal (acquisition frequency, for example, 250Hz) and a hand operation signal (acquisition frequency, for example, 100Hz). Accordingly, the first type of target slice signal has multiple first target sub-slice signals, namely target physiological signals and target hand operation signals. As another example, the second sub-slice signal in the second type of signal is an analog voltage signal corresponding to sitting pressure (acquisition frequency, for example, 10Hz) and a CAN signal (acquisition frequency, for example, 50Hz). Accordingly, the second type of target slice signal has multiple second target sub-slice signals, namely target analog voltage signals and target CAN signals.

[0075] The second feature vector then corresponds one-to-one with the slice signal of each target subclass. Thus, in another embodiment, fusing the second feature vector to obtain the first feature vector includes: determining target weight coefficients for each target subclass slice signal. The weight coefficients correspond to the quality of the corresponding initial slice signal. Based on the target weight coefficients, the second feature vectors corresponding to each subclass are weighted and concatenated according to the arrangement order in the aforementioned preset rules to obtain the first feature vector. The first feature vector has the same dimension as the second feature vector. For example, the second feature vector of each target subclass slice signal is represented as... , , ,…, The corresponding weighting coefficient is , , ,…, Accordingly, the first eigenvector is represented as, ; i is the index corresponding to each subtype.

[0076] In one embodiment, the weight coefficients of each second feature vector can be the uncertainty of each target sub-slice signal. Therefore, to determine the first feature vector, the uncertainty of each target sub-slice signal can be determined first, and then the weights corresponding to the second feature vectors can be determined based on the uncertainty of each target sub-slice signal. Finally, the second feature vector can be weighted, that is, the weights are multiplied by the second feature vector, and then concatenated according to the aforementioned preset rules to obtain the first feature vector.

[0077] In order to determine the weights corresponding to the second feature vector, in one embodiment, the uncertainty of the slice signal of each target subclass can be used as the weight.

[0078] Alternatively, in another embodiment, to further improve the expressive power and accuracy of the first feature vector, the uncertainty can be normalized so that the sum of the weights is 1. More specifically, the uncertainty can be processed using a normalized exponential function to obtain the weights of the target subclass slice signals. : ;in, This represents the weighting coefficient of the target subclass slice signal to be determined, where γ can be selected from 1.0, 2.0, or 3.0. The uncertainty of the slice signal for the target subclass to be determined. The index of the target subclass slice signal in the baseline window. For the first in the baseline window Uncertainty of a target subclass slice signal.

[0079] The above weights In the formula, the larger γ is, The higher the sensitivity to the uncertainty of the corresponding target sub-slice signal, the more sensitive the mode becomes. When γ is 3.0, it is a high-sensitivity mode; when γ is 2.0, it is a default mode balancing sensitivity and stability; and when γ is 1.0, it is a stationary mode. Optionally, γ can be determined as follows: First, calculate the average value of the uncertainty of each target sub-slice signal and calculate the standard deviation of each target sub-slice signal. Then, calculate the ratio of the standard deviation to the average value to obtain the dispersion D. By querying the preset mapping relationship, determine the mode corresponding to the dispersion and the value of γ. For example, D < a1, it can be determined as a stationary mode with γ = 1; a1 ≤ D < a2, it can be determined as the default mode with γ = 2; D ≥ a2, it can be determined as a high-sensitivity mode with γ = 3; where a2 < 1.

[0080] Furthermore, the following examples illustrate the uncertainty of the slice signals for each target subclass: (1) Uncertainty of the target analog voltage signal This can be represented by the percentage of frames showing abnormal jumps in the pressure center within a preset time window. The expression is: Where M represents the number of times the displacement between the center trajectories of adjacent frames exceeds a preset displacement threshold. This is the number of time frames within a preset time window. The preset displacement threshold can be the product of the diagonal length of the effective sensor area on the driver's seat and a preset coefficient (less than or equal to 10%). This preset coefficient can be, for example, 5%. For instance, if the preset time window is 30 seconds, then the sampling frequency of the posture pressure sensor used to acquire the target analog voltage signal is 10Hz, and N is 300.

[0081] (2) Uncertainty of the aforementioned target physiological signal It can be determined based on the quality index of the target heart rate signal and the baseline drift rate of the skin conductance signal. The expression is: ; , The preset coefficients, and > , The complement of the heart rate signal quality index SQI_hr for a preset time window (e.g., 30 seconds); This is a mapping value for the baseline drift rate of the skin conductance signal. (The above...) The sharpness of the R-wave peak can be determined. This can be obtained by normalizing the amplitude by which the baseline drift rate of the skin conductive signal exceeds a preset threshold. When the baseline drift is greater than 5 μS / min, the uncertainty increases, so the preset threshold can be 5 μS / min.

[0082] (3) Uncertainty of the target CAN signal It can be determined based on the number of 3σ abnormal sampling points within a preset time window. ;in, This represents the number of 3σ anomaly sampling points for all CAN channels within a preset time window. The total number of channels. The total number of sampling points within a preset time window. For example, if the preset time window is 30 seconds, and the sampling rate of the CAN signal corresponding to the target CAN signal is 50Hz, then the total number of sampling points is 30. 50 = 1500.

[0083] In the above formula, the CAN channel corresponds to the vehicle speed signal, steering wheel angle signal, steering wheel angular velocity signal, and vehicle yaw angle signal in the CAN signals mentioned in step 101 above.

[0084] (4) Uncertainty of the target hand operation signal The coefficient values ​​can be obtained by performing a normalized cross-correlation calculation between the steering wheel grip force and a preset grip force template. The expression is as follows: ;in, The steering wheel grip force carried by the target hand operation signal. This is a preset grip strength template corresponding to the driver.

[0085] Step 105: Determine the driver's state information based on the first feature vector.

[0086] The driver's status information includes the driver's fatigue level.

[0087] In one embodiment, the driver's fatigue level can be obtained by quantum encoding the first feature vector and determining the Pauli Z operator of the encoded output quantum state.

[0088] The following embodiment provides a detailed explanation of the generation of the output quantum state through quantum encoding and the determination of the fatigue level. Please refer to [reference needed]. Figure 4 : Step 1501: Preprocess the first feature vector to obtain the first vector to be processed.

[0089] This preprocessing includes scaling and normalization.

[0090] In one embodiment, the first feature vector can be scaled first so that all data points in the first feature vector are located between [0, 1]. Then, the scaled first feature vector is normalized to obtain a first vector to be processed, so that the sum of the squares of all data points in the first vector to be processed is 1. The formula for this normalization process is as follows: ; For the i-th data point in the first vector to be processed, For the i-th data point in the first feature vector, Let N be the j-th data point in the first feature vector, and N be the dimension of the first feature vector (i.e., the number of data points).

[0091] Step 1502: Determine the target combined quantum state corresponding to the first vector to be processed.

[0092] The target combined quantum state can be obtained by angular encoding of the principal components obtained from quantum phase estimation. An embodiment is provided below for detailed explanation; please refer to [reference needed]. Figure 5 , Step 501: Encode the data points in the first vector to be processed to obtain the first combined quantum state.

[0093] The first combined quantum state can be obtained by encoding a pre-constructed quantum circuit. Specifically, N qubits are allocated based on the number of data points in the first vector to be processed. N is the number of data points in the first vector to be processed.

[0094] Then, angle encoding is performed on each qubit to obtain the single-qubit encoding. i is the index of a single quantum bit.

[0095] Data points in the first vector to be processed Mapped to angle Then, the angles obtained from the mapping are encoded. For example, since R... y The revolving door operates on all real numbers, offering advantages such as fast convergence and high hardware simulation efficiency. Therefore, the revolving door R... y Angle encoding is performed. Where R... y The function of a revolving door is as follows: ; The corresponding single-qubit representation is as follows: .

[0096] The first combined quantum state is obtained by combining single qubits through tensor product. The specific expression is as follows: .

[0097] Step 502: Based on the first combined quantum state, determine the density matrix, and based on the density matrix, determine the principal components.

[0098] In one embodiment, the density matrix ρ can be generated by the first combined quantum states through an outer product.

[0099] In one embodiment, principal components in the density matrix can be extracted using QPCA (Quantum Principal Component Analysis). More specifically, multiple second vectors to be processed in the density matrix, and second eigenvalues ​​to be processed, can be determined using QPE (Quantum Phase Estimation). Then, based on the relative magnitudes of the eigenvalues, the first... The second vector to be processed corresponding to the larger eigenvalue is used as the third vector to be processed. Based on the third vector... The principal components are determined. This is specifically represented as follows: ;in, is the index of the k largest eigenvalues ​​of the third vector to be processed, and X represents the aforementioned first vector to be processed. This embodiment utilizes QPCA, requiring only one auxiliary bit for quantum phase estimation to determine the eigenvalues ​​of the density matrix, thereby reducing experimental difficulty and quantum resource requirements.

[0100] Step 503: Encode the data points in the principal components to obtain the target combined quantum state.

[0101] This encoding method is consistent with the encoding method in the first vector to be processed in step 501. The only difference is that the encoded data points are different. For the specific implementation method, please refer to step 501.

[0102] Step 1053: Input the target combined quantum state into the pre-constructed quantum neural network to obtain the fatigue identifier.

[0103] The quantum neural network comprises H sequentially connected quantum layers and a measurement layer connected to the Hth quantum layer, where H is a positive integer; The value of H can be, for example, 1, 2, or 3. The aforementioned quantum layer includes a quantum convolutional layer and a quantum pooling layer. In one embodiment, in the quantum neural network, adjacent bits in the target combined quantum state are convolved by a quantum convolutional layer, and then bits are discarded from the output of the quantum convolutional layer using a quantum pooling layer to achieve information compression and obtain the output quantum state. For example, H=2. The output quantum state from the second quantum layer is then input to the measurement layer, where it is converted into a fatigue identifier.

[0104] The training of machine models (or components thereof) such as quantum convolutional layers and quantum pooling layers in the embodiments of this application is based on existing technology, and specific details can be found in existing technology.

[0105] For example, for n qubits, a quantum convolutional layer can be represented as: ;in, Let θ be a quantum convolutional layer, and θ be the pre-trained parameters in the quantum convolutional layer.

[0106] The quantum layer can be represented as: ;in, For the quantum states input into the quantum convolutional layer, It is the quantum state output by the quantum pooling layer. It is a quantum pooling layer.

[0107] In one embodiment, the aforementioned fatigue identifier can be a Pauli Z operator; then, the Pauli Z operator obtained by compression through the quantum pooling layer in the two quantum layers is 1 bit. This Pauli Z operator is represented as follows: .

[0108] Step 1054: Determine the fatigue level corresponding to the fatigue identifier based on the preset mapping relationship.

[0109] Thus, in one embodiment, determining the fatigue level corresponding to the fatigue identifier based on a preset mapping relationship includes: querying the target interval corresponding to the Pauli Z operator in the preset mapping relationship, and reading the fatigue level corresponding to the target interval. The aforementioned 1-bit Pauli operator is a scalar, and its symbolic representation is...<z>The aforementioned preset mapping relationship can be... <z>∈ (+0.5, 1], the fatigue level can be determined as L0, indicating that the driver is awake; <z>∈ (0, 0.5], the fatigue level can be determined as L1, indicating mild driver fatigue; when <z>∈ (-0.5, 0], the fatigue level can be determined as L2, indicating moderate driver fatigue; when <z>∈ (-1, 0.5], the fatigue level can be determined as L3, which indicates that the driver is severely fatigued.

[0110] In one embodiment, a fully connected layer can be connected after the measurement layer to determine the driver's fatigue probability based on the Pauli Z operator. The driver's fatigue level can then be determined based on this fatigue probability. Since this fully connected layer can be trained along with the quantum neural network, it reduces the influence of individual differences on fatigue level judgment, thereby improving the accuracy of fatigue detection.

[0111] Furthermore, after determining the driver's status information, alert strategies can be obtained and implemented based on the fatigue level to improve driving safety. Table 1 below provides examples of alert strategies and their relationship with fatigue levels.

[0112] Table 1

[0113] To further optimize the user experience and mitigate the repeated switching of alert strategies caused by multiple jumps in fatigue level due to the probability corresponding to the fatigue identifier or the Pauli Z operator being near the threshold, in one embodiment, when two consecutive fatigue levels are detected differently, and the latest detected fatigue level is higher and is L1, the alert strategy corresponding to L1 must be executed if the cumulative number of times the fatigue level of L1 is detected within a preset time window (e.g., 15 seconds) reaches a preset cumulative threshold (e.g., 3 times).

[0114] In one embodiment, when two consecutive fatigue levels are different, and the latest detected fatigue level is higher and is L2, the cumulative number of times the fatigue level of L2 is detected within a preset time window (e.g., 15 seconds) reaches a preset cumulative threshold (e.g., 2 times) before the reminder strategy corresponding to L2 is executed.

[0115] In one embodiment, when two consecutive fatigue levels are detected differently, and the latest detected fatigue level is higher and is L3, the alert strategy corresponding to L3 can be directly executed to ensure driving safety. And / or, when two consecutive fatigue levels are detected differently, and the latest detected fatigue level is lower, the current alert strategy can only be switched to the alert strategy for the lower fatigue level if a preset number of lower fatigue levels are detected within a preset time window, to prevent triggering again after a brief recovery to the lower fatigue level.

[0116] It should be understood that, although Figure 1 , Figure 3 , Figure 4 , Figure 5 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 , Figure 3 , Figure 4 , Figure 5 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0117] Based on the same inventive concept, such as Figure 6 As shown in the figure, this application provides a driver state detection device, including: a data acquisition unit 601, a classification unit 602, a slicing unit 603, a vector unit 604, and a state unit 605, wherein: The acquisition unit 601 is used to acquire vehicle driving signals; wherein the vehicle driving signals correspond to the driver's driving status. The classification unit 602 is used to classify the vehicle driving signal based on the acquisition frequency of the vehicle driving signal to obtain multiple types of signals; The slicing unit 603 is used to slice the multiple types of signals to obtain multiple types of target slice signals; wherein the time window of the target slice signal is negatively correlated with the acquisition frequency; Vector unit 604 is used to determine the first feature vector of the target slice signal; State unit 605 is used to determine the driver's state information based on the first feature vector; wherein the state information includes the driver's fatigue level.

[0118] In some embodiments, the slicing unit 603 is specifically used for: Obtain preset slicing parameters; the slicing parameters correspond to the types of the multiple signal types, and the slicing parameters include a preset time sliding window and a preset sliding step size; based on the slicing parameters, slice the multiple signal types to obtain multiple initial slice signals; the time window of the initial slice signal corresponds to the time sliding window; based on the time window, time-align the multiple initial slice signals to obtain the target slice signal.

[0119] In some embodiments, the vector unit 604 is specifically used for: According to preset rules, a second feature vector corresponding to each type of target slice signal is determined; wherein, the preset rules include, using the time window as a processing unit, extracting the time domain features and / or frequency domain features of the target slice signal; and fusing the second feature vector to obtain the first feature vector.

[0120] In some embodiments, the second feature vector corresponds to the target sub-slice signal in the target slice signal; the vector unit 604 is further configured to: Determine the uncertainty of the target subclass slice signal; based on the uncertainty, determine the weight corresponding to the second feature vector; weight the second feature vector and then concatenate them to obtain the first feature vector.

[0121] In some embodiments, the state unit 605 is specifically used for: The first feature vector is preprocessed to obtain a first vector to be processed; the preprocessing includes scaling and normalization; a target combined quantum state corresponding to the first vector to be processed is determined; the target combined quantum state is input into a pre-constructed quantum neural network to obtain a fatigue identifier; wherein the quantum neural network includes H sequentially connected quantum layers and a measurement layer connected to the Hth quantum layer, where H is a positive integer; the fatigue level corresponding to the fatigue identifier is determined based on a preset mapping relationship.

[0122] In some embodiments, the state unit 605 is further configured to: The data points in the first vector to be processed are quantum encoded to obtain a first combined quantum state; based on the first combined quantum state, a density matrix is ​​determined, and principal components are extracted from the density matrix; the data points in the principal components are quantum encoded to obtain the target combined quantum state.

[0123] In some embodiments, the quantum layer includes a quantum convolution layer and a quantum pooling layer; the state unit 605 is further configured to perform convolution operations on adjacent bits in the target combined quantum state through the quantum convolution layer, and to discard bits using the quantum pooling layer to obtain an output quantum state; and to convert the output quantum state into the fatigue identifier using the measurement layer.

[0124] In some embodiments, the fatigue identifier is a Pauli Z operator; the state unit is specifically used to query the target interval corresponding to the Pauli Z operator in a preset mapping relationship, and read the fatigue level corresponding to the target interval.

[0125] Specific limitations regarding the suspension control device can be found in the above description of the driver state detection method, and will not be repeated here. Each module in the aforementioned suspension control device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the electronic device, or stored in the memory of the electronic device as software, so that the processor can call and execute the corresponding operations of each module.

[0126] Based on the same inventive concept, please refer to Figure 7 This application also provides an electronic device. In one embodiment, the electronic device, as shown in the figure, may include a memory 701, a communication module 703, and one or more processors 702.

[0127] The memory 701 is used to store computer programs executed by the processor 702. The memory 701 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, and the data storage area may store various operation instruction sets, etc.

[0128] Memory 701 may be volatile memory, such as random-access memory (RAM); memory 701 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 701 may be any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 701 may be a combination of the above-described memories.

[0129] The processor 702 may include one or more central processing units (CPUs) or digital processing units, etc. The processor 702 is used to implement the aforementioned driver state detection method when it calls the computer program stored in the memory 701.

[0130] The communication module 703 is used to communicate with terminal equipment, site equipment or other network equipment.

[0131] This application embodiment does not limit the specific connection medium between the memory 701, communication module 703, and processor 702 described above. This application embodiment... Figure 7 The memory 701 and the processor 702 are connected via a bus 704, and the bus 704 is in Figure 7 The diagram uses thick lines to describe the connections between other components; these are for illustrative purposes only and should not be considered limiting. The 704 bus can be divided into address bus, data bus, control bus, etc. For ease of description, Figure 7 It is described using only a thick line, but does not indicate that there is only one bus or one type of bus.

[0132] The memory 701 stores a computer storage medium containing computer-executable instructions for implementing the driver state detection method of the embodiments of this application. The processor 702 executes the driver state detection methods of the embodiments described above using the computer-executable instructions.

[0133] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0134] Based on the same inventive concept, embodiments of this application provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the driver state detection method as described above.

[0135] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0136] Based on the same inventive concept, embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the driver state detection method described in any of the above claims.

[0137] The program code for executing the computer program product of this application can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0138] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0139] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0140] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0141] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of user-operated steps to be executed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0142] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.< / z> < / z> < / z> < / z> < / z>

Claims

1. A driver state detection method, characterized in that, include: Collect vehicle driving signals; wherein the vehicle driving signals correspond to the driver's driving status; Based on the acquisition frequency of the vehicle driving signals, the vehicle driving signals are classified to obtain multiple types of signals; The multiple types of signals are sliced ​​to obtain multiple types of target slice signals; wherein, the time window of the target slice signal is negatively correlated with the acquisition frequency; Determine the first feature vector of the target slice signal; Based on the first feature vector, the driver's state information is determined; wherein, the state information includes the driver's fatigue level.

2. The method as described in claim 1, characterized in that, The process of slicing the multiple types of signals to obtain multiple types of target slice signals includes: Obtain preset slice parameters; the slice parameters correspond to the types of the multiple signal types, and the slice parameters include preset time sliding windows and preset sliding step sizes; Based on the slicing parameters, the multiple types of signals are sliced ​​to obtain multiple types of initial slice signals; the time window of the initial slice signal corresponds to the time sliding window; Based on the time window, the initial slice signals of multiple types are time-aligned to obtain the target slice signal.

3. The method as described in claim 1, characterized in that, Determining the first feature vector of the target slice signal includes: According to preset rules, a second feature vector corresponding to each type of target slice signal is determined; wherein, the preset rules include extracting the time domain features and / or frequency domain features of the target slice signal using the time window as a processing unit; The first feature vector is obtained by fusing the second feature vector.

4. The method as described in claim 3, characterized in that, The second feature vector corresponds to the target subclass slice signal in the target slice signal; The process of fusing the second feature vector to obtain the first feature vector includes: Determine the uncertainty of the target sub-class slice signal; Based on the uncertainty, determine the weights corresponding to the second feature vector; The second feature vector is weighted and then concatenated to obtain the first feature vector.

5. The method according to any one of claims 1 to 4, characterized in that, Determining the driver's state information based on the first feature vector includes: The first feature vector is preprocessed to obtain a first vector to be processed; the preprocessing includes scaling and normalization. Determine the target combined quantum state corresponding to the first vector to be processed; The target combined quantum state is input into a pre-constructed quantum neural network to obtain a fatigue identifier; wherein, the quantum neural network includes H sequentially connected quantum layers, and a measurement layer connected to the Hth quantum layer, where H is a positive integer; Based on a preset mapping relationship, the fatigue level corresponding to the fatigue identifier is determined.

6. The method as described in claim 5, characterized in that, Determining the target combined quantum state corresponding to the first vector to be processed includes: The data points in the first vector to be processed are quantum-encoded to obtain the first combined quantum state; Based on the first combined quantum state, the density matrix is ​​determined, and principal components are extracted from the density matrix; The data points in the principal components are quantum encoded to obtain the target combined quantum state.

7. The method as described in claim 5, characterized in that, The quantum layer includes a quantum convolutional layer and a quantum pooling layer; The step of inputting the target combined quantum state into a pre-constructed quantum neural network to obtain a fatigue identifier includes: The quantum convolution layer performs convolution operations on adjacent bits in the target combined quantum state, and the quantum pooling layer discards bits to obtain the output quantum state. The output quantum state is converted into the fatigue identifier using the measurement layer.

8. A driver status detection device, characterized in that, include: The acquisition unit is used to acquire vehicle driving signals; wherein the vehicle driving signals correspond to the driver's driving status. The classification unit is used to classify the vehicle driving signal based on the acquisition frequency of the vehicle driving signal to obtain multiple types of signals; A slicing unit is used to slice the multiple types of signals to obtain multiple types of target slice signals; wherein the time window of the target slice signal is negatively correlated with the acquisition frequency; A vector unit is used to determine the first feature vector of the target slice signal; A state unit is used to determine the driver's state information based on the first feature vector; wherein the state information includes the driver's fatigue level.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and for running on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images